Download Life Cycle Assessment: Quantitative Approaches for Decisions That Matter...
1
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter
H. Scott Matthews Chris T. Hendrickson Deanna H. Matthews
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
2
Copyright/permissions/creative commons/etc on this page
Microsoft, Encarta, MSN, and Windows are either registered trademarks or trademarks of Microsoft Corporation in the United States and/or other countries. MATLAB is a registered trademark of The MathWorks, Inc. in the United States and/or other countries.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
3
Dedication
To Lester Lave
Who taught us to work on problems that matter
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
4
Table of Contents Dedication ................................................................................................................... 3 Table of Contents ......................................................................................................... 4 Preface ........................................................................................................................ 7 Chapter 1 : Life Cycle and Systems Thinking ............................................................... 10 Learning Objectives for the Chapter .................................................................................... 10 Overview of Life Cycles ....................................................................................................... 10 A Brief History of Engineering and The Environment ........................................................... 12 Life Cycle Thinking .............................................................................................................. 14 Systems Thinking in the Life Cycle ....................................................................................... 18 A History of Life Cycle Thinking and Life Cycle Assessment .................................................. 18 Decisions Made Without Life Cycle Thinking ....................................................................... 21 Inputs and Outputs of Interest in Life Cycle Models ............................................................ 22 From Inputs and Outputs to Impacts ................................................................................... 25 The Role of Design Choices ................................................................................................. 27 What Life Cycle Thinking and Life Cycle Assessment Is Not .................................................. 28 Chapter Summary ............................................................................................................... 29 Chapter 2 : Quantitative and Qualitative Methods Supporting Life Cycle Assessment 32 Learning Objectives for the Chapter .................................................................................... 32 Basic Qualitative and Quantitative Skills ............................................................................. 32 Working with Data Sources ................................................................................................. 33 Accuracy vs. Precision ......................................................................................................... 37 Uncertainty and Variability ................................................................................................. 38 Management of Significant Figures ..................................................................................... 39 Ranges ................................................................................................................................ 41 Units and Unit Conversions ................................................................................................. 44 Considerations for Energy Unit Conversions ........................................................................ 45 Use of Emissions or Resource Use Factors ........................................................................... 47 Estimations vs. Calculations ................................................................................................ 48 Attributes of Good Assumptions ......................................................................................... 53 Validating your Estimates ................................................................................................... 54 Building Quantitative Models ............................................................................................. 56 A Three-‐step method for Quantitative and Qualitative Assessment .................................... 58 Chapter Summary ............................................................................................................... 59 Chapter 3 : Life Cycle Cost Analysis ............................................................................ 62 Learning Objectives for the Chapter .................................................................................... 62 Life Cycle Cost Analysis in the Engineering Domain ............................................................. 62 Discounting Future Values to the Present ........................................................................... 64 Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
5
Life Cycle Cost Analysis for Public Projects .......................................................................... 67 Deterministic and Probabilistic LCCA ................................................................................... 69 Chapter Summary ............................................................................................................... 72
Chapter 4 : The ISO LCA Standard – Goal and Scope ................................................... 77 Learning Objectives for the Chapter .................................................................................... 77 Overview of ISO and the Life Cycle Assessment Standard .................................................... 77 ISO LCA Study Design Parameters ....................................................................................... 80 Chapter Summary ............................................................................................................... 92 Chapter 5 : Data Acquisition and Management for Life Cycle Inventory Analysis ....... 96 Learning Objectives for the Chapter .................................................................................... 96 ISO Life Cycle Inventory Analysis ......................................................................................... 97 Life Cycle Interpretation .................................................................................................... 108 Identifying and Using Life Cycle Data Sources .................................................................... 109 Details for Other Databases .............................................................................................. 119 LCI Data Module Metadata ............................................................................................... 120 Referencing Secondary Data ............................................................................................. 124 Additional Considerations about Secondary Data and Metadata ....................................... 125 Chapter Summary ............................................................................................................. 128 Advanced Material for Chapter 5 ...................................................................................... 131 Section 1 -‐ Accessing Data via the US LCA Digital Commons .............................................. 131 Section 2 – Accessing LCI Data Modules in SimaPro ........................................................... 136 Section 3 – Accessing LCI Data Modules in openLCA .......................................................... 142 Chapter 6 : Analyzing Multifunctional Product Systems ........................................... 155 Learning Objectives for the Chapter .................................................................................. 155 Allocation of Flows for Processes with Multiple Products .................................................. 157 Allocation Example from LCI Databases ............................................................................. 163 Chapter Summary ............................................................................................................. 179 Further Reading ................................................................................................................ 180 Chapter 7 Another chapter TBA? .............................................................................. 182 Chapter 8 : LCA Screening via Economic Input-‐Output Models ................................. 185 Learning Objectives for the Chapter .................................................................................. 185 Input-‐Output Tables and Models ....................................................................................... 185 Input–Output Models Applied to Life Cycle Assessment .................................................... 193 Introduction to the EIO-‐LCA Input-‐Output LCA Model ....................................................... 197 EIO-‐LCA Example: Automobile Manufacturing ................................................................... 200 Beyond Cradle to Gate Analyses with IO-‐LCA .................................................................... 205 Chapter Summary ............................................................................................................. 207 Homework Questions for Chapter 8 .................................................................................. 210 Advanced Material for Chapter 8 -‐ Overview ..................................................................... 213 Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
6
Section 1 -‐ Linear Algebra Derivation of Leontief (Input-‐Output) Model Equations ............ 213 Section 2 – Commodities, Industries, and the Make-‐Use Framework of EIO Methods ....... 215 Section 3 – Further Detail on Prices in IO-‐LCA Models ....................................................... 218 Section 4 – Mapping Examples from Industry Classified Sectors to EIO Model Sectors ...... 224 Section 5 – Spreadsheet and MATLAB Methods for Using EIO Models ............................... 229
Chapter 9 : Advanced Life Cycle Models .................................................................. 241 Learning Objectives for the Chapter .................................................................................. 241 Process Matrix Based Approach to LCA ............................................................................. 241 Connection Between Process-‐ and IO-‐Based Matrix Formulations .................................... 246 Extending process matrix methods to post-‐production stages .......................................... 254 Categories of Hybrid LCA Models ...................................................................................... 258 Chapter Summary ............................................................................................................. 264 Advanced Material for Chapter 9 – Section 1 – Process Matrix Models in MATLAB ........... 267 Advanced Material for Chapter 9 – Section 2 – Process Matrix Models in SimaPro ............ 269 Advanced Material for Chapter 9 – Section 3 – Process Matrix Models in openLCA ........... 273 Chapter 10 : Life Cycle Impact Assessment .............................................................. 278 Learning Objectives for the Chapter .................................................................................. 278 Why Impact Assessment? ................................................................................................. 278 Overview of Impacts and Impact Assessment ................................................................... 279 Chapter Summary ............................................................................................................. 300 Homework Questions for Chapter 10 ................................................................................ 300 Chapter 11 : Uncertainty and Variability Assessment in LCA ..................................... 302 Learning Objectives for the Chapter .................................................................................. 302 Methods to Address Uncertainty and Variability ............................................................... 316 Quantitative Methods to Address Uncertainty and Variability .......................................... 320 Deterministic and Probabilistic LCCA ................................................................................. 328 Chapter Summary ............................................................................................................. 329 Homework Questions for Chapter 11 ................................................................................ 330 Chapter 12 : Advanced Hybrid Hotspot and Path Analysis ........................................ 341 Learning Objectives for the Chapter .................................................................................. 341 Results of Aggregated LCA Methods ................................................................................. 341 A Disaggregated Two-‐Unit Example .................................................................................. 344 Structural Path and Network Analysis ............................................................................... 345 Web-‐based Tool for SPA ................................................................................................... 354 Chapter Summary ............................................................................................................. 365 Homework Questions for Chapter 12 ................................................................................ 366 Advanced Material for Chapter 12 – Section 1 -‐ MATLAB Code for SPA ............................. 367
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
7
Preface After finishing our book on Economic Input Output Life Cycle Assessment (EIO-LCA) (Hendrickson, Lave and Matthews 2006) with help from our colleagues Arpad Horvath, Satish Joshi, Fran McMichael, Heather MacLean, Gyorgyi Cicas, Deanna Matthews and Joule Bergerson, we assumed that would be our final word. We did not imagine writing another book on the topic. The 2006 book successfully demonstrated the EIO-LCA approach and demonstrated various applications. At Carnegie Mellon University (CMU), we had a sustainability sequence of four half-semester courses in our graduate program in Civil and Environmental Engineering. Only one of those courses was on environmental life cycle assessment (LCA), and over the course of a seven-week term there was only so much material that could be covered. Also, that LCA follows an established process set by the International Organization for Standardization (ISO) (and other similar agencies) meant that it was hard to justify writing a book that teaches you how to use an existing recipe. Imagine writing a cookbook that intends to teach you how to read other cookbooks! But after using the book for a few years, we realized how much other material was needed and how the book had only limited value as a textbook (which was not even the intent of the book in the first place). Our half-semester graduate LCA course grew to a full semester. We supplemented readings from our book with many other resources – to the point that as of a few years ago we were only assigning a few of the original book chapters. So while this book was not really planned, the preparations for it have been happening for the last five years. Another driving force is that LCA has changed since 2006. From our observations as educators, researchers, practitioners, and peer reviewers in the LCA community, there are trends that concern us. One of the trends is that practitioners are depending too much on LCA software features (i.e., pressing buttons) without fully understanding the implications of simply pressing buttons in existing software tools and reporting the results. In particular, many practitioners accept calculations without considering the large amount of underlying uncertainty in the numbers. These observations are especially concerning as LCA (as the title of the book implies) is increasingly being used to support "big decisions" rather than simple decisions such as whether to use paper or plastic bags (we actually favor cloth bags). And thus we have prepared this free e-book to help educate you about LCA. Let us clearly note that this book should be a supplement to, not a substitute for, acquiring, reading, and learning the established LCA standards that we do not re-publish here. This book is intended to be a companion to an organized tour of those standards, not a replacement. In addition, we have organized chapters in a consistent way so that it can be used for undergraduate or graduate audiences. For many of the chapters, there are sections at the end of each chapter that we expect an undergraduate course may skip but that a graduate course may dive into quite deeply. We use the book in this serial format in our own undergrad and graduate LCA courses at CMU. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
8
This book (like its predecessor) is about life cycle thinking and acquiring and generating the information to make sound decisions to improve environmental quality and sustainability. How can we design products, choose materials and processes, and decide what to do at the end of a product's life in ways that produce fewer environmental discharges and use less material and energy? We also should add that pursuing environmental improvement is only one of many social objectives. In realistic design situations, costs and social impacts are also important to consider. This book focuses on environmental impacts, although life cycle costs are discussed in Chapter 3. Readers are encouraged to also seek out material on life cycle costs and social impacts. A good starting point is our free, online book on Civil Systems Planning, Investment and Pricing (Hendrickson and Matthews, 2013). We expect that readers of this book (and thus students in courses using the book) are generally knowledgeable about environmental and energy issues, are comfortable with probability, statistics, and building small quantitative models, and are willing to learn new methods that will help organize broad thoughts about how products, processes, and systems can be assessed. In summary, we consider this a "take two" of our original purpose – to have a unified resource for use in our own courses. This book's significantly expanded scope benefits from our collective 40 years of experience in LCA. We overview the ISO LCA Framework, but spend most of the time and space discussing the needs and practices associated with assembling, modeling, and analyzing the data that will support assessments. We thank our colleagues Xiaoju (Julie) Chen, Gwen DiPietro, Rachel Hoesly, and Francis McMichael from CMU, Vikas Khanna and Melissa Bilec at the University of Pittsburgh, and Joyce Cooper of the University of Washington for their many thoughts, comments, and contributions to make this book project a success. A special thanks to Cate Fox-Lent and Michael M. Whiston who provided substantial proofreading assistance for drafts. We also thank dozens of students and colleagues for many interactions, questions and inspirations over the years. We hope that our experiences, as represented here in this free e-book, will make you a more informed and educated teacher and practitioner of LCA and allow you to learn it and apply it right the first time - as you are introduced to the topic. H. Scott Matthews Deanna H. Matthews Chris T. Hendrickson July 2014 Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
9
References Hendrickson, Chris T., Lester B. Lave, and H. Scott Matthews. Environmental life cycle assessment of goods and services: An input-output approach. RFF Press, 2006. Hendrickson, Chris T. and H. Scott Matthews, Civil Infrastructure Planning, Investment and Pricing. http://cspbook.ce.cmu.edu/ (accessed July, 2013).
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
10
Chapter 1: Life Cycle and Systems Thinking
Chapter 1 : Life Cycle and Systems Thinking In this chapter, we introduce the concept of "thinking" about life cycles. Whether or not you become a practitioner of LCA, this skill of broadly considering the implications of a product or system is useful. We first provide definitions of life cycles and a short history of LCA as it has grown and developed over the past decades and then give some examples where life cycle thinking (not full-blown LCAs) has demonstrated where analyses can lead (or has already led) to poor decisions. The goal is to learn how to think about problems from a system wide perspective.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. State the concept of a life cycle and its various stages as relevant to products. 2. Illustrate the complexity of life cycles for even simple products. 3. Explain why environmental problems, like physical products, are complex and require broad thinking and boundaries that include all stages of the life cycle. 4. Describe what kinds of outcomes we might expect if we fail to use life cycle thinking.
Overview of Life Cycles We first learn about life cycles at a young age – the butterfly's genesis from egg to larva to caterpillar to chrysalis to butterfly; the path of water from precipitation into bodies of water, then evaporation or transpiration back into the air. Frogs, tomatoes in the garden, seasons throughout the year – all life cycles we know or experience in our own life cycle. Each individual stage along the cycle is given a distinct term to distinguish it from the others, yet each stage flows seamlessly into the next often with no clear breaks. The common theme is a continuous stepwise path, one stage morphing into the next, where after some time period we are back to the initial starting point. A dictionary definition of life cycle might be "a series of stages or changes in the life of an organism". Here we consider this definition for products, physical processes, or systems. While we often are taught or consider life cycles as existing in the natural world, we can just as easily apply the concept to manmade products or constructs: aluminum's journey from beverage can to recycle bin back to beverage can; a cellphone we use for our 2-year contract period then hold onto (because it must have some value!) before donating to a good cause Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
11
where (we presume) it is used again before…being recycled? … being thrown away? The same common theme – a continuous stepwise path, one stage morphing into the next, where after some time we are (or may be) back to the initial starting point. It is these kinds of life cycles for manmade products and systems that are the focus of this book. As the domain of sustainable management has taken root, increasingly stakeholders describe the need for decision making that considers the "life cycle". But what does that mean? Where does that desire and intent come from? The entire life cycle for a manmade product goes from obtaining everything needed to make the product, through manufacturing it, using it, and then deciding what to do with it once it is no longer being used. Returning to the natural life cycles described above this means going from the birth of the product to its death. As such, this kind of view is often called a "cradle to grave" view of a product, where the cradle represents the birthplace of the product and the grave represents what happens to it when we are done with it – often to be thrown into a landfill. Some life cycles may focus on the process of making the product (up to the point of leaving the factory) and have a "cradle to gate" view, where the word gate refers to the factory gate. If we have a fairly progressive view, we might think about alternatives to a "grave". That might mean recycling of some sort, or taking back the product and using it again. Building on this alternative terminology, proponents have also referred to the complete recycling of products as going from "cradle to cradle". Consider some initial product life cycle views: •
A piece of fruit is grown on a farm which uses water and perhaps various fertilizers and equipment to bring it to market. There it is sold to either a food service business or an individual consumer. While much of it is hopefully eaten, some of it will not be edible and the remainder will be disposed of as food waste – either as compost or in the trash.
•
A tuxedo is sewn together at a factory and then distributed and sold. It is purchased either for personal use (perhaps only being used once or twice a year), or for the purposes of renting it out for profit to people who need it only once, and maybe cannot justify the cost of buying one. The rental tuxedo will be rented several times a month, and after each rental it is cleaned and prepared for the next rental. Eventually the tuxedo will either be too worn to use, or the owner will grow out of it. At that point it is likely donated or thrown away.
•
A car is put together from components at a factory. It is then delivered to a dealer, purchased by a consumer, and driven for a number of years. At some point the owner decides to get rid of the car – perhaps selling it to another driver who uses it
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
12
Chapter 1: Life Cycle and Systems Thinking
for several years. Eventually its owner finds no sufficient value for it, and it will likely be shredded into small pieces and useful metals reclaimed. •
A computer is assembled from components manufactured across the world (all of which are shipped to an assembly line). It is bought and plugged in by the owner, consuming electricity for several years before becoming obsolete. At the end of its useful life it might be sold for a fraction of its purchase price, or may be donated to a party that still finds value in it, or it may be stored under a desk for several years. Like the car example above, though, eventually the owner will find no sufficient value for it and want to get rid of it.
We can already start to think about some implications of these basic life cycles. Using fuels and electricity generates pollution. Applying fertilizers results in runoff and stream contamination. Washing a tuxedo releases chemicals into wastewater systems that need to be removed. Making semiconductor chips consumes large amounts of water and uses hazardous chemicals. Finally, putting items in landfills minimizes our opportunity to continue extracting usefulness from those value-added items, takes up land that we cannot then use for other purposes, and, if the items contain hazardous components, leaks may eventually contaminate the environment. This is a modern view of a product. We have not always been so broad and comprehensive in thinking about such things. In the next few sections we briefly talk about the related history of this kind of thinking, and also give some sobering examples of decisions and products that were made (or promoted) that had not fully considered the life cycle.
A Brief History of Engineering and The Environment Before we further motivate life cycle thinking, let's briefly talk about the history of industrial production, environmental engineering, science, and management as it applies to managing the impacts of products. While engineers and others have been creating production or manufacturing processes for products for centuries, nearly all of the production systems we have created in that time are "linear", i.e., we need to keep feeding the system with input at one end to create output at the other. We design such linear processes independently of whether we will have long-lasting supplies of the needed inputs, and certainly have not made contingencies for how to change the process should we begin to run out of those resources. We also have thought quite linearly in terms of how well the natural environment could deal with any potential wastes from the production systems we have designed. It is worth realizing that environmental engineering (i.e., the integration of science to improve our natural environment) is a fairly young discipline. While there is evidence of ancient civilizations making interesting and innovative solutions to dealing with wastes, the
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
13
establishment of a real environmental engineering profession was not really formalized until around 1900. Initially, what we now call environmental engineering grew out of the need to better manage urban wastes, and thus most of the activity was originally referred to as "sanitary engineering". Such activities involved diversion of waste streams to distant sinks to avoid local health problems, such as sewer systems (Tarr 1996). Eventually, end of pipe treatment emerged. By end of pipe, we mean that the engineering problem was focused on what to do with the waste of a system (e.g., a factory or a social waste collection system) after it has already been produced. Releases of wastes and undesirable outputs to the environment are also called emissions. Another historical way of dealing with environmental problems has been through remediation. Remediation occurs after the pollution has already occurred, and may involve cleaning up a toxic waste dump, dredging a river to remove long-buried contaminants that were dumped there via an effluent pipe, or converting contaminated former industrial sites (brownfields) into new developments. The remediation activities may occur soon after or even decades after the initial pollution occurred. An alternative paradigm was promoted in the 1980s, referred to as pollution prevention (P2, or cleaner production). It is probably obvious that the whole point of this alternative paradigm was to make stakeholders realize that it is costly and late in the process to wait until the end of the pipe to manage wastes. If we were to think about the inevitable waste earlier in the process chain, we could create a system that produces less (or ideally, no) waste. A newer paradigm is to promote sustainability. Achieving sustainability refers to the broader balancing of social, economic, and environmental aspects within the planet's ability to provide. The United Nations' Brundtland Commission (1987) suggested "sustainable development is development that meets the needs of the present without compromising the ability of future generations to meet their own needs". Almost all people in developed nations share the goals of improving environmental quality and making sure that future generations have sufficient resources. Unfortunately consumers, business leaders, and government officials do not have the information required to make informed decisions. We need to develop tools that tell these decision makers the life cycle implications of their choices in selecting materials, products, or energy sources. These decisions are complicated: they depend on the environmental and sustainability aspects of all products and services that contribute to making, operating, and disposing of those materials, products, or energy sources. They also depend on being able to think non-linearly about our production systems and envision the possibilities of resource scarcity or a lack of resilience in the natural environment. Accomplishing these goals requires life cycle thinking, or thinking about environmental problems from a systems perspective. Nowadays all of these activities are part of what we refer to as environmental engineering. Despite trends towards pollution prevention and sustainability, basic challenges remain to design better end of pipe systems even in the developed world where pollution prevention is Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
14
Chapter 1: Life Cycle and Systems Thinking
well known but is deemed as too expensive for particular processes (or where all costeffective P2 solutions have already been implemented). But the general goal of the field is to reduce pollution in our natural environment, and a primary objective is to encourage broader thinking and problem solving that goes back before the end of the pipe and prevents pollution generation. Practically, we will not achieve a pollution-free world in our lifetimes. But we can help get there by thinking about environmental problems in a life cycle context, and ideally identify solutions that focus on stages earlier in the life cycle than the point where the waste pipe interfaces with our natural environment.
Life Cycle Thinking Now that we have introduced the idea of a life cycle, and motivated why thinking about products as systems or life cycles is important, we can dive deeper into the ways this kind of thinking is defined and how it has evolved. Much of this development has come in the engineering and science communities, and thus the views and representations of life cycles are fairly technical. That said, given the typically focused and detailed views of scientists and engineers, you will see that the way these systems are studied is quite broad. A conceptual view of the stages of such life cycles is in Figure 1-1. Beginning with the linear path along the top, we first extract raw materials from the ground, such as ores or petroleum. Second, these are processed, transformed or combined to make basic material or substance building blocks, such as metals, plastics or fuels. These materials are combined to manufacture a product such as an automobile. These final products are then shipped (while not shown) by some mode of transport to warehouses and/or stores to be purchased and used by other manufacturers or consumers. During a product's use phase it may be used to make life easier, provide services, or make other products, and this stage may require use of additional energy or other resources (e.g., water). When the product is no longer needed, it enters its "end of life" which means managing its disposition, possibly treating it as waste.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
15
Figure 1-1: Overview of a Physical Product Life Cycle (OTA, 1992)
As Figure 1-1 also shows, at the end of life phase there are alternatives to treating a product as waste. The common path (linear path across the top) is for items to be thrown away, a process that involves collection in trucks and putting the item as waste in a landfill. However, the bottom row of lines and arrows connect the end of life phase back to previous stages of the typical life cycle through alternative disposition pathways. Over the course of a life cycle, products, energy and materials may change form but will not disappear. Reuse takes the product as is (or with very minor effort) and returns it to the use phase, such as a tuxedo. Remanufacturing returns the product to the manufacturing stage, which may mean partially disassembling the product but then re-assembling it into a new final product to be delivered, such as a power tool or a photocopier. Finally, recycling involves taking a product back to its raw materials, which can then be processed into any of a number of other products, such as aluminum beverage cans or cardboard boxes. This bottom row also reminds us that despite the colloquial use of the word "recycling" in society, recycling has a very distinct definition, as noted above. Other disposition options have their own terms. An Internet search would turn up hundreds more pictures of life cycles, but for our introductory purposes these will suffice. Once we discuss the actual ISO LCA Framework in Chapter 4 we will see the standard figures and some additional useful ones. If you are from an engineering background, you might be asking where the other traditional product stages fit in to the product life cycle described above. In engineering, the typical product life cycle starts with initiation of an idea, as well as research and design iterations that lead to multiple prototypes, and eventually, mass production. One could classify all such activities as research and development (or R&D) that would come to the left of all activities (or perhaps in parallel with some activities such as material extraction) in Figure 11. We could imagine a reverse flow arrow for "Re-design" going along the bottom of Figure 1-1 to represent product failures or iterations. While not represented in the figure above, all of these R&D-like activities are relevant stages in the life cycle. As we will see, though, when analyzing life cycles for environmental impact, these stages are typically ignored.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
16
Chapter 1: Life Cycle and Systems Thinking
Simple and Complex Life Cycles Before we go further in our discussion of life cycles, it is useful to pause and think about all of the components of something with a very simple life cycle, like a paper clip. Get a blank sheet of paper, and write "paper clip" in a corner of the sheet. If we think very simply about its life cycle (e.g., using Figure 1-1 as a guide), we can work backwards from the paper clip we are used to. To get its shape, it is coiled with machinery. We can write "coiling" and draw an arrow from the words "coiling" to "paper clip". Before coiling it is just a straight wiry piece of steel. Steel is made from iron and carbon. We can write "steel" and draw an arrow to "coiling". Iron ore and the carbon source both need to be extracted from the ground. All of these components and pieces are shipped between factories by truck, rail, or other modes of transportation. Any or all of these stages of the life cycle could be added to the diagram. Putting all these materials and processes into a diagram is not so simple. Even that description above for a paper clip was very terse. If we think a little more, we realize that all of those stages have life cycles of their own. For example, the machinery that coils the steel wire into a paper clip must be manufactured (its use phase is making the paper clip!). The metal and other parts needed to make the machine also must be processed and extracted. The same goes for all of the transportation vehicles and the infrastructure they travel on and the factories to make iron and steel, etc. Figure 1-2 shows what the diagram might look like at this point.
Figure 1-2: Exploded View Diagram of Production of Paper Clip
This chain goes back, almost infinitely, and the sheet of paper is quickly filled with words and arrows. Even a product as simple as a paper clip has a complex life cycle. Thus a
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
17
product that we consider to be "complex" (for example a car) has a ridiculously complex life cycle! Now that we can appreciate the complexity of all life cycles, you can begin to understand why our thought processes and models need to be sufficiently complex to incorporate them. Without going in to all of the required detail, but to impress upon you the complexity of LCA for more complex products, consider that a complete LCA of an automobile would require careful energy and materials balances for all the stages of the life cycle: 1. the facilities extracting the ores, coal, and other energy sources; 2. the vehicles, ships, pipelines, and other infrastructure that transport the raw materials, processed materials, and subcomponents along the supply chain to manufacture the consumer product, and that transport the products to the consumer: iron ore ships, trucks carrying steel, engines going to an automobile assembly plant, trucks carrying the cars to dealers, trucks transporting gasoline, lubricating oil, and tires to service stations; 3. the factories that make each of the components that go into a car, including replacement parts, and the car itself; 4. the refineries and electricity generation facilities that provide energy for making and using the car; and 5. the factories that handle the vehicle at the end of its life: battery recycling, shredding, landfills for shredder waste. Each of these tasks requires energy and materials. Reducing requirements saves energy, as well as reducing the environmental discharges, along the entire supply chain. Often a new material requires more energy to produce, but promises energy savings or easier recycling later. Evaluating whether a new material helps improve environmental quality and sustainability requires an examination of the entire life cycles of the alternatives. To make informed decisions, consumers, companies, and government agencies must know the implications of their choices for environmental quality and sustainability. Having good intentions is not sufficient when a seemingly attractive choice, such as a battery-powered car, can wind up harming what the manufacturer and regulator were trying to protect. This book provides some of the tools that allow manufacturers and consumers to make the right choices.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
18
Chapter 1: Life Cycle and Systems Thinking
Systems Thinking in the Life Cycle All of this discussion of increasingly larger scales of problems requires us to be more explicit in discussing an issue of critical importance in LCA studies that relates to system boundaries. Of course a system is just a collection or set of interconnected parts, and the boundary is the subset of the overall system that we care to focus on. Our chosen system boundary helps to shape and define what the appropriate parts are that we should study. Above we suggested that the entire life cycle boundary goes from cradle to grave or cradle to cradle. Either choice means that we will have a very large system boundary, and maintaining that boundary (as we will see later) will require a significant amount of effort to complete a study. Due to this effort requirement, or because of different interests, we may instead choose a smaller system boundary. If we are a manufacturer, perhaps our focus is only the cradle to gate impacts. If so, our boundary would include only the stages up to manufacturing. It is also possible that the boundary of our interest lies only in our factory, which further constrains the system boundary. Life cycle thinking is not restricted to manufactured products. Services, systems, and even entire urban areas can be better understood via life cycle thinking. Services are particularly interesting because such activities are typically considered as having very low impacts (e.g., consulting or banking) because there is no physical good being created, but in reality the same types of effects are uncovered across the life cycle through the service sector's dependence on fuels and electricity. Entire systems (e.g., a roadway network or the electric power grid) can be considered from building all of the equipment components and also then thinking about its design and disposition. At an even higher level, the life cycle of cities includes the life cycles of all of the resources consumed by residents of the city, not just the activities they do within the city's borders. Finally, life cycle thinking is often useful when making comparisons, such as paper vs. plastic bags or cups, cloth vs. disposable diapers, or retail shopping vs. e-commerce. The relevant issues to deal with in such comparisons would be whether one option is more useful than another, whether they are equal, whether they have similar production processes, etc. In fact as we will see some of the great classic comparisons that have been done in the life cycle analysis domain were very simple comparisons.
A History of Life Cycle Thinking and Life Cycle Assessment We will discuss the formal methods that apply life cycle thinking to real questions in future chapters (called life cycle analysis or assessment). In a life cycle analysis or assessment, the total and comparative impacts of the life cycle stages are considered, with or without quantification of those impacts. But to start, let us talk about some of the original studies
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
19
that inspire the field of life cycle thinking (before we even knew there was a field for such things). Most people attribute the first life cycle assessment (LCA) to Coca-Cola in 1969. At the time, Coca-Cola sold its product to consumers in individual glass bottles. Coca-Cola was trying to determine whether to use glass or plastic containers to deliver their beverage product, and wanted to formally support a decision given the tradeoffs between the two materials. Glass is a natural material, but Coca-Cola suggested switching to plastic bottles. They reasoned that this switch would be desirable for the ability to produce plastics in their own facilities, the lower weight of plastic to reduce shipping costs, and the recyclability of plastic versus glass at the time. No specific form of this study has been publicly released but we can envision the considerations that would have been made. More recently, in the early 1990s, there were various groups of researchers debating the question of "Paper or plastic?" This simple question, which you might get at the grocery store checkout counter or coffee shop, turned into relatively complex exchanges of ideas and results. We may think that we know that the correct answer is "paper," because it is a "natural" product rather than some chemical based material like plastic. We can feel selfsatisfied, even if the bag gets wet and tears, spilling our purchases on the ground because we made the natural and environmentally friendly decision. But even these simple questions can, and should, be answered by data and analysis, rather than just a feeling that the natural product is better. The ensuing analysis ignited a major controversy over how to decide which product is better for the environment, beginning with an analysis of paper versus polystyrene cups (Hocking 1991). Hocking's initial study was focused on energy use and estimated that one glass cup used and rewashed 15 times required the same amount of energy as manufacturing 15 paper cups. He also estimated break-even use values for ceramic and plastic cups. The response generated many criticisms and spawned many follow-up studies (too many to list here). In the end, though, what was clear at the time of these studies was that there was no single agreed upon answer to the simple question of "paper vs. plastic". Even now, any study using the best data and methods available today, will still conclude with an answer along the line of "it depends". This is a sobering outcome for a discipline (life cycle thinking) trying to gain traction in the scientific community. Beyond these studies, other early analyses surprised people, since they found that paper bags, paper cups (or even ceramic cups), and cloth diapers were not obviously superior to their maligned alternatives (i.e., plastic bags, styrofoam cups and disposable diapers) in terms of using less energy and materials, producing less waste, or even disposal at the end of life. •
Paper for bags requires cutting trees and transporting them to a paper mill, both of which use a good deal of energy. Paper-making results in air emissions and water discharges of chlorine and biological waste. After use, the bag goes to a landfill where it gradually decays, releasing methane.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
20
Chapter 1: Life Cycle and Systems Thinking
•
A paper hot-drink cup generally has a plastic coating to keep the hot liquid from dissolving the cup. The plastic coating introduces the same problems as the foam plastic cup. The plastic is made from petroleum with relatively small environmental discharges. Perhaps most surprising, washing a single ceramic cup by hand uses a good deal of hot water and soap, resulting in discharges of waste water that has to be treated and the expenditure of a substantial amount of fuel to heat the water, although washing the cup in a fully loaded dish washer uses less soap and hot water per cup.
•
The amount of hot water and electricity required to wash and dry cloth diapers is substantial. If water is scarce or sewage is not treated, washing cloth diapers is likely to cause more pollution than depositing disposable diapers in a landfill. The best option depends on the issue of water availability (washing uses much more water) and heating the water.
In short, it is not obvious which product is more environmentally benign and more sustainable. Such results are counterintuitive, but they reinforce the importance of life cycle thinking. The analyses found that the environmental implications of choosing paper versus plastic were more similar than people initially thought. Which is better depends on how bad one thinks water pollution is compared to air pollution compared to using a nonrenewable resource. Perhaps most revealing was the contrast between plants and processes to make paper versus plastic. The best plant-process for making paper cups was much better than the worst plant-process; the same was true for plastic cups. Similarly, the way in which the cups were disposed of made a great deal of difference. Perhaps the most important lesson for consumers was not whether to choose one material over another, but rather to insist that the material chosen be made in an environmentally friendly plant. The original analyses showed that myriad processes are used to produce a material or product, and so the analyst has to specify the materials, design, and processes in great detail. This led to another problem: in a dynamic economy, materials, designs, and processes are continually changing in response to factor prices, innovation, regulations, and consumer preferences. For example, in a life cycle assessment of a U.S.-manufactured automobile done in the mid-1990s, the design and materials had changed significantly by the time the analysis was completed years later. Still another problem is that performing a careful material and energy balance for a process is time-consuming and expensive. The number of processes that are practical to analyze is limited. Indeed, the rapid change in designs, materials, and processes together with the expense of analyzing each one means that it is impractical and inadvisable to attempt to characterize a product in great detail. The various dependencies, rationales, and assumptions used all make a great deal of difference in the studies mentioned
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
21
above (for which we have provided no real detail yet). LCA has a formal and structured way of doing the analysis, which we will begin to discuss in Chapter 4.
Decisions Made Without Life Cycle Thinking Hopefully you are already convinced that life cycle thinking is the appropriate way of thinking about problems. But this understanding is certainly not universal, and there are various examples of not taking a life cycle view that led to poor (albeit well intentioned) decisions being made. A useful example is the consideration of electric vehicles in the early 1990s. At the time, California and other estates were interested in encouraging the adoption of vehicles with no tailpipe emissions in an effort to reduce emissions in Southern California and to gain the associated air quality benefits. Policymakers at the time had a specific term for such vehicles – "zero emissions vehicles (ZEVs)". The thought was that getting a small but significant chunk of the passenger vehicle fleet to have zero emissions could yield big benefits. Regulations at the time sought to get 2% of new vehicles sold to be ZEVs by 1998. In parallel, manufacturers such as General Motors had been designing and developing the EV-1 and similar cars to meet the mandated demand for the vehicles (see Figure 1-3). Figure 1-3: General Motors' EV-1 (Source: motorstown.com)
So why did we refer to this case as one about life cycles? The electric vehicles to be produced at the time were much different than the electric vehicles of today that include hybrids and plug-in hybrids. These initial cars were rechargeable, but the batteries were lead-acid batteries – basically large versions of the starting and ignition batteries we use in all cars (by large, we mean the batteries were 1,100 pounds!). Let us go back to Figure 1-1 and use life cycle thinking to briefly consider such a system. How would the cars be recharged? They would run on electricity, which even in a progressive state like California leads to various emissions of air pollutants. Similarly, the batteries would have large masses of lead that would need to be processed efficiently. Lead must be extracted, smelted, and processed before it can be used in batteries and then, old lead-acid batteries are often collected and recycled. None of these processes are 100% efficient, despite the claims at the time by industry that it was the case. Would these vehicles be produced in factories with no pollution? It is hard to consider that these vehicles would really have "zero emissions" – but then again, zero is a very small number! There would be
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
22
Chapter 1: Life Cycle and Systems Thinking
increased emissions in the life cycle of these electric vehicles – the question was whether those increases would fully offset the potential gains of reduced tailpipe emissions. Aside from the perils of considering anything as having zero emissions, various parties began to question whether these vehicles would in fact have any positive improvement on air quality in California, and further, given the need for more electricity and lead, whether one could even consider them as beneficial. In a study published by Lave et al. (to whom this book is dedicated) in Science in 1995, the authors built a simple but effective model of the life cycle of these vehicles that estimated that generating the electricity to charge the batteries would result in greater emissions of nitrogen oxide pollution than gasoline-powered cars. Eventually, California backed off of its mandate for ZEVs, partly because of such studies, and policymakers learned important lessons about considering whole life cycles as well as casual use of the number zero. The policymakers had been so focused on the problem of reducing tailpipe emissions that they had overlooked the back-end impacts from lead and increased electricity generation. It is fair to say this was one of the first instances of life cycle thinking being used to change a "big decision". The lesson again is that life cycle thinking is needed to make informed decisions about environmental impacts and sustainability. Being prepared to use life cycle thinking and analysis to support big decisions is the focus of this book. A more recent example of life cycle thinking in big decisions is the case of compact fluorescent lamps (CFLs), which were heavily promoted as energy efficient alternatives to incandescent bulbs. While CFLs use significantly less electricity in providing the same amount of light (and thus cost less in the use phase) as traditional bulbs, their disposal represented a problem due to the presence of a small amount of mercury in the lamps (about 4mg per bulb). This amount of mercury is not generally a problem for normal, intact, use of the lamps (and is less mercury than would be emitted from electric power plants to power incandescent bulbs). However, broken CFLs could pose a hazard to users due to mercury vapor – and the DOE Energy Star guide to CFLs has somewhat frightening recommendations about evacuating rooms, using sealed containers, and staying out of the room for several hours. None of this information was good news for consumers thinking about a choice of incandescent vs. CFL lighting choices. The examples and discussion above hopefully reveal that you can think about life cycles qualitatively or quantitatively, meaning with or without numbers (more on that in Chapter 2).
Inputs and Outputs of Interest in Life Cycle Models Above we have suggested that there is a need to think about products, services, and other processes as systems by considering the life cycle. We have also mentioned some popular
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
23
examples of the kinds of life cycle thinking studies that have been done. It is also worth discussing the types of effects across a life cycle that we might be interested in tracking or accounting for. By 'effects' we mean what happens as a result of a product being manufactured, or a service being provided, etc. There are likely economic costs incurred, for example by paying for the parts and labor needed for assembly. There are interesting and relevant issues to consider when focused purely on economic factors, and Chapter 3 discusses this type of thinking. In many cases, the 'effects' of producing or using a product mean consuming energy in some way. Likewise, there may be emissions of pollution to the air, water, or land. There are many such effects that one might be interested in studying, and more importantly, in being able to detect and measure. Thus we can already create a list of potential effects that one might be concerned about in a life cycle study. In terms of effects associated with inputs to life cycle systems, we could be concerned about: •
Use of energy inputs, including electricity, as well as solid, liquid, and gaseous fuels.
•
Use of resources as inputs, such as ores, fertilizers, and water.
Note that our concern with energy and resource use as inputs may be in terms of the quantities of resources used and/or the extent to which the use of these resources depletes the existing stock of that resource (i.e., are we consuming a significant share of the available resource?). We may also be concerned with whether the energy or resources being consumed are renewable or non-renewable. In terms of effects associated with outputs of life cycle systems, we could be concerned about: •
The product created as a result of an activity, such as electricity from a power plant.
•
Emissions of air pollution, for example conventional air emissions such as sulfur dioxide, nitrogen oxides, and carbon monoxide.
•
Emissions of greenhouse gases, such as carbon dioxide, methane, and nitrous oxide.
•
Emissions to fresh or seawater, including solid waste, chemical discharges, toxics, and warming.
•
Other emissions of hazardous or toxic wastes to air, land, water, or recycling facilities.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
24
Chapter 1: Life Cycle and Systems Thinking
In short, there is no shortage of energy, environmental, and other effects that we may care about and which may be estimated as part of a study. As we will see later in the book, we may have interest in many effects but only be able to get quality data for a handful of them. We can choose to include any effect for which we think we can get data over as many of the parts of the life cycle as possible. One could envision annotating the paper clip life cycle diagram created above with colored bars representing activities in the life cycle we anticipate have significant inputs or outputs associated with them. For example, activities that we expect to consume significant quantities of water could have a blue box drawn around them or to have a blue square icon placed next to them. Activities we expect to release significant quantities of air pollutants could have black boxes or icons. Activities we expect to create a large amount of solid waste could be annotated with brown. While simplistic (and not informed by any data) such diagrams can be useful in terms of helping us to look broadly at our life cycle of interest and to see where in the life cycle we anticipate the problems to occur. Aside from simply keeping track of (accounting for) all of these effects across the life cycle, a typical reason for using life cycle thinking is to not just measure but prioritize. Another way of referring to this activity might be hot spot analysis, where we look at all of the effects and decide which of the life cycle stages contributes most to the total (where "hot spots" appear). Our colored box or icon annotation above could be viewed as a crude hot spot analysis, because it is not informed by actual data yet. For most cars, the greatest energy use happens during the use phase. Cars in the United States are typically driven more than 120,000 miles over their useful lives. Even fairly fuelefficient cars will use more energy there than at any other stage of their life cycle. This is a seemingly obvious example but it illustrates the reason we use life cycle thinking – as we have shown above our intuition is not sufficient in assessing where effects occur, and only by actually collecting data and estimating the effects can we effectively identify hot spots. This use of life cycle thinking to support hot spot analysis helps us identify where we need to focus our attention and efforts to improve our engineering designs. If done in advance, it can have a significant benefit. If done too late, it can lead to designs such as large lead-acid battery vehicles. Likewise if we create a plan to generate numerical values representing several of these life cycle effects, we will eventually have to make decisions about how to compare them or prioritize them. Such a decision process will be complicated by needing to compare releases of the same type of pollution across various media (air, water, or land) and also by needing to compare releases of one pollutant against another, comparing pollution and energy, etc. While complicated, the process of making all of these judgments and choices will assist with making a study that we can use to help our decision process. Chapter 12 overviews the types of methods used to support such assessments.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
25
From Inputs and Outputs to Impacts It is appropriate early on in this textbook to briefly discuss the kinds of uses, emissions, and releases discussed above in connection with the types of environmental or resource use problems they create. The new concept in this section is the idea of an environmental impact. Unlike the underlying inputs and outputs of interest such as resource use or emissions, an environmental impact exists when the underlying flows cause an environmental problem. One can think of the old phrase "if a tree falls in the forest but no one is there to hear it, does it make a sound?" This is similar to the connection between environmental releases and environmental impacts. It is possible that a release of a specific type and quantity of pollutant into the environment could have little or no impact. But if the release is of sufficient quantity, or occurs in a location near flora or fauna (especially humans) it is likely that there will be measurable environmental impact. Generally, our concerns are motivated by the impacts but are indicated by the uses or releases because most of us can not directly estimate the impacts. In other words, we often look at the quantities of inputs and outputs as a proxy for the impacts themselves that need to be estimated separately. This brief section is not a substitute for a more rigorous introduction to such environmental management issues, and should be supplemented with external work or reading if this is not an area of your expertise. One could easily spend a whole semester learning about these underlying connections before attempting to become an expert in life cycle thinking.
Example Indicators for Impacts that Inspire Life Cycle Thinking In this section, we present introductory descriptions of several prominent environmental impacts considered in LCA studies as exemplars and discuss how various indicators can guide us to the actual environmental problems created. If interested, there are more detailed summaries available elsewhere from agencies, such as the US Environmental Protection Agency, US Geological Survey, the Department of Energy, and we will circle back to discussing them in Chapter 12. Impact: Fossil fuel depletion – Use of energy sources like fossil fuels is generally an easy to measure activity because energy costs us to acquire, and there are billing records and energy meters available to give specific quantities. Beyond the basic issue of using energy, much of our energy use comes from unsustainable sources such as fossil fuels that are finite in supply. We might care simply about the finiteness of the energy resource availability as a reason to track energy use across the life cycle. As mentioned above, we might seek to separately classify our use of renewable and non-renewable energy. We might also care about whether a life cycle system at scale could consume significant amounts of the available resources. If so, the use of energy by our life cycle could be quite significant. In the context of our descriptions above, some quantity of fossil energy use (e.g., in BTU or MJ) may be an
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
26
Chapter 1: Life Cycle and Systems Thinking
indicator for the impact of fossil fuel depletion. Of course, all of the energy extraction, conversion, and combustion processes may lead to other types of environmental impacts (like those detailed below). Impact: Global Warming / Climate Change – Most people know that there is considerable evidence suggesting that manmade emissions of greenhouse gases (GHGs) lead to global warming or climate change. The majority of such GHG emissions come from burning fossil fuels. While we might already be concerned with the use of energy (above), caring more specifically about how our choices of energy sources may affect climate change is an additional impact to consider. Carbon dioxide (CO2) is the most prominent greenhouse gas, but there are other GHGs that are emitted from human activities that also lead to warming of the atmosphere such as methane (CH4) and nitrous oxide (N2O). These latter GHGs have far greater warming effects per unit than carbon dioxide and are emitted from systems such as oil and gas energy infrastructure systems and agricultural processes. GHGs are inevitably global pollutants, as increasing concentrations of them in the atmosphere lead to impacts all over the planet, not just in the region or specific local area where they are emitted. These impacts may eventually manifest as increases in sea levels, migration of biotic zones, changes in local temperatures, etc. Our concern about climate change may be rooted in a desire to assess which stage or component of our product or process has the highest carbon footprint, and thus all else equal, the biggest contributor to climate change. The GHG emissions are indicators of the impacts of global warming and climate change. Impact: Ozone Depletion – In the early 1970s, scientists discovered that human use of certain substances on the earth, specifically chlorofluorocarbons (CFCs), led to reduction in the quantity of ozone (O3) in the stratosphere for a period of 50-100 years. This phenomenon is often tracked and referred to as "holes in the ozone layer". The ozone layer, amongst other services, keeps ultraviolet rays from reaching the ground, preserving plant and ocean life and avoiding impacts such as skin cancers. The Montreal Protocol called for a phase out of chemicals that deplete the ozone layer, but not all countries ratified it, not all relevant substances were included, and not all uses were phased out. Consequently, while emissions of many of these substances have been dramatically reduced in the past 30 years, they have not been eliminated, and given the 50-100 year lifetime, ozone depletion remains an impact of concern. Thus, releases of the various ozone-depleting substances can be indicators of potential continued impacts of ozone depletion. Note that there is also "ground level" ozone that is created by interactions of local pollutants and helps to create smog, which, when breathed in, can affect human health. This is an entirely different but important potential environmental impact related to ozone. Impact: Acid Rain – Releases of various chemicals or chemical compounds lead to increased levels of acidity in a local or regional environment. This acidity penetrates the water cycle and can eventually move into clouds and rain droplets. In the developed world the key linkage was between emissions of sulfur dioxide (SO2) and acidity of freshwater systems. One of the original points of concern was emissions of sulfur dioxide by coal-fired power Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
27
plants because they were large single sources and also because they could be fairly easily regulated. Emissions of these pollutants are an indicator of the potential impacts of more acidic environments such as plants and aquatic life destroyed. While in this introduction we have only listed acid rain as an impact, acid rain is part of a family of environmental impacts related to acidification, which we will discuss in more detail later. In short, other non-sulfur compounds like nitrogen oxides can also lead to acidification of waterways, and systems other than freshwater can be affected. Acidification of water occurs due to global uptake of carbon dioxide and is of increasing concern in oceans where acidification affects coral reefs and thus the entire ocean ecosystem. There are various other environmental impacts that have been considered in LCA studies, such as those associated with eutrophication, human health, and eco-toxicity, but we will save discussion of them for later in the text. These initial examples, though, should demonstrate that there are a wide variety of local and global, small and large scale, and scientifically relevant indicators that exist to help us to assess the many potential environmental impacts of products and systems.
The Role of Design Choices The principles of LCA can help to build frameworks that allow us to consider the implications of making design (or re-design) decisions and to track the expected outcomes across the life cycle of the product. For example, deciding whether to make a car out of aluminum or steel involves a complicated series of analyses: • Would the two materials provide the same level of functionality? Would structural strength or safety be compromised with either material? Lighter vehicles have been found to be less safe in crashes, although improved design and new automation technology might remove this difference (NRC 2002, Anderson 2014). A significant drop in safety for the lighter vehicles would outweigh the energy savings, depending on the values of the decision maker. • Are there any implications for disposal and reuse of the materials? At present, about 60% of the mass of old cars is recycled or reused. Moreover, motor vehicles are among the most frequently recycled of all products since recycling is usually profitable; both aluminum and steel are recycled and reused from automobiles (Boon et al. 2000). It takes much less energy to recycle aluminum than to refine it from ore. The advantage for recycling steel is smaller. • What is the relative cost of the two materials, both for production and over the lifetime of the vehicle? An aluminum vehicle would cost more to build, but be lighter than a comparable steel vehicle, saving some gasoline expenses over the lifetime of
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
28
Chapter 1: Life Cycle and Systems Thinking
the vehicle. Do the gasoline savings exceed the greater cost of manufacturing? Of energy? Of environmental quality? In this example, steel, aluminum, copper, glass, rubber, and plastics are the materials, while electricity, natural gas, and petroleum are the energy that go into making, using, and disposing of a car. The vehicle runs on gasoline, but also needs lubricating oil and replacement parts such as tires, filters, and brake linings. At the end of its life, the typical American car is shredded; the metals are recycled, and the shredder waste (plastic, glass, and rubber) goes to a landfill.
What Life Cycle Thinking and Life Cycle Assessment Is Not The purpose of this chapter has been to motivate life cycle thinking, and why it should be chosen to ensure broadly scoped analysis of issues with potential environmental impacts – i.e., we have been introducing "what life cycle thinking is". We end the chapter by briefly summarizing what life cycle thinking (and, by extension, life cycle assessment) is not able to achieve. First, life cycle thinking will not ensure a path to sustainability. If anything, thinking more broadly about environmental problems has the potential side effect of making environmental problems seem even more complex. At the least it will typically lead to greater estimates of environmental impact as compared to studies with more limited scopes. But life cycle thinking can be a useful analytical and decision support tool for those interested in promoting and achieving sustainability. Second, life cycle thinking is not a panacea - a magic pill or remedy that solves all of society's problems. It is merely a way of structuring or organizing the relevant parts of a life cycle and helping to track performance. Addressing the economic, environmental, and social issues in the context of sustainability can be done without using LCA. To reduce energy and environmental impacts associated with product or process life cycles, we must want to take action on the findings of our studies. By taking action we decide to improve upon the current impacts of a product and make changes to the design, manufacture, or use of the current systems so that future impacts are reduced. LCA is not a single model solution to our complex energy and environmental problems. It is not a substitute for risk analysis, environmental impact assessment, environmental management, benefit-cost analysis, etc. All of these related methods have been developed over many years and may still be useful in bringing to the table to help solve these problems. LCA can in most cases interact with these alternative methods to help make decisions.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 1: Life Cycle and Systems Thinking
29
Chapter Summary Life cycle assessment (LCA) is a framework for viewing products and systems from the cradle to the grave. The key benefit of using such a perspective is in creating a "systems thinking" view that is broadly encompassing and can be analyzed with existing methods. When a life cycle perspective has not been used, unexpected but predictable environmental impacts have occurred. As we will see in the chapters to come, even though there is a standard for applying life cycle thinking to problem solving, it is not a simple recipe. There are many study design choices, variations, and other variables in the system. One person may apply life cycle thinking in one way, and another in a completely different way. We cannot expect then that simply using life cycle thinking will lead to a single right answer that we can all agree on.
References for this Chapter Anderson, J., Kalra, N., Stanley, K., Sorensen, P., Samaras, C., Oluwatola, O. Autonomous Vehicle Technology: A Guide for Policymakers, Santa Monica, CA: RAND Corporation, RR-443-RC, 2014. Boon, Jane E., Jacqueline A. Isaacs, and Surendra M. Gupta, "Economic Impact of Aluminum-Intensive Vehicles on the U.S. Automotive Recycling Infrastructure", Journal of Industrial Ecology 4(2), pp. 117–134, 2000. Hocking, Martin B. "Paper versus polystyrene: a complex choice." Science 251.4993 (1991): 504-505. Lave, Lester, Hendrickson, Chris, and McMichael, Francis, "Environmental implications of electric cars", Science, Volume 268, Issue 5213, pp. 993-995, 1995. Mihelcic, James R., et al. "Sustainability science and engineering: The emergence of a new metadiscipline." Environmental Science and Technology 37.23 (2003): 5314-5324. Tarr, Joel, The Search for the Ultimate Sink, University of Akron Press, 1996. United Nations General Assembly (1987) Report of the World Commission on Environment and Development: Our Common Future. Transmitted to the General Assembly as an Annex to document A/42/427 - Development and International Co-operation: Environment. United States Office of Technology Assessment (OTA), “Green Products by Design: Choices for a Cleaner Environment”, OTA-E-541, 1992.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
30
Chapter 1: Life Cycle and Systems Thinking
End of Chapter Questions 1. On a sheet of paper, draw by hand or software a diagram of a life cycle for a simple product other than a paper clip, with words representing the various activities in the life cycle needed to make the product, and arrows representing connections between the activities. Annotate the diagram with colors or shading to try to represent hot spots for two inputs or outputs that you believe are relevant for decisions associated with the product. 2. Do the same exercise as in Question 1, but for a school or university, which is a service not a physical product. 3. Describe the major activities in each of the five life cycle stages of Figure 1 for a soft drink beverage container of your choice. Describe also the activities needed to support reuse, remanufacturing, and recycling activities for the container chosen. 4. Power plants (especially fossil-fuel based coal and gas-fired units) are frequently mentioned sources of environmental problems. List three specific types of outputs to the environment resulting from these fossil plants. Which other parts of the life cycle of producing electricity from fossil plants also contribute to these problems? 5. Suppose that a particular truck requires diesel fuel to provide freight transportation (that is, moving tons of freight over some distance). In the process, carbon dioxide is emitted from the truck. a. In the terminology of life cycle thinking presented in this chapter, what does the diesel fuel represent? b. What do the freight movement and carbon dioxide emissions represent? c. What stage of the truck life cycle is being presented in this problem so far? What other truck life cycle stages might be important to consider? d. In considering the environmental impacts of trucks, would it be advisable to expand our system of thinking to include providing roadways? Why or why not? 6. Across the life cycle of a laptop computer, discuss which life cycle stages might contribute to the environmental impact categories discussed in the chapter (global warming, ozone depletion, and acid rain). Are there other classes of environmental impact you can envision for this product?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
Dana Fradon, The New Yorker May 17, 1976 (Permission pending but NOT granted to use this figure yet)
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
31
32
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
Chapter 2 : Quantitative and Qualitative Methods Supporting Life Cycle Assessment In this chapter, we introduce basic quantitative skills needed to perform successful work in LCA. The material is intended to build good habits in critically thinking about, assessing, and documenting your work in the field of LCA (or, for that matter, any type of systems analysis problem). First we describe good habits with respect to data acquisition and documentation. Next we describe skills in building and estimating simple models. These skills are not restricted to use in LCA and should be broadly useful for business, engineering, and policy modeling tasks. As this book is intended to be used across a wide set of disciplines and levels of education, we write as if aimed at undergraduates who may not be familiar with many of these concepts. It may be a cursory review for many graduate students. Regardless, improving such skills will make your LCA work even more effective.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. Apply appropriate skills for qualitative and quantitative analysis. 2. Document values and data sources in support of research methods. 3. Improve your ability to perform back of the envelope estimation methods. 4. Approach any quantitative question by means of describing the method, providing the answer, and describing what is relevant or interesting about the answer.
Basic Qualitative and Quantitative Skills To be proficient in any type of systems analysis, you need to have sharp analytical skills associated with your ability to do research, and much of this chapter is similar to what one might learn in a research methods course. While the skills presented here are generally useful (and hopefully will serve you well outside of the domain of LCA) we use examples relevant to LCA to emphasize and motivate their purpose. Much of LCA involves doing "good research" and communicating the results clearly. That is why so many people with graduate degrees are able to learn LCA quickly – because they already have the base of skills needed to be successful, and just need to learn the new domain knowledge. Amongst the most important skills are those associated with your quantitative and qualitative abilities. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
33
Quantitative skills are those associated with your ability to create numerical manipulations and results, i.e., using and applying math and statistics. Qualitative skills are those related to your ability to think and write about your work beyond numbers, and to describe the relevance of your results. While this textbook is more heavily geared towards improving your quantitative skills, there are many examples and places of emphasis throughout the text that are intended to develop your qualitative skills. You will need to be proficient at both to successfully appreciate and perform LCA work. Identifying your own weaknesses in these two areas now can help you improve them while you are also learning new material relevant to the domain. Your quantitative skills are relatively easy to assess – e.g., if you can correctly answer a technical or numerical question by applying an equation or building a model, you can "pass the test" for that quantitative skill. Qualitative skills are not as easy to evaluate and so must be assessed in different ways, e.g., your ability to synthesize or summarize results or see the big picture could be assessed by using a rubric that captures the degree to which you put your findings into context. In the remainder of this chapter, we'll first review some of the key quantitative types of skills that are important (and which are at the core of life cycle studies) and then discuss how to mix qualitative and quantitative skills to produce quality LCA work. One of the most important skills is identifying appropriate data to use in support of analyses.
Working with Data Sources Most data are quantitative, i.e., you are provided a spreadsheet of numerical values for some process or activity and you manipulate the data in some quantitative way (e.g., by finding an average, sorting it, etc.). But data can also be qualitative – you may have a description of a process that discusses how a machine assembles inputs, or you may generally know that a machine is relatively old (without knowing an exact date of manufacture). Being able to work with both types of data is useful when performing LCA. As we seek to build a framework for building quantitative models, inevitably one of the challenges will be to find data (and in LCA, finding appropriate data will be a recurring challenge). But more generally we need to build skills in acquiring and documenting the data we find. As we undertake this task, it is important to understand the difference between primary and secondary sources. A primary source of data comes directly from the entity collecting the data and/or analyzing it to find a result. It is thus generally a definitive source of information, which is why you want to find it. A secondary source is one that cites or reuses the information from the primary source. Such sources may use the information in different ways inconsistent with the primary source's stated goals and intentions, and may incorporate biases. It is thus good practice to seek the primary source of the information and not merely a source that makes use of it. Finding (and reading, if necessary) the primary Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
34
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
source also allows you to gain appreciation for the full context that reported the result. This context may include the sponsor of the study, any time or data constraints, and perhaps caveats on when or how the result should be used. In today's Internet search-enabled world, secondary sources are far more prominent. Search engines are optimized to find often linked to and repeated sources, not necessarily primary sources. As an example, the total annual emissions of greenhouse gases in the US are prepared in a study and reported every year by the US Environmental Protection Agency (EPA). The EPA spends a substantial amount of time - with the assistance of government contractors - each year refining the methods and estimates of emissions to be reported. Given their official capacity and the work done, the reporting of this annual estimate (i.e., "the number") is a primary source. This number, which is always for a prior period and is therefore a few years old, gets noticed and reported on by hundreds of journalists and media outlets, and thousands of web pages or links are created as a result. A web search for "annual US GHG emissions" turns up millions of hits. The top few may be links to the latest EPA report or the website that links to the report. The web search may also point to archived EPA reports of historical emissions published in previous years. But there is only a single primary source for each year's emissions estimate – the original study by EPA. The vast majority of the web search results lead to studies "re-reporting" the original published EPA value. It is possible that the primary source is not even in the top 10 of the ordered websites of a web search. This phenomenon is important because when looking for data sources, it is easy to find secondary sources, but there is often a bit of additional work needed to track backwards to find and cite the primary source. It is the primary source that one should use in any model building and documentation efforts (even if you found it via finding a secondary source first). A primary source of data is typically from a credible source, and citing "US EPA" instead of "USA Today" certainly improves the credibility of your work. Backtracking to find these primary sources can be tricky because often newspaper articles will simply write "EPA today reported that the 2011 emissions of greenhouse gases in the United States were 7 billon metric tons" without giving full references within the article. Blogs on the other hand tend to be slightly more academic in nature and may cite sources or link to websites (and of course they still might link to a secondary source). If your secondary sources do not link to the EPA report directly, you need to do some additional searching to try to find the primary source. It will help your search that you know the numerical value that will be found in the primary source (but of course you should confirm that the secondary source used the correct and most up to date value). With some practice you will become adept at quickly locating primary sources. The relevant contextual information that may appear in the official EPA source includes things like how the estimate was created, what year it is for, what the year-over-year change was, and which activities were included. All of that contextual information is important. A more frequently reported estimate of US GHG emissions (only a few months old when reported) comes from the US Department of Energy, but only includes fossil fuel Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
35
combustion activities, which are far easier to track because power plants annually report their fuel use to the Department. If you were looking for a total inventory of US greenhouse gas emissions, the EPA source is the definitive source. After finding appropriate data, it is essential to reference the source adequately. It is assumed that you are generally familiar with the basics of creating footnote or endnote references or bibliographical references to be used in a report. You can see short bibliographical reference lists at the end of each of the chapters of this textbook. Primary data sources should be completely referenced, just as if you were excerpting something from a book. That means you need to give the full bibliographic reference as well as point to the place inside the source where you found the data. That might be the page number if you borrow something from the middle of a report, or a specific Table or Figure within a government report. For example, if you needed data about the electricity consumption per square foot for a commercial building, the US Department of Energy's Energy Information Administration 2003 Commercial Buildings Energy Consumption Survey (CBECS) suggests the answer is 14.1 kWh/square foot (for non-mall buildings). The summary reports for this survey are hundreds of pages in length. The specific value of 14.1 kWh/sf is found (on page 1) of Table C14. By referencing this source specifically, you allow others to reproduce your study quickly. You also are allowing others (who may stumble upon your own work when looking for something else) to use your work as a secondary source. The full primary source reference for the CBECS data point could look like this: US Dept. of Energy, 2003 Commercial Buildings Energy Consumption Survey (CBECS), Table C14. "Electricity Consumption and Expenditure Intensities for Non-Mall Buildings, 2003", 2006, http://www.eia.gov/consumption/commercial/data/2003/pdf/c14.pdf, last accessed July 5, 2013.
What is unfortunately common is to see very loose or abbreviated referencing of data sources, such as "DOE CBECS". Such casual referencing is problematic for many reasons. The DOE has done at least four CBECS surveys, roughly four years apart, since 1992, for which they have made the results available online. If one finds a single data point on the Energy Information Administration's website and uses it in a study, that data point might come from any of these four surveys, which span 20 years of time, from any of the thousands of pages of data summaries. With only a reference to "CBECS", one would have no way of knowing how recent, relevant, or useful is your data point. Beyond the examples above, one might be interested in the population of a country, the average salary of workers, or other fundamental data. You are likely (and encouraged) to find and report multiple primary sources. These multiple sources could come from independent agencies or groups who sought to find answers to the same or very similar questions. A rule of thumb is to seek and report results from at least three such sources if possible. In the best case, the primary sources yield the same (or nearly equal) data. In reality, they will likely disagree to a small or large extent. There may be very easy explanations for why they differ, such as using different assumptions or methods. By noting Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
36
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
and representing that you have found multiple data points, and summarizing reasons for the differences, you gain the ability to judge whether to simply use an assumption based on the three sources, or need to use a range or an average. The practice of seeking multiple sources will sometimes even uncover errors in original studies or data reports, or at the least make you realize that a primary source found is not appropriate to use in your own work given differences in how the result was made. "When we look up a number in more than one place, we may get several different answers, and then we have to exercise care. The moral is not that several answers are worse than one, but that whatever answer we get from one source might be different if we got it from another source. Thus we should consider both the definition of the number and its unreliability." -- Mosteller (1977) If you end up with several values, it may be useful to summarize them in a table. If you had been trying to find the total US greenhouse gas emissions as above, you might summarize it like in Figure 2-1. Additional rows could be added for other primary or secondary sources. A benefit of organizing these summary tables is that it allows the audience to better understand your underlying data sources as well as potential issues with applying them. Value (million metric tons CO2)
Source
Type of Source
Comments
6,702
US EPA, Inventory Of U.S. Greenhouse Gas Emissions And Sinks: 1990-2011
Primary
Value is for 2011.
5,471
US DOE, U.S. Energy-Related Carbon Dioxide Emissions, 2011
Primary
Value is for 2011. Only counts energy-related emissions.
6,702.3
Environmental News Network, US Greenhouse Gas Emissions are Down, April 21, 2013.
Secondary
Specifically references EPA.
Figure 2-1: Summary of Sources for US Greenhouse Emissions
A final note about seeking data sources pertains to the use of statistical abstracts. Such references exist for many countries, states and organizations like universities. These abstracts are valuable reference materials that are loaded with many types of summary data. They are typically organized by sections or chapter of related data. For example, the Statistical Abstract of the United States (2011) has sections on agriculture, manufacturing, energy, and transportation (all of which are potentially relevant for LCA studies). Each of the sections contains a series of data tables. The Agriculture section has, amongst other interesting facts, data on the number of farms and area of cropland planted for many types of crops. The Table (Number 823) of farms and cropland has a footnote showing the primary source of the data, in this case the 2007 Census of Agriculture. Such abstracts may also have other footnotes that need to be considered when using them as a source, such as noting the units of presentation (e.g., dollar values in millions), or the boundaries considered.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
37
This example is intended to reinforce two important facts of using statistical abstracts. First, it is important to realize that while generally statistical abstracts may be a convenient "go to" reference source, they are not a primary source. The best practice is to use statistical abstracts as links to primary sources – and then go read the primary source. Re-publication of data sometimes leads to errors, or omissions of important footnotes like units or assumptions used. Second, despite the "2012" in the title of the abstract, it is generally not true that all data within is from the year 2012. Generally though, any data contained within is the most recent available. Abstracts for states and other organizations are organized in similar ways and with similar source referencing. Finally, it is worth noting that in the age of Google, statistical abstracts are no longer the valuable key reference sources that they once were. Nonetheless, they are still a great first "one stop" place to look for information, especially if doing research in a library with an actual book.
Accuracy vs. Precision We seek primary sources (and multiple primary sources) because we want to get credible values to use. Depending on the kind of model we are building, we may simply need a reasonable estimate, or we may need a value as exact as possible. This raises the issue of whether we are seeking accuracy or precision in our search for sources and/or our model building efforts. While the words accuracy and precision are perhaps synonyms to lay audiences, the "accuracy versus precision" dialogue is a long-standing one in science. We are often asked to clarify our goals more clearly in terms of what we are seeking – accuracy or precision (or both)—in our system of measurement. The accuracy of a measurement system is the degree to which measurements made are close to the actual value (of course, as measured by some always correct system or entity). The precision of a measurement system is the degree to which repeated measurements give the same results. Precision is thus also referred to as repeatability or reproducibility. In addition to physical measurement systems, these features are relevant to computational methods on data, such as statistical transformations, Microsoft ® Excel ®1 models, etc. Figure 2-2 summarizes the concepts of accuracy and precision within the context of aiming at a target, but could be analogously used to consider our measurements of a value.
1
Microsoft and Excel are registered trademarks of Microsoft Corporation. In the rest of the book, just "Microsoft Excel" will be used.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
38
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
Accurate
Inaccurate
Precise
Imprecise
Figure 2-2: Comparison of accuracy and precision. Source: NOAA 2012
Systems can thus be accurate but not precise, precise but not accurate, or neither, or both. Systems are considered valid when they are both accurate and precise. With respect to our CBECS example above, the survey used could provide an inaccurate (but precise) result if mall and non-mall buildings are included in an estimate of retail building effects. It could produce an imprecise (but accurate) result if samples from different geographical regions do not align with the actual geographical mix of buildings. Performing mathematical or statistical operations (e.g., averages) on imprecise values may not lead to a value that is credible to use in your work. When a measurement system is popular and needs to be known to be accurate and precise, typically a standard is made for all parties to agree upon how to test and formalize the features of the system (e.g., how to perform the test many times and assess the results).
Uncertainty and Variability As we seek to find multiple sources for our data needs, inevitably we will come across situations where the data do not agree to the extent that we would hope. This will lead us to situations of dealing with uncertainty and variability of our data. While the ways in which we Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
39
work with and model uncertain and variable data are similar, we first separately define each condition. These simple definitions will be used here, with further detail in later chapters as needed. Variability exists because of heterogeneity or diversity of a system. It may be, for example, that the energy used to manufacture an item differs between the morning and afternoon shift in a factory. Uncertainty exists because we either are unable to precisely measure a value, or lack full information, or are ignorant of some state. It is possible that if we did additional research or improved our measurement methods, we could reduce the uncertainty and narrow in on a most likely outcome or value. Variability, on the other hand, is not likely to be reducible – it may exist purely due to natural or other factors outside of our control.
Management of Significant Figures Beyond thinking that we have created a way of accurately and precisely measuring a quantity, we also want to ensure that we appropriately represent the result of our measurement. Many of us learned of the importance of managing the use of significant figures (or digits) in middle school. Two important lessons learned that merit mention in this context relate to leading and trailing zeros and reporting the results of mathematical operations. Remember that trailing zeros are always significant and indicate the level of precision of the measurement. Leading zeros (after a decimal point), however, are not significant. This means that a value like 0.00037 still has only two significant digits because scientific notation would refer to it as 3.7E-04 and the first component of the notation (3.7) represents all of the significant digits. Also take care not to introduce extra digits in the process of adding, subtracting, multiplying, or dividing significant figures. That means, for example, not perpetuating a result from a calculator or spreadsheet that multiplies two 2-digit numbers and reporting 4 digits. The management of significant digits means reporting only 2 digits from such a result, even if it means rounding off to achieve the second digit. Recall that the basis for such directives is that our measurement devices are calibrated to a fixed number of digits. A graduated cylinder used to measure liquids in a laboratory usually shows values in 1 ml increments (e.g., 10, 11, or 12 ml). We then attempt to estimate the level of the liquid to the nearest 10th of an increment. As an example, when measuring a liquid we would report values like 10.2 ml – with three significant figures - which expresses our subjective view that the height of the liquid is approximately 2/10ths of the way between the 10 and 11 ml lines. Given our faith in the measurement system, we are quite sure of the first 2 digits to the left of the decimal point (e.g., 10), and less sure of the digit to the right of the decimal point as it is our own estimate given the uncertainty of the measurement device, and thus is the least significant figure.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
40
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
When counting significant figures, think about scientific notation. •
All nonzero digits are significant
•
Zeroes between nonzero digits are significant
•
Trailing zeroes that are also to the right of a decimal point in a number are significant
Digits do not increase with calculations. •
When adding and subtracting, the result is rounded off to have the same number of decimal places as the measurement with the least decimal places.
•
When multiplying and dividing, the result is rounded off to have the same number of significant figures as in the component with the least number of significant figures. Figure 2-3: Summary of Rules of Thumb for Managing Significant Figures
Inevitably, our raw measurements will be used in additional calculations. For example our graduated cylinder observation of volume can then be used to find mass, molarity, etc. If those subsequent calculations are presented with five significant figures (since that's what the calculator output reads), such results overstate the accuracy of the calculations based on the original data, and by implication understate their uncertainty. Figure 2-3 summarizes rules for managing significant figures. We will circle back to discussing data acquisition in the context of life cycle assessment in a later chapter. Going back to our CBECS example, the published average electricity use of 14.1 kWh/square foot is a ratio with three significant figures. That published value represents an average of many buildings included in the survey. The buildings would give a wide range of electricity consumption values in the numerator. However, the three significant figures reported are likely because some relatively small buildings led to a value with only three significant figures. If not concerned about managing significant figures, DOE could have reported a value of 14.1234 kWh/sf. This result would have led to negligible modeling errors, but would have added extraneous digits for no reason. One of the main motivations for managing the number of significant digits is in considering how to present model results of an LCA. As many LCAs are done in support of a comparison of two alternatives, an inevitable task is comparing the quantitative results of the two. For such a comparison to be valid, it is important not to report more significant figures in the result than were present in the initial measured values. A common output of an LCA, given the need to maintain assumptions between the modeling of various alternatives, is that the alternatives would have very similar effects across at least one metric. Consider a Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
41
hypothetical result where the energy use Alternative A is found to be 7.56 kWh and for Alternative B is 7.57 kWh. Would one really expect a decision maker to prioritize one over the other because of a 0.01 kWh reduction in energy use, which is a 0.1% difference, or a savings worth less than 0.1 cents at current US electricity prices? Aside from the fact that it is a trivial amount, it is likely outside of the range of measurement available. In LCA, we do not have the same "measurement device" issues used to motivate a middle school introduction to significant digits. Instead, the challenge lies in understanding the uncertainty of the "measurement process" or the "method" used to generate the numerical values needed for a study. So while we do not worry about the number of digits on a graduated cylinder, we need to consider that the methods are uncertain. Thus you will see many studies create internally consistent rules that define "significance" in the context of comparing alternatives. These rules of thumb are rooted in the types of significance testing done for statistical analyses, but which are generally not usable given the small number of data points used in such studies. Often used rules will suggest that the uncertainty of values such as energy and carbon emissions are at least 20%, with even higher percentages for other metrics. When implemented, that means our values for Alternatives A and B would need to be at least 20% different for one to consider the difference as being meaningful or significant. The comparative results would be "inconclusive" for energy use using such a study's rules of thumb. In the absence of study rules of thumb for significance, what would we recommend? Returning to our discussion above an LCA practitioner should seek to minimize the use of significant digits. We generally recommend reporting no more than 3 digits (and, ideally, only 2 given the potential for a 20% consideration of uncertainty). In the example of the previous paragraph that would mean comparing two alternatives with identical energy use – i.e., 7.6 kWh. The comparison would thus have the appropriate outcome – that the alternatives are equivalent. Ranges If you are able to find multiple primary sources, it is typically more useful to fully represent all information you have than to simply choose a single point as a representation. If you use a single value, you are making a conscious statement that one particular value is the most correct and the others are irrelevant. In reality, you may have more than one value being potentially correct or useful, e.g., because you found multiple credible primary sources. By using ranges, you can represent multiple data points, or a small set or subset of data. While individual data points are represented by a single number (e.g., 5), a range is created by encapsulating your multiple data points, and may be represented with parentheses, such as (0,5) or (0-5). A range represented as such could mean "a number somewhere from 0 to 5". The values used as the limits of a range may be created with various methods. Often used parameters of ranges are the minimum and maximum values of a dataset. In an energy Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
42
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
technology domain, you might want to represent a range of efficiency values of an electricity generation technology, such as (30%, 50%). If you have a large amount of data, then it might be more suitable to use the 5th and 95th percentile values as your stated range. While this may sound like an underhanded way of ignoring data, it can be appropriate to represent the underlying data if you believe some of the values are not representative or are overly extreme. Using the same technology efficiency example, you may find data on efficiencies of all coal-fired or gas-fired power plants in the US, and decide that the lowest efficiency values (in the teens) are far outside of the usual practice because they represent the efficiencies of plants that are used very infrequently or are using extremely out of date technology. There could be similarly problematic values at the high end of the full range of data if the efficiency for a newer plant has be estimated by the manufacturer, but the plant has not been in service long enough to measure the true efficiency. Using these percentile limits in the ranges can help to slightly constrain the potential values in the data. Ranges can be used to represent upper and lower bounds. Bounding analysis is useful when you do not actually have data but have a firm (perhaps even qualitative) belief that a value is unlikely to be beyond a certain quantity. A bounding analysis of energy technology might lead you to conclude that given other technologies, it is unlikely that an efficiency value could be less than 20% or greater than 90%. Using a range in this way constrains your data to values that you feel are the most realistic or representative. Finally, ranges can be used to represent best or worst case scenarios. The limit values chosen for the stated ranges are thus subjectively chosen although perhaps by building on some range limits derived from some of the other methods above. For example, you might decide that a "best case value" for efficiency is 100% and "worst case" value is 0% (despite potentially being unrealistic). Best and worst case limits are typically most useful when modeling economic parameters, e.g., representing the highest salary you might need to pay a worker or the lowest interest rate you might be able to get for a bank loan. Best and worst cases, by their nature, are themselves unlikely. It is not very probable that all of your worst parameters will occur, just as it is improbable that all best parameters will occur. Thus you might consider the best-worst ranges as a type of bounding analysis. Another way of implementing a range is by using statistical information from the data, such as the variance, standard error, or standard deviation. You may recall from past statistics courses that the variance is the average of the squared differences from the mean, and the standard deviation (how much you expect one of the data points to be different from the mean) is the square root of the variance. The standard error (the "precision of the average", or how much you might expect a mean of a subsample to be different from the mean of the entire sample) is the standard deviation divided by the square root of the number of samples of data. Either of these values if available can be used to construct confidence intervals to give some sense of the range of the underlying data. A related statistical metric is the relative Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
43
standard error (RSE), which is defined as the standard error divided by the mean and multiplied by 100, which gives a percentage-like range variable. Another way to think about the RSE is as a metric representing the standard error relative to the mean estimate on a scale from zero to 100. As the RSE increases, we would tend to believe our mean estimate is less precise when referring to the true value in the population being studied. Of course when found in this way, the range will be symmetric around the mean. A 95-percent confidence range is calculated for a given survey (mean) estimate from the RSE via a three-step process. First, divide the RSE by 100 and multiply by the mean presented by the survey to get the standard error. Second, multiply the standard error by 1.96 to generate the confidence error (recall from statistics that the value 1.96 comes from the shape and structure of an arbitrarily assumed normal distribution and its 0.975 quantile). Finally, add and subtract the confidence error to the survey estimate from the second step to create the 95% confidence range. Note that a 95% confidence range is not the same as a 5th95th percentile range. A 95% confidence range represents the middle 95% of a normal distribution, or a 2.5th-97.5th percentile range, leaving only 2.5% of the distribution at the top and bottom. A 5th-95th percentile range leaves 5% on the top and bottom. Example 2-1: Question: Develop a 95% confidence interval around the 2003 CBECS estimate of US commercial building electricity consumption per square foot (14.1 kWh/sf) given the stated RSE (3.2). Answer: Given the RSE definition provided above, the standard error is (3.2/100)*14.1 = 0.45 kWh/square foot, and the confidence error is 0.88 kWh/square foot. Thus, the 95th percentile confidence interval would be 14.1 +- 0.88 kWh/square foot. Note that this range seems to contradict the 25th-75th percentile range of 3.6-17.1 provided directly by the survey (it is a much tighter distribution around the mean of 14.1). However the confidence interval is representing something different –how confident we should be that the average electricity use of all of the buildings surveyed (as if we re-did the survey multiple times) would be approximately 14.1, not trying to represent the underlying range of actual electricity use of the buildings! If you are making a model that needs to represent the range of electricity use, the provided 25th-75th percentile values are likely much more useful. Source: US Dept. of Energy, 2003 Commercial Buildings Energy Consumption Survey (CBECS), RSE Tables for Consumption and Expenditures for Non-Mall Buildings, http://www.eia.gov/consumption/commercial/data/2003/pdf/c1rse-c38rse.pdf, page 94.
A main benefit of using ranges instead of single point estimates is that the range boundaries can be used throughout a model. For example one can propagate the minimum values of ranges through all calculations to ensure a minimum potential result, or the maximum values to get a maximum potential result. One word of caution when using ranges as suggested Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
44
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
above is to maintain the qualitative sense of the range boundaries. If you are envisioning a best-worst kind of model, then the "minimum" value chosen in your range boundary should consistently represent the worst case possible. This is important because you may have a parameter in your model that is very high but represents a worst case, for example, a loss factor from a production process. In a best-worst range type of model, you want to have all of your best and worst values ordered in this way so that your final output range represents the worst and best case outputs given all of the worst possible variable values, and all possible best values.
Units and Unit Conversions In quantitative analysis, it is critical to maintain awareness of the unit of analysis. That might mean noting grams or kilograms, short tons or metric tons (a.k.a. tonnes). While conversions can be simple, such as multiplying or dividing by 1000 in SI units, this is an area where many errors occur, especially when done manually. It is easy to make errors by not thinking out the impacts and accidentally multiply instead of divide, or vice versa. Thus a good practice is to ask yourself whether the resulting conversion makes sense. This is also known as applying a reasonableness test, or a sanity check. Some refer to it as a "sniff test", suggesting that you might be able to check whether the number smells right. To convert from kilograms to grams, we multiply by 1000 - the result should be bigger because we should have many more grams than we do kilograms. If we accidentally divide by 1000 (an error the authors themselves have made many times in the rush of getting a quick answer) the number gets smaller and the sniff test would tell us it must be an error. In the context of finding sources for data, simple changes of unit scales, such as grams to kilograms, don't require extensive referencing. When performing simple unit conversions like this, it is typical that instead of seeking external data sources you would simply document the step used (e.g., you would state that you "converted to kilograms"). There are however more complex unit conversions that change the entire basis of comparison (not just kg to g). If you are changing more than just the scale, such as switching from British Thermal Units (BTU) to megajoules (MJ), this is referred to as performing physical or energy unit conversions. A unit conversion factor is just a mathematical relation between the same underlying phenomena but with different measurement scales, such as English and SI (metric) units. For example you may find a data source expressing emissions in pounds but need to report it in kilograms (or metric tons). This type of conversion does not require much documentation either, e.g., you could write that you "assumed 2.2 pounds per kilogram". Such conversions still need to be done and managed correctly. In 1999, NASA famously lost the Mars Climate Orbiter after a ninemonth mission when navigation engineers gave commands in metric units to the spacecraft,
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
45
whose software engineers had programmed to it to operate with English units, causing the vehicle to overshoot the planet. If you do not know the conversion factors needed, then you will need to search for sources of your conversion factors using the same methods discussed above. If you were to do a search for unit conversions with the many tools and handbooks available, you will certainly find slightly different values in various sources, although most of these differences are simply due to rounding off or reducing digits. One source may say 2.2 pounds per kg, another 2.20462, and yet another 2.205. Practically speaking any of these unit conversions will lead to the same result (they would be at most 0.2% apart) and quantity aside, in the big picture they are all the same number, i.e., 2.2. The existence of multiple conversion factors is the reason why to state the one you used. Without stating the actual conversion factor used, someone else may not be able to reproduce your study results (or may assume an alternative unit conversion factor and not understand why your results are different). Given the scientific and engineering basis of unit conversion factors, you do not typically need to cite specific 'sources' for them, just the numbers used. As you build your models, your calculations will become increasingly complex. You can double-check your calculations by tracing your units. As a simple example, assume you have tugboat transit time data for a stretch of river between two locks. You know the transit time in minutes (140), and the distance between locks in miles (6.1). Equation 2-1 shows how to calculate the tugboat speed in kilometers/hour, which could later allow you to calculate power load and emissions rates. Getting the speed units wrong, despite being a trivial conversion, will have disastrous effects on your overall model results. Tracing the units confirms that you have used all of the necessary conversion factors, and used them appropriately and in the right order. !"
!.! !"#$% !"#$""% !"#
%$𝑥 !!"# = !"# !"#$%&' !"#$%&! !"#$ ×
! !"#$%&'&( !.!"# !"#$
×
!" !"#$%&' ! !!"#
= 4.2 𝑘𝑚/ℎ𝑟
(2-1)
We end this section by briefly discussing the need to manage units in calculations. Note that when solving equation 2-1, your calculator would suggest that the speed is actually 4.2098 km/hr, a level of accuracy that would be impossible to achieve (and silly to present). The reason to document the units is so that when we are using them in calculations that we do the mathematical operations correctly, i.e., adding kg to kg, not kg to g. The graphic made in 1976 for The New Yorker (presented at the beginning of this chapter) is a reminder of this. Considerations for Energy Unit Conversions Sometimes changing units involves more than applying a single conversion factor. You may recall from a physics course that energy is a measure of the amount of work done or generated (typical units are joules, BTU, or kilowatt-hours). On the other hand, power is the rate at which energy is used or transformed (typical units are watts or joules/second). Unit conversions in the energy domain, e.g., between BTU and kWh, can be more complicated Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
46
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
than they appear. Unlike physical unit conversions that are just different ways of measuring or weighing, there can be different interpretations or contexts about use of energy sources. The quantity of energy used locally for a specific task is typically referred to as site energy, such as the electricity we use for recharging a laptop or mobile phone. However, site uses of energy typically lead to an even greater use of energy elsewhere, such as at a power plant. The energy conversion performance of a coal-fired power plant and losses from the power grid means that for every 3 units of energy in the coal burned at a plant we can use only about 1 unit of energy at our electrical outlet. That amount of original energy needed, such as at a power plant, is referred to as primary or source energy. A conversion between English and metric units (e.g., BTU and MJ) for primary energy is straightforward because BTU and MJ both represent energy content (e.g., the quantity of BTUs in a gallon of gasoline). However, our assessment of energy use should certainly include consideration for the inefficiencies in the conversion processes of our chosen methods, as discussed below. A related concept that is more specific to the modeling of fuel use pertains to the heating value of the fuel, which refers to the energy released from combusting the fuel, with units such as kJ/kg or BTU/lb. Of particular importance is which heating value— the lower or higher heating value—is used. The difference between the lower heating value (LHV) and the higher heating value (HHV) is whether the energy used to vaporize liquid water in the combustion process is included or not. While the difference between HHV and LHV is typically only about 10%, you can often argue that the HHV is a more inclusive metric, consistent with the system and life cycle perspectives relevant to LCA. Regardless, this is yet another example of why all relevant assumptions need to be explicit in energy analysis. We may also need to make assumptions about the conversion process. A difficulty in converting from BTU to kWh can depend on whether an intermediate thermodynamic process is involved. For example, many engineering reference manuals suggest the conversion factor "1 kWh is equal to 3,413 BTU". But this assumes a perfect conversion with no losses and thus is pure energy equivalence. The likely context behind such a conversion is an energy process where a fuel input is used to generate a quantity of electricity, known as a heat rate. However, in describing the conversion of fossil energy from a fuel in a power plant, the heat rate for a typical coal–fired plant may be 10,000 BTU (of coal input) to generate 1 kWh (electricity output). The reason that power plant heat rate is so much larger than the pure engineering conversion factor is that converting coal to electricity requires burning the coal and then using the produced heat to turn water into steam, and then using the pressurized steam to spin a turbine, which is connected to a generator. There are losses throughout all of these steps, and thus far more than the 3,413 BTU are needed to make 1 kWh. The overall difference between the 3,413 BTU and the 10,000 BTU is expressed as a ratio representing the efficiency of the power plant, which is 3,413 BTU per kWh / 10,000 BTU per kWh, or 34%. While this may sound like a convenient example with rounded off numbers, it is quite common for a traditional coalfired power plant to have this approximate efficiency. Natural gas plants can have Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
47
efficiencies of about 50%. While not comprised of burning fuels, solar PV cells are about 10% efficient. It may be surprising to you to learn that in the 21st century we rely on such inefficient methods to make our electricity! The important point of this example is that in such contexts, you cannot use or assume the basic BTU to kWh conversion factor. You also need to know either the heat rate or efficiency. Careful management of units and the conversion process is generally needed when working with fuels. Fuels can be inputs to a variety of processes, not just making electricity. For example, when used to make heat in a building, natural gas with an energy content of 40 MJ/m3 may be used in a furnace that is 72% efficient to produce 29 MJ of heat/m3. Overall, while the same documentation guidelines apply, in these cases it is even more important to document all conversion factors and assumptions used, as other authors might choose different conversions or efficiencies as a result of personal or domain-specific knowledge. Many external references detail the various conversions available and needed to work in the energy domain. As a final note, "converting" between energy and power is not appropriate for LCA analyses, but is often done to provide examples or benchmarks to lay audiences. For example, 300 kWh of electricity may be referred to as the quantity such that a 30-Watt light bulb is used for 10,000 hours (a bit more than a complete year).
Use of Emissions or Resource Use Factors Many production processes have releases to the environment, such as the various types of pollutants mentioned in Chapter 1. For many analyses, an emissions factor is needed to represent the units of emissions released as a function of some level of activity. We will discuss specific data sources for emissions factors in later chapters, but most emissions factors can be found using the same type of methods needed to find primary data sources or unit conversions. Emissions factors may be sourced from government databases or reports (e.g., the US EPA's AP-42 database) or technical specifications of a piece of equipment and as such should be explicitly cited if used. Given the potential for discrepancies in emissions factors, you should look for multiple sources of emissions factors and represent them with a range of values. Beyond finding sources, knowledge of existing physical quantities and chemical processes can be used to find emissions factors. Equation 2-2 can be used to generate a CO2 emissions factor for a combusted fuel based on its carbon content (as found by laboratory experiments) and an assumed oxidation rate of carbon (the percent of carbon that is converted into CO2 during combustion): CO2 emissions from burning fuel (kg / MMBTU) = Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
48
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
Carbon Content Coefficient (kg C / MMBTU) * Fraction Oxidized * (44/12)
(2-2)
where the 44/12 parameter in Equation 2-2 is the ratio of the molecular weight of CO2 to the molecular weight of carbon, and MMBTU stands for million BTU. If we were doing a preliminary analysis and only needed an approximate emissions factor, we could assume the fraction oxidized is 1 (100% or complete oxidation). In reality, the fraction oxidized could be closer to 0.9 than 1 for some fuels. For an example of coal with a carbon content of 25 kg C per MMBTU, and assuming perfect oxidation, the emissions factor would be 92 kg CO2 / MMBTU. Various emissions factors can be developed through similar methods by knowing contents of elements (such as for SO2), however, other emissions factors are a function of the choice of combustion and emissions control technologies used (such as for nitrogen oxide or particulate matter emissions) In LCA, we will also discover resource use factors, such as material input factors, that are used and developed in similar ways as emissions factors. The main difference is that resource use factors are made as a function of input rather than output.
Estimations vs. Calculations "It is the mark of an instructed mind to rest satisfied with the degree of precision which the nature of the subject permits and not to seek an exactness where only an approximation of the truth is possible." - Aristotle "God created the world in 4004 BC on the 23rd of October." – Archbishop James Ussher of Ireland, The Annals of the Old Testament, in 1650 AD ".. at nine o'clock in the morning." –John Lightfoot of Cambridge, in 1644 AD Most courses and textbooks teach you how to apply known equations and methods to derive answers that are exact and consistent (and selfishly, easy to grade). These generally are activities oriented towards teaching calculation methods. Similarly, methods as described above can assist in finding and documenting data needed to support calculations. A simple example of a calculation method is applying a conversion factor (e.g., pounds to kilograms). More complex calculation methods may involve solving for distance traveled given an equation relating distance to velocity and time. As solving LCA problems seldom requires you to learn a completely new calculation method, we presume you have had sufficient exposure to doing calculations.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
49
But what if all else fails and you cannot find a primary source or a needed unit conversion? What if we are unable to locate an appropriate calculation method? An alternative method must be found that assists in finding a quantitative answer, and which preserves a scientific method, but is flexible enough to be useful without all needed data or equations. Such an alternative could involve conducting a survey of experts or non-experts, or guessing the answer. It is this idea of "guessing" the answer that is the topic of this section. Here we assume that there is a time-critical aspect to the situation, and that you require a relative guess in lieu of investing a substantial more amount of time looking for a source, conducting a complete survey, etc. Estimation methods use a mix of qualitative and quantitative approaches to yield a "ballpark", or "back of the envelope", or order of magnitude assessment of an answer. These are not to be confused with the types of estimation done in statistical analyses that are purely quantitative in nature (e.g., estimating parameters of a regression equation). With estimation methods, we seek an approximately right answer that is adequate for our purpose – thus the concept that we are merely looking for an order of magnitude result, or one that we could do in the limited space of an envelope. The quotations at the beginning of this section are given here to represent the spectrum of the exact versus approximate methods being contrasted. Estimation methods are sometimes referred to as educated guessing or opportunistic problem solving. As you will see, the intent is to create educated guesses that do not sound like guesses. The references at the end of this chapter from Koomey, Harte (both focused on environmental issues), Weinstein and Adam, and Mahajan are popular book-length resources and are highly recommended reading if you find this topic interesting. Estimation methods succeed by using a structured approach of creating and documenting assumptions relevant to the question rather than simply plugging in known values into an equation. In this context, you need to adjust your expectations (and those of your audience) to reflect the fact that you are not seeking a calculated value. You may be simply trying to correctly represent the sign and/or the order of magnitude of the result. "Getting the sign right" is fairly straightforward but still often difficult. Approximating the order of magnitude means generating a value where only one significant figure is needed and the "power of 10" behind it gives a sense of how large or small it is (i.e., is the value in the millions or billions?). If you come from a "hard science" discipline such as chemistry or physics, the thought of generating an answer without an equation may sound like blasphemy. But recall the premise of estimation methods – that you do not have access to, are unable to acquire, or unfamiliar with the data and equations needed for a calculated result. We are not suggesting you need to use estimations to find the force of gravity, the number of molecules per mole, etc. Many students may have encountered these methods in the form of classroom exercises known as "Fermi Problems". Furthermore, such estimation challenges are being used more and more frequently as on-the-spot job interview questions for those entering technical fields. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
50
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
While the mainstream references mentioned above give many examples of applying estimation methods, other references are useful for learning the underlying methods. Mosteller (1977) lists several building block-type methods that can be used and intermixed to assist in performing estimation. You are likely familiar with many or all of them, but may not have considered their value in improving you estimation skills: •
Rules of thumb – Even a relative novice has various numbers and knowledge in hand that can help to estimate other values. For example, if performing a financial analysis it is useful to know the "rule of 72" that defines when an invested amount will double in value. Likewise, you may know of various multipliers used in a domain to account for waste, excess, or other issues (e.g., contingency or fudge factors). The popular Moore's Law for increases in integrated circuit densities over time is an example. Any of these can be a useful contributor to a good estimation. Also realize that one person's rule of thumb may be another's conversion factor.
•
Proxies or similarity – Proxy values in estimation are values we know in place of one we do not know. Of course the needed assumption is that the two values are expected to be similar. If we are trying to estimate the total BTU of energy contained in a barrel of diesel fuel, but only had memorized data for gasoline, we could use the BTU/gallon of gasoline as a proxy for diesel fuel (in reality the values are quite close, as might be expected since they are both refined petroleum products). Beyond just straight substitution of values via proxy, we can use similarity methods to reuse datasets from other purposes to help us. For example if we wanted to know estimates of leakage rates for natural gas pipelines in the US, we might use available data from Canada which has similar technologies and environmental protection policies.
•
Small linear models – Even if we do not have a known equation to apply to an estimation, we can create small linear models to help us. If we seek the total emissions of a facility over the course of a year, we could use a small linear model (e.g., of the form y = mx + b) that estimates such a value (y) by multiplying emissions per workday (m) by number of work days (x). In a sense we are creating shortcut equations for our needs. Of course, these small linear models could be even more complicated, for example by having the output of one equation feed into another. In the example above, we could have a separate linear model to first estimate emissions per day (perhaps by multiplying fuel use by some factor). Another way of using such models is to incorporate growth rates, e.g., by having b as some guess of a value in a previous year, and mx the product of an estimated growth rate and the number of years since.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
51
•
Factoring – Factoring is similar to the small linear models mentioned above, except in purely multiplicative terms. Factoring seeks to mimic a chain of unitized conversions (e.g., in writing out all of the unitized numerators and denominators for converting from days in a year to seconds in a year, which looks similar to Equation 2-1). As above, the goal here is to estimate the individual terms and then multiply them together to get the right value with the right units. The factors in the equation used may be comprised of constants, probabilities, or separately modeled values.
•
Bounding – Upper and lower bounds were discussed in the context of creating ranges for analysis purposes, but can also be used in estimations. Here, we can use bounds to help set the appropriate order of magnitude for a portion of the analysis and then use some sort of scaling or adjustment factor to generate a reasonable answer. For example if we were trying to estimate how much electricity we could generate via solar PV panels, using the entire land mass of the world would give us an upper bound of production. We could then scale down such a number by a guess at the fraction of land that is highly urbanized or otherwise not fit for installation.
•
Segmentation or decomposition – In this type of analysis, we break up a single part into multiple but distinct subparts, and then separately estimate a value for each subpart and then report a total. If we were trying to estimate fossil-based carbon dioxide emissions for the US, we could estimate carbon dioxide emissions separately for fossil fueled power plants, transportation, and other industries. Each of these subparts may require its own unique estimation method (e.g., a guess at kg of CO2 per kWh, per vehicle mile traveled, etc.) that are added together to yield the original unknown total emissions of CO2.
•
Triangulation – Using triangulation means that we experiment in parallel with multiple methods to estimate the same value, and then assess whether to use one of the resulting values or to generate an average or other combination of them. Triangulation is especially useful when you are quite uncertain of what you are estimating, or when the methods you are otherwise choosing have many guesses in them. You can then control whether to be satisfied with one of your results, or to use a range. Of course if your various parallel estimates are quite similar you could just choose a consensus value.
While Mosteller summarized these specific building blocks, you should not feel limited by them. Various other kinds of mathematical functions, convolutions, and principles could be brought to bear to aid in your estimation efforts. Beyond these building blocks, you should try to create ranges (since you are estimating unknown quantities) by assuming ranges of constants in your methods or by using ranges created from triangulation. Do not assume that you can never "look up" a value needed within the scope of your estimation. There may be some underlying factor that could greatly help you find the unknown value you seek, such Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
52
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
as the population of a country, the total quantity of energy used, etc. You can use these to help you reach your goal, but be sure to cite your sources for them. It might be useful to avoid using these reference source values while you are first learning how to do estimation, and then incorporate them when you are more experienced. As expressed by several of the building block descriptions, a key part of good estimations is using a "divide and conquer" method. This means you recursively decompose a high-level unknown value as an expression of multiple unknown values and estimate each of them separately. A final recommendation is that you should be creative and also to consider "outside the box" approaches that leverage personal knowledge or experience. That may mean using special rules of thumb or values that you already know, or attempting methods that you have good experience in already. Now that we have reviewed the building blocks, Example 2-2 shows how to apply them in order to create a simple estimate. Example 2-2: Estimating US petroleum consumption per day for transportation Question: Given that the total miles driven for all vehicles in the US is about 3 trillion miles per year, how many gallons of petroleum are used per day in the US for transportation? Answer: If we assume an average fuel economy figure of about 20 miles per gallon we can estimate that 150 billion gallons (3 trillion miles / 20 miles per gallon) of fuel are consumed per year. That is about 400 million gallons per day.
You might also develop estimations to serve a specific purpose of explaining a result to be presented to a general audience. In these cases you might want to find a useful visual or mental reference that the audience has, and place a result in that context. Example 2-3 shows how you might explain a concentration of 1 ppb (1 part per billion). Example 2-3: Envisioning a one part per billion concentration Question:
How many golf balls would it take to encircle the Earth?
Answer: Assume that the diameter of a golf ball is approximately 1.5 inches, and that the circumference of the Earth is about 25,000 miles (roughly 10x the distance to travel coast to coast in the United States). We can convert 25,000 miles to 1.6 x 109 inches. Thus there would be 1.6 x 109 inches / 1.5 inches, or ~1 billion golf balls encircling the Earth. Thus, if trying to explain the magnitude of a 1 part per billion (ppb) concentration, think about there being one red golf ball along the equator that has 1 billion white balls lined up!
Acknowledgment to "Guesstimation" book reference for motivating this example. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
53
Attributes of Good Assumptions One of the key benefits of becoming proficient in estimation is that your skills in documenting the assumptions of your methods will improve. As application of estimation methods requires you to make explicit assumptions about the process used to arrive at your answer, it is worth discussing the attributes of good assumptions. You may have the impression that making assumptions is a bad thing. However, most research has at its core a structured set of assumptions that serve to refine and direct the underlying method. Your assumptions may refer specifically to the answer you are trying to find, as well as the measurement technologies used, the method, or the analysis. You might think of your assumptions as setting the "ground rules" or listing the relevant information that is believed to be true. You should make and write assumptions with the following attributes. 1. Clarify and Simplify - First, realize that the whole point of making an assumption is to help to clarify the analysis (or at the least to rule out special cases or complications). Assumptions ideally also serve to refine and simplify your analysis. It is not useful to have an assumption that makes things harder either for your analysis or for the audience to follow your process. For example, if you were trying to estimate the number of power plants in the US, you might first assume that you are only considering power plants greater than 50 MW in capacity. Or you might assume that you are only considering facilities that generate and sell electricity (which would ignore power plants used by companies to make their own power). By making these assumptions, you are ruling out a potentially significant number of facilities (leading to an undercount of the actual), but you have laid out this fact explicitly at the beginning as opposed to doing it without mention. It is possible that an assumption may be required in order to make any estimate at all. For example, you might need to assume that you are only estimating fossil-based power plants, because you have no idea of the capacities, scale, or processes used in making renewable electricity. 2. Correct, credible and feasible - If it is not obviously true (i.e., you are not stating something that is a well known fact), your audience should read an assumption and feel that it is valid - even if hard to believe or agree with. For example, you should not assume a conversion factor inconsistent with reality, such as there being only four days in a week or 20 hours in a day. 3. Not a shortcut - While assumptions help to narrow down and refine the space in which you are seeking an answer, they should not serve to merely carve out an overly simple path towards a trivial solution. Your audience should not be left with the impression that you ran out of time or interest in finding the answer and that you substituted a good analysis with a convenient analysis. For example, you might assume that you were only counting privately owned power plants. This is a Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
54
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
narrowing of the boundaries of the problem, but does not sound like you are purposely trying to make the problem trivial. 4. Unbiased – Your assumptions should not incorporate a degree of connection to some unrelated factor. For example, in estimating the number of power plants you do not want to rely on a geographical representation associated with the number of facilities that make ethanol, which are highly concentrated in areas where crops like corn grow. Beyond listing them, it is good practice to explicitly write a justification for your assumptions. In the power plant example above, the justification for why you will only count relatively large (> 50 MW) facilities might be "because you believe that the number of plants with smaller capacities is minimal given the demands of the power grid". Since you're looking for an order of magnitude estimate, neglecting part of the solution space should have no practical effect. In the case of assuming only privately owned facilities, the justification might simply specify that you are not estimating all plants, just those that are privately owned. In Example 2-2, the 20 miles per gallon assumed fuel economy is appropriate for passenger vehicles, but not so much for trucks or buses that are pervasive. In that example, it would be useful to state and justify an assumption explicitly, such as "Assuming that most of the miles traveled are in passenger vehicles, which have a fuel economy of 20 miles per gallon, …" Writing out the thought process behind your assumptions helps to develop your professional writing style, and it helps your audience to more comfortably follow and appreciate the analysis you have done. Furthermore, by becoming proficient at writing up the assumptions and process used to support back of the envelope calculations, you become generally proficient at documenting your methods. Hopefully you will leverage these writing skills in other tasks. In the alternative where you do not state all of your assumptions, the readers are left to figure them out themselves, or to create their own assumptions based on your incomplete documentation. Needless to say either of those options raises the possibility that they make bad assumptions about your work.
Validating your Estimates When you have to estimate a quantity, it is important that you attempt to ensure that the value you have estimated makes sense (see the discussion earlier in this chapter about reasonableness tests). Even though you have estimated a quantity that you were unable to find a good citation for originally, you should still be able to validate it by comparing it to other similar values. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
55
As a learning experience, you might try to estimate a value with a known and available number that you know can be found but that you do not already know the answer to (e.g., the number of power plants in the US, or a value that you could look up in a statistical abstract). Doing so helps you to hone your skills with little risk, meaning that you can try various methods and directly observe which assumptions help you arrive at values closest to the "real answer" and track the percentage error in each of your attempts before looking at the real answer. The goals in doing so are explicitly to learn from doing many estimates of various quantities (not just 5 attempts at the same unknown value) and to increasingly understand why your estimates differ from the real answers. You may not be making good assumptions, or you might be systematically always guessing too high or too low. It is not hard to become proficient after you have tried to estimate 5-10 different values on your own. When doing so, try to apply all of the building block methods proposed by Mosteller. Example 2-4: Validating Result found in Example 2-2 In Example 2-2, we quickly estimated that the transportation sector consumes 400 million gallons per day of petroleum. The US Energy Information Administration reports that about 7 billion barrels of crude oil and other petroleum products were consumed in 2011. About 1 billion barrels equivalent was for natural gas liquids not generally used in transportation. That means about 17 million barrels per day (about 850 million gallons per day at about 50 gallons per barrel) was consumed. That is roughly twice as high as our estimate in Example 2-2, but still in the same order of magnitude. Let's think more about the reasons why we were off by a factor of two. First off, we attempted an estimate in one paragraph with two assumptions. The share of passenger vehicles in total miles driven is not 100%, and heavy trucks represent 10% of the miles traveled and about onefourth of fuel consumed (because their fuel economies are approximately 5 mpg, not 20). Considering these deviations our original estimate, while simplistic, was useful. Sources: US DOE, EIA, Annual Petroleum and Other Liquids Consumption Data http://www.eia.gov/dnav/pet/pet_cons_psup_dc_nus_mbbl_a.htm US Department of Transportation, Highway Statistics 2011, Table VM-1.
Beyond validation of your own estimates, you might also want to do a reasonableness test on someone else's value. You will often find numbers presented in newspapers or magazines as well as scholarly journals that you are curious about or fail a quick sniff test. You can use the same estimation methods to validate those numbers. Just because something is published does not necessarily mean it has been extensively error-checked. Mistakes happen Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
56
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
all the time and errata are sometimes (but not always) published to acknowledge them. Example 2-5 shows a validation of values published in mainstream media pertaining to EPA's proposed 2010 smog standard.
Example 2-5: Validating a comparative metric used in a policy discussion Question: Validate the number of tennis balls in the following CBS News excerpt (2010) pertaining to the details of EPA's proposed 2010 smog standard. "The EPA proposal presents a range for the allowable concentration of ground-level ozone, the main ingredient in smog, from 60 parts per billion to 70 parts per billion. That's equivalent to 60 to 70 tennis balls in an Olympic-sized swimming pool full of a billion tennis balls." Answer: Suppose your sniff test fails because you realize a billion tennis balls is a very large number of balls for this pool. A back of the envelope estimate suggests the approximate size of an Olympic pool is 50m x 25m * 2m = 2500 cubic meters. Similarly, assume a tennis ball occupies a 2.5 inch (70 mm or 0.07m) diameter cube so it thus has a volume of 0.00034 m^3. Such a pool holds only about 7 million tennis balls, almost three orders of magnitude less than the 1 billion suggested in the excerpt. Of course, we could further refine our assumptions such that the pool can be uniformly deeper, or that the tennis ball fully occupies that cube (to consider that adjacent tennis balls could fill in some of the voids when stacked) but none would fully account for the several orders of magnitude difference. You cannot put a billion tennis balls in an Olympic-sized pool, thus the intended reference point for the lay audience was erroneous. It is likely an informal reference from the original EPA Fact Sheet was copied badly in the news article (e.g., "60-70 balls in a pool full of balls"). Thanks to Costa Samaras of Carnegie Mellon University for this example.
Now that we have built important general foundations for working with and manipulating data, we turn our attention to several concepts more specific to LCA.
Building Quantitative Models Given all of the principles above, you should now be prepared to build the types of models needed for robust life cycle thinking. These models have inputs and outputs. The inputs are the various parameters, variables, assumptions, etc., and the output is the result of the equation or manipulation performed on the inputs. In a typical model, we have a single set of input values and a single output value. If we have ranges, we might have various sets of
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
57
inputs and multiple output values. Beyond these typical models there are other types of models we might choose to build that are less straightforward. In a breakeven analysis, you solve for the input value that has a prescribed effect on your model. A classic example, and where the name "breakeven" comes from is if you are building a profit or loss model, where your default model may suggest that profits are expected to be positive (i.e., the result is greater than $0). A relevant breakeven analysis may assess the input value (e.g., price of electricity or number of units sold) needed to lead to a no-profit outcome, i.e., a $0 (or negative) value. It is what you need to "break even" or make profit. This is simply back-solving to find the input required to meet the specified conditions of the result. Not all breakeven analyses need to be about monetary values, and do not need to be set against zero. Using the example of Equation 2-1, you could backsolve for the transit time for a tugboat moving at a speed of 5 km/hr. While the math is generally easy for such analyses, common software like Microsoft Excel have built-in tools (Goal Seek) to automate them. Goal Seek is quite comprehensive in that it can solve for a breakeven value across a fairly complicated spreadsheet of values. The final quantitative skill in this chapter is about identifying how robust your results are to changes in the parameters or inputs of your model. In a sensitivity analysis, you consider various inputs individually into the model, and assess the degree to which changing the value of those inputs has meaningful effects on the results (it is called a sensitivity analysis because you are seeing how "sensitive" the output is to changes in the inputs). By meaningful, you are, for example, assessing whether the sign of the result changes from positive to negative, or whether it changes significantly, e.g., by an order of magnitude, etc. If small changes in input values have a big effect on the output, you would say that your output is sensitive. If even large changes in the inputs have modest effect on the output, then the output is not sensitive. If any such results occur across the range of inputs used in the sensitivity analysis, then your qualitative analysis should support that finding by documenting those outcomes. Note that a sensitivity analysis changes each of your inputs independently (i.e., changing one while holding all other inputs constant). You perform a sensitivity analysis on all inputs separately and report when you identify that the output is sensitive to a given input. Again referring to the tugboat example (Equation 2-1) we could model how the speed varies as the time in transit varies over a range of 20 minutes to more than 4 hours. Figure 2-4 shows the result of entering values for transit time in increments of 20 minutes into Equation 2-1. It suggests that the speed is not very sensitive to large transit times, but changes significantly for small transit times. We will show more examples of breakeven and sensitivity analyses in Chapter 3.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
58
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
Figure 2-4: Sensitivity of Tugboat Speed to Transit Time
A Three-step method for Quantitative and Qualitative Assessment We conclude the chapter with suggestions on how to qualitatively and quantitatively answer questions. LCA is about life cycle assessment. While we have not yet demonstrated the method itself, it is important to develop assessment skills. If you are doing quantitative work (as you will need to do to successfully complete an LCA), a general guideline is that you should think of each task as having three parts: (1) A description of the method used to complete the task, (2) The result or output (quantitative or qualitative) of the task, and (3) A critical assessment, validation, or thought related to the result. The amount of time and/or text you develop to document each of these 3 steps varies based on the expectations and complexity of the task (and perhaps within the constraints of the size of a study). In step one, you should describe any assumptions, data sources found, equations needed, or other information required to answer the question. In step two, you state the result, including units if necessary. In step three, you somehow comment on, validate, or otherwise reflect on the answer you found. This is an important step because it allows you to both check your work (see the example about unit conversions above) and to convince the reader that you have not only done good work but have also spent some time thinking about the implication of the result. For example, a simple unit conversion might be documented with the three-step method as follows: "Inputs of plastic were converted from kg to pounds (2.2 lbs. per 1 kg) yielding 100 kg of inputs. This value represents 20% of the mass flows into the system." Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
59
Each of the three expected steps is documented in those 2 sentences: the method (a basic unit conversion), the result (100 kg), and an assessment (20% of the total flows). If this were part of an assignment, you could envision the instructor deciding on how to give credit for each part of the question, e.g., 3 points for the method, 2 points for the result, and 2 points for the assessment. Such a rubric would emphasize the necessity of doing each part, and could also formalize the expectations of working in this manner and forming strong model building habits. For many types of problem solving—especially those related to LCA, where many answers are possible depending on how you go about modeling the problem—the emphasis may be on parts 1 and 3, relatively de-emphasizing the result found in part 2. In other domains, such as in a mathematics course, the result (part 2) may be the only significant part in terms of how you are assessed. Regardless, you probably still used a method (and may have briefly shown it by writing an equation and applying it), and hopefully tried to quickly check your result to ensure it passed a reasonableness test, even if you did not in detail write about each of those steps. A way of remembering the importance of this three-step process is that your answer should never simply be a graph or a number. There is always a need to discuss the method you used to create it, as well as some reflection on the value. Regardless of the grading implications and distributions, hopefully you can see how this three-step process always exists – it is just a matter of translating the question or task presented to determine how much effort to make in each part, and how much documentation to provide as an answer. You will find that performing LCA constitutes assembling many small building block calculations and mini-models into an overall model. If you have mostly ignored how you came up with these building block results, it will be difficult to follow your overall work, and to follow how the overall result was achieved.
Chapter Summary In LCA, any study will be composed of a collection of many of the techniques above. You'll be piecing together emissions factors and small assumption-based estimates, generating new estimates, and summarizing your results. A frequently stated reason for why people enter the field of science or engineering is that they are more comfortable with numbers or equations than they are with "writing". But communicating your method, process, and results via writing is an especially important skill in conducting life cycle assessment.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
60
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
References for this Chapter CBS News, "Reversing Bush, EPA Toughens Smog Rules", via Internet, http://www.cbsnews.com/news/reversing-bush-epa-toughens-smog-rules/, last accessed July 20, 2014. Harte, John , Consider a Spherical Cow: A Course in Environmental Problem Solving, University Science Books, 1988. Koomey, Jonathan, Turning Numbers into Knowledge, Analytics Press, 2008. Mahajan, Sanjoy, Street-Fighting Mathematics: The Art of Educated Guessing and Opportunistic Problem Solving, MIT Press, 2010. Mosteller, Frederick, "Assessing Unknown Numbers: Order of Magnitude Estimation", in Statistics and Public Policy, William Fairley and Frederick Mosteller, editors, AddisonWesley, 1977. NOAA 2012, Surveying: Accuracy vs. Precision, via Internet, http://celebrating200years.noaa.gov/magazine/tct/tct_side1.html U.S. Census Bureau, Statistical Abstract of the United States: 2012 (131st Edition) Washington, DC, 2011; available at http://www.census.gov/compendia/statab/ Weinstein, Lawrence, and Adams, John A., Guesstimation: Solving the World's Problems on the Back of a Cocktail Napkin, Princeton University Press, 2008.
End of Chapter Questions 1. Find and reference three primary sources for the amount of energy used in residences in the United States. Validate your findings as possible. 2. Find the fraction of the population that lives in cities versus rural areas in the US, or in your home state. Validate your findings as possible. 3. Estimate the total weight of the population in your home state. 4. Estimate the number of hairs on your head. 5. Estimate the number of swimming pools in Los Angeles.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 2: Quantitative and Qualitative Methods Supporting Life Cycle Assessment
THIS PAGE INTENTIONALLY BLANK TO PRESERVE FORMATTING
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
61
62
Chapter 3: Life Cycle Cost Analysis
Chapter 3 : Life Cycle Cost Analysis In this chapter, we begin our discussion of life cycle analytical methods by overviewing the long-standing domain of life cycle cost analysis (LCCA). It is assumed that the reader already understands the concepts of costs and benefits – if not, a good resource is our companion e-book on civil systems planning (Hendrickson and Matthews 2013). The methods and concepts from this domain form the core of energy and environmental life cycle assessment that we will introduce in Chapter 4. We describe the ideas of "first cost" and "recurring costs", as well as methods to put all of the costs over the life of a product or project into the financial-only basis of common monetary units.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. Describe the types of costs that are included in a life cycle cost analysis. 2. Assess the difference between one-time (first) costs and recurring costs. 3. Select a product or project amongst alternatives based on life cycle cost. 4. Convert current and future monetary values into common monetary units.
Life Cycle Cost Analysis in the Engineering Domain Material, labor, and other input costs have been critical in the analysis of engineered systems for centuries. Studies of costs are important to understand and make decisions about product designs or decisions as these will inevitably allow you to profit from a successful one. Separate from cost is the concept of benefit, which includes the value you would receive from an activity such as using a product. Many of the cost models used to support engineering decisions have been relatively simple, for example, summing all input costs and ensuring they are less than the funds budgeted for a project. Engineers have been estimating the whole life cycle cost of built infrastructure for decades. Life cycle cost analysis (LCCA) has been used to estimate and track lifetime costs of bridges, highways, other structures and manufactured goods because important and costly decisions need to be made for efficient management of social resources used by these structures and goods. Early design and construction decisions are often affected by future maintenance costs. Given this history, LCCA has most often been used for decision support Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 3: Life Cycle Cost Analysis
63
on fairly large-scale projects. It has also, however, been applied to individual products. LCCA is often performed as a planning tool early in a project but is also done during a project's lifetime. We focus on LCCA because its economic focus across the various life cycle phases is very similar to the frameworks we will need to build for our energy and environmental life cycle models. If you can understand the framework, and follow the quantitative inputs and models used, you will better be able to understand LCA. The project that is already being undertaken or is already in place is typically referred to as the status quo. Key to the foundation of LCCA is a set of alternative designs (or alternatives) to be considered, which may vary significantly or only slightly from one another. These alternatives may have been created specifically in an attempt to reduce costs, or may simply be alternatives deviating from an existing project design along non-cost criteria. With respect to the various costs that may be incurred across the life cycle, first (or initial) cost refers to costs incurred at the beginning of a project. First cost generally refers only to the expense of constructing or manufacturing as opposed to any overhead costs associated with designing a product or project – it is the "shovel in the ground" or factory costs. While design and other overhead costs may be routinely ignored in cost analyses, and in LCA, they are real costs that are within the life cycle. Future costs refer to costs incurred after construction/manufacture is complete and typically occur months to years after. Recurring costs are those that happen with some frequency (e.g., annually) during the life of a project. In terms of accounting and organization, these costs are often built into a timeline, with first costs represented as occurring in "year 0" and future/recurring costs mapped to subsequent years in the future. The sum of all of these costs is the total cost. The status quo will often involve using investments that have already been made. The original costs of these investments are termed sunk costs and should not be included in estimation of the new life cycle cost of alternatives from the present period. Beyond civil engineering, LCCA also presents itself in concepts such as whole-life cost or total cost of ownership, which consumers may be more familiar with. Total cost of ownership (TCO) is used in the information technology industry to capture costs of purchasing, maintaining, facilities, training users, and keeping current a hardware-software system. TCO analyses have been popular for comparisons between proprietary software and open source alternatives (e.g., Microsoft Office vs. OpenOffice) as well as for operating systems (Mac vs. Windows). However not all decisions get made on the basis of knowing the minimum TCO - despite many TCO studies showing lower costs, neither Mac nor OpenOffice have substantial market share. Before discussing LCCA in the context of some fairly complex settings, let us first introduce a very simple but straightforward example that we will revisit throughout the book.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
64
Chapter 3: Life Cycle Cost Analysis
Example 3-1: Consider a family that drinks soda. Soda is a drink consisting of carbonated water, flavoring and (usually) sweetener. The family's usual way of drinking it is buying 2-liter bottles of soda from a store at a price of $1.50 each. An alternative is to make soda on demand with a home soda machine. The machine carbonates a 1-liter bottle of water, and the user adds a small amount of flavor syrup (with or without sweetener) to produce a 1-liter bottle of flavored soda. An advantage of a home soda machine is that it can be easily stored and use of flavor bottles removes the need to purchase in advance and store soda bottles (which are mostly water). Soda machines cost $75 and come with several 1-liter empty bottles and a carbonation canister for 60 liters of water. Flavor syrup bottles cost $5 and make 50 8-ounce servings (12 liters) of flavored soda. Additional carbonation canisters cost $30. Question: If the family drinks 2 liters of soda per week (52 per year), compare the costs of 2-liter soda bottles with the purchase of a soda maker and flavor bottles over a one-year period. Answer: The cost of soda from a store is $1.50 * 1 = $1.50 per week, or $78 per year. Note that this cost excludes any cost of gasoline or time required for shopping. For the soda machine option, we need a soda machine ($75) and sufficient flavor syrup bottles to make 104 liters of soda (about 9 bottles or $45), and would use the entire first (free) carbonation canister and most of a second ($30). Thus the soda machine cost for a year is $150. This cost also excludes any cost of water, gasoline or time (as well as unused syrup or carbonation). Over a one-year period, the life cycle cost or total cost of ownership for a soda machine is almost double that of store-bought bottles. The soda machine provides additional benefit for those who dislike routine shopping or have a high value of their time, which we noted has not been included.
We can use the methods of Chapter 2 to find breakeven values for the soda machine. Example 3-2: Find the breakeven price of soda bottles in Example 3-1 compared to buying a soda machine over one year, without considering discounting. Answer: The breakeven price is the price such that when we consider all of the costs for each option, they are exactly equal. The cost of soda bottles is $72 per year less expensive than the machine. You could either divide $72 by 52 bottles ($1.38 cents per bottle) and add it to the current price, or about $2.88/bottle) or explicitly solve for the price per bottle using the equation $150=52 bottles * p, where p is the price per bottle. Again we find that, at $2.88 per bottle, purchased soda will cost the same as homemade soda over a one-year period.
Discounting Future Values to the Present While a full discussion of the need to convert future values to present values as a common monetary unit is beyond the scope of this chapter, this activity is shown to ensure that the Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 3: Life Cycle Cost Analysis
65
time value of money is represented. There are many resources available to better understand the theory of such adjustments, including Au and Au's Engineering Economics book (1992). In short, though, just like other values that are increased or decreased over time due to growth or decay rates, financial values can and should be adjusted if some values are in current (today's) dollars and values in the future are given in then-current values. If that is the case, there is a simple method to adjust these values, as shown in Equation 3-1: F = P (1 + r) n ó P = F (1 + r) -n
(3-1)
where P represents a value in Present (today's) dollars, F represents a value in Future dollars, r is the percentage rate used to discount from future to present dollars, and n is the number of years between present and future. Equation 3-1 can be used to convert any future value into a present value. Equation 3-1 is usually used with constant dollars, which have been adjusted for the effects of inflation (not shown in this chapter). When values are plugged into Equation 3-1 the P or F results are referred to as present or future discounting factors. Thus if r=5%, n=1, and a future value (F) of $100 into Equation 3-1, we would get a present discounting factor of 0.952, which means the future value would be discounted by 4.8%. Example 3-3: What is the present total cost over 5 years of soda made at home using the approximated costs above (ignoring unused syrup and carbonation) at a discount rate of 5%? Answer:
The table below summarizes the approximated costs for each of the five years. Year 0
Year 1
Year 2
Year 3
Year 4
Year 5
$75
0
0
0
0
0
Flavor
-
$45
$45
$45
$45
$45
Carbonators
-
$30
$60
$60
$60
$60
$75
$75
$105
$105
$105
$105
Soda Machine
Total
The soda machine is bought at the beginning of Year 1 (a.k.a. Year 0) and costs $75. It does not need to be discounted as that is already in present dollars. It would cost $45 for flavor bottles in each Year 1 through 5. The first carbonator is free, but the second costs $30 in Year 1. Two are needed ($60) in every subsequent year. Thus the present cost (rounded off to 2 significant digits with present discounting factors of 0.952, 0.907, 0.864, 0.823, and 0.784) is: Present Value of Cost = $75 + $75/1.05 + $105/1.052 + $105/1.053 + $105/1.054 + $105/1.055 = $75 + $71 + $95 + $91 + $86 + $82 = about $500
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
66
Chapter 3: Life Cycle Cost Analysis
Now that we have a slightly more rigorous way of dealing with costs over time, let us consider the advanced Example 3-4 comparing life cycle costs of two different new cars. Example 3-4: Consider a new car purchase decision for someone deciding between a small sedan or a small hybrid electric vehicle. A key part of such a comparison is to assume that the buyer is considering otherwise equivalent vehicles in terms of features, size, and specifications. Given this constraint we compare a 2013 Toyota Corolla with a 2013 Toyota Prius. While the engines are different, the seating capacity and other functional characteristics are quite similar. Question: What are the total costs over 5 years for the two cars assuming 15,000 miles driven per year? Answer: Edmunds.com (2013) has a "True Cost to Own" calculator tool that makes this comparison trivial. Note that the site assumes the equivalent of financing the car, and the values are not discounted. Selecting just those vehicles and entering a zip code gives values that should look approximately like the values listed below. Even driving 15,000 miles per year, the Prius would be $5,000 more expensive than just buying a small, fuel-efficient Corolla. 2013 Toyota Corolla Depreciation Taxes & Fees Financing Fuel Insurance Maintenance Repairs Tax Credit True Cost to Own
Year 1 $2,766 $1,075 $572 $1,892 $2,288 $39 $0 $0 $8,632
Year 2 $1,558 $36 $455 $1,949 $2,368 $410 $0
Year 3 $1,370 $36 $333 $2,007 $2,451 $361 $89
Year 4 $1,215 $36 $205 $2,068 $2,537 $798 $215
Year 5 $1,090 $36 $74 $2,130 $2,626 $1,041 $314
$6,776
$6,647
$7,074
$7,311
Total $7,999 $1,219 $1,639 $10,046 $12,270 $2,649 $618 $0 $36,440
2013 Toyota Prius Depreciation Taxes & Fees Financing Fuel Insurance Maintenance Repairs Tax Credit True Cost to Own
Year 1 $7,035 $1,919 $1,045 $1,098 $1,920 $39 $0 $0 $13,056
Year 2 $2,820 $36 $830 $1,131 $1,987 $423 $0
Year 3 $2,481 $36 $608 $1,165 $2,057 $381 $89
Year 4 $2,200 $36 $375 $1,200 $2,129 $786 $215
Year 5 $1,973 $36 $134 $1,236 $2,203 $1,784 $314
$7,227
$6,817
$6,941
$7,680
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Total $16,509 $2,063 $2,992 $5,830 $10,296 $3,413 $618 $0 $41,721
Chapter 3: Life Cycle Cost Analysis
67
Life Cycle Cost Analysis for Public Projects The examples above were all centered on life cycle costing for personal or individual decisions. However, as introduced, generally LCCA is applied to public projects such as buildings or infrastructure. Life cycle stages of infrastructure systems are similar to those we discussed in Chapter 1. They also rely on resource extraction and assembly, although infrastructure is generally constructed rather than manufactured. The use phase is occupation or use by the public. The use phase also involves maintenance, repair, or rehabilitation activities. The end of life phase is when it is demolished, either because it is no longer needed or is being replaced. LCCA is a useful tool to help assess how various decisions will affect cost. For example, a particular design may be adjusted, resulting in increased initial cost, but as a means to reduce planned maintenance costs. The design change could take the form of a planned increase in the expected time until rehabilitation, reduction in the actual expenditure at time of maintenance, or by changing the cost structure. LCCA also has a fairly large scope of stakeholder costs to include, accounting for both owner costs and user costs over the whole life cycle, as shown in Equation 3-2. Life Cycle Costs (LCC) =
! ! 𝑂𝑤𝑛𝑒𝑟 𝐶𝑜𝑠𝑡!
+
! ! 𝑈𝑠𝑒𝑟 𝐶𝑜𝑠𝑡!
(3-2)
Owner costs are those incurred by the party responsible for the product or project, while user costs are incurred by the stakeholders who make use of it. For example, a state or local department of transportation may own a highway, but local citizens will be the users. The owner costs are straightforward to consider – they are the cost of planning and building the highway. The user costs might include the value of drivers' time spent waiting in traffic (and thus, we have incentive to choose options which would minimize this cost). User costs may be quite substantial and a multiple or order of magnitude higher than the owner costs. Figure 3-1 organizes the various types of life cycle costs in rows and columns and shows example costs for a highway project. For products purchased by private parties, owner and user costs are the same category and do not need to be distinguished.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
68
Chapter 3: Life Cycle Cost Analysis
Owner
Recurring (Year n)
First (Year 0)
Recurring (Year 1)
Recurring (Year 2)
Design Construction
Financing Maintenance
Financing Maintenance
Financing Rehabilitation
Vehicle Use Tolls Cost of Time Driving
Vehicle Use Tolls Cost of Time Driving
Vehicle Use Tolls Cost of Time Driving
Category
User
Figure 3-1: Example Life Cycle Cost Table for Highway Project
LCCA focuses only on costs. While differences in costs between two alternatives may be considered "benefits", true benefit measures are not used. In the end, LCCA generally seeks to find the least costly project alternative over the life cycle considering both owner and user costs. But since agency decision makers are responsible for the project over the long run, they could be biased towards selecting projects with minimum owner life cycle costs regardless of user costs since user costs are not part of the agency budget. This stakeholder difference will also manifest itself when we discuss LCA later in the book, as there may be limited benefits to a company making their product have a lower environmental impact if the consumer is the one who will benefit from it (e.g., if it costs more for the company to produce and perhaps reduces profits but uses less electricity in the use phase). Various government agencies suggest and expect LCCA practices to be part of the standard toolbox for engineers, planners, and decision makers. The US Federal Highway Administration (FHWA) has promoted LCCA since 1991 and the US Department of Transportation (DOT) created a Life Cycle Cost Analysis Primer (2002) to formalize their intentions to have engineers and planners use the tool in their practice. In this document they describe the following steps in LCCA: 1. Establish design alternatives, including status quo 2. Determine activity timing 3. Estimate costs (agency and user) 4. Compute life-cycle costs (LCC) 5. Analyze the results While we have described most of these steps already, we emphasize them to demonstrate that LCCA does not end with simply determining the life cycle costs of the various alternatives. It is a multi-step process and it ends with an expected conclusion and analysis, Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 3: Life Cycle Cost Analysis
69
building upon the three-part system we introduced in Chapter 2. Such an analysis may reveal that the LCC of one of the alternatives is merely 1% less than the next best alternative, or that it is 50% less. It might also indicate that there is too much uncertainty in the results to make any conclusion. In the end, the analyst's result may not be the one chosen by the decision maker due to other factors such as budgets, politics, or different assessments of the relative worth of the various cost categories. Regardless, the act of analyzing the results is a critical component of any analytical framework.
Deterministic and Probabilistic LCCA Our examples so far, as well as many LCCAs (and LCAs, as we will see later) are deterministic. That means they are based on single, fixed values of assumptions and parameters but more importantly it suggests that there is no chance of risk or uncertainty that the result might be different. Of course it is very rare that there would be any big decision we might want to make that lacks risk or uncertainty. Probabilistic or stochastic models are built based on some expected uncertainty, variability, or chance. Let us first consider a hypothetical example of a deterministic LCCA as done in DOT (2002). Figure 3-2 shows two project alternatives (A and B) over a 35-year timeline. Included in the timeline are cost estimates for the life cycle stages of initial construction, rehabilitation, and end of use. An important difference between the two alternatives is that Alternative B has more work zones, which have a shorter duration but that cause inconvenience for users, leading to higher user costs as valued by their productive time lost. Following the five-step method outlined above, DOT showed these values:
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
70
Chapter 3: Life Cycle Cost Analysis
Figure 3-2: Deterministic LCCA for Construction Project
Without discounting, we could scan the data and see that Alternative A has fewer periods of disruption and fairly compact project costs in three time periods. Alternative B's cost structure (for both agency and user costs) is distributed across the analysis period of 35 years. Given the time value of money, however, it is not obvious which might be preferred. At a 4% rate, the discounting factors using Equation 3-1 for years 12, 20, 28, and 35 are 0.6246, 0.4564, 0.3335, and 0.2534, respectively. Thus for Alternative A the discounted life cycle agency costs would be $31.9 million and user costs would be $22.8 million. For Alternative B they would be $28.3 million and $30.0 million, respectively. As DOT (2002) noted in their analysis, "Alternative A has the lowest combined agency and user costs, whereas Alternative B has the lowest initial construction and total agency costs. Based on this information alone, the decision-maker could lean toward either Alternative A (based on overall cost) or Alternative B (due to its lower initial and total agency costs). However, more analysis might prove beneficial. For instance, Alternative B might be revised to see if user Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 3: Life Cycle Cost Analysis
71
costs could be reduced through improved traffic management during construction and rehabilitation." Even though this was a hypothetical example created to demonstrate LCCA to the civil engineering audience, presumably you are already wondering how robust these numbers are to other factors and assumptions. DOT also noted "Sensitivity analysis could be performed based on discount rates or key assumptions concerning construction and rehabilitation costs. Finally, probabilistic analysis could help to capture the effects of uncertainty in estimates of timing or magnitude of costs developed for either alternative." While engineers have been collecting data on their products for as long as they have been designing products, the types of data required to complete LCCA analyses are generally much different than the usually collected data. LCCA can require planners to have estimates of future construction or rehabilitation costs, potentially a decade or more from the time of construction. These are obviously uncertain values (and further suggests the need for probabilistic methods). For big decisions like that in the DOT example, one would want to consider the ranges of uncertainty possible to ensure against a poor decision. Building on DOT's recommendation, we could consider various values of users' time, the lengths of time of work zone closures, etc. If we had ranges of plausible values instead of simple deterministic values, that too could be useful. Construction costs and work zone closure times, for example, are rarely much below estimates (due to contracting issues) but in large projects have the potential to go significantly higher. Thus, an asymmetric range of input values may be relevant for a model. We could also use probability distributions to represent the various cost and other assumptions in our models. By doing this, and using tools like Monte Carlo simulation, we could create output distributions of expected life cycle cost for use in LCCA studies. We could then simulate costs of the alternatives, and choose the preferred alternative based on combinations of factors such as the lowest mean value of cost and the lowest standard deviation of cost. Finally, probabilistic methods support the ability to quantitatively assess the likelihood that a particular value might be achieved. That means you might be able to assess how likely each Alternative is to be greater than zero, or how likely it is that the cost of Alternative A is less than Alternative B. It is by exploiting such probabilistic modeling that we will be able to gain confidence that our analysis and recommendations are robust to various measures of risk and uncertainty, and hopefully, support the right decisions. We will revisit these concepts in Chapter 11 after we have learned a bit more about LCA models.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
72
Chapter 3: Life Cycle Cost Analysis
Chapter Summary As introduced in Chapter 1, sustainability involves social, economic, and environmental factors. We can track costs over the life cycle of products or projects and use it as a basis for making decisions regarding comparative economic performance. There are various methods and applications to perform life cycle cost analysis (LCCA) in support of decisions for basic products, choices, and for infrastructure systems. Depending on the complexity of the project, we may want to adjust for the time value of money by using discounting methods that normalize all economic flows as if they occurred in the present. A benefit of using such methods is that they allow incorporation of costs by both the owner as well as other users. Beyond deterministic methods, LCCA can support probabilistic methods to ensure we can make robust decisions that incorporate risk and uncertainty. Now that you have been exposed to the basics of LCCA, you can appreciate how building on the straightforward idea of considering costs over the life cycle can broaden the scope involved in life cycle modeling. As we move forward in this textbook to issues associated with energy or environmental life cycle assessment, concepts of life cycle cost analysis should remain a useful part of LCA studies.
References for this Chapter Hendrickson, Chris T. and H. Scott Matthews, Civil Infrastructure Planning, Investment and Pricing. http://cspbook.ce.cmu.edu/ (accessed July, 2013). Tung Au and Thomas P. Au, Engineering Economics for Capital Investment Analysis, 2nd edition, Prentice-Hall, 1992. Available at http://engeconbook.ce.cmu.edu. edmunds.com, website, www.edmunds.com, last accessed January 2, 2013. US Department of Transportation Office of Asset Management, "Life-Cycle Cost Analysis Primer", FHWA-IF-02-047 2002. Available at http: http://www.fhwa.dot.gov/infrastructure/asstmgmt/lcca.cfm
End of Chapter Questions 1. Building on Example 3-1, find the total cost of buying 2-liter bottles of soda over a 5-year period, with and without discounting at a rate of 5%.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 3: Life Cycle Cost Analysis
73
2. How would your results change for Example 3-1 if you spent 3 minutes per shopping trip buying soda, and that time spent had a cost of $20 per hour? What is this breakeven cost of time per hour? 3. Building on Example 3-1, but where you must drive 5 miles to the store in a vehicle that gets 25 miles per gallon (at a gasoline price of $3.50 per gallon) in order to buy the soda machine, as well as the flavor bottles or to purchase two-liter bottles every time you want to drink soda, what are the total costs in the first year? What are the total discounted costs over 5 years at a 5% rate? Discuss qualitatively how your model results might change if you were buying other items on your shopping trips. 4. Combine the original Example 3-1 data and assumptions, as well as the additional information from Questions 1 through 3 above. Calculate total life cycle costs over 5 years for each option and create a visual to summarize your results. Which alternative should be chosen over 5 years, buying soda from a store or buying a soda machine? Which should be chosen over 10 years? 5. Compared to the result in Example 3-2, does the breakeven price of soda bottles change over a 5-year period if you do not consider discounting? Does the breakeven price change over 5 years if you discount at 5%? 6. Generate a life cycle cost summary table (using Figure 3-1 as a template) for the following: a. A privately purchased computer b. A public airport c. A sports arena or stadium 7. How sensitive (quantitatively and qualitatively) is the decision in Example 3-4 to the annual cost of fuel? Create a graphic to show your result. 8. What are the total costs to own for the two vehicles in Example 3-4 with a 5% discount rate? Which vehicle would you choose? Does your decision ever change if the discount rate varies from 0 to 20%? 9. A household is considering purchasing a washing machine and has narrowed the choice to two alternatives. Machine 1 is a standard top-loading unit with a purchase cost of $500. This machine uses 40 gallons of water and 2 kilowatt-hours of electricity per load (assuming an electric water heater). The household would do roughly 8 loads of laundry per week with this machine. Machine 2 is a front-loading unit; it costs $1,000, but it can wash double the amount of clothes per load, and each Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
74
Chapter 3: Life Cycle Cost Analysis
load uses half the water and electricity. Assume that electricity costs 8 cents/kWh and water is $2 per 1,000 gals. a. Generate a life cycle cost summary table for the two washing machines b. Develop a life cycle cost comparison of the two machines over a 10-year life period without discounting. Which machine should be chosen if considering only cost? c. Which would you choose over a 10-year period with a 3% discount rate? 10. How sensitive (quantitatively and qualitatively) is the choice of washing machines to the discount rate, price of electricity, and price of water? 11. A recent and continuing concern of automobile manufacturers is to improve fuel economy. One of the easiest ways to accomplish this is to make cars lighter. To do this, vehicle manufacturers have substituted specially strengthened— but lighter— aluminum for steel (they have also experimented with carbon fibers). Unfortunately, processed aluminum is more expensive than steel - about $3,000 per ton instead of $750 per ton for steel. Aluminum-intensive vehicles (AIVs) are expected to weigh less by replacing half of the steel in the car with higher-strength aluminum on a 1 ton of steel to 0.8 ton of aluminum basis. This is expected to reduce fuel use 20%. Assume: • Current cars can travel 25 miles per gallon of gasoline and gasoline costs $3.50 per gallon • Current [steel] cars cost $20,000 to produce, of which $1,000 is currently for steel and $250 for aluminum • AIVs are equivalent to current cars except for substitution of lighter aluminum for steel • All cars are driven 100,000 miles • All tons are short tons (2,000 pounds) a) Of current cars and AIVs, which is cheaper over the life cycle (given only the information above)? Develop a useful visual aid to compare life cycle costs across steel vehicles and AIVs.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 3: Life Cycle Cost Analysis
75
b) How uncertain would our cost estimates for steel, aluminum, and gas have to be to reverse your opinion on which car was cheaper over the life cycle? c) Do your answers above give you enough information to determine whether we should produce AIVs? What other issues might be important?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Photo of nuclear electricity generation facility in France prominently showing its certification to the ISO 14001 Environmental Management Standard. Photo credit: By Pierre-alain dorange (Own work) [CC-BY-SA-3.0 (http://creativecommons.org/licenses/by-sa/3.0)], via Wikimedia Commons http://upload.wikimedia.org/wikipedia/commons/0/08/Centrale_Nucl%C3%A9ai re_du_Blayais.jpg
Chapter 4: The ISO LCA Standard – Goal and Scope
77
Chapter 4 : The ISO LCA Standard – Goal and Scope We have discussed many of the skills that are necessary to complete a rigorous LCA. Now we present the standard framework for planning and organizing such a study. In this chapter, we supplement information found in the official ISO Standard for LCA. We only summarize and expand on the most critical components, thus this chapter is not intended to be a substitute for reading and studying the entire ISO Standard (likely more than once to gain sufficient understanding). The rationale for studying the ISO Standard is to build a solid foundation on which to understand the specific terminology used in the LCA community and to learn directly from the Standard and from our collective experience what is required in an LCA, and what is optional. We use excerpts and examples from completed LCA studies to highlight key considerations since examples are generally lacking in the Standard. As such, the purpose of this chapter is not to re-define the terminology used but to help you understand what the terms mean from a practical perspective.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: •
Describe the four major phases of the ISO LCA Standard
•
List all of the ISO LCA Standard study design parameters (SDPs)
•
Review SDPs given for an LCA study and assess their appropriateness and anticipate potential challenges in using them
•
Generate specific SDPs for an LCA study of your choosing
Overview of ISO and the Life Cycle Assessment Standard Before we specifically discuss the LCA Standard, we review standards in general. Standards are created to make some activity or process consistent, or at least to be done using common guidelines or methods. They might also be created to level the playing field in a particular market by ensuring that everyone does things the same way. Standards are made for a variety of reasons, and exist at many levels, from local building codes all the way up to global standards. In civil engineering and construction, there are standards for concrete; for example a request for proposals could require that the product meet "ASTM C94 concrete". This means that any concrete used in the project must meet the testing standard defined in ASTM C94, developed by the ASTM International organization. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
78
Chapter 4: The ISO LCA Standard – Goal and Scope
There are many organizations around the world that work on developing and promoting the use of standards. ASTM International, mentioned above, has been developing standards for specific tests and materials for more than 100 years. ISO (the International Organization for Standardization – the acronym makes sense in French!) is an international organization that creates standards geared more towards safety, quality, and management standards, and various companies and entities around the world follow these standards. The actual processes used by each organization to create a standard vary, but for ISO the process has the following components: it (1) responds to a market need; (2) is based on expert opinion; (3) is developed by a multi-stakeholder team; and (4) is ratified by consensus. The actual standard is drafted, edited, and revised by a technical committee of global experts based on comments until consensus (75% agreement) is reached (ISO 2012). There are various frameworks for performing life cycle assessment (LCA) but the primary and globally accepted way of doing it follows the ISO LCA Standard (which is comprised primarily of two related standards, 14040:2006 and 14044:2006), which we assume you have accessed and read separately. We will refer to both underlying standards as the ISO Standard. The notation "14040:2006" means that the ISO LCA Standard is in the "ISO 14000" family of standards, which are global standards for environmental management and encompass various other processes to track and monitor emissions and releases. The version current as of the time of writing this book was most recently updated in 2006. The first version of the ISO LCA Standard was published in 1997. One thing that you may now realize is that many of the foundational LCA studies mentioned in Chapter 1 (e.g., by Hocking, Lave, etc.) were completed before the LCA Standard was formalized. That does not mean they were not legitimate studies – it just means that in today's world these could not be referred to as "ISO compliant", where ISO compliant means that the work conforms to the Standard as published. While it may seem trivial, compliance with the many ISO standards is typically a goal of an entity looking for global acceptance and recognition. This is not just in the LCA domain – firms in the automotive supply chain seek "ISO 9000 compliance" to prove they have quality programs in place at their companies that meet the standard set by ISO, so that they are able to do business in that very large global market. Chapter 13 in this book will discuss more about peer review and assessing ISO compliance for an LCA study. It should be obvious why a standard for LCA is desirable. Without a formal set of requirements and/or guidelines, anyone could do an LCA according to her own views of how a study should be done and what methods would be appropriate to use. In the end, 10 different parties could each perform an LCA on the same product and generate 10 different answers. The LCA Standard helps to normalize these efforts. However, as we will see below, its rules and guidelines are not overly restrictive. Simply having 10 parties conforming to the Standard does not guarantee you would not still generate 10 different answers! One could alternatively argue that in a field like LCA, a diversity of thoughts and Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
79
approaches is desirable, and thus, that having a prescriptive standard stifles development of methods or findings. As you have read separately, the ISO LCA Standard formalizes the quantitative modeling and accounting needs to implement life cycle thinking to support decisions. ISO 14040:2006 is the current "principles and framework" of the Standard, and is written for a managerial audience while ISO 14044:2006 gives the "requirements and guidelines" as for a practitioner. Given that you have already read the Standard (and have their glossaries of defined terms to help guide you), you are already familiar with the basic ideas of inputs, outputs, and flows. At a high level, Figure 4-1 summarizes the ISO LCA Standard's 4 phases: goal and scope definition, inventory analysis, impact assessment, and interpretation. The goal and scope are statements of intent for your study, and part of what we will refer to as the study design parameters (discussed below). They explicitly note the reason why you are doing the study, as well as the study reach. In the inventory analysis phase, you collect and document the data needed (e.g., energy use and emissions of greenhouse gases) to meet the stated goal and scope. In the impact assessment phase you transition from tracking simple inventory results like greenhouse gas emissions to impacts such as climate change. Finally, the interpretation phase looks at the results of your study, puts them into perspective, and may recommend improvements or other changes to reduce the impacts.
Figure 4-1: Overview of ISO LCA Framework (Source: ISO 14040:2006) Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
80
Chapter 4: The ISO LCA Standard – Goal and Scope
It is important to recognize that all of the double arrows mean that the four phases are iterative, i.e., you might adjust the goal and scope after trying to collect inventory data and realizing there are challenges in doing so. You may get to the interpretation phase and realize the data collected does not help answer the questions you wanted and then revise the earlier parts. You may get unexpected results that make reaching a conclusion difficult, and need to add additional impact assessments. Thus, none of the phases are truly complete until the entire study is complete. From experience, every study you do will be modified as you go through it. This is not a sign of weakness or failure; it is the prescribed way of improving the study as you learn more about the product system in question. As ISO mentions, it is common that studies following the Standard do not include an impact assessment phase, and these studies are simply called life cycle inventory studies (LCIs). That is, their final results are only the accounting-like exercise of quantifying total inputs and outputs without any consideration of impact. You could interpret this to mean that impact assessment is not a required component, but more correctly it is required of an LCA study but not an LCI. That said, we will generally use the phrase "LCA" to refer either to an LCA or an LCI, as is common in the field. The right hand side of Figure 4-1 gives examples of how LCA might be used. The first two, for product improvement and strategic planning, are common. In this book we focus more on "big decisions" and refer to activities such as informing public policy (e.g., what types of incentives might make paper recycling more efficient?) and assessing marketing claims. In these domains the basis of the study might be in comparing between similar products or technologies. In the rest of this chapter, we focus on the goal and scope phases of LCA. Subsequent chapters discuss the inventory, interpretation, and impact assessment phases in greater detail.
ISO LCA Study Design Parameters As noted above, ISO requires a series of parameters to be qualitatively and quantitatively described for an LCA study, which in this text we refer to as the study design parameters (SDPs), listed in Figure 4-2. In this section we provide added detail and discussion about the underlying needs of each of these parameters and discuss hypothetical parameter statements and values in terms of their ISO conformance.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
Goal
81
Scope Items: Product System System Boundary Functional Unit Inventory Inputs and Outputs LCIA Methods Used
Figure 4-2: Study Design Parameters (SDPs) of ISO LCA Framework
Think of the SDPs as a summary of the most important organizing aspects of an LCA. The SDPs are a subset of the required elements in an LCA study, but are generally the most critical considerations and thus those that at a glance would tell you nearly everything you needed to know about what the study did and did not seek to do. Thus, these are items that need to be chosen and documented very well so there is no confusion. In documenting each in your studies, you should specifically use the keywords represented in the Standard (e.g., "the goal of this study is", "the functional unit is", etc.) Expanding on what is written in the ISO LCA Standard we discuss each of the items in the SDP below. SDP 1. Goal The goal of an LCA, like the goal of any study, must be clearly stated. ISO requires that the goal statement include unambiguous statements about: (1) the intended application, (2) the reasons for carrying out the study, (3) the audience, and (4) whether the results will be used in comparative assertions released publicly. An easy way to think about the goal statement of an LCA report is that it must fully answer two questions: "who might care about this and why?" and "why we did it and what will we do with it?". As noted above, the main components of an LCA are iterative. Thus, it is possible you start an LCA study with a goal, and by going through the effort needed to complete it, the goal is changed because more or less is possible than originally planned. Below are excerpts of the goal statement from an LCA study comparing artificial and natural Christmas trees bought in the US2 (PE Americas 2010). "The findings of the study are intended to be used as a basis for educated external communication and marketing aimed at the American Christmas tree consumer." "The goal of this LCA is to understand the environmental impacts of both the most common artificial Christmas tree and the most common natural Christmas tree, and to analyze how their environmental impacts compare." "This comparative study is expected to be released to the public by the ACTA to refute myths and misconceptions about the relative difference in environmental impact by real and artificial trees." 2
In the interest of full disclosure, one of the authors of this book (HSM) was a paid reviewer of this study.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
82
Chapter 4: The ISO LCA Standard – Goal and Scope
From these three sentences, we clearly can understand all 4 of the ISO-required components of the goal statement. The intended application is external marketing. The reasons are to refute misconceptions. The audience is American tree consumers. Finally, the study was noted to be planned for public release (and it is available on a website). We will discuss further implications of public studies later in the book. The examples above help constitute a good goal statement. It should be clear that skipping any of the 4 required parts or trying to streamline the goal for readability could lead to an inappropriate goal statement. For example, the sentence "This study seeks to find the energy use of a power plant" is clear and simple but only addresses one of the four required elements of a goal. It also never uses the word "goal" which could be perceived as stating no goal. Beyond the stated goals, we could consider what is not written in the goals. From the above statements, there would be no obvious use of the study by a retailer, e.g., to decide whether to stock one kind of tree over another. It is useful to consider what a reader or reviewer of the study would think when considering your goal statement. A reviewer would be sensitive to biases and conflicts, as well as creative use of assumptions in the SDP that might favor one alternative over others. Likewise, they may be sensitive to the types of conclusions that may arise from your study given your chosen goals. You want to write so as to avoid such interpretations. One of the primary reasons that scientists seek to use LCA is to make a comparative assertion, which is when you compare multiple products or systems, such as two different types of packaging, to be able to conclude and state (specifically, to make a claim) that one is better than the other (has lower impacts). As noted above, the ISO LCA Standard requires that such an intention be noted in the goal statement. Scope Although ISO simply lists "goal and scope", a goal statement is just a few sentences while the scope may be several pages. The study scope is not a single statement but a collection of qualitative and quantitative information denoting what is included in the study, and key parameters that describe how it is done. Most of the SDPs are part of the scope. There are 14 separate elements listed in ISO's scope requirements, but our focus is on five of them that are part of the SDPs: the product system studied, the functional unit(s), system boundaries, and the inventory and/or impact assessments to be tracked. The other ten are important (and required for ISO compliance) but are covered sufficiently either in the ISO Standard or elsewhere in this book. While these five individual scope SDPs are discussed separately below, they are highly dependent on each other and thus difficult to define separately. We acknowledge that this Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
83
interdependency of terminology typically confuses most readers, as every definition of one of the scope SDPs contains another SDP term. However, a clear understanding of these terms is crucial to the development of a rigorous study and we recommend you read the following section, along with the ISO Standard, multiple times until you are comfortable with the distinctions. SDP 2. Functional Unit While we list only the functional unit as an SDP, the ISO Standard requires a discussion of the function of the product system as well. A product system (as defined in ISO 14040:2006 and expanded upon below) is a collection of processes that provide a certain function. The function represents the performance characteristics of the product system, or in layman's terms, "what does it do?" A power plant is a product system that has a function of generating electricity. The function of a Christmas tree product system is presumably to provide Christmas joy and celebrate a holiday. The function of a restroom hand dryer is drying hands. The function of a light bulb is providing light. In short, describing the function is pretty straightforward, but is done to clarify any possible confusions or assumptions that one might make from otherwise only discussing the product system itself. The functional unit, on the other hand, must be a clearly and quantitatively defined measure relating the function to the inputs and outputs to be studied. Unfortunately, that is all the description the ISO Standard provides. This ambiguity is partly the reason why the expressed functional units of studies are often inappropriate. A functional unit should quantify the function in a way that makes it possible to relate it to the relevant inputs and outputs (imagine a ratio representation). As discussed in Chapter 1, inputs are items like energy or resource use, and outputs are items like emissions or waste produced. You thus need a functional unit that bridges the function and the inputs or outputs. Your functional unit should explicitly state units (as discussed in Chapter 2) and the results of your study will be normalized by your functional unit. Building on the examples above, a functional unit for a coal-fired power plant might be "one kilowatt-hour of electricity produced". Then, an input of coal could be described as "kilograms of coal per one kilowatt-hour of electricity produced (kg coal/kWh)" and a possible output could be stated as "kilograms of carbon dioxide emitted per kilowatt-hour of electricity produced (kg CO2/kWh)." For a Christmas tree the functional unit might be "one holiday season" because while one family may leave a tree up for a month and another family for only a week, both trees fulfill the function of providing Christmas joy for the holiday season. For a hand dryer it might be "one pair of hands dried". For a light bulb it might be "providing 100 lumens of light for one hour (a. k. a. 100 lumen-hours)". All of these are appropriate because they discuss the function quantitatively and can be linked to study results. Figure 4-3 summarizes the bridge between function, functional units, and possible LCI results for the four product systems discussed. While not explicit to function, you could Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
84
Chapter 4: The ISO LCA Standard – Goal and Scope
have a study where your functional unit was "per widget produced" which would encompass the cradle to gate system of making a product. Product System
Function
Functional Unit
Example LCI Results
Power Plant
Generating electricity
1 kWh of electricity generated
kg CO2 per kWh
Christmas Tree
Providing holiday joy
1 undecorated tree over 1 holiday season
MJ energy per undecorated tree per holiday season
Hand Dryer
Drying hands
1 pair of hands dried
MJ energy per pair of hands dried
Light Bulb
Providing light
100 lumens light for 1 g Mercury per 100 hour (100 lumen-hrs) lumen-hrs Figure 4-3: Linkages between Function, Functional Unit, and Example LCI Results for hypothetical LCA studies
Now that we have provided some explicit discussion of functional units, we digress to discuss common problems with statements of functional units in studies. One common functional unit problem is failure to express the function quantitatively or without units. Often, suggested functional units sound more like a function description, e.g., for a power plant "the functional unit is generating electricity". This cannot be a viable functional unit because it is not quantitative and also because no unit was stated. Note that the units do not need to be SI-type units. The unit can be a unique unit relevant only for a particular product system, as in "1 pair of hands dried". Another common problem in defining a study's functional unit is confusing it with the inputs and outputs to be studied. For example, "tons of CO2" may be what you intend to use in your inventory analysis, but is not an appropriate functional unit because is not measuring the function, it is measuring the greenhouse gas emission outputs of the product system. Likewise, it is not appropriate to have a functional unit of "kg CO2 per kWh" because the CO2 emissions, while a relevant output, have nothing to do with the expression of the function. Further, since results will be normalized to the functional unit, subsequent emissions of greenhouse gas emissions in such a study would be "kg CO2 per kg CO2 per kWh", which makes no sense. Thus, product system inputs and outputs have no place in a functional unit definition. For LCA studies that involve comparisons of product systems, choices of functional units are especially important because the functional unit of the study needs to be unique and Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
85
consistent across the alternatives. For example, an LCA comparing fuels needs to compare functionally equivalent units. It would be misleading to compare a gallon of ethanol and a gallon of gasoline (i.e., a functional unit of gallon of fuel), because the energy content of the fuels is quite different (gasoline is about 115,000 BTU/gallon while ethanol (E100) is about 75,000 BTU/gallon). In terms of function or utility, you could drive much further with a gallon of gasoline than with ethanol. You could convert to gallons of gasoline equivalent (GGE) or perhaps use a functional unit based on energy content (such as BTU) of fuel. Likewise, if comparing coal and natural gas to make electricity, an appropriate functional unit would be per kWh or MWh, not per MJ of fuel. Hopefully it is clear that using an inappropriate function or functional unit could lead to lots of wasted effort if a study were later reviewed and found to be faulty. If you were to use functional units that, for example, had no actual units, you would create results that were not normalized to anything. Having to go back and correct that after a study is done is effectively an entirely new study. SDPs 3 and 4. Product System and System Boundary Before discussing an ISO LCA product system, we first discuss products, which can be any kind of good or service. This could mean a physical object like a component part, or software or services. Processes, similarly are activities that transform inputs to outputs. As already mentioned, an ISO LCA product system is the definition of the relevant processes and flows related to the chosen product life cycle that lead to one or more functions. Even virtual products like software (or cloud services) have many processes needed to create them. Products are outputs of such systems, and a product flow represents the connection of a product between product systems (where it may be an output of one and an input of another). For example, the product output of a lumber mill process—wood planks—may be an input to a furniture manufacturing process. Similarly, petroleum is the product output of an oil extraction process and may be an input into a refinery process that has product outputs like gasoline and diesel fuels. A product system is comprised of various subcomponents as defined below, but is generally comprised of various processes and flows. The system boundary notes which subset of the overall collection of processes and flows of the product system are part of the study, in accordance with the stated study goals. While not required, a diagram is crucial in helping the audience appreciate the complexity of the product system and its defined system boundary. The diagram is created by the study authors (although it may be generated by the LCA software used in the study). This diagram should identify the major processes in the system and then explicitly note the system boundary chosen, ideally with a named box "system boundary" around the processes included in the study. Alternatively, some color-coded representation could be used to Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
86
Chapter 4: The ISO LCA Standard – Goal and Scope
identify the processes and flows contained within the boundary. Even with a great product system diagram, the study should still discuss in detailed text the various processes and flows. Figure 4-4 shows the generic product system and system boundary example provided in ISO 14040:2006. If your study encompasses or compares multiple products, then you have to define several product systems.
Figure 4-4: ISO 14040 Product System and System Boundary example
There are a few key components of a product system diagram (also called a process flow diagram). Boxes in these diagrams represent various forms of processes, and arrows represent flows, similar to what might be seen in a mass balance or materials flow analysis. Boxes (or dashed lines) may represent system boundaries. At the highest level of generality (as in Figure 4-4) the representation of a product system may be such that the process boxes depicted correspond to entire aggregated life cycle stages (raw materials, production, use, etc.) as discussed in Chapter 1. In reality each of these aggregated stages may be comprised of many more processes, as we discuss below. Before going into more detail, it is worth discussing the art of setting a system boundary in more detail. Doing a complete LCA (one that includes every process in the product system) of a complicated product is impossible. An automobile has roughly 30,000 components. Tens of thousands of processes are involved in mining the ores, making the ships, trucks, and railcars used to transport the materials, refining the materials, making the components, and assembling the vehicle. A "complete" ISO LCA requires information on the materials and energy flows for each of these processes. Compiling and updating such detailed information for all of these processes and flows is all but impossible. Furthermore, each of Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
87
the processes directly involved in producing the components requires inputs from other processes. LCA models are able to capture direct and indirect effects of systems. In general, direct effects are those that happen directly as a result of activities in the process in question. Indirect effects are those that happen as a result of the activities, but outside of the process in question. For example, steel making requires iron ore and oxygen directly, but also electricity, environmental consulting, natural gas exploration, production, and pipelines, real estate services, and lawyers. Directly or indirectly, making cars involves the entire economy, and getting detailed mass and energy flows for the entire economy is impossible. Since conducting a complete LCA is impossible, what can we do? As we will see below, the ISO Standard provides for ways of simplifying our analyses so as not to require us to track every possible flow. But we still need to make key decisions (e.g., about stages to include) that can eventually lead to model simplifications. Focusing on the product itself while ignoring all other parts of the life cycle would lead to inaccurate and biased results, as shown in the example of the battery-powered car in Chapter 1. An LCA of a generic American passenger automobile was undertaken by representatives of the three major automobile manufacturing companies, aka the "big three", in the US in the mid-1990s. This study looked carefully at the processes for extracting ore and petroleum and making steel, aluminum, and plastic for use in vehicles. It also looked carefully at making the major components of a car and assembling the vehicle. Given the complexity described above, the study was forced to compromise by selecting a few steel mills and plastics plants as "representative" of all plants. Similarly, only a few component and assembly plants were analyzed. Whether the selected facilities were really representative of all plants cannot be known. Finally, many aspects of a vehicle were not studied, such as much of the transportation of materials and fuels and "minor" components. Nonetheless, the study was two years in duration (with more than 10 person years of effort) and is estimated to have cost millions of dollars. Thus, system boundaries need to be justified. Beyond the visual display and description of the boundary used in the study, the author should also explain choices and factors that led to the boundary as finally chosen and used. As mentioned above, significant effort looking for data could fail, and a process may have to be excluded from a study. Such an outcome should be discussed when defining the boundary, and may lead otherwise skeptical readers to realize that a broader boundary was originally attempted but found to be too challenging. By justifying, you allow the audience to better appreciate some of the challenges faced and tradeoffs made in the study. Other justifications for system boundary choices may include statements about a process being assumed or found to have negligible impact, or in the case of a comparative study, that identical processes existed in both product systems and thus would not effect the comparison. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
88
Chapter 4: The ISO LCA Standard – Goal and Scope
Process Flows Product systems have elementary flows into and out of them. As defined by ISO 14044, elementary flows are "material or energy entering the system being studied that has been drawn from the environment without previous human transformation, or material or energy leaving the system being studied that is released into the environment without subsequent human transformation." Translating, that means pure flows that need no other process to represent them on the input or output side of the model. For the sake of discussion, assume that Figure 4-4 is the product system and boundary diagram for a mobile phone. The figure shows that the product system for the mobile phone as defined with its rectangular boundary has flows related to input products and elementary flows. The input product (on the left side of the figure) is associated with another product system and is outside of the system boundary. Likewise on the right side of the figure, the mobile phone "product flow" is an input to another system. As an example, the left side of the figure product flow could represent that the mobile phone comes with paper instructions printed by a third party (but which are assumed to not be part of the study) and on the right side could be noting that the mobile phone as a device can be used in wireless voice and data network systems (the life cycles of such equipment also being outside the scope of the study). That's not to say that no use of phones is modeled, as Figure 4-4 has a "use phase" process box inside the boundary, but which may only refer to recharging of the device. The study author may have chosen the boundary as such because they are the phone manufacturer and can only directly control the processes and flows within the described boundary. As long as their goal and scope elements are otherwise consistent with the boundary, there are no problems. However, if, for example, the study goal or scope motivated the idea of using phones to make Internet based purchases for home delivery, then the current system boundary may need to be modified to consider impacts in those other systems, for example, by including the product system box on the right. Figure 4-4 might be viewed as implying that the elementary flows are not part of the study since they are outside of the system boundary. This is incorrect, however, because these elementary flows while not part of the system are the inputs and outputs of interest that may have motivated the study, such as energy inputs or greenhouse gas emission outputs. In short, they are in the study but outside of the system. Product system diagrams may be hierarchical. The high level diagram (e.g., Figure 4-4) may have detailed sub-diagrams and explanations to describe how other lower-level processes interact. These hierarchies can span multiple levels of aggregation. At the lowest such level, a unit process is the smallest element considered in the analysis for which input and output data are quantified. Figure 4-5 shows a generic interacting series of three unit processes that may be a subcomponent of a product system.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
89
Figure 4-5: Unit Processes (Source: ISO 14040:2006)
Figure 4-6 gives an example of how one might detail the high level "Waste Treatment" process from Figure 4-4 in the manner of Figure 4-5, where the unit processes are one of the three basic steps of collecting, disassembling, and sorting of e-waste. Additional unit processes (not shown) could exist for disposition of outputs.
Figure 4-6: Process Diagram for E-waste treatment
It is at the unit process level, then, that inputs and outputs actually interact with the product system. While already defined in Chapter 1, ISO specifically considers them as follows. Inputs are "product, material or energy flows that enter a unit process" and may include raw materials, intermediate products and co-products. Outputs are "products, material or energy flows that leave a unit process" and may include raw materials, intermediate products, Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
90
Chapter 4: The ISO LCA Standard – Goal and Scope
products, and releases e.g., emissions, and waste. Raw materials are "primary or secondary material that is used to produce a product" and waste is "substances or objects to be disposed of. Intermediate products flow between unit processes (such as cumulatively assembled components). Co-products are two or more products of the same process or system. The overall inputs and outputs to be measured by the study should be elementary flows. This is why "electricity" is not typically viewed as an input, i.e., it has not been drawn from the environment without transformation. Electricity represents coal, natural gas, sunlight, or water that has been transformed by generation processes. "MJ of energy" on the other hand could represent untransformed energy inputs. In the Christmas tree LCA mentioned above, which compares artificial and natural trees, the following text was used (in addition to a diagram): "For the artificial tree the system boundary includes: (1) cradle-to-gate material environmental impacts; (2) the production of the artificial tree with tree stand in China; (3) transportation of the tree and stand to a US retailer, and subsequently a customer's home; and (4) disposal of the tree and all packaging." SDP 5. Inventory Inputs and Outputs The definition of your study needs to explicitly note the inputs and/or outputs you will be focusing on in your analysis. That is because your analysis does not need to consider the universe of all potential inputs and outputs. It could consider only inputs (e.g., an energy use footprint), only outputs (e.g., a carbon emissions footprint), or both. The input and output specification part of the scope is not explicitly defined in the ISO Standard. It is presumably intended to be encompassed by the full product system diagram with labeled input and output flows. Following the example above, your mobile phone study could choose to track inputs of water, energy, or both, but needs to specify them. By explicitly noting which inputs and/or outputs you will focus on, it helps the audience better understand why you might have chosen the selected system boundary, product system, functional unit, etc. If you fail to explicitly note which quantified inputs and outputs you will consider in your study (or, for example, draw a generic product system diagram with only the words "inputs" and "outputs") then the audience is left to consider or assume for themselves which are appropriate for your system, which could be different than your intended or actual inputs and outputs. Chapter 5 discusses the inventory analysis component of LCA in more detail. SDP 6. Impact Assessment ISO 14040 requires you to explicitly list "the impact categories selected and methodology of impact assessment, and subsequent interpretation to be used". While we save more detailed discussion of impact assessment for Chapter 12, we offer some brief discussion and
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
91
examples here so as to help motivate how and why your choice of impact assessment could affect your other SDP choices. As we discussed in Chapter 1, there is a big difference between an inventory (accounting) of inputs and outputs and the types of impacts they can have. While we may track input and output use of energy and/or greenhouse gas emissions, the impacts of these activities across our life cycle could be resource depletion, global warming, or climate change. In impact assessment we focus on the latter issues. Doing so will require us to use other methods that have been developed in conjunction with LCA to help assess impacts. Specifically, there are impact assessment methods to consider cumulative energy demand (CED) and to assess the global warming potential (GWP) of emissions of various greenhouse gases. If we chose to consider these impacts in our study, then we explicitly state them and the underlying methods in the SDP. Again, the point of doing so explicitly is to ensure that at a glance a reader can appreciate decisions that you have made up front before having to see all of your study results. There are other required elements for the goal and scope, as noted above, but the SDPs are the most important and time consuming. They are the scope elements that need to be most carefully worded and evaluated. A Final Word On Comparative Assertions And Public Studies Comparative studies can only be done if the life cycle models created for each compared product use the same study design parameters, such as the same goal and scope, functional unit, and system boundary. The ISO Standard in various places emphasizes what needs to be done if you are going to make comparative assertions. By making such assertions you are saying that applying the ISO Standard has allowed you to make the claim. For example, ISO requires that for comparative assertions, the study must be an LCA and not simply an LCI, and that a special sensitivity analysis is done. The additional rules related to when you intend to make comparative assertions are in place both to ensure high quality work and to protect the credibility of the Standard. If several high visibility studies were done without all of these special considerations, and the results were deemed to be suspicious, the Standard itself might be vulnerable to criticism. Similarly, ISO requires an LCA to be peer reviewed if the comparative results are intended for public release. This means that a team of experts (typically three) needs to review the study, write a report of its merits, and assess whether it is compliant with the ISO Standard (i.e., whether all of the goal, scope, etc., elements have been done in accordance with what is written in the Standard). The vast majority of completed LCAs are not seen by the public, and therefore have not been peer-reviewed. That does not mean they are not ISO compliant, just that they have not been reviewed as such and designated as compliant. We will discuss more issues about peer review in Chapter 13. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
92
Chapter 4: The ISO LCA Standard – Goal and Scope
E-resource: On the www.lcatextbook.com website, in the Chapter 4 folder, is a spreadsheet listing publicly available LCA studies from around the world for many different products. Amongst other aspects, this spreadsheet shows whether studies were peer reviewed (which is interesting because they have all been "released to the public" but not all have been peer reviewed). PDF files of most of the studies listed are also available. The icon to the left will be used in the remainder of the book to designate resources available on the textbook website. Readers are urged to read one or more of these public studies that are of interest to them as a means of becoming familiar with LCA studies.
Chapter Summary The ISO LCA Standard is an internationally recognized framework for performing life cycle assessment, and has been developed and revised over time to guide practitioners towards making high-quality LCA studies. Any LCA practitioner should first read and know the requirements of the Standard. This chapter has focused on a subset of the Standard, namely the so-called study design parameters (SDPs) which comprise the main high level variables for a study and which when presented allow the audience to quickly appreciate the goals and scope of the study. The chapter focused on practical examples of SDPs from actual studies and seeks to demonstrate the importance of the bridge between product systems and their functional units and LCI results. When the integrity of this bridge is maintained, and common mistakes avoided, high-quality results can be expected.
References for this Chapter ISO 2013 http://www.iso.org/iso/home/standards_development.htm, February 1, 2013.
last
accessed
PE Americas, "Comparative Life Cycle Assessment of an Artificial Christmas Tree and a Natural Christmas Tree", November 2010. http://www.christmastreeassociation.org/pdf/ACTA%20Christmas%20Tree%20LCA%20 Final%20Report%20November%202010.pdf Life Cycle Assessment: Principles And Practice, United States Environmental Protection Agency, EPA/600/R-06/060, May 2006.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
93
End of Chapter Questions 1. Consider the following examples of goal statements for three different hypothetical LCA studies. Answer the questions (a-b) for each goal statement below. •
"The goal of this study is to find the energy use of making ice cream."
•
"The goal of this study is to produce an LCA for internal purposes."
•
"This study seeks to do a life cycle assessment of a computer to be used for future design efforts."
a. Briefly discuss the ISO compliance of the stated goal as written. b. Propose revisions if needed for the hypothetical goal statement to meet ISO requirements. 2. Consider the examples of study design parameters (SDPs) for four hypothetical LCA studies in the table below. Assess the partially provided entries in the table, and fill in or correct the rest of the columns for each product system with examples of relevant SDPs that bridge the various elements of the study using appropriate values (i.e., correct a functional unit that seems inappropriate). Product System
Function
Printed book
Collect 100 pages of printed text
Portable flash memory drive
Storing electronic content
Functional Unit
energy per gigabyte
E-book reader Automobile
LCI Results
GHG per reader bought 1 mile driven
3. Draw a product system diagram for a paper clip labeling inputs, outputs, intermediate flows, etc., as in Figure 4-4.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
94
Chapter 4: The ISO LCA Standard – Goal and Scope
4. Draw a product system diagram for the purchase of an airplane ticket via an electronic commerce website, labeling inputs, outputs, intermediate flows, etc., as in Figure 4-4. 5. Read one of the LCA studies found by using the E-resource link at the end of the chapter. Summarize the study design parameters of the chosen study, and discuss any discrepancies or problems found, and how they could be improved.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 4: The ISO LCA Standard – Goal and Scope
PAGE ADDED TO PRESERVE FORMATTING
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
95
96
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Chapter 5 : Data Acquisition and Management for Life Cycle Inventory Analysis Now that the most important elements of the LCA Standard are better understood, we can begin to think about the work needed to get data for your study. In this chapter, we introduce the inventory analysis phase of the LCA Standard, as well acquiring and using data needed for the inventory phase of an LCA or an LCI study. As data collection, management, and modeling are typically the most time-consuming components of an LCA, understanding how to work with data is a critical skill. We build on concepts from Chapter 2 in terms of referencing and quantitative modeling. Improving your qualitative and quantitative skills for data management will enhance your ability to perform great LCAs. While sequentially this chapter is part of the content on process-based life cycle assessment, much of the discussion is relevant to LCA studies in general.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: •
Describe the workflow of the life cycle inventory phase
•
Recognize how challenges in data collection may lead to changes in study design parameters (SDPs), and vice versa
•
Map information from LCI data modules into a unit process framework
•
Explain the difference between primary and secondary data, and when each might be appropriate in a study
•
Document the use of primary and secondary data in a study
•
Create and assess data quality requirements for a study
•
Perform an interpretation analysis on LCI results
•
Extract data and metadata from LCI data modules and use them in support of a product system analysis
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
97
ISO Life Cycle Inventory Analysis After reviewing the ISO LCA Standard and its terminology in Chapter 4, you should be able to envision the level and type of effort needed to perform an inventory analysis of a chosen product system. Every study using the ISO Standard has an inventory analysis phase, but as discussed above, many studies end at this phase and are called LCI studies. Those that continue on to impact assessment are LCAs. That does not mean that LCI studies have better inventory analyses than LCAs, in fact LCAs may require more comprehensive inventory analyses to support the necessary impact assessment. Figure 5-1, developed by the US EPA, highlights the types of high-level inputs and outputs that we might care to track in our inventory analysis. As originally mentioned in Chapter 1, we may be concerned with accounting for material, energy, or other resource inputs, and product, intermediate, co-product, or release outputs. Recall that based on how you define your goal, scope, and system boundary, you may be concerned with all or some of the inputs and outputs defined in Figure 5-1.
Figure 5-1: Overview of Life Cycle Assessment (Source: US EPA 1993)
Inventory analysis follows a straightforward and repeating workflow, which involves the following steps (as taken from ISO 14044:2006) done as needed until the inventory analysis matches the then-current goal and scope: •
Preparation for data collection based on goal and scope
•
Data Collection
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
98
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
•
Data Validation (do this even if reusing someone else's data)
•
Data Allocation (if needed)
•
Translating Data to the Unit Process
•
Translating Data to the Functional Unit
•
Data Aggregation
As the inventory analysis process is iterated, the system boundary and/or goal and scope may be changed (recall the two-way arrows in Figure 4-1). The procedure is as simple as needed, and gets more complex as additional processes and flows are added. Each of the inventory analysis steps are discussed in more detail below, with brief examples for discussion. Several more detailed examples are shown later in the chapter. Step 1 - Preparation for data collection based on goal and scope The goal and scope definition guides which data need to be collected (noting that the goal and scope may change iteratively during the course of your study and thus may cause additional data collection effort or previously collected data to be discarded). A key consideration is the product system diagram and the chosen system boundary. The boundary shows which processes are in the study and which are not. For every unit process in the system boundary, you will need to describe the unit process and collect quantitative data representing its transformation of inputs to outputs. For the most fundamental unit processes that interface at the system boundary, you will need to ensure that the inputs and outputs are those elementary flows that pass through the system boundary. For other unit processes (which may not be connected to those elementary flow inputs and outputs) you will need to ensure they are connected to each other through non-elementary flows such as intermediate products or co-products. When planning your data collection activities, keep in mind that you are trying to represent as many flows as possible in the unit process shown in Figure 5-2. Choosing which flows to place at the top, bottom, left, or right of such a diagram is not relevant. The only relevant part is ensuring inputs flow into and outputs flow out of the unit process box. You want to quantitatively represent all inputs, either form nature or from the technosphere (defined as the human altered environment, thus flows like products from other processes). By covering all natural and human-affected inputs, you have covered all possible inputs. You want to quantitatively represent outputs, either as products, wastes, emissions, or other releases. Inputs from nature will come from resources in the ground or water. Outputs to nature will be in the form of emissions or releases to "compartments" in the ground, air, or water.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
99
Figure 5-2: Generalized Unit Process Diagram
As a tangible example, imagine a product system like the mobile phone example in Chapter 4 where we have decided that the study should track water use as an input. Any of the unit processes within the system boundary that directly uses water will need a unit process representation with a quantity of water as an input and some quantitative measure of output of the process. For mobile phones, such processes that use water as a direct input from nature may include plastic production, energy production, and semiconductor manufacturing. Other unit processes within the boundary may not directly consume water, but may tie to each other through flows of plastic parts or energy. They themselves will not have water inputs, but by connecting them all together, in the end, the water use of those relevant sectors will still be represented. The final overall accounting of inventory inputs and/or outputs across the life cycle within the system boundary is called a life cycle inventory result (or LCI result). The unit process focus of LCA drives the need for data to quantitatively describe the processes. If data is not available or inaccessible, then the product system, system boundary, or goal may need to be modified. Data may be available but found not to fit the study. For example, an initial system boundary may include a waste management phase, but months of effort could fail to find relevant disposition data for a specific product of the process. In this case, the system boundary may need to be adjusted (made smaller) and other SDPs edited to represent this lack of data in the study. On the other hand, data that is assumed to not be available at first may later be found, which would allow an expansion of the system boundary. In general, system boundaries are made smaller not larger over the course of a study. Step 2 - Data Collection For each process within the system boundary, ISO requires you to "measure, calculate, or estimate" data to quantitatively represent the process in your product system model. In LCA, the "gold standard" is to collect your own data for the specific processes needed, called primary data collection. This means directly measuring inputs and outputs of the process
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
100
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
on-site for the specific machinery use or transformation that occurs. For example, if you required primary data for energy use of a process in an automobile assembly line that fastens a component on to the vehicle with a screw, you might attach an electricity meter to the piece of machinery that attaches the screw. If you were trying to determine the quantity of fuel or material used in an injection molding process, you could measure those quantities as they enter the machine. If you were trying to determine the quantity of emissions you could place a sensor near the exhaust stack. If you collect data with methods like this, intended to inventory per-unit use of inputs or outputs, you need to use statistical sampling and other methods to ensure you generate statistically sound results. That means not simply attaching the electricity meter one time, or measuring fuel use or emissions during one production cycle (one unit produced). You should repeat the same measurement multiple times, and perhaps on multiple pieces of identical equipment, to ensure that you have a reasonable representation of the process and to guard against the possibility that you happened to sample a production cycle that was overly efficient or inefficient with respect to the inputs and outputs. The ISO Standard gives no specific guidance or rules for how to conduct repeated samples or the number of samples to find, but general statistical principles can be used for these purposes. Your data collection summary should then report the mean, median, standard deviation, and other statistical properties of your measurements. In your inventory analysis you can then choose whether to use the mean, median, or a percentile range of values. Note that many primary data collection activities cannot be completed as described above. It may not be possible to gain access to the input lines of a machine to measure input use on a per-item processed basis. You thus may need to collect data over the course of time and then use total production during that time to normalize the unit process inventory. For the examples in the previous paragraph, you might collect electricity use for a piece of machinery over a month and then divide by the total number of vehicles that were assembled. Or you may track the total amount of fuel and material used as input to the molding machine over the course of a year. In either case, you would end up with an averaged set of inputs and/or outputs as a function of the product(s) of the unit process. The same general principles discussed above apply here with respect to finding multiple samples. In this case you could find several monthly values or several yearly values to find an average, median, or range. The ISO Standard (14044:2006, Annex A) gives examples of "data collection sheets" that can support your primary data collection activities. Note that these are only examples, and that your sheets may look different. The examples are provided to ensure, among other things, that you are recording quantities and units, dates and locations of record keeping, and descriptions of sampling done. The most likely scenario is that you will create electronic data collection sheets by recording all information in a spreadsheet. This is a fair choice because from our perspective, Microsoft Excel is the most popularly used software tool in support of LCA. Even practitioners using other advanced LCA software packages still typically use Microsoft Excel for data management, intermediate analysis, and graphing. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
101
Collecting primary data can be difficult or impossible if you do not own all the equipment or do not have direct access to it either due to geographical or organizational barriers. This is often the case for an LCA consultant who may be tasked with performing a study for a client but who is given no special privileges or access to company facilities. Further, you may need to collect data from processes that are deemed proprietary or confidential by the owner. This is possible in the case of a comparative analysis with some long-established industry practice versus a new technology being proposed by your client or employer. In these cases, the underlying data collection sheets may be confidential. Your analysis may in these cases only "internally use" the average data points without publicly stating the quantities found in any subsequent reports. If the study is making comparative assertions, then it may be necessary to grant to third-party reviewers (who have signed non-disclosure agreements) access to the data collection sheets to appreciate the quality of the data and to assess the inventory analysis done while maintaining overall confidentiality. Beyond issues of access, while primary data is considered the "gold standard" there are various reasons why the result may not be as good as expected in the context of an LCA study. First, the data is only as good as the measurement device (see accuracy and precision discussion in Chapter 2). Second, if you are not able to measure it yourself then you outsource the measurement, verification, and validation to someone else and trust them to do exactly as you require. Various problems may occur, including issues with translation (e.g., when measuring quantities for foreign-owned or contracted production) or not finding contacts with sufficient technical expertise to assist you. Third, you must collect data on every input and output of the process relevant to your study. If you are using only an electric meter to measure a process that also emits various volatile organic compounds, your collected data will be incomplete with respect to the full litany of inputs and outputs of the process. Your inventory for that process would undercount any other inputs or outputs. This is important because if other processes in your system boundary track volatile organics (or other inputs and outputs) your primary data will undercount the LCI results. The alternative to primary data collection is to use secondary data (the "calculating and estimating" referenced above). Broadly defined, secondary data comes from life cycle databases, literature sources (e.g., from searches of results in published papers), and other past work. It is possible you will find data closely, but not exactly, matching the required unit process. Typical tradeoffs to accessibility are that the secondary data identified is for a different country, a slightly different process, or averaged across similar machinery. That does not mean you cannot use it – you just need to carefully document the differences between the process data you are using and the specific process needed in your study. While deemed inferior given the use of the word secondary, in some cases secondary data may be of comparable or higher quality than primary data. Secondary data is typically discoverable because it has been published by the original author who generated it as primary data for their own study (and thus is typically of good quality). In short, one analyst's primary data may be another's secondary data. Again, the "secondary" designation is simply recognition
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
102
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
that it is being "reused" from a previously existing source and not collected new in your own study. Many credible and peer reviewed studies are constructed mostly or entirely of secondary data. More detail on identifying and using secondary data sources like LCI databases is below. For secondary data, you should give details about the secondary source (including a full reference), the timestamp of the data record, and when you accessed it. In both cases you must quantitatively maintain the correct units for the inputs and outputs of the unit process. While not required, it is convenient to make tables that neatly summarize all of this information. Regardless of whether your data for a particular process comes from a primary or secondary source, the ISO Standard requires you to document the data collection process, give details on when data have been collected, and other information about data quality. Data quality requirements (DQRs) are required scope items that we did not discuss in Chapter 4 as part of the SDP, but characterize the fundamental expectations of data that you will use in your study. As specified by ISO 14044:2006, these include statements about your intentions with respect to age of data, geographical reach, completeness, sources, etc. Data quality indicators are summary metrics used to assess the data quality requirements. For example, you may have a data quality requirement that says that all data will be primary, or at least secondary but from peer-reviewed sources. For each unit process, you can have a data quality indicator noting whether it is primary or secondary, and whether it has been peer-reviewed. Likewise, you may have a DQR that says all data will be from the same geographical region (e.g., a particular country like the US or a whole region like North America). It is convenient to summarize the DQRs in a standardized tabular form. The first two columns of Figure 5-3 show a hypothetical DQR table partly based on text from the 2010 Christmas tree study mentioned previously. The final column represents how the requirements might be indicated as a summary in a completed study. The indicated values are generally aligned with the requirements (as they should be!). Data Quality Category Temporal Geographical Technological
Requirement
Data Quality Indicator Artificial trees: 2009 data Natural trees: 2002-2009 data Data within 10 years of study Artificial trees: China Natural trees: US Data matches local production All processes used in study are Most common production process representative of most common practices basis Figure 5-3: Sample Data Quality Requirements (DQR) Table
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
103
Beyond using primary or secondary data, you might need to estimate the parameters for some or all of the input and outputs of a unit process using methods as introduced in Chapter 2. Your estimates may be based on data for similar unit processes (but which you deem to be too dissimilar to use directly), simple transformations based on rules of thumb, or triangulated averages of several unit processes. From a third-party perspective, estimated data is perceived as lower quality than primary or secondary sources. However when those sources cannot be found, estimating may be the only viable alternative. Example 5-1: Estimating energy use for a service Question: Consider that you are trying to generate a unit process associated with an internal corporate design function as part of the life cycle "overhead" of a particular product and given the scope of your study need to create an input use of electricity. Your company is all located in one building. There is no obvious output unit for such a process, so you could define it to be per 1 product designed, per 1 square foot of design space, etc., as convenient for your study. Answer: You could estimate the input electricity use for a design office over the course of a year and then try to normalize the output. If you only had annual electricity use for the entire building (10,000 kWh), and no special knowledge about the energy intensity of any particular part of the building as subdivided into different functions, you could find the ratio of the total design space in square feet (2,000 sf) as compared to the total square feet of the building (50,000 sf), and use that ratio (2/50) to scale down the total consumption to an amount used for design over the course of a year (400 kWh). If your output was per product, you could then further normalize the electricity used for the design space by the unique number of products designed by the staff in that space during the year.
You could add consideration of non-electricity use of energy (e.g., for heating or cooling) with a similar method. Note that such ancillary support services like design, research and development, etc., generally have been found to have negligible impacts, and thus many studies exclude these services from their system boundaries. Step 3 - Data Validation Chapter 2 provided some general guidance on validating research results. With respect to validating LCI data, you generally need to consider the quantitative methods used and ensure that the resulting inventories meet your stated DQRs. Data validation should be done after data is collected but before you move on to the actual inventory modeling activities of your LCA.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
104
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
As an example of validation, it may be useful to validate energy or mass balances of your processes. Using the injection molding process example from Step 2, one would expect that the total input mass of material to be greater than (but approximately equal to) the output mass of molded plastic. You can ensure that the total mass input of plastic resin, fuels, etc., is roughly comparable to the mass of molded plastic (subject to reasonable losses). If the balances are deemed uneven, you can assess whether the measured process is merely inefficient or whether there is a problem in your data collection, and thus resample. You can use available secondary data to validate primary data collection. If you have chosen to collect your own data for a process that is similar to processes for which there is already secondary data available, you can quantitatively compare your measured results with the published data. Again, if there are significant differences then you will need to determine the source of the discrepancy. You can validate secondary data that you have chosen to use against other sources in similar ways. The results of validation efforts can be included in the main text of your report or in an included Appendix, depending on the level of detail and explanation needed. If you collected primary data and compared it to similar data from the same industry, the following text might be included to show this: "Collected data from the year 2012 on the technology-specific process used in this study was compared to secondary data on the similar injection molding process from 2005 (Reference). The mean of collected data was about 10% lower than the secondary data. This difference is not significant, and so the collected data is used as the basis for the process in the study." If validation suggests the differences are more substantial, that does not automatically mean that the data is invalid. It is possible that there are no good similar data sources to compare against, or that the technology has changed substantially. That too could be noted in the study, such as: "Collected data from the year 2012 on the technology-specific process used in this study was compared to secondary data on the similar injection molding process from 2005 (Reference). The mean of collected data was about 50% lower than the secondary data. This difference is large and significant, but is attributed to the significant improvements in the industry since 2005, and so the collected data is still chosen as the basis for the process in the study." As noted above, the validation step is where you re-assess whether the quantitatively sound data you want to use also is within the scope of your DQRs. Many studies state DQRs to use all primary data at the outset, but subsequently realize it is not possible. Likewise studies may not be able to find sufficient geographically focused data. In both cases, the DQRs would need to be iteratively adjusted as the study continues. This constant refining of the
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
105
initial goal and scope may sound like "cheating", but the purpose of the DQRs is as a single and convenient summary of the study goals for data quality. It allows a reader to quickly get a sense of how relevant the study results are given the final DQRs. While not required, you can state initial goal DQRs alongside final DQRs upon completion of the study. Step 4 - Data Allocation (if needed) Allocation will be discussed more in Chapter 6, but in short, allocation is the quantitative process done by the study analyst to assign specific quantities of inputs and outputs to the various products of a process based on some mathematical relation between the products. For example, you may have a process that produces multiple outputs, such as a petroleum refinery process that produces gasoline, diesel, and other fuels and oils. Refineries use a significant amount of energy. Allocation is needed to quantitatively connect the energy input to each of the refined products. Without specified allocation procedures, the connections between those inputs and the various products could be done haphazardly. The ISO Standard suggests that the method you use to perform the allocation should be based on underlying physical relationships (such as the share of mass or energy in the products) when possible. For example, if your product of interest is gasoline, you will need to determine how much of the total refinery energy was used to make the gasoline. For a mass allocation, you could calculate it by using the ratio of the mass of the gasoline produced to the total mass of all of the products. You may have to further research the energetics of the process to determine what allocation method is most appropriate. If physical relationships are not possible, then methods such as economic allocation—such as by eventual sale price— could be used. ISO also says that you should consistently choose allocation methods as much as possible across your product system, meaning that you should try not to use a mass-based allocation most of the time and an energy-based allocation some of the time. This is because mixing allocation methods could be viewed by your audience or reviewers as a way of artificially biasing the results by picking allocations that would provide low or high results. Allocation is conceptually similar to the design space electricity Example 5-1. Most allocations are just linear transformations of effects. When performing allocation, the most important considerations are to fully document the allocation method chosen (including underlying allocation factors) and to ensure that total inputs and outputs are equal to the sum of the allocated inputs and outputs. It is possible that none of your unit processes have multiple products, and thus you do not need to perform allocation. You might also be able to avoid allocation entirely, as we will see later. Step 5 - Translating Data to the Unit Process In this step you convert the various collected data into a representation of the output of the unit process. Regardless of how you have defined the study overall, this step requires you to collect all of the inputs and outputs as needed for 1 unit output from that process. From
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
106
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Example 5-1, you would ensure that the electricity input matched the unit basis of your product flow (e.g., per 1 product designed). This result also needs to be validated. Step 6 - Translating Data to the Functional Unit The reason why this step is included in the ISO LCA Standard is to remind you that you are doing an overall study on the basis of 1 functional unit of product output. Either during the data collection phase, or in subsequent analysis, you will need to do a conversion so that the relative amount of product or intermediate output of the unit process is related to the amount needed per functional unit. Eventually, all of your unit process flows will need to be converted to a per-functional unit basis. If all unit processes have been so modified, then finding the total LCI results per functional unit is a trivial procedure. From Example 5-1, the design may be used to eventually produce 1 million of the widgets. The electricity use for one product design must be distributed to the 1 million widgets so that you will then have the electricity use for a single widget in the design phase (a very small amount). This result also needs to be validated. Step 7 - Data Aggregation In this step, all unit process data in the product system diagram are combined into a single result for the modeled life cycle of the system. What this typically means is summing all quantities of all inputs and outputs into a single total result on a functional unit basis. Aggregation occurs at multiple levels. Figure 4-4 showed the various life cycle stages within the view of the product system diagram. A first level of aggregation may add all inputs and outputs under each of the categories of raw material acquisition, use, etc. A second level of aggregation may occur across all of these stages into a final total life cycle estimate of inputs and outputs per functional unit. Aggregated results are often reported in a table showing total inputs and outputs on per-process, or per stage, values, and then a sum for the entire product system. Example 5-2 shows aggregated results for a published study on wool from sheep in New Zealand. The purpose of such tables is to emphasize category level results, such as that half of the life cycle energy use occurs on farm. Results could also be graphed. Example 5-2: Aggregation Table for Published LCA on Energy to Make Wool (Source: The AgriBusiness Group, 2006) Life Cycle Stage
Energy Use (GJ per tonne wool)
On Farm
22.6
Processing
21.7
Transportation
1.5
Total
45.7
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
107
Beyond such tables, product system diagrams may be annotated with values for different levels of aggregation by adding quantities per functional unit. Example 5-3 shows a diagram for a published study on life cycle effects of bottled water and other beverage systems performed for Nestle Waters. Such values can then be aggregated into summary results. Example 5-3: Aggregation Diagram for Bottled Water (Source: Quantis, 2010)
We have above implied that aggregation of results occurs over a relatively small number of subcomponents. However, a product system diagram may be decomposed into multiple sets of tens or hundreds of constituent pieces that need to be aggregated. If all values for these subcomponents are on a functional unit basis, the summation is not difficult, but the bookkeeping of quantities per subcomponent remains an issue. If the underlying subcomponent values are not consistently on a per functional unit basis, units of analysis should be double checked to ensure they can be reliably aggregated.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
108
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Life Cycle Interpretation Because some studies only include an inventory (LCI), we discuss Interpretation, the final step for all LCAs and LCIs, now. For those studies (LCAs) that also include an impact assessment, the procedures for the assessment will be discussed in Chapter 10). There is little detail provided in the ISO Standard on what must be done in this phase, but in short, interpretation is similar to the last step of the "three step" method introduced in Chapter 2. The interpretation phase refers to studying the results of the goal and scope, inventory analysis, and impact assessment, in order to make conclusions and recommendations that can be reported. As shown in Figure 4-1, interpretation is iterative with the three other phases. As this chapter is focused on inventory analysis, much of the discussion and examples provided relate to interpreting inventory results, but the same types of interpretation can be done with impact assessment results (to be discussed in Chapter 10). A typical first task in interpretation is to study your results to determine whether conclusions can be made based on the inventory results that are consistent with the goal and scope. One of the most common and important interpretation tasks involves discussing which life cycle stage leads to the largest share of LCI results. A high-level summary helps to set the stage for subsequent analyses. For example, an LCA of a vehicle will likely show that the use phase (driving the car) is the largest energy user, as compared to manufacturing and recycling. An interpretation task could involve creating a tabular or graphical summary showing the energy use contributions for each of the stages. Part of your goal statement may have been to do a comparison between two types of products and assess whether the life cycle energy use of one is significantly less than the other. If your inventory results for the two products are nearly identical (say only 1% different) then it may be difficult to scientifically conclude that one is better than the other given the various uncertainties involved. Such an interpretation result could cause you to directly state that no appreciable difference exists, or it may cause you to change the system boundary in a way that ends up making them significantly different. A key part of interpretation is performing relevant sensitivity analyses on your results. The ISO Standard does not require specific sensitivity analysis scenarios as part of interpretation, but some consideration of how alternative parameters for inputs, outputs, and methods used (e.g., allocation) would affect the final results is necessary. As discussed in Chapter 2, a main purpose of sensitivity analysis is to help assess whether a qualitative conclusion is affected by quantitative changes in the parameters of the study. For example, if your general qualitative conclusion is that product A uses significantly less energy than product B, the sensitivity analysis may test whether different quantitative assumptions related to A or B lead to results where energy use of A is roughly equal to B, or where A is greater than B. Any of the latter two outcomes is qualitatively different than the initial conclusion, and it would be important for the sensitivity results to be stated so that it was clear that there is a variable that if credibly changed by a specified amount, has the potential to alter the study conclusions.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
109
While on the subject of assessing comparative differences, it is becoming common for practitioners in LCA to use a "25% rule" when testing for significant differences. The 25% rule means that the difference between two LCI results, such as for two competing products, must be more than 25% different for the results to be deemed significantly different, and thus for one to be declared as lower than the other. While there is not a large quantitative framework behind the choice of 25% specifically, this heuristic is common because it roughly expresses the fact that all data used in such studies is inherently uncertain, and by forcing 25% differences, then relatively small differences would be deemed too small to be noted in study conclusions. We will talk more about modeling and assessing uncertainties in Chapter 11 on uncertainty. Interpretation can also serve as an additional check on the goal and scope parameters. This is where you could assess whether a system boundary is appropriate. As an example, while the ISO Standard encourages full life cycle stage coverage within system boundaries, it does not require that every LCA encompass all stages. One could try to defend the validity of a life cycle study of an automobile that focused only on manufacturing, or only on the use stage. The results of the interpretation phase could then internally weigh in on whether such a decision was appropriate given the study goal. If a (qualified) conclusion can be drawn, the study could be left as-is, if not, a broader system boundary could be chosen, with or without preliminary LCI results. Regardless, the real purpose of interpretation is to improve the quality of your study, especially the quality of the written conclusions and recommendations that arise from your quantitative work. As with other quantitative analysis methods, you will need to also improve your qualitative skills, including documentation, to ensure that your interpretation efforts are respected.
Identifying and Using Life Cycle Data Sources In support of modeling the inputs and outputs associated with unit processes, you will need a substantial amount of data. Even studies of simple product systems may require data on 10 different unit processes. While this may sound like a small amount of effort, as you will see below, the task of finding, documenting, manipulating, validating and using life cycle data is time consuming. The text above gave a fair amount of additional detail related to developing your own primary data via collection and sampling efforts. This section is related to the identification and use of secondary data. One prominent source of secondary data is the thousands of peer-reviewed journal papers done over time by the scientific community, also known as literature sources. Some of these papers have been explicitly written to be a source of secondary data, while authors of other papers developed useful data in the course of research (potentially on another topic)
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
110
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
and made the process-level details available as part of the paper or in its supporting information. Sometimes the study authors are not just teams of researchers, but industry associations or trade groups (e.g., those trying to disseminate the environmental benefits of their products). Around the world, industry groups like Plastics Europe, the American Chemistry Council, and the Portland Cement Association have sponsored projects to make process-based data available via publications. It is common to see study authors citing literature sources, and doing so requires you to simply use a standard referencing format like you would for any source. Unfortunately, data from such sources is typically not available in electronic form, and thus there are potentials for data entry or transcription errors as you try to make use of the published data. It is due to issues like these that literature sources constitute a relatively small share of secondary data used in LCA studies. There is a substantial amount of secondary data available to support LCAs in various life cycle databases. These databases are the main source of convenient and easy to access secondary data. Some of the data represented in these databases are from the literature sources mentioned above. Since the first studies mentioned in Chapter 1, various databases comprised of life cycle inventory data have been developed. The original databases were sold by Ecobilan and others in the mid-1990s. Nowadays the most popular and rigorously constructed database is from ecoinvent, developed by teams of researchers in Switzerland and available either by paying directly for access to their data website or by an add-on fee to popular LCA system tools such as SimaPro and GaBi (which in turn have their own databases). None of these databases are free, and a license must be obtained to use them. On the other hand, there are a variety of globally available and publicly accessible (free) life cycle databases. In the US, LCI data from the National Renewable Energy Laboratory (NREL)'s LCI database and the USDA's LCA Digital Commons are popular and free3. Figure 5-4 summarizes the major free and paid life cycle databases (of secondary data) in the world that provide data at the unit process level for use in life cycle studies. Beyond the individual databases, there is also an "LCA Data Search Engine," managed by the United Nations Environmental Programme (UNEP), that can assist in finding available free and commercial unit process data (LCA-DATA 2013). All of the databases have their own user's guides that you should familiarize yourself with before searching or using the data in your own studies.
Data from the US NREL LCI Database has been transferred over to the USDA LCA Digital Commons as of 2012. Both datasets can now be accessed from that single web database. 3
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Database
Approximate Cost
Number of processes
Notes
ecoinvent
2,500 Euros ($3,000 USD)
4,000+
Has data from around the world, but majority is from Europe. Available directly, or embedded within LCA software.
US NREL LCI Database
Free (companies, agencies pay to publish data)
600+
US focused. Now hosted by USDA LCA Digital Commons.
USDA LCA Digital Commons
Free (manufacturers and agencies pay to publish data)
300+
Focused on agricultural products and processes. Geographically specific unit processes for specific US states.
ELCD
Free
300+
Relatively few processes, spread across various sectors. Additional data being added rapidly.
BEES
Free
GaBi
$3,000 USD
111
Focused on building and construction materials. 5,000+
Database made by PE International. Global, but heavily focused on European data. Figure 5-4: Summary of Data Availability for Free and Licensed LCA Databases (Sources provided at end of chapter)
These databases can be very comprehensive, with each containing data on hundreds to thousands of unique processes, with each process comprised of details for potentially hundreds of input or output flows. Collecting the various details of inputs and outputs for a particular unit process (which we refer to as an LCI data module but which are referred to as "datasets" or "processes" by various sources) requires a substantial amount of time and effort. This embedded level of effort for unit process data is important because even though it represents a secondary data source, to create a superior set of primary data for a study, you might need to collect data for 100 or more input and output flows for the process. Of course your study may have a significantly smaller scope that includes only 5 flows, and thus your data collection activities would only need to measure those. The databases do highlight an ongoing conundrum in the LCA community – the naïve stated preference for primary data when substantial high-quality secondary data is pervasive. Another benefit of these databases is that subsets of the data modules are created and maintained consistently, thus a common set of assumptions or methods would be associated with hundreds of processes. This is yet another difference to primary data which could have a set of ad-hoc assumptions used in its creation.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
112
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Now that the availability of vast secondary data sources has been introduced, we discuss the data structures typical of these LCI data modules. As with many facets of LCA, there is a global standard for storing information in LCI data modules, known as EcoSpold. The EcoSpold format is a structured way of storing and exchanging LCI data, where details such as flows and allocation methods are classified for each process. There is no requirement that LCA tools use the EcoSpold format, but given its popularity and the trend that all of the database sources in Figure 5-4 use this format, it is worth knowing. Instead of giving details on the format (which is fairly technical and generally only useful for personnel involved in creating LCA software) we instead will demonstrate the way in which LCI data modules are typically represented in the database and allow you to think about the necessary data structures separately. In the rest of this chapter we consider an LCI of the CO2 emitted to generate 1 kWh of coalfired electricity in the United States. Our system boundary for this example (as in Figure 5-5) has only three unit processes: mining coal, transporting it by rail, and burning it at a power plant. The refinery process that produces diesel fuel, an input for rail, is outside of our boundary, but the effects of using diesel as a fuel are included. We can assume, beyond the fact that this is an academic example, that such a tight boundary is realistic because these are known to be significant parts of the supply chain of making coal-fired power. We will discuss the use of screening methods to help us set such boundaries in Chapter 8.
Figure 5-5: Product System Diagram for Coal-Fired Electricity LCI Example
To achieve our goal of the CO2 emitted per kWh, we will need to find process-level data for coal mining, rail transportation, and electricity generation. In the end, we will combine the
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
113
results from these three unit processes into a single estimate of total CO2 per kWh. This way of performing process-level LCA is called the process flow diagram approach. We will focus on the US NREL LCI database (2013) in support of this relatively simple example. This database has a built-in search feature such that typing in a process name or browsing amongst categories will show a list of available LCI data modules (see the Advanced Material at the end of this chapter for brief tutorials on using the LCA Digital Commons website, that hosts the US LCI data, as well as other databases and tools). Searching for "electricity" yields a list of hundreds of processes, including these LCI data modules: •
Electricity, diesel, at power plant
•
Electricity, lignite coal, at power plant
•
Electricity, natural gas, at power plant
•
Electricity, anthracite coal, at power plant
•
Electricity, bituminous coal, at power plant
The nomenclature used may be confusing, but is somewhat consistent across databases. The constituents of the module name can be deciphered as representing (1) the product, (2) the primary input, and (3) the boundary of the analysis. In each of the cases above, the unit process is for making electricity. The inputs are various types of fuels. Finally, the boundary is such that it represents electricity leaving the power plant (as opposed to at the grid, or at a point of use like a building). Once you know this nomenclature, it is easier to browse the databases to find what you are looking for specifically. Given the above choices, we want to use one of the three coal-fueled electricity generation unit processes in our example. Lignite and anthracite represent small shares of the generation mix, so we choose bituminous coal as the most likely representative process and use the last data module in the list above (alternatively, we could develop a weighted-average process across the three types that would be useful). Using similar brief search methods in the US NREL website we would find the following unit processes as relevant for the other two pieces of our system: •
Bituminous coal, at mine
•
Transport, train, diesel powered
These two processes represent mining of bituminous coal and the transportation of generic product by diesel-powered train.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
114
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Figure 5-6 shows an abridged excerpt of the US NREL LCI data module for Electricity, bituminous coal, at power plant. The entire data module is available publicly4. Within the US NREL LCI database website, such data is found by browsing or searching for the process name and then viewing the "Exchanges". These data modules give valuable information about the specific process chosen as well as other processes they are linked to. While here we discuss viewing the data on the website, it can also be downloaded to a Microsoft Excel spreadsheet or as XML. It is noted that this is an abridged view of the LCI data module. The complete LCI data module consists of quantitative data for 7 inputs and about 60 outputs. For the sake of the example in this section, we assume the abridged inventory and ignore the rest of the details. Most of the data modules in databases have far more inputs and outputs than in this abridged module; it is not uncommon to find data modules with hundreds of outputs (e.g., for emissions of combustion processes). If you have a narrow scope that focuses on a few air emission outputs, many of the other outputs can be ignored in your analysis. However if you plan to do life cycle impact assessment, the data in the hundreds of inputs and/or outputs may be useful in the impact assessment. If your study seeks to do a broad impact assessment, collecting your own primary data can be problematic as your impact assessment will all but require you to broadly consider the potential flows of your process. If you focus instead on just a few flows you deem to be important, then the eventual impact assessment could underestimate the impacts of your process. This is yet another danger of primary data collection (undercounting flows).
Data from the NREL US LCI database in this chapter are as of July 20, 2014. Values may change in revisions to the database that cannot be expressed here. 4
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Flow
Category
Type
Unit
Amount
bituminous coal, at mine
root/flows
ProductFlow
kg
4.42e-01
transport, train, diesel powered
root/flows
ProductFlow
t*km
4.61e-01
electricity, bituminous coal, at power plant
root/flows
ProductFlow
kWh
1.00
carbon dixoide, fossil
air/unspecified
ElementaryFlow
kg
9.94e-01
115
Comment
Inputs
Transport from mine to power plant
Outputs
Figure 5-6: Abridged LCI data module from US NREL LCI Database for bituminous coal-fired electricity generation. Output for functional unit italicized. (Source: US LCI Database 2012)
Figure 5-6 is organized into sections of data for inputs and outputs. At the top, we see the abridged input flows into the process for generating electric power via bituminous coal. Recalling the discussion of direct and indirect effects from Chapter 4, the direct inputs listed are bituminous coal and train transport. The direct outputs listed are fossil CO2 emissions (which is what results when you burn a fossil fuel) and electricity. Before discussing all of the inputs and outputs, we briefly focus on the output section to identify a critical component of the data module – the electricity output is listed as a product flow, with units of 1 kWh. Every LCI process will have one or more outputs, and potentially have one or more product flows as outputs, but this module has only one. That means that the functional unit basis for this unit process is per (1) kWh of electricity. All other inputs and outputs in Figure 5-6, representing the US NREL LCI data module for Electricity, bituminous coal, at power plant are presented as normalized per 1 kWh. You could think of this module as providing energy intensities or emissions factors per kWh. Thinking back to the discussion above on data collection, its unlikely that the study done to generate this LCI data module actually measured the inputs and outputs needed to make just 1 kWh of electricity at a power plant – it is too small a value. In reality, it is likely that the inputs and outputs were measured over the course of a month or year, and then normalized by the total electricity generation in kWh to find these normalized values. It is the same process you would do if you were making the LCI data module yourself. We will discuss how to see the assumptions and boundaries for the data modules later in this chapter. We now consider the abridged data module in more detail. In Figure 5-6, each of the input flows are a product flow from another process (namely, the product of bituminous coal mining and the product of train transportation). The unit basis assumption for those inputs is also given – kg for the coal and ton-kilometers (t*km) for the transportation. A tonLife Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
116
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
kilometer is a compound unit (like a kilowatt-hour) that expresses the movement of 1 ton of material over the distance of 1 kilometer. Both are common SI units. Finally the amount of input required is presented in scientific notation and can be translated into 0.442 kg of coal and 0.46 ton-km of train transport. Likewise, the output CO2 emissions to air are estimated at 0.994 kg. All of these quantities are normalized on a per-kWh generated basis. The comment column in Figure 5-6 (and which appears in many data modules) gives brief but important notes about specific inputs and outputs. For example, the input of train transportation is specified as being a potential transportation route from mine to power plant, which reminds us that the unit process for generating electricity from coal is already linked to a requirement of a train from the mine.5 Now that we have seen our first example of a secondary source LCI data module, Figure 5-7 presents a graphical representation of the abridged unit process similar to the generic diagram of Figure 5-2. The direct inputs, which are product flows from other man made processes, are on the left side as inputs from the technosphere. The abridged unit process has no direct inputs from nature. The direct CO2 emissions are at the top. The output product, and functional unit basis of the process, of electricity is shown on the right. All quantitative values are representative of the functional unit basis of the unit process.
Figure 5-7: Unit Process Diagram for abridged electricity generation unit process
Returning to our example LCA problem, we now have our first needed data point, that the direct CO2 emissions are 0.994 kg / kWh generated. Given that we have only three unit processes in our simple product system, we can work backwards from this initial point to get estimated CO2 emissions values from mining and train transport. Again using the NREL LCI database, Figure 5-8 shows abridged data for the data module bituminous coal, at mine. The unabridged version of the module has several other averaged transport inputs in ton-km, such as truck, barge, etc. Overall, the module gives a "weighted average" transport input to get the coal from the mine to the power plant. Since we are only using the abridged (and unedited) version, we will otherwise undercount the upstream CO2 emissions from delivering coal since we are skipping the weighted effects from those other modes. 5
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
117
The output and functional unit is 1 kg of bituminous coal as it leaves the mine. Two important inputs are diesel fuel needed to run equipment, and coal. It may seem odd to see coal listed as an input into a coal mining process, but note it is listed as a resource and as an elementary flow. As discussed in Chapter 4, elementary flows are flows that have not been transformed by humans. Coal trapped in the earth for millions of years certainly qualifies as an elementary flow by that definition! Further, it reminds us that there is an elementary flow input within our system boundary, not just many product flows. This particular resource is also specified as being of a certain quality, i.e., with energy content of about 25 MJ per kg. Finally, we can see from a mass balance perspective that there is some amount of loss in the process, i.e., that every 1.24 kg of coal in the ground leads to only 1 kg of coal leaving the mine. Flow
Category
Type
Unit
Amount
Coal, bituminous, 24.8 MJ per kg
resource/ground
ElementaryFlow
kg
1.24
Diesel, combusted in industrial boiler
root/flows
ProductFlow
l
8.8e-03
Comment
Inputs
Outputs Bituminous coal, at mine root/flows ProductFlow kg 1.00 Figure 5-8: Abridged LCI data module from US NREL LCI Database for bituminous coal mining. Output for functional unit italicized. (Source: US LCI Database 2012)
Figure 5-9 shows the abridged NREL LCI data module for rail transport (transport, train, diesel powered). The output / functional unit of the process is 1 ton-km of rail transportation service provided. Providing that service requires 0.00648 liters of diesel fuel and emits .0189 kg of CO2, both per ton-km.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
118
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Flow
Category
Type
Unit
Amount
root/flows
ProductFlow
l
6.48e-03
air/unspecified
ElementaryFlow
kg
1.89e-02
Comment
Inputs Diesel, at refinery Outputs Carbon dixoide, fossil
transport, train, diesel root/flows ProductFlow t*km 1 powered Figure 5-9: Abridged LCI data module from US NREL LCI Database for rail transportation. Output for functional unit italicized. (Source: US LCI Database 2012)
To then find the total CO2 emissions across these three processes, we can work backwards from the initial process. We already know there are 0.994 kg/kWh of CO2 emissions at the power plant. But we also need to mine the coal and deliver it by train for each final kWh of electricity. The emissions for those activities are easy to associate, since Figure 5-6 provides us with the needed connecting units to estimate the emissions per kWh. Namely, that 0.442 kg of coal needs to be mined and 0.461 ton-km of rail transport needs to be used per kWh of electricity generated. We can then just use those unit bases to estimate the CO2 emissions from those previous processes. Figure 5-8 does not list direct CO2 emissions from coal mining, although it does list an input of diesel used in a boiler6. If we want to assume that we are only considering direct emissions from each process, we can assume the CO2 emissions from coal mining to be zero7, or we could expand our boundary and acquire the LCI data module for the diesel, combusted in industrial boiler process. Our discussion below follows the assumption that direct emissions are zero. Figure 5-9 notes that there are 0.0189 kg of CO2 emissions per ton-km of rail transported. Equation 5-1 summarizes how to calculate CO2 emissions per kWh for our simplistic product system. Other than managing the compound units, it is a simple solution: about 1 kg CO2 per kWh. If we were interpreting this result, we would note that the combustion of coal at the power plant is about 99% of the total emissions. 0.994 kg CO2 /kWh + 0.442 kg * 0 + (0.461 ton-km / kWh)*(0.0189 kg CO2 / ton-km) = 0.994 kg CO2 / kWh + 0.0087 kg CO2 / kWh = 1.003 kg CO2 / kWh
(5-1)
The estimated CO2 emissions for coal-fired electricity of 1 kg / kWh was obtained relatively easily, requiring only three steps and queries to a single database (US NREL LCI). As always This particular input of "diesel, combusted in industrial boiler" may not be what you would expect to find in an LCI data module, since it is a description of how an input of diesel is used. Such flows are fairly common though. 7 Also, the unabridged LCI data modules list emissions of methane to air, which could have been converted to equivalent CO emissions. 2 Doing so would only change the result above by about 10%. 6
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
119
one of our first questions should be "is it right?" We can attempt to validate this value by looking at external references. Whitaker et al (2012) reviewed 100 LCA studies of coal-fired electricity generation and found the median value to be 1 kg of CO2 per kWh, thus we should have reasonable faith that the simple model we built leads to a useful result. Of course we can add other processes to our system boundary (such as other potential transportation modes) but we would not appreciably change our simple result of 1 kg/kWh. Note that anecdotally experts often refer to the emissions from coal-fired power plants to be 2 pounds per kWh, which is a one significant digit equivalent to our 1 kg/kWh result. Process-based life cycle models are constructed in this way. For each unit process within the system boundary, data (primary or secondary) is gathered and flows between unit processes are modeled. Since you must find data for each process, such methods are often referred to as "bottom up" studies because you are building them up from nothing, as you might construct a building on empty land. Beyond validating LCI results, you should also try to validate the values found in any unit process you decide to use, even if sourced from a well-known database. That is because errors can and do exist in these databases. It is easy to accidentally put a decimal in the wrong place when creating a digital database. As an example, the US NREL LCI database had an error in the CO2 emissions of its air transportation process, of 53 kg per 1000 ton-km (0.053 kg per ton-km) for several years before it was fixed. This error was brought to their attention because observant users noted that this value was less than the per-ton-km emissions for truck transportation, which went against common sense. Major releases of popular databases are also imperfect. It is common to have errors found and fixed, but this may happen months after licenses have been purchased, or worse, after studies have been completed. These are additional reasons why despite being of high quality, you need to validate your data sources.
Details for Other Databases The discussion above was focused on the US NREL LCI Database, which contains only process data for US-based production, yet there are other considerations both for data access and metadata for the other databases. As noted in Figure 5-4, the ecoinvent database is far more geographically diverse. While generally focused on Europe, data can be found in ecoinvent for other regions of the world as well. This fact creates a new challenge in interpreting available process data modules, namely, determining the country of production basis assumption for the data. While examining the metadata can be useful, ecoinvent and other databases typically summarize the country used within the process naming convention. For example, a process you might find within ecoinvent might be called electricity, hard coal, at power plant, DE, where the first part is the process name formatted similar to the NREL database, and at the end is an abbreviated term for the country or region to which that
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
120
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
process is representative. Figure 5-10 summarizes some of the popular abbreviations used for country basis within ecoinvent. Country or Region
Abbreviation
Country or Region
Abbreviation
Norway
NO
Japan
JP
Australia
AU
Canada
CA
India
IN
Global
GLO
China
CN
Europe
RER
Germany
DE
Africa
RAF
United States
US
Asia
RAS
Netherlands
NL
RU
Hong Kong
HK
France
FR
Russian Federation Latin America and the Caribbean North America
RLA RNA
United Kingdom GB Middle East RME Figure 5-10: Summary of abbreviations for countries and regions in ecoinvent
Ecoinvent has substantially more available metadata for its data modules, including primary sources, representative years, and names of individuals who audited the datasets. While ecoinvent data are not free, the metadata is freely accessible via the database website. Thus, you could do a substantial amount of background work verifying that ecoinvent has the data you want before deciding to purchase a license. A particular feature of ecoinvent data is its availability at either the unit process or system process level. Viewing and using ecoinvent system processes is like using already rolled-up information (and computations would be faster), while using unit processes will be more computationally intensive. This will be discussed more in Chapter 9.
LCI Data Module Metadata Our example using actual LCI data modules from the US NREL LCI database jumped straight into extracting and looking at the quantitative data. However, all LCI data modules provide some level of metadata, which is information regarding how the data was collected, how the modules were constructed, etc. Metadata is also referred to as "data about data". The metadata that we care about for our unit processes are elements such as the year the data was collected, where it was collected, whether the values are single measurements or averages, and whether it was peer reviewed. To understand metadata more, we can look at the metadata available for the processes we used above. The US NREL LCI Database has three different metadata categories as well as the Exchanges information shown above.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
121
Figure 5-11 shows metadata from the Activity metadata portion of the US NREL LCI database for the Electricity, bituminous coal, at power plant process used above. This metadata notes that the process falls into the Utilities subcategory (used for browsing on the website) and that it has not yet been fully validated. It applies to the US, and thus it is most appropriate for use in studies looking to estimate impacts of coal-fired electricity generation done within the United States. Note that this does not mean that you can only use it for that geographical region. A process like coal-fired generation is quite similar around the world; although factors such as pollution controls may differ greatly by region. However, since capture of carbon is basically non-existent, if we wanted to use this process to estimate CO2 emissions from coal-fired generation in other regions it might still be quite useful. The metadata field for "infrastructure process" notes whether the process includes estimated infrastructure effects. For example, one could imagine two parallel unit processes for electricity generation, where one includes estimated flows from needing to build the power plant and one does not (such as the one referenced above). In general, infrastructure processes are fairly rare, and most LCA study scopes exclude consideration of infrastructure for simplicity. Name
Electricity, bituminous coal, at power plant
Category
Utilities - Fossil Fuel Electric Power Generation
Description
Important note: although most of the data in the US LCI database has undergone some sort of review, the database as a whole has not yet undergone a formal validation process. Please email comments to
[email protected].
Location
US
Geography Comment
United States
Infrastructure Process
False
Quantitative Reference Electricity, bituminous coal, at power plant Figure 5-11: Activity metadata for Electricity, bituminous coal, at power plant process
Figure 5-12 shows the Modeling metadata for the coal-fired generation unit process. There is no metadata provided for the first nine categories of this category, but there are ten references provided to show the source data used to make the unit process. While a specific "data year" is not dictated by the metadata, by looking at the underlying data sources, the source data came from the period 1998-2003. Thus, the unit process data would be most useful for analyses done with other data from that time period. If we wanted to use this process data for a more recent year, we would either have to look for an LCI data module
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
122
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
that was newer, or verify that the technologies have not changed much since the 1998-2003 period. LCI Method Modelling constants Data completeness Data selection Data treatment Sampling procedure Data collection period Reviewer Other evaluation Sources
U.S. EPA 1998 Emis. Factor AP-42 Section 1.1, Bituminus and Subbituminus Utility Combustion U.S. Energy Information Administration 2000 Electric Power Annual 2000 Energy Information Administration 2000 Cost and Quality of Fuels for Electric Utility Plants 2000 Energy Information Administration 2000 Electric Power Annual 2000 U.S. EPA 1998 Study of Haz Air Pol Emis from Elec Utility Steam Gen Units V1 EPA-453/R-98-004a U.S. EPA 1999 EPA 530-R-99-010 unspecified 2002 Code of Federal Regulations. Title 40, Part 423 Energy Information Administration 9999 Annual Steam-Electric Plant Operation and Design Data Franklin Associates 2003 Data Details for Bituminous Utility Combustion Figure 5-12: Modeling metadata for Electricity, bituminous coal, at power plant process
Finally, Figure 5-13 shows the Administrative metadata for the Electricity, bituminous coal, at power plant process. There are no explicitly-defined intended applications (or suggested restrictions on such applications), suggesting that it is broadly useful in studies. The data are
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
123
not copyrighted, are publicly available, and were generated by Franklin Associates, a subsidiary of ERG, one of the most respected life cycle consulting business in the US. The "Data Generator" is a significant piece of information. You may opt to use or not use a data source based on who created it. A reputable firm has a high level of credibility. A listed individual with no obvious affiliation or reputation might be less credible. Finally, the metadata notes that it was created and last updated in October 2011, meaning that perhaps it was last checked for errors on this date, not that the data is confirmed to still be valid for the technology as of this date. Intended Applications
"
Copyright
false
Restrictions
All information can be accessed by everybody.
Data Owner Data Generator
Franklin Associates
Data Documentor
Franklin Associates
Project Version Created
2011-10-24
Last Update 2011-10-24 Figure 5-13: Administrative metadata for Electricity, bituminous coal, at power plant process
Our metadata examples have focused on the publicly available US NREL LCI Database, but other databases like ELCD and ecoinvent have similar metadata formats. These other databases typically have more substantive detail, in terms of additional fields and more consistent entries in these fields. Since these other data sources are not public, we have not used examples here. You should browse through the available metadata for any of the databases that you have access to, so that you can better appreciate the records that may exist within various metadata records. Remember that the reason for better appreciating the value of the metadata is to help you with deciding which secondary data sources to use, and how compatible they are with your intended goal and scope.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
124
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Referencing Secondary Data When you use secondary data as part of your study it must be appropriately referenced, as with any other source. Referencing data sources was first mentioned in Chapter 2, but here we discuss several important additions for referencing data from LCA databases. As an example, the US NREL LCI database explicitly suggests the following referencing style for use of its data modules: When referencing the USLCI Database, please use the following format: U.S. Life Cycle Inventory Database. (2012). National Renewable Energy Laboratory, 2012. Accessed November 19, 2012: https://www.lcacommons.gov/nrel/search However, this is the minimum referencing you should provide for process data. First of all, you can not simply reference the database. You need to ensure that the specific unit process from which you have used data is clear to the reader, for example if they would like to validate your work. That means you need to explicitly reference the name of the process (either obviously in the text or in the reference section). In the US NREL database and other sources, there may be hundreds of LCI data modules for electricity. Thus, the danger is that in the report you loosely reference data for coal-fired electricity generation as being from "the NREL database", but do not provide enough detail for the reader to know which electricity process was used. Unfortunately, this is a common occurrence in LCA reports. This situation can be avoided by explicitly noting the name of the process used in the reference, such as: U.S. Life Cycle Inventory Database. Electricity, bituminous coal, at power plant unit process (2012). National Renewable Energy Laboratory, 2012. Accessed Nov. 19, 2012: https://www.lcacommons.gov/nrel/search A generic reference to the database, as given at the top of this section, may be acceptable if the report separately lists all of the specific processes used in the study, such as in an inventory data source table listing all of the processes used. You will likely use multiple unit processes from the same database. You can either create additional references like the one above for each process, or use a combined reference that lists all processes as part of the reference, such as: U.S. Life Cycle Inventory Database. Electricity, bituminous coal, at power plant; bituminous coal, at mine; transport, train, diesel powered unit processes (2012). National Renewable Energy Laboratory, 2012. Accessed Nov. 19, 2012: https://www.lcacommons.gov/nrel/search The greater the number of similar processes, the greater the need to specify which specific data module you used in your analysis. This becomes especially important if you are using LCI data modules from several databases.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
125
A final note about referencing is that the LCA databases are generally not primary sources, they are secondary sources. Ideally, sources would credit the original author, not the database owner who is just providing access. If the LCI data module is taken wholesale from another source (i.e., if a single source were listed in the metadata), it may make sense to also reference the primary source, or to add the primary source to the database reference. In this case the reference might look like one of the following: RPPG Of The American Chemistry Council 2011. Life Cycle Inventory Of Plastic Fabrication Processes: Injection Molding And Thermoforming. http://plastics.americanchemistry.com/Education-Resources/Publications/LCI-ofPlastic-Fabrication-Processes-Injection-Molding-and-Thermoforming.pdf. via U.S. Life Cycle Inventory Database. Injection molding, rigid polypropylene part, at plant unit process (2012). National Renewable Energy Laboratory, 2012. Accessed November 19, 2012: https://www.lcacommons.gov/nrel/search U.S. Life Cycle Inventory Database. Injection molding, rigid polypropylene part, at plant unit process (2012). National Renewable Energy Laboratory, 2012. Accessed November 19, 2012: https://www.lcacommons.gov/nrel/search (Primary source: RPPG Of The American Chemistry Council 2011. Life Cycle Inventory Of Plastic Fabrication Processes: Injection Molding And Thermoforming. http://plastics.americanchemistry.com/Education-Resources/Publications/LCI-ofPlastic-Fabrication-Processes-Injection-Molding-and-Thermoforming.pdf) As noted in Chapter 2, ideally you would identify multiple data sources (i.e., multiple LCI data modules) for a given task. This is especially useful when using secondary data because you are not collecting data from your own controlled processes. Since the data is secondary, it is likely that there are slight differences in assumptions or boundaries than what you would have used if collecting primary data. By using multiple sources, and finding averages and/or standard deviations, you could build a more robust quantitative model of the LCI results. We will discuss such uncertainty analysis for inventories in Chapter 10.
Additional Considerations about Secondary Data and Metadata Given the types and classes of data we are likely to find in life cycle studies, we introduce in this subsection a few more considerations to ensure you are finding and using appropriate types of data to match the needs of your study. These considerations are in support of the data quality requirements.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
126
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Temporal Issues In creating temporal data quality requirements, you will set a target year (or years) for data used in your study. For example, you might have a DQR of "2005 data" or "data from 2005-2007" or "data within 5 years of today". After setting target year(s) you then must do your best to find and use data that most closely matches the target. It is likely that you will not be able to match all data with the target year(s). When setting and evaluating temporal DQRs, the following issues need to be understood. You may need to do some additional work to guarantee you know the basis year of the data you find, but this is time well spent to ensure compatibility of the models you will build. You will need to distinguish between the year of data collection and year of publication. In our CBECS example in Chapter 2, the data were collected in the year 2003 but the study was not published by DOE until December 2006 (or, almost 2007). It is easy to accidentally consider the data as being for 2006 because the publication year is shown throughout the reports. But the data were representative of the year 2003. If your temporal DQR was set at "2005", you might still be able to justify using the 2003 CBECS data, but would need to assess whether the electricity intensity of buildings likely changed significantly between 2003 and 2005. The same types of issues arise when using sources such as US EPA's AP-42 data, which are compilations of (generally old) previously estimated emissions factors. Other aspects of your DQRs may further help decide the appropriateness of data newer or older than your target year. The same is true of dates given in the metadata of LCI data modules. You don't care about when you accessed the database, or when it was published in the database. You care about the primary source's years of analysis. Figure 5-12 showed metadata on the coal-fired electricity generation process where the underlying data was from 1998-2003, and which was put in the US LCI database in 2011. An appropriate "timestamp" for this process would be 1998-2003. While on the topic of temporal issues, we revisit the point about age of data in databases. The US LCI database project started in the mid-2000s. Looking at the search function in that database, you can find a distribution of the "basis year" of all of the posted data modules. This is a date that is not visible within the metadata, but is available for downloaded data modules and summarized in the web server. Figure 5-14 shows a graph of the distribution of the years. In short, there is a substantial amount of relatively old data, and a substantial amount of data where this basis year is not recorded (value given as '9999'). Half of the 200 data modules updated in 2010 are from an update to the freight transportation datasets. These could be key considerations when considering the suitability of data in a particular database.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
127
Figure 5-14: Frequency Distribution of Data Years in US NREL LCI Database (as of August 15, 2013)
Geographical Issues You must try to ensure that you are using data with the right geographical scope to fit your needs. If you are doing a study where you want to consider the emissions associated with producing an amount of electricity, then you will find many potential data sources to use. The EIA has data that can give you the average emissions factors for electricity generation across the US. E-GRID (a DOE-EPA partnership) can give you emissions factors at fairly local levels, reflecting the types of power generation used within a given region. The question is the context of your study. Are you doing a study that inevitably deals with national average electricity? Then the EIA data is likely suitable. Or are you doing a study that needs to know the impact of electricity from a particular factory's production? In that case you likely want a fairly local data source, e.g., from E-GRID. An alternative is to leverage the idea of ranges, presented in Chapter 2, to represent the whole realm of possible values for electricity generation, including various local or regional averages all the way up to the national average.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
128
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Uncertainty and Variability Sadly, in the field of LCA there are many practitioners who actively or passively ignore the effects of uncertainty or variability in their studies. They treat all model inputs as single values and generate only a single result. The prospect of uncertainty or variability is lost in their model, and typically then that means those effects are lost on the reader of the study. How can we support a big decision (e.g., paper vs. plastic?) if there is much uncertainty in the data but we have completely ignored it? We are likely to end up supporting poor decisions if we do so. We devote Chapter 11 to methods of overcoming and structuring uncertainty in LCA models.
Chapter Summary Typically, the most time consuming aspect of an LCA (or LCI) study relates to the data collection and management phase. While the LCA Standard encourages practitioners to collect primary data for the product systems being studied, typically secondary data is used from prior published studies and databases. Using secondary data requires being knowledgeable and cognizant of issues relating to the sources of data presented and also requires accurate referencing. Data quality requirements help to manage expectations of the study team as well as external audiences pertaining to the goals of your data management efforts. Studies done with robust LCI data management methods lead to excellent and wellreceived studies.
References for this Chapter BEES LCA Tool, website, http://ws680.nist.gov/Bees/Default.aspx, last accessed August 12, 2013. ecoinvent website, www.ecoinvent.ch, last accessed August 12, 2013. ELCD LCA Database, website, http://lca.jrc.ec.europa.eu/lcainfohub/, last accessed August 12, 2013. Environmental Protection Agency. 1993. Life Cycle Assessment: Inventory Guidelines and Principles. EPA/600/R-92/245. Office of Research and Development. Cincinnati, Ohio, USA. Gabi Software, website, http://www.gabi-software.com/, last accessed August 12, 2013.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
129
LCA-DATA, UNEP, website, http://lca-data.org:8080/lcasearch, last accessed August 12, 2013. Quantis, "Environmental Life Cycle Assessment of Drinking Water Alternatives and Consumer Beverage Consumption in North America", LCA Study completed for Nestle Waters North America, 2010, http://www.beveragelcafootprint.com/wpcontent/uploads/2010/PDF/Report_NWNA_Final_2010Feb04.pdf, last accessed September 9, 2013. The Agribusiness Group, "Life Cycle Assessment: New Zealand Merino Industry Merino Wool Total Energy Use and Carbon Dioxide Emissions", 2006, http://www.agrilink.co.nz/Portals/Agrilink/Files/LCA_NZ_Merino_Wool.pdf, last accessed September 1, 2013. US NREL LCI Database, website, http://www.nrel.gov/lci/, last accessed August 12, 2013. U.S. Life Cycle Inventory Database. Electricity, bituminous coal, at power plant, bituminous coal, at mine, and transport, train, diesel powered unit processes (2012). National Renewable Energy Laboratory, 2012. Accessed August 15, 2013: https://www.lcacommons.gov/nrel/search USDA LCA Digital Commons, website, http://www.lcacommons.gov, last accessed August 12, 2013. Whitaker, Michael, Heath, Garvin A., O'Donoughue, Patrick, and Vorum, Martin, "Life Cycle Greenhouse Gas Emissions of Coal-Fired Electricity Generation: Systematic Review and Harmonization", Journal of Industrial Ecology, 2012. DOI: 10.1111/j.15309290.2012.00465.x
Questions for Chapter 5 1. Using the US NREL LCI Database (from the USDA Digital Commons) or another LCI database, search or browse amongst the available categories. For each of the following broadly defined processes in the list below, discuss how many different LCI data modules are available and qualitatively discuss what different assumptions have been used to generate the data modules. a. Refining of petroleum b. Generating electricity from fossil fuel c. Truck transportation
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
130
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
2. If you had data quality requirements stating that you wanted data that was national (US) in scope, and from within 5 years of today, how many of the LCI data modules from Question 1 would be available? Which others might still be relevant? Justify your answer. 3. The data identified in part 1c above would be secondary data if you were to use it in a study. If you instead wanted primary data for a study on trucking, discuss what methods you might use in order to get the data. 4. Using an LCI database available to you, search for one LCI data module in each of the following broad categories - energy, agriculture, and transportation. For each of the three, do the following: a. List the name of the process. b. Identify the functional unit. c. Draw a unit process diagram. d. Try to do a brief validation of the data reported. e. Comment briefly on an example LCA study that this process might be appropriate for, and one where it would not be appropriate. f. Show how to appropriately reference the LCI data module in a study. 5. Redo the Figure 5-5 example but include the diesel, combusted in industrial boiler process within the system boundary. What is your revised estimate of CO2 emissions per kWh? How different is your estimate compared to Equation 5-1? 6. Redo the Figure 5-5 example but include within the system boundary refining of the diesel used in the coal mining and rail transportation processes (and assume you have LCI flow data that there are 2.5 E-04 kg CO2 emissions per liter of diesel fuel). How is your revised estimate of CO2 emissions per kWh compared to Equation 5-1?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
131
Advanced Material for Chapter 5 The advanced material in this chapter will demonstrate how to find and access LCI data modules from various popular databases and software tools, and how to use the data to build simple models like the main model presented in the chapter related to coal-fired electricity. Not all databases and software tools are discussed; however, access methods are generally very similar across tools. For consistency, we will demonstrate how to find the same process data as used in the chapter so that you can learn about the different options and selections needed to find equivalent data and metadata across tools. Specifically, we will demonstrate how to find data from the US LCI database by using the LCA Digital Commons Website, SimaPro (a commercial LCA tool) and openLCA (a free LCA tool). The databases and tools use different terminology, categories, etc., to organize LCI data, but can all lead to the same data. Seeing how each of the tools categorizes and refers to the data is an important concept to understand.
Section 1 - Accessing Data via the US LCA Digital Commons The LCA Digital Commons is a free, US government-sponsored and hosted web-based data resource. Given that all of its data are publicly available, it is a popular choice for practitioners. Thus, it is also a great resource for learning about what LCI data looks like, how to access it, and how to build models. The main purpose of the Digital Commons is to act as a resource for US Department of Agriculture (USDA) agricultural data and, as a result, accessing the home page (at https://www.lcacommons.gov/discovery) will filter access to those datasets. However, the US LCI database previously hosted by NREL (at http://www.nrel.gov/lci/), and mentioned extensively in Chapter 5, is also hosted via the Digital Commons website (at https://www.lcacommons.gov/nrel/search). Given its comprehensiveness, most of the discussion in this book is related to use of the NREL data. The examples provided below are for accessing the NREL data source, which has slightly different metadata and contents than the USDA data but a similar method for searching and viewing. The LCI data modules on the Digital Commons website can be accessed via searching or browsing. Brief overviews are provided for both options, followed by how to view and download selected modules. Before following the tutorial below, you should consider registering for an account on the Digital Commons website (you will need separate accounts for the USDA and NREL data). While an account is not required to view all of the data, it is required if you wish to download the data. You can copy and paste the data from a web browser instead of downloading but this sometimes leads to formatting errors.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
132
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Browsing for LCI Data Modules on the Digital Commons (NREL) Figure 5-15 shows the NREL Digital Commons home page, where the left hand side shows how the data modules are organized, including dataset type (elementary flows or unit processes), high-level categories (like transportation and utilities), and year of data8.
Figure 5-15: Excerpt of LCA Digital Commons Website Home Page
Clicking on the + icon next to the categories generally reveals one or more additional subcategories. For example, under the Utilities category there are fossil-fired and other generation types. Clicking on any of the dataset type, category/subcategory or year checkboxes will filter the overall data available. The "order by" box will sort the resulting modules. Filtering by (checking) Unit processes and the Fossil fuel electric power generation category under Utilities, and ordering by description will display a subset of LCI data modules, as shown in Figure 5-16. A resulting process module can be selected (see below for how to do this and download the data).
Figure 5-16: Abridged View of LCA Digital Commons Browsing Example Results 8
The examples of the NREL US LCI Database in this section are as of July 2014, and may change in the future.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
133
Searching for an LCI data module via keyword The homepage has a search feature, and entering a keyword such as electricity and pressing the Go button on the right hand side, as shown in Figure 5-17, will return a list of data modules within the NREL LCI database that have that word in the title or category, as shown in Figure 5-18.
Figure 5-17: Keyword search entry on homepage of NREL engine of LCA Digital Commons Website
Figure 5-18: Abridged Results of electricity keyword search
Figure 5-18 indicates that the search engine returns more than 100 LCI data modules (records) that may be relevant to "electricity". Some were returned because electricity is in the name of the process and others because they are in the Electric power distribution data category. When searching, you can order results by relevance, description, or year. Once a
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
134
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
set of search results is obtained, results can be narrowed by filtering via the options on the left side of the screen. For example, you could choose a subset of years to be included in the search results, which can help ensure you use fairly recent instead of old data (as discussed along with Figure 5-14). You can also filter based on the LCI data categories available, in this case by clicking on the + icon next to the high-level category for Utilities, which brings up all of the subcategories under utilities. Figure 5-19 shows the result of a keyword search for 'electricity', ordered by relevance, and filtered by the Utilities subcategory of Fossil fuel electric power generation and by data for year 2003. The fifth search result listed is the same one mentioned in the chapter that forms the basis of the process flow diagram example.
Figure 5-19: Abridged Results of electricity keyword search, ordered and filtered
Selecting and viewing an LCI data module When you have searched or browsed for a module and selected by clicking on it, the module detail summary is displayed, as in Figure 5-20.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
135
Figure 5-20: Details for Electricity, bituminous coal process on LCA digital commons
The default result is a view of the Activity tab, which was shown in Figure 5-11. The information available under the Modeling and Administrative tabs was presented in Figure 5-12 and Figure 5-13. Finally, an abridged view of the information available on the Exchanges tab was also shown in Figure 5-6. Not previously mentioned is that the module can be downloaded by first clicking on the shopping cart icon in the top right (adjacent to the "Next item" tag). This adds it to your download cart. Once you have identified all of the data you are interested in, you can view your current cart (menu option shown in Figure 5-21) and request them all to be downloaded (Figure 5-22).
Figure 5-21: Selection of Current Cart Download Option on LCA Digital Commons
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
136
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Figure 5-22: Cart Download Screen on LCA Digital Commons
After clicking download, you will be sent a link via the e-mail in your account registration. As noted, the format will be an Ecospold XML file. For novices, viewing XML files can be cumbersome, especially if just trying to look at flow information. While less convenient, the download menu (All LCI datasets submenu) will allow you to receive a link to a ZIP file archive containing all of the NREL modules in Microsoft Excel spreadsheet format (or you can receive all of the modules as Ecospold XML files). You can also download a list of all of the flows and processes used across the entire set of about 600 modules. A spreadsheet of all flows and unit processes in the US LCI database (and their categories) is on the www.lcatextbook.com website in the Chapter 5 folder. When uncompressed the Electricity, bituminous coal, at power plant module file has four worksheets, providing the same information as seen in the tabs of the Digital Commons/NREL website above. The benefit of the spreadsheet file, though, is the ability to copy and paste that values into a model you may be building. We will discuss building spreadsheet models with such data in Section 4 of this advanced material.
Section 2 – Accessing LCI Data Modules in SimaPro As mentioned in the chapter, SimaPro is a popular commercial software program specifically aimed at building quantitative LCA models. Its value lies both in these model-building support activities as well as in being able to access various datasets from within the program. Commercial installations of SimaPro cost thousands of dollars, but users may choose commercial databases (e.g., ecoinvent) to include in the purchase price. Regardless of which databases are chosen, SimaPro has the ability to use various other free datasets (e.g., US NREL, ELCD, etc.). This tutorial assumes that such databases have already been installed and will demonstrate how to find the same US NREL-based LCI data as in Section 1.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
137
This tutorial also does not describe any of the initial steps needed to purchase a license for or install SimaPro on your Windows computer or server. It will only briefly mention the login and database selection steps, which are otherwise well covered in the SimaPro guides provided with the software. Note that SimaPro refers to the overall modeling environment of data available as a "database" and individual LCI data sources (e.g., ecoinvent) as "libraries". After starting SimaPro, selecting the database (typically called "Professional"), and opening or creating a new project of your choice, you will be presented with the screen in Figure 5-23. On the left side of the screen are various options used in creating an LCA in the tool. By default the "processes" view is selected, showing the names and hierarchy of all processes in the currently selected libraries of the database. This list shows thousands of processes (and many of those will be from the ecoinvent database given its large size).
Figure 5-23: Default View of Processes in Libraries When Starting SimaPro
You can narrow the processes displayed by clicking on "Libraries" on the left hand side menu, which will display Figure 5-24. Here you can select a subset of the available libraries for use in browsing (or searching) for process data. You can choose "Deselect all" and then to follow along with this tutorial, click just the "US LCI" database library in order to access only the US NREL LCI data.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
138
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Figure 5-24: List of Various Available Libraries in SimaPro
If you then click the "Processes" option on the left hand side, you return to the original screen but now SimaPro filters and shows only processes from the selected libraries, as in Figure 5-25. Many of the previously displayed processes are no longer displayed.
Figure 5-25: View of Processes and Data Hierarchy for US-LCI Library in SimaPro
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
139
Now that you have prepared SimaPro to look for the processes in a specific database library, you can browse or search for data. Browsing for LCI Data Modules in SimaPro Looking more closely at Figure 5-25, the middle pane of the window shows the categorized hierarchy of data modules (similar to the expandable hierarchy list in the Digital Commons tool). However, these are not the same categories used on the NREL LCA Digital Commons website. Instead, they are the standard categories used in SimaPro for processes in any library. Clicking on the + icon next to any of the categories will expand it and show its subcategories. To find the Electricity, bituminous coal process, expand the Energy category then expand Electricity by fuel, then expand coal, resulting in a screen like Figure 5-26. Several of the other processes burning coal to make electricity and mentioned in the chapter would also be visible.
Figure 5-26: Processes Shown by Expanding Hierarchy of Coal-Sourced Electricity in SimaPro
The bottom pane shows some of the metadata detail for the selected process. By browsing throughout the categories (and collapsing or expanding as needed) and reading the metadata you can find a suitable process for your model. The tutorial will demonstrate how to view or download such data after briefly describing how to search for the same process.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
140
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Searching for a process in SimaPro Once libraries have been specified as noted above, clicking on the magnifying glass icon in the toolbar brings up the search interface as shown in Figure 5-27. You enter your search term in the top box, and then choose from several search options. If you are just looking for process data (as in this tutorial) then you would want to restrict your choice of where to look for the data to only libraries you have currently chosen (i.e., via the interface in Figure 5-24) rather than all libraries. This will also make your search return results more quickly. Note the default search only looks in the names of processes, not in the metadata (the "all fields" option changes this behavior).
Figure 5-27: Search Interface in SimaPro
Figure 5-28 shows the result of a narrowed search on the word "electricity" in the name of processes only in "Current project and libraries" and sorted by the results column "Name". Since we have already selected only the US LCI database in libraries, the results will not include those from ecoinvent, etc. One of the results is the same Electricity, bituminous coal, at power plant process previously discussed.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
141
Figure 5-28: Results of Modified Search for Electricity in SimaPro
By clicking "Go to" in the upper right corner of the search results box, SimaPro "goes to" the same place in the drill-down hierarchy as shown in Figure 5-26. Viewing process data in SimaPro To view process data, choose a process by clicking on it (e.g., as in Figure 5-26) and then click the View button on the right hand side. This returns the process data and metadata overview shown in Figure 5-29. Similar to the Digital Commons website, the default screen shows high-level summary information for the process. Full information is found in the documentation and system description tabs.
Figure 5-29: Process Data and Metadata Overview in SimaPro
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
142
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Clicking on the input-output tab displays the flow data in Figure 5-30, which for this process is now quite familiar. If you need to download this data, you can do so by choosing "Export" in the File menu, and choosing to export as a Microsoft Excel file.
Figure 5-30: View of Process Flow Data (Inputs and Outputs) in SimaPro
Section 3 – Accessing LCI Data Modules in openLCA openLCA is a free LCA modeling environment (available at http://www.openlca.org/) available for Windows, Mac, and Linux operating systems. While installation and configuration can be quite complicated (and is not detailed here), various datasets are available. The tutorial assumes you have access to a working openLCA installation with the US LCI database, and discusses how to find the same US NREL-based LCI data as in Section 1. After launching openLCA and connecting to your data source you should see a list of all of your databases, as shown in Figure 5-31. If you do not see the search and navigation tabs, you may add them under the "Window menu -> Show views option" to add them. If you have installed the US LCI database, it should be one of the options available.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
143
Figure 5-31: List of Data Connections in openLCA
Browsing for process data in openLCA Clicking on the triangle to the left of the folder allows you to open it and see the standard hierarchy of information for all data sources in openLCA, like in Figure 5-32. This is where you could see the process data, types of flows, and units.
Figure 5-32: Hierarchical Organization of Information for openLCA Databases
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
144
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
If you double click on the "Processes" folder it will display the same sub-hierarchy of processes (not shown here) that we saw in the NREL/Digital Commons website in Section 1. All of the data for unit processes are contained under that folder. If you click on the "Utilities" subcategory folder, then the "Fossil Fuel Electric Power Generation" folder, you will see the Electricity, bituminous coal, at power plant seen above, as shown in Figure 5-33. Several of the other processes burning coal to make electricity and mentioned in the chapter would also be visible.
Figure 5-33: Expanded View of Electricity Processes in Fossil Fuel Generation Category
Searching for a process in openLCA Instead of using the Navigation tab, a search for process data can be done using the Search tab. Clicking on the search tab brings up the search interface, as shown in Figure 5-34.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
145
Figure 5-34: Default Search Interface in openLCA
In the first search option, you may search in all databases or narrow the scope of your search to only a single database (e.g., to the US-LCI database). In the second option, you may search all object types, or narrow the scope of your search to just "Processes", etc. Finally, you can enter a search term, such as "electricity". If you choose to search for "electricity" only in your US LCI database (note you may have named it something different), and only in processes, and click search you will be presented with the results as in Figure 5-35. Note that these results have been manually scrolled down to show the same Electricity, bituminous coal, at power plant process previously identified.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
146
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Figure 5-35: Search Results for Electricity in US-LCI Database in OpenLCA
Unlike the other tools, there is no quick and easy way to skim metadata to ensure which process you want to use.
Viewing process data in openLCA To view process data, choose a process by double-clicking on it from either the browse or search interface. This opens a new pane of the openLCA environment and returns the process data and metadata overview, as shown in Figure 5-36. Similar to the Digital Commons website, the default screen shows high-level summary information for the process (not all of the information is shown in the Figure). Additional information is available in the Inputs/Outputs, Administrative information, other tabs at the bottom of this pane.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Figure 5-36: Process Data and Metadata Overview in SimaPro
Clicking on the Inputs/Outputs tab displays the flow data in Figure 5-37, which for this process is now quite familiar.
Figure 5-37: View of Process Flow Data (Inputs and Outputs) in openLCA
If you need to download this data, you can do so by choosing "Export" in the File menu, but you cannot export it as a Microsoft Excel file.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
147
148
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Section 4 – Spreadsheet-based Process Flow Diagram Models Now that process data has been identified, quantitative process flow diagram-based LCI models can be built. Amongst the many tools to build such models, Microsoft Excel is one of the most popular. Excel has many built-in features that are useful for organizing LCI data and calculating results, and is already familiar to most computer users. To make these examples easy to follow, we repeat the core example from Chapter 5 (and shown in Figure 5-5) involving the production of coal-fired electricity via three unit processes in the US LCI database. The US LCI database is used since it is freely available and indicative of many other databases (e.g., ELCD). To replicate the structure of the core model from Chapter 5, we need to manage our process data in support of our process flow diagram. The following steps illustrate the quantitative structure behind a process-flow diagram based LCI model. 1) Find all required process data In the first few sections of the advanced material for this chapter, we showed how to find the required process data from the US LCI database via several different tools. Using similar browse and search methods, you can find the LCI data for the other two processes so that you have found US LCI data for these three core processes: •
Electricity, bituminous coal, at power plant
•
Bituminous coal, at mine
•
Transport, train, diesel powered
Depending on which tool you used to find the US LCI process data, it may be easy to export the input and output flows for the functional unit of each process into Excel. If not, you may need to either copy/paste, or manually enter, the data. Recall that accessing the US LCI data directly from the LCA Digital Commons can yield Microsoft Excel spreadsheet files. 2) Organize the data into separate worksheets A single Microsoft Excel spreadsheet file can contain many underlying worksheets, as shown in the tabs at the bottom of the spreadsheet window. For each of the downloaded or exported data modules, copy / paste the input/output flows into a separate Microsoft Excel worksheet. If you downloaded the US LCI process data directly from the lcacommons.gov website, the input/output flow information is on the "X-Exchange" worksheet of the downloaded file (the US LCI data in other sources would be formatted in a similar way). The Transport, train, diesel powered process has 1 input and 9 outputs (including the product output), as shown in Figure 5-38.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
149
Figure 5-38: Display of Extracted flows for Transport, train, diesel powered process from US LCI
3) Create a separate "Model" worksheet in the Microsoft Excel file This Model worksheet will serve as the primary workspace to keep track of the relevant flows for the process flow diagram. This sheet uses cell formulas to reference the flows on the other worksheets that you created from the process LCI datasets. Beyond just referencing the flows in the other worksheets, the Model worksheet must scale the functional unit-based results as needed based on the process flow diagram. For example, in Equation 5-1, results were combined for 1 kWh of electricity from bituminous coal, 0.46 ton-km of train transportation, and from 0.44 kg of coal mining. Since the process LCI data modules are generally normalized on a basis of a functional unit of 1, we need to multiply these LCI results by 1, 0.46, or 0.44. Basic LCI Spreadsheet Example In this example, a basic cell formula is created on the Model worksheet to add the output flows of CO2 from the three separate process worksheets. We first make a summary output result cell for each of the three processes where we multiply the CO2 emissions value from each worksheet (e.g., the rounded value 0.019 in cell G8 of Figure 5-38) by the functional unit scale factor listed above. Then we find the sum of CO2 emissions across the three processes by typing = into an empty cell and then successively clicking on the three scaled process emissions values. The Chapter 5 folder has a "Simple and Complex LCI Models from US LCI" spreadsheet file following the example as shown in the Chapter (which only tracked emissions of fossil CO2). Figure 5-39 shows an excerpt of the "Simple Model" worksheet in the file. The same result as shown in the chapter (not rounded off) is visible in cell E8, with the cell formula =B8+C8+D8.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
150
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Figure 5-39: Simple Spreadsheet-Based Process LCI Model
This simple LCI model shows a minimal effort result, such that using a spreadsheet is perhaps overkill. Tracking only CO2 emissions means that we only have to add three scaled values, which could be accomplished by hand or on a calculator. However this spreadsheet motivates the possibility that a slightly more complex spreadsheet could be created that tracks all flows, not just emissions of CO2. Complex LCI Spreadsheet Example Beyond the assumptions made in the simple model above, in LCA we often are concerned with many (or all) potential flows through our product system. Using the same underlying worksheets from the simple spreadsheet example, we can track flows of all of the outputs listed in the various process LCI data modules (or across all potential environmental flows). This not only allows us a more complete representation of flows, but better prepares us for next steps such as impact assessment. In this complex example, we use the same three underlying input/output flow worksheets, but our Model worksheet more comprehensively organizes and calculates all tracked flows from within a dataset. Instead of creating cell formulas to sums flows for each output (e.g., CO2) by clicking on individual cells in other worksheets, we can use some of Excel's other built-in functions to pull data from all listed flows of the unit processes into the summary Model worksheet. An example file is provided, but the remaining text in this section describes in a bit more detail how to use Excel's SUMPRODUCT function for this task. The SUMPRODUCT function in Microsoft Excel, named as such because it finds the sum of a series of multiplied values, is typically used as a built-in way of finding a weighted average. Each component of the function is multiplied together. For example, instead of the method shown in the Simple LCI spreadsheet above, we could have copied the CO2
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
151
emissions values from the three underlying worksheets into the row of cells B8 through D8, and then used the function =SUMPRODUCT(B4:D4*B8:D8) to generate the same result. The "Simple and Complex LCI Models" file has a worksheet "Simple Model (with SUMPRODUCT)" showing this example in cell E8, yielding the same result as above. However the SUMPRODUCT function can be more generally useful, because of how Excel manages TRUE and FALSE values and the fact that the "terms" of SUMPRODUCT are multiplied together. In Excel, TRUE is represented as 1 and FALSE is represented as 0 (they are Booleans). So if we have "terms" in the SUMPRODUCT that become 1 or 0, we can use SUMPRODUCT to only yield results when all expressions are TRUE, else return 0. This is like achieving the mathematical equivalent of if-then statements on a range of cells. The magic of this SUMPRODUCT function for our LCI purposes is that if we have a master list of all possible flows, compartments, and sub-compartments, we can find whether flow values exist for any or all of them. On the US LCI Digital Commons website, a text file can be downloaded with all of the nearly 3,000 unique compartment flows present in the US LCI database. This master list of flows can be pasted into a Model worksheet and then used to "look up" whether numerical quantities exist for any of them. A representative cell value in the complex Model worksheet, which has similar cell formulas in the 3,000 rows of unique flows, looks like this (where cells A9, B9, and C9 are the flow, compartment, and subcompartment values we are trying to match in the process data): =E$4*SUMPRODUCT((Electricity_Bitum_Coal_Short!$A$14:$A$65=A 9)*(Electricity_Bitum_Coal_Short!$C$14:$C$65=B9)*(Electrici ty_Bitum_Coal_Short!$D$14:$D$65=C9)*Electricity_Bitum_Coal_ Short!$G$14:$G$65) This cell formula multiplies the functional unit scale factor in cell E4 by the SUMPRODUCT value of: •
whether the flow name, compartment, and subcompartment in the unit flows for the coal-fired electricity process match every item in the master list of flows.
•
and, if the flow/compartment/subcompartment values match, the inventory value for the matched flow.
Within the SUMPRODUCT, if the flow/compartment/subcompartment in the unit process data doesn't match the flow/compartment/subcompartment on the row of the Model worksheet, the Boolean values are all 0's and the result is 0. If they all match, the Boolean results are 1, and the final part of the SUMPRODUCT expression (the actual flow quantity) is returned.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
152
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
Figure 5-40: Complex Spreadsheet-Based Process LCI Model
The Chapter 5 folder on the textbook website has spreadsheets with all of the flows and processes in the US LCI database, as downloaded from the LCA Digital Commons website. The "Simple and Complex LCI Models" file has a worksheet "Complex Model" which shows how to use the SUMPRODUCT function to track all 3,000 flows present in the US LCI database (from the flow file above). Of course the results are generally zero for each flow due to data gaps, but this example model expresses how to broadly track all possible flows. You should be able to follow how this spreadsheet was made and, if needed, add additional processes to this spreadsheet model.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 5: Data Acquisition and Management for Life Cycle Inventory Analysis
(the rest of these will be done, in order shown, but in no hurry to finish yet)
Section – Ecoinvent website
Section – Accessing LCI Data Modules in ILCD?
Reorder these Sections (e.g., ILCD after NREL since so similar)?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
153
154
Chapter 6: Analyzing Multifunctional Product Systems
Photo Credit: © Chris Goldberg, 2009, via Creative Commons license (CC BY-NC 2.0)
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
155
Chapter 6 : Analyzing Multifunctional Product Systems In Chapter 5, we showed the relatively simple steps of building a process flow diagram-based LCA model where there was only one product in the system. However, product systems in LCA studies may have multiple products, providing multiple functions. Analyzing these systems introduces new complexities, and this chapter demonstrates various methods (referenced in the Standard) for overcoming or addressing these challenges. The methods described herein modify either the systems studied or the input and output flow values so that the multifunction systems can be quantitatively assessed.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. Discuss the challenges presented by processes and systems with multiple products and functions. 2. Perform allocation of flows to co-products from unallocated data. 3. Replicate results from database modules containing unallocated and allocated flows. 4. Estimate inventory flows from a product system that has avoided allocation via system expansion.
Multifunction Processes and Systems Many processes and product systems are simple enough that they have only a single product output that provides a single function. However, even when tightly scoped, there are also many processes and systems that will have multiple products that each provide their own function. A good example is a petroleum refinery that has outputs of gasoline, diesel, and other products. LCA studies typically have function and functional unit definitions related to the life cycle effects of only one product. As such, a method is needed to connect input and output flow data with a desired functional unit, subject to the data associated with multiple products. The method chosen can have a significant effect on the results, and thus, the choice of method is controversial. How to deal with such systems is subject to much debate. Building on the example figures and discussion in Chapters 4 and 5, Figure 6-1 shows a generic view of a unit process with multiple product outputs that each provides their own Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
156
Chapter 6: Analyzing Multifunctional Product Systems
function. In this case, there are three Products, A, B, and C, associated with functions 1, 2, and 3, respectively.
Figure 6-1: Generic Illustration of a Unit Process with Multiple Products or Co-Products
Co-products exist when a process has more than one product output – which is a fairly common outcome, given the complexity of many industrial processes. If the goal of our study is to assess the effects associated with Product A, which provides Function 1, we need to find a way to deal with the provision of Products B and C, which provide Functions 2 and 3. In the context of a particular study, typically the product of primary interest (i.e., Product A above) is referred to as the product, and any other products (i.e., Products B and C above) as co-products, but this is not a standard terminology. The Standard suggests two ways of approaching this problem: either by partitioning the process so that a set of quantitative connections are derived between the inputs and outputs and the various products (known as allocation), or by changing the way in which we have defined our system so that we can clearly show just the effects associated with Product A and its associated function (known as system expansion). While system expansion is the preferred method, we discuss allocation first because it is simpler to understand and also helps to frame the broader discussion.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
157
Allocation of Flows for Processes with Multiple Products For a unit process, the goal of allocation is to assign a portion of each input and nonproduct (e.g., emission) output to each of the various products, such that the sum of all product shares equals the total input and output flows for the process. Allocation is also referred to as partitioning. In Chapter 5, we accessed the US LCI database information and directly used all of the data without modification in our model for the bituminous coal-fired electricity process flow diagram. We even used some of the information in the process data to decide the multipliers needed in using our other process data sources. The reason we would directly use all of the data is because the Electricity, bituminous coal, at power plant process listed only one product: electricity (see Figure 5-6). For other LCA models, we may have to manipulate the data in some way to make it fit the needs of our study. However, one could envision an alternative process where aside from generating electricity, the process also produced heat (e.g., a combined heat and power, or CHP system). Such a process has multiple products, heat and power, each of which has a different function. Furthermore, we might want to derive a mathematical means of associating a relevant portion of the quantified inputs and outputs to each of the products (i.e., to know how much pollution we associate to each product of the system). This association is called allocation. The data associated with processes or systems having multiple product outputs may be organized in several ways. Their most raw form (i.e., as collected) will be an inventory of unallocated inputs and outputs, and relative quantities of co-product outputs. These unallocated flows represent a high level view of the process, representing all flows as measured but without concern for how those flows may connect to specific co-products. An example would be process data for an entire refinery that tracks all inputs (crude oil, energy, etc.) and quantifies all outputs (e.g., diesel, gasoline, etc.). Alternatively, process data may consist of already allocated flow estimates of inputs and outputs for each co-product. For instance, the refinery process data would contain estimates of crude oil and energy inputs used for each unit of gasoline, diesel, etc. In allocation, the key concern is determining the appropriate mathematical relationship to transform the unallocated flows to allocated flows. The Standard gives specific, yet somewhat vague directions on the appropriate methods of allocation. First off, ISO says that, if possible, allocation should be avoided, which we will assume has been deemed not possible. But, for the sake of discussion, if allocation is needed, then the Standard says that the inputs and outputs of the process should be partitioned between its products based on a physical relationship. It states that the physical relationship "should reflect the way in which the Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
158
Chapter 6: Analyzing Multifunctional Product Systems
inputs and outputs are changed by quantitative changes in the products or functions delivered by the system." Further, if the physical relationship alone is not sufficient to perform the allocation, then other relationships should be used, such as economic factors. Commonly used allocation methods include mass-based, energy-based, and economic-based methods. Not all systems will be able to be allocated in these ways, as some products have no mass, may differ in utility of energy (e.g., heat versus electricity), and economic features like market value may fluctuate widely. It is important to observe that the Standard does not prescribe which allocation method to use, i.e., it does not say to always use a mass-basis or an energy-basis for allocation, or to never use an economic basis. The only specifications provided pertain to reuse and recycling, where the Standard gives an ordering of preference for allocation methods, specifically, physical properties (e.g., mass), economic values, and number of uses. Practitioners often use this same ordering for allocating processes other than reuse and recycling, which may be a useful heuristic, but it is not prescribed by the Standard. As with other work, choices and methods behind allocation should be justified and documented. In addition, the Standard requires that when several possible allocations seem reasonable, sensitivity analysis should be performed to show the relative effects of the methods on the results (see section at the end of the chapter). For example, we might compare the results of choosing a mass-based versus an energy-based allocation method. The following example does not use a traditional production process with various coproducts, but it will help to motivate and explain allocation. In this example, consider a truck transporting different fruits and vegetables. The truck is driven from a farm to a market, as in the photo at the beginning of this chapter. For this one-way trip, the truck consumes 5 liters of diesel fuel, and it emits various pollutants (not quantified in this example). If apples, watermelons, and lettuce were the only three produce items delivered, the collected data might show that the truck delivered produce in your measured trip with the values shown in Figure 6-2. "Per type" values are the per item values multiplied by the number of items for each type of fruit or vegetable.
Apples Watermelon Lettuce Total
Items Number 100 25 50 175 items
Per Item 0.2 kg 2 kg 0.4 kg -
Mass Per Type 20 kg 50 kg 20 kg 90 kg
Market Value Per Item Per type $0.40 $40 $4 $100 $1 $50 $190
Figure 6-2: Summary Information of Fruits and Vegetables Delivered (per item and per type)
If we focus on determining how to allocate the diesel fuel, our LCA-supported question becomes, "how much of the diesel fuel use is associated with each item of fruit and vegetable?" To answer this, we need to do an allocation, which requires only simple math. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
159
The allocation process involves determining the allocation factor, or the quantitative share of flows to be associated with each unit of product, and then multiplying the unallocated flow (in this case, 5 liters of diesel fuel) by these allocation factors to find the allocated flows. Before we show the general equation for doing this, we continue with the produce delivery truck example. We use significant digits casually here to help follow the methods. If the allocation method chosen was based on number of items in the truck, from Figure 6-2, there are 175 total items, so the flow per unit (items) is 1/175 items. Each item of produce in the truck, regardless of the type of produce, would be allocated (1/175 items)*(1 item) *(5 liters) of diesel, or 0.029 liters. The value (1/175 items)*(1 item) is the allocation factor, 5 liters is the unallocated flow, and 0.029 liters is the allocated flow. Alternatively, in a mass-based allocation, the total mass transported was 90 kg. The flow per unit of mass is 1/90 kg, and each apple would be allocated 0.2 kg*(1/90 kg)*5 liters, or 0.011 liters of diesel. Finally, the total market value of all produce on the truck is $190. The flow per dollar is 1/$190, and each apple would be allocated $0.4*(1/$190)*5 liters, or 0.01 liters of diesel. The allocation factors and allocated flows of diesel fuel for each fruit and vegetable are shown in Figure 6-3. The results show that the diesel fuel allocated is quite sensitive to the type of allocation chosen for apples and watermelon – the diesel fuel allocated varies for apples from 0.01 to 0.029 liters (a factor of 3), and for watermelon, from 0.029 to 0.11 liters (a factor of 4). The allocated flow of diesel for lettuce is much less sensitive – varying from 0.022 to 0.029 liters (only about 30%). Item Allocation Allocated factor flow (liters) Apples Watermelon Lettuce
1 item * 1/175 item
0.029
Mass Allocation Allocated factor flow (liters) 0.2 kg * 1/90 kg 0.011 2 kg * 1/90 kg 0.11 0.4 kg * 1/90 kg 0.022
Economic Allocation Allocated factor flow (liters) $0.40 *1/$190 0.01 $4 *1/$190 0.11 $1 *1/$190 0.026
Figure 6-3: Allocation Factors and Allocated Flows Of Diesel Fuel per Type of Produce
To validate that our math is correct, we check that the sum of the allocated flows equals the unallocated value (5 liters). For allocation by items, 0.029 l/item * 175 items = 5.075 liters. By mass, the check is 0.011*100 + 0.11*25 + 0.022*50 = 4.95 liters. For price, the check is 0.01*100 + 0.11*25 + 0.026*50 = 1+2.75+1.3 = 5.05 liters. The allocations appear correct, and the slight discrepancies from 5 liters are due to rounding. The estimates from Figure 6-3 could be used to support a cradle to consumer LCI of energy use for bringing fruit to market. If you had process data on energy use for producing (growing) an apple, for instance, you could expand the scope of your study by adding one of the allocated flows, i.e., 0.029, 0.011, or 0.01 liters of diesel fuel for transport. As noted above, a key concern is the choice of allocation method (or methods) in support of such a study. While the Standard says the first priority is to use a physical relationship-based factor, the larger issue is whether any of the allocation methods would individually lead to a Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
160
Chapter 6: Analyzing Multifunctional Product Systems
different result. If in the apple LCI, for instance, you chose the economic allocation over the item-based allocation, you would be choosing a factor that represents 3 times less estimated transport energy. If the energy used to grow the apple was otherwise comparable in magnitude to the energy required for transport, then the choice of transportation allocation method could have a significant effect on the overall result. In this case, the choice of allocation may be construed as biasing the overall answer. Since the Standard suggests using sensitivity analyses, the best option may be to show the cradle to consumer results using all three types of allocation. The same math could be used to allocate other flows if available, such as data on an unallocated output flow of 10 kg of CO2 emissions from the truck. Since the allocation factors represent flow shares of individual pieces of fruit, we use the same allocation factors as we did for diesel fuel to allocate the CO2 emissions. In this case, the item-based allocation flow would be 1/175th of the 10 kg of CO2, or 0.057 kg of CO2 for any type of produce. A mass-based allocation for each apple would distribute 0.2 kg/90 kg * 10kg = 0.022 kg of CO2. Of course, all of the allocated flows of CO2 would have a value exactly double those of diesel (since there are 10 kg versus 5 liters of unallocated flow). The relative sensitivities of the various allocation choices would be the same. Note that in the delivery truck example, it was implicitly assumed that all of the produce were sold at market - we expected to get $190 in revenue. Aside from being a convenient assumption, it also implies that the truck would return back to the farm with no produce. One could argue that this empty return trip (referred to as backhaul in the transportation industry) requires additional consumption of fuel and generates additional air emissions that should be allocated to the produce sold at market. Given the weight of the fruit compared to the total weight of the truck, it's likely the backhaul consumed a similar amount of fuel, and thus, adding the backhaul process might double the allocated flows of diesel for delivery in an updated cradle to consumer LCA. For larger trucks or ocean freighters, an empty backhaul may consume significantly less fuel. Regardless, these considerations represent potential additions to the system boundary compared to the delivery alone. The delivery truck example is not just an example chosen to simplify the discussion of allocation. Indeed, similar process data would have to be allocated to support different LCIs and LCA studies. For example, a study on the LCA of making purchases online versus in retail stores might allocate the energy required for driving a UPS or FedEx delivery truck amongst the packages delivered that day. It is not obvious which of the allocation methods is best, nor is it obvious how the allocated results might change with the change of the method. The mass of the boxed products is potentially a bigger factor in how much fuel is used, and the variation in the value of the boxes is likely much higher (especially on a per unit mass basis!) than in our simple produce delivery truck example.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
161
Now that the general quantitative methods of allocation have been discussed, Equation 6-1 represents a general allocation equation, recalling that unit processes generally have multiple unallocated flows (in this case, indexed by i). Every allocated flow can be associated with each of the n co-products (indexed by j) using the product of the unallocated flow and the allocation factor for each co-product. The fraction on the right hand side of Equation 6-1 is the previously defined allocation factor, which divides the unit parameter of the co-product j, wj (e.g., mass per unit), for the chosen allocation method by the product of the number of units (m) and the unit parameters for all n co-products (with the sum indexed by k). 𝐴𝑙𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑓𝑙𝑜𝑤!,! = 𝑈𝑛𝑎𝑙𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑓𝑙𝑜𝑤! ∗
𝑤! (6 ! !!! 𝑚! 𝑤!
− 1)
Applying this equation to the truck delivery example, the unallocated flow of diesel fuel is 5 liters, the mass-based allocation factor for apples is the mass per apple divided by the sum of mass of all of the produce in the truck, or 0.2 kg / (0.2 kg*100+2 kg*25+0.4 kg*50) = 0.00222, so the allocated flow per apple is 0.011 liters. Using these values in Equation 6-1 generates all of the results in Figure 6-3. While this transportation example is useful for explaining allocations, the equation and general method are useful in deriving allocations for other unit processes. In a subsequent section, we discuss the ways in which allocated flows are implemented and documented in existing LCI data modules by looking at actual data modules. Allocation methods within the scope of a study should be as consistent as possible, and comparative studies should use the same allocation methods (e.g., allocating all flows on a mass basis) for each system or process. Due to challenges associated with data acquisition, this may prove difficult, so at least analogous processes should be allocated in the same way (e.g., all refining processes on a mass basis). Regardless, all allocation methods and deviations from common allocation assumptions should be documented.
Allocation in Resource Recovery Systems The Standard provides additional detail for allocation in processes and systems where resources are recovered. Resource recovery processes are those where resources are reused, recycled, composted, etc. Additional detail is provided in the Standard because recovery systems have inputs and output flows that are shared across multiple systems. For example, virgin plastic may be manufactured and then recovered and used in various iterations of recycled plastic products, such as plastic fasteners, bottles, and packaging. This view may imply that the flows from the virgin production of plastic are allocated across all subsequent product systems. However, resource recovery systems lead to various effects, such as
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
162
Chapter 6: Analyzing Multifunctional Product Systems
potential changes in material properties (e.g., the durability of the material may be adversely affected due to reprocessing) that need to be considered. One way that recovery systems can be characterized is as to whether the resource is recovered into the same kind of product system or not, i.e., whether open or closed loop recovery occurs. In closed loop systems, the recovered resource is recovered into the same product system (e.g., aluminum cans are recycled into aluminum cans). In open loop systems, the initial material or resource is subsequently used in alternative systems that, over time, are different than the initial system (e.g., plastic is recovered into different products, or in various recovery systems related to paper and paperboard products). Such systems are often associated with changes in material properties, as motivated above. A popular form of open loop procedure is cascade recycling, where an initial virgin material is continually used in various processes, losing some quality in each process, until it eventually can no longer be recovered into another system. At the end of life, such materials must be landfilled, burned for energy, etc. The Standard provides guidance for both open and closed loop systems. For closed loop systems (or open loop systems where material properties do not change), allocation is not needed because there are no processes whose flows must be shared across the systems in the extended life cycle. For open loop systems, on the other hand, allocation is needed, and the set of processes whose flows need to be shared are explicitly identified and allocated (e.g., the material recovery process from the initial product life to the second use of the material). As mentioned above, in open-loop systems, allocation occurs, and in order of preference the method should consider mass, economic, and the number of uses as the basis. Recall that the Standard defines a product as, "any good or service," which implies each product has value, else it might be classified more generally as a release or waste. There are interesting implications when the perspective changes regarding an output of a process from a product to a waste, or vice versa. For example, fly ash releases from electricity generation have historically been impounded but now may have value as a feedstock for alternative materials like cement or wallboard. Another example pertains to zinc production. Historically, the heavy metals lead and cadmium have been co-products of zinc mining, and LCA studies allocated flows across the three products. Increasing global regulation has suppressed demand for lead and cadmium, and they are now generally waste products. Results of past LCAs would need to be updated to assign no flows to the wastes, since only products are subject to allocation.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
163
Allocation Example from LCI Databases Existing LCI databases have information on many unit processes with co-product outputs. When used with search engines and LCA software, the unallocated and allocated flows may be found, and comparing them helps to understand the link between the unallocated and allocated data. The US LCI database provides various process data modules where there is already available information on products and co-products, and thus excellent examples to learn about allocation. Two prime examples are for the two refinery processes in the US LCI database, named Petroleum refining, at refinery and Crude oil, in refinery9. These two unit processes provide very similar, yet different, LCIs for a refinery. If you search for "refinery" or "refining" in the Digital Commons/NREL website (as demonstrated in Chapter 5), various unit processes and allocated co-products are returned, including: •
Petroleum refining, at refinery
•
Diesel, at refinery (Petroleum refining, at refinery)
•
Gasoline, at refinery (Petroleum refining, at refinery)
•
Crude oil, in refinery
•
Diesel, at refinery (Crude oil, in refinery)
•
Gasoline, at refinery (Crude oil, in refinery)
The co-products in the returned result can be identified because the name of the co-product as well as the name of the unallocated refinery process model (in parentheses) is given. The connection between these two types of data is discussed in more detail below. Following the format of Chapter 5, Figure 6-4 shows the data available for the US LCI unit process Petroleum refining, at refinery. The table has been abridged by removing various elementary flow inputs to save space and reduced to 3 significant figures. The 'category' and 'comment' fields have also been removed. This crude oil refining process shows 9 italicized product flows representing various fuels and related refinery outputs, e.g., diesel fuel, bitumen and refinery gas. The last row also shows a functional unit basis for the unit process, i.e., per 1 kg of petroleum refining, which is just a bookkeeping reference entry and does not represent an additional product. Unlike other process data modules we discovered in Chapter 5, the outputs of this crude oil process are not all in a singular unit such as 1 gallon or 1 kg. Instead, the product flow quantities are 0.252 liters of diesel, 0.57 liters of gasoline, etc. The reason for this difference should be clear – there are multiple products! It 9
Other quite accessible examples in the US LCI database include agricultural and forestry products.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
164
Chapter 6: Analyzing Multifunctional Product Systems
is not possible to have a single set of raw inputs and outputs all be normalized to '1 unit of product' when more than one product exists. This further motivates the notion that the refinery inputs and outputs would need to be allocated. Flow
Type
Unit
Amount
Electricity, at grid, US, 2008
ProductFlow
kWh
0.143
Natural gas, combusted in industrial boiler
ProductFlow
m3
0.011
Residual fuel oil, combusted in industrial boiler
ProductFlow
L
0.027
Liquefied petroleum gas, combusted in industrial boiler
ProductFlow
L
0.001
Transport, barge, diesel powered
ProductFlow
tkm
0.000
Transport, barge, residual fuel oil powered
ProductFlow
tkm
0.001
Transport, ocean freighter, diesel powered
ProductFlow
tkm
0.490
Transport, ocean freighter, residual fuel oil powered
ProductFlow
tkm
4.409
Transport, pipeline, unspecified petroleum products
ProductFlow
tkm
0.652
Crude oil, extracted
ProductFlow
kg
1.018
Dummy_Disposal, solid waste, unspecified, to sanitary landfill
ProductFlow
kg
0.006
Benzene
ElementaryFlow
kg
1.08E-06
Carbon dioxide, fossil
ElementaryFlow
kg
2.51E-04
Carbon monoxide
ElementaryFlow
kg
4.24E-04
Methane, chlorotrifluoro-, CFC-13
ElementaryFlow
kg
2.18E-08
Methane, fossil
ElementaryFlow
kg
3.70E-05
Methane, tetrachloro-, CFC-10
ElementaryFlow
kg
1.36E-09
Particulates, < 10 um
ElementaryFlow
kg
3.15E-05
Particulates, < 2.5 um
ElementaryFlow
kg
2.31E-05
SO2
ElementaryFlow
kg
2.47E-04
Diesel, at refinery
ProductFlow
L
0.252
Liquefied petroleum gas, at refinery
ProductFlow
L
0.049
Gasoline, at refinery
ProductFlow
L
0.57
Residual fuel oil, at refinery
ProductFlow
L
0.052
Bitumen, at refinery
ProductFlow
kg
0.037
Kerosene, at refinery
ProductFlow
L
0.112
Petroleum coke, at refinery
ProductFlow
kg
0.060
Refinery gas, at refinery
ProductFlow
m3
0.061
Petroleum refining coproduct, unspecified, at refinery
ProductFlow
kg
0.051
Petroleum refining, at refinery
ProductFlow
kg
1
Inputs
Outputs
Figure 6-4: US LCI Database Module for Petroleum refining, at refinery
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
165
The comment fields in the Petroleum refining, at refinery process, which were removed in Figure 6-4, are excerpted in Figure 6-5. They contain notes related to how the input and output flows could be allocated to the co-products, and how the creators of this data module derived the converted allocated values for the specific co-products on a 'per unit of product' basis. The result of this procedure is the various co-product based LCI modules listed above. Product Diesel, at refinery Liquefied petroleum gas, at refinery Gasoline, at refinery Residual fuel oil, at refinery Bitumen, at refinery Kerosene, at refinery Petroleum coke, at refinery Refinery gas, at refinery Petroleum refining co-product, at refinery
Comment Mass (0.2188 kg/kg output) used for allocation. Mass (0.0266 kg/kg output) used for allocation. Mass (0.4213 kg/kg output) used for allocation. Mass (0.0489 kg/kg output) used for allocation. Mass (0.0372 kg/kg output) used for allocation. Mass (0.0910 kg/kg output) used for allocation. Mass (0.0596 kg/kg output) used for allocation. Mass (0.0451 kg/kg output) used for allocation. Mass (0.0515 kg/kg output) used for allocation.
Figure 6-5: Comments Related to Allocation for Co-Products Of Petroleum refining, at refinery
The available US LCI database Microsoft Excel spreadsheet for the Petroleum refining, at refinery process module shows these allocation factors in the X-Exchange worksheet, in columns to the right of the comment fields. The summary product flow of "per kg of petroleum refining" and the comment fields make clear that allocation is based on the physical relationship of mass of the various products. For example, the first row of Figure 6-5 shows the data needed to create an allocation factor for the diesel co-product. The comment in Row 1 says that 0.2188 kg diesel is produced per kg refinery output, or in other words, 21.88% of the refinery product represented in the data for this unit process becomes diesel on a mass basis. The value 21.88% is the allocation factor for diesel. Likewise, 2.66% by mass becomes LPG, and 42.13% becomes gasoline. The sum of all of the mass fractions provided is 1 kg, or 100% of the mass of total refinery output. With these values, you could use the information in Figure 6-5 to transform the unallocated inputs and outputs into allocated flows for your desired co-product. Management of different units and conversions can be a complicating factor in transforming from unallocated to allocated values in data modules. Figure 6-4 shows all of the inputs and outputs connected to the refinery on a basis of a functional unit of 1 kg of refined petroleum product (which we noted was just a book-keeping entry). Note that the co-products have multiple units– liters, cubic meters, and kilograms. But the product flow of diesel given in the unit process is 0.252 liter, rather than 1 liter, or rather than 1 kg. We are likely interested in an allocated flow per 1 unit of co-product, which means our previous allocation equation needs an additional term, as in the generalized Equation 6-2. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
166
Chapter 6: Analyzing Multifunctional Product Systems
𝐴𝑙𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑓𝑙𝑜𝑤!,! =
𝑈𝑛𝑎𝑙𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑓𝑙𝑜𝑤! ∗ 𝑢𝑛𝑎𝑙𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑓𝑙𝑜𝑤 𝑢𝑛𝑖𝑡𝑠
𝑤! ! !!! 𝑚! 𝑤!
∗
𝑢𝑛𝑎𝑙𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑓𝑙𝑜𝑤 𝑢𝑛𝑖𝑡𝑠 (6 − 2) 𝑐𝑜– 𝑝𝑟𝑜𝑑𝑢𝑐𝑡! 𝑢𝑛𝑖𝑡𝑠
Since the allocation factor for diesel in the refinery on a mass basis is 0.2188 (21.88%), that amount of each of the inputs of the refinery process would be associated with producing 0.252 liters of diesel. Using Equation 6-2, the allocated flow of crude oil (row 10 of Figure 6-4) needed to produce 1 liter of diesel fuel is: 𝐴𝑙𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑓𝑙𝑜𝑤!"#$%,!"#$#% =
1.018 𝑘𝑔 𝑐𝑟𝑢𝑑𝑒 𝑜𝑖𝑙 1 𝑘𝑔 𝑟𝑒𝑓𝑖𝑛𝑒𝑟𝑦 𝑜𝑢𝑡𝑝𝑢𝑡 0.884 𝑘𝑔 𝑐𝑟𝑢𝑑𝑒 𝑜𝑖𝑙 ×0.2188 × = 1 𝑘𝑔 𝑟𝑒𝑓𝑖𝑛𝑒𝑟𝑦 𝑜𝑢𝑡𝑝𝑢𝑡 0.252 𝑙𝑖𝑡𝑒𝑟𝑠 𝑑𝑖𝑒𝑠𝑒𝑙 𝑙𝑖𝑡𝑒𝑟𝑠 𝑑𝑖𝑒𝑠𝑒𝑙
Other comments in the US LCI data module (not shown here) note that the assumed density of diesel is 1.153 liter/kg, so this allocated result means 0.884 kg crude oil is needed to produce 1 liter /(1.153 liter/kg) = 0.867 kg of diesel. This mass-based ratio is comparable to the 1.018 kg crude oil needed to produce the overall 1 kg of refinery product10. While the 1.153 liter/kg value may or may not be consistent with a unit conversion factor you might find on your own, it is important to use the same ones as used in the study, else you may derive odd results, such as requiring less than 1 kg of crude to produce 1 kg of diesel. Similarly, the amount of crude oil needed to make 1 liter of gasoline would be 1.018 𝑘𝑔 𝑐𝑟𝑢𝑑𝑒 𝑜𝑖𝑙 1 𝑘𝑔 𝑟𝑒𝑓𝑖𝑛𝑒𝑟𝑦 𝑜𝑢𝑡𝑝𝑢𝑡 0.752 𝑘𝑔 𝑐𝑟𝑢𝑑𝑒 𝑜𝑖𝑙 × 0.4213 × = 1 𝑘𝑔 𝑟𝑒𝑓𝑖𝑛𝑒𝑟𝑦 𝑜𝑢𝑡𝑝𝑢𝑡 0.57 𝑙𝑖𝑡𝑒𝑟𝑠 𝑔𝑎𝑠𝑜𝑙𝑖𝑛𝑒 𝑙𝑖𝑡𝑒𝑟𝑠 𝑔𝑎𝑠𝑜𝑙𝑖𝑛𝑒
or, using the US LCI module's assumed density of 1.353 liter/kg, we need 0.752 kg of crude oil to produce 1/1.353 = 0.739 kg of gasoline. The same allocation factors are used to transform the other unallocated flows into allocated flows (e.g., the many other inputs and outputs listed in Figure 6-4) per unit of fuel. Some of the results above may be unintuitive – i.e., that you started with a process making 0.252 liters of diesel, but ended up needing 0.884 kg of crude petroleum to produce 1 liter of diesel. Or, if the refinery has multiple products, why is more than 1 unit of crude oil (rather than just a fraction of 1) needed to produce a unit of refined fuel? Figure 6-6 tries to rationalize the potential sources of confusion above by showing how much crude oil is needed to produce varying quantities and units of the 9 co-products. This information is based on the input flow of crude oil into the unit process (1.018 kg crude / kg refinery product), the allocation factors from Figure 6-5, and the unit conversions provided in the US LCI data module. The results show the allocated mass of crude oil per the flows given in the original unit process from Figure 6-4 (e.g., per 0.252 liters of diesel as above),
10
This could also be represented as adding yet another unit conversion at the end of Equation 6-2, from liters to kg of diesel.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
167
per the varying unitized flow of each product (e.g., per 1 liter of diesel or 1 kg of bitumen), and with all flows converted on a per kg of co-product basis. Allocated Crude oil (kg) per Process Units
Norm Process Units
kg product
Diesel (liters)
0.219 / .252 l
0.884 / l
1.018
Liquefied petroleum gas (liters)
0.027 / .049 l
0.552 / l
1.016
Gasoline (liters)
0.421 / .57 l
0.752 / l
1.018
0.049 / .052 l
0.962 / l
1.019
0.037 / .037 kg
1.018 / kg
1.018
Co-product (units given in US LCI module)
Residual fuel oil (liters) Bitumen (kg) Kerosene (liters)
0.091 / .112 l
0.824 / l
1.018
Petroleum coke (kg)
0.060 / .060 kg
1.018 / kg
1.018
Refinery gas (m3)
0.045 / .061 m3
0.751 / m3
1.019
Petroleum refining coproduct, unspec. (kg)
0.051 / .051 kg
1.018 / kg
1.018
Figure 6-6: Comparison of Allocated Quantities of Crude Oil for Nine Co-products of Petroleum refining, at refinery Process, for Various Co-product Units (l = liter).
The first two result columns summarize results using the methods just demonstrated for gasoline and diesel. The main difference between them is whether the unit basis of the coproduct is the fractional unit value given in the process data, or whether it has been converted to a per 1 unit of co-product basis. There were three different units of product presented (liter, kg, and m3), which may otherwise distort a consistent view of the effects of allocation. The final column of Figure 6-6 may be surprising, as all of the co-products have the same requirement of crude oil per kg of product (about 1.018, the original unit process flow). If the whole point of allocation was to assign flows to the different products, why does it appear that they all have the same allocation value? In this case, it is because a massbased allocation was used, thus the appearance of the crude oil per kg is constant (the same effect would be have been seen in Figure 6-3 – the mass based allocation factor was 0.0055 liters per kg for all produce). Since we were looking at an energy system where the default units were liters or m3, it disguised this result. It is pervasive, though, in LCA, and is an expected result. If, for instance, we performed an economic-based allocation, the column 'crude oil per dollar of product' would have constant values. Regardless, you can follow the use of allocated input and output flows for a co-product based on data modules in the US LCI model by exploring their data and metadata in openLCA, SimaPro, etc., using the same methods shown in the Advanced Material for Chapter 5.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
168
Chapter 6: Analyzing Multifunctional Product Systems
Avoiding Allocation When allocation was introduced, it was stipulated that the Standard says that the main goal should be to avoid it. More specifically, the Standard says that allocation should be avoided in one of two ways, each of which has the goal of making more direct connections between the input and output flows of a process and its products, so as to remove the need for allocating those flows. The first recommended alternative to allocation is to sub-divide or disaggregate the unit process into a sufficient number of smaller sub-processes, until none of these processes have multiple products. Thus the link can be made between inputs and outputs and a single product for each process (and thus the system). While this solution is very attractive, it requires being able to collect additional data for all of the new sub-processes. It may also be a puzzling suggestion because the Standard elsewhere defines a unit process as "the smallest element in the inventory analysis for which input and output data are quantified." If a unit process can be broken into sub-processes, then it was not the smallest possible element in the first place. However, this reminds the analyst to create processes as distinct and small in boundary as possible given access to data, so as to avoid allocation issues. In this case, it means trying to collect data at sufficient resolution as to be able to have processes and systems with singular product outputs. Figure 6-7 illustrates a simple case of disaggregation, where a process similar to Figure 6-1 with multiple product outputs is subdivided into multiple processes (in this case, 1 and 2), with distinct Products A and B. In reality, the disaggregated system may need to have more than 2 processes, and some processes may have only intermediate flows leading to an eventual product output. Generally, creating process models at a lower level (alternatively called a higher level of resolution) is no different than creating a higher-level process model that requires more effort and data. In short, the goal of this 'divide and conquer' style approach is to ensure the result is a set of unit processes with single outputs and no coproducts.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
169
Figure 6-7: Disaggregation into Sub-processes
There are other possible ways of avoiding allocation in this way without disaggregating processes, such as by drawing different explicit boundaries around the system, so that coproducts are not included. For example, an overall oil refinery process has hundreds of processes and many products. If only a particular product is of interest, then it may be possible to draw a smaller and more explicit boundary around the inputs, outputs, and processes in a refinery needed just for that product (and that has no connection to other products). Thus, allocation may be avoided. Such an exercise might be impossible in a complex system as a refinery since simultaneous production of outputs is common. The second alternative to allocation recommended in the Standard is system expansion, or, "expanding the product system to include the additional functions related to the coproducts." The Standard offers little detail on this method, so practitioners and researchers have developed various interpretations and demonstrations of system expansion, such as Weidema (2000). System expansion leverages the facts that systems with multiple product outputs are typically multifunctional, and that LCA requires definition and comparison of systems on a functional unit basis. System expansion adds production of outputs to product systems so that they can be compared on the basis of having equivalent function. Going back to the earlier example of a system producing heat and electric power, each of these products provides a different function – the ability to provide warmth and the ability to provide power. In a hypothetical analysis based only on a functional unit of electricity, comparing a CHP system with a process producing only electricity would be unequal. Figure 6-8 generalizes the unequal comparison of the two different systems, one with a single product and function and one with two products and functions. Expansions would be similar for more than two products. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
170
Chapter 6: Analyzing Multifunctional Product Systems
Figure 6-8: Initial Comparison of Multifunctional Product Systems (Source: Tillman 1994)
Note that while the two systems provide the identical Function 1 (providing electricity in our example), the technological process behind providing that function is not generally assumed to be identical (one is Product A and the other is Product C). If the functional unit of a study is, for instance, 'moving a truck 1 km', the product to support that function might be gasoline or diesel fuel. Likewise, the product could be the same, but the technology behind making it different. In such an example, the product could be electricity, but generated using renewable or non-renewable generation technology. In a comparative LCA, consideration of functions is important. Product systems with a different number of functions can be analyzed by adding functions and products across systems as needed until the systems are equal (i.e., by 'expanding the system boundary' for systems that do not have enough function outputs). Considering Function 1 to be providing power and Function 2 to be providing heat, for instance, system expansion allows the product systems in Figure 6-8 to be compared by adding processes representing the production of heat to the system (making Product C). Figure 6-9 shows the result of such a system expansion.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
171
Figure 6-9: System Expansion by Adding Processes (Source: Tillman 1994)
The system expansion focuses on ensuring that various systems provide the same functions. However, the additional functions may be provided via various products, and thus a variety of alternative technologies or production processes. When considering how to model additional function in the expanded process, it may be done by modeling the identical product as in the multifunction products (e.g., natural-gas fired electricity), or with an alternative product and/or technology providing the same function (e.g., gasoline or diesel, renewable or non-renewable electricity). Of course, using different production technologies in system expansion may to lead to significantly different results, but the reality of markets and practice may support this case. Alternative technology assumptions for the expanded system should be considered in a sensitivity analysis. Alternative technologies may be hard to determine or justify. They should be reasonable, typical in a market, and not overly bias the results. For example, in expanding a system, electricity produced as a co-product by burning lignin in a biofuel production plant may be alternatively produced by US average electricity. Likewise, solar cells may be a poor choice to expand a system in comparison of a fossil-based process that produces electricity. The general example presented in Figure 6-9 is straightforward, and system expansion is not necessarily more complex than disaggregation – in fact, it can often be quite simple. One should not view either disaggregation or system expansion as the preferred alternative to allocation, and likewise not generalize one or the other as being more difficult. In a particular study, performing disaggregation could be time and/or data prohibitive and thus system expansion is the only alternative to allocation. On the other hand, system expansion may be hard to motivate or justify given challenges in identifying alternative technologies for the expanded system, which leads to spending time and effort in disaggregating the processes. The refinery example discussed earlier is a useful example. The high-level refinery process data module shown is highly aggregated (and as a result it has many product Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
172
Chapter 6: Analyzing Multifunctional Product Systems
outputs). Far more detailed process models of refineries have been developed and could be used if needed for a study in lieu of using the allocated refinery module presented. While this may take substantial effort, it avoids the need for the system expansion approach that would require creating larger comparative product systems with added production for many more refinery outputs. It is hopefully intuitive that the math behind the system expansion example in Figure 6-9 that 'adds to process 2' would lead to the same relative result as 'subtracting' (i.e., crediting) the same process data from the results of Process 1. This equivalent method is shown in Figure 6-10, and it is referred to as the avoided burden approach. This is still considered system expansion because the boundary was expanded for one of the systems (but the numerical results are credited instead of added).
Figure 6-10: System Expansion of a Multifunction Process via Subtraction
The result of either allocation or system expansion methods may cause other issues in your study, since modifying the original model may mean your study scope changes (and potentially your goal as well). In the heat and power example, a study originally seeking to compare only the effects of 'generating 1 kWh of electricity' in two different systems (i.e., Product A vs. Product C) would need to have its study design parameters adjusted to consider systems providing power and an equivalent amount of heat (via Product B), e.g., 'generating 1 kWh of electricity and producing 100 MJ of heat'. For the heat and power example, the US LCI database has various data to support a system expansion effort. Figure 6-11 shows abridged US LCI data for CHP.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
Flow
173
Type
Unit
Amount
ProductFlow
kg
4E-02
ProductFlow
m3
0.004
Carbon dioxide, fossil
air/unspecified
g
8
Carbon dioxide, biogenic
Inputs Wood, NE-NC hardwood Natural gas, combusted in industrial boiler Outputs air/unspecified
g
70
Heat, onsite boiler, hardwood mill average, NE-NC
ProductFlow
MJ
1
Electricity, onsite boiler, hardwood mill average, NE-NC
ProductFlow
kWh
5.0E-05
Figure 6-11: Process Data for Hardwood Combined Heat and Power (CHP) (abridged, adapted from various US LCI database modules)
The US LCI database also has process data for gas-fired electricity, as in Figure 6-12. Flow
Type
Unit
Amount
ProductFlow
m3
0.3
air/unspecified
kg
0.6
ProductFlow
kWh
1
Inputs Natural gas, processed, at plant Outputs Carbon dioxide, fossil Electricity, natural gas, at power plant
Figure 6-12: Process Data for Gas-Fired Electricity Generation (abridged, adapted from US LCI database module Electricity, natural gas, at power plant)
Biogenic emissions occur as a result of burning or decomposing bio-based products. The 'biogenic' CO2 emissions in the CHP process data arise from burning the wood, and the fossil CO2 emissions come from burning the gas. The carbon in the wood comes during its growth cycle via natural uptake of carbon (in this case through photosynthesis). In lieu of crediting the natural product for this same uptake of carbon, which would lead to a net of zero emissions, the biogenic emissions of CO2 are considered to be neutral. As such, biogenic carbon emissions are accounted for but not generally added to fossil or other sources. Example 6-1 demonstrates how to use the US LCI data for a system expansion involving CHP.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
174
Chapter 6: Analyzing Multifunctional Product Systems
Example 6-1: Compare the direct CO2 emissions of the electricity from a hardwood mill combined heat and power (CHP) process with US natural gas-fired electricity. Answer: As motivated above, the requested comparison is problematic, as the CHP process has two outputs with two functions and the second has only one with one function. This is a good case for using system expansion, and alternative process data is needed for producing heat to add to the existing gas-fired electricity process, as in Figure 6-9. Using an alternative technology assumption, we can say that heat is often provided by burning natural gas, e.g., in a furnace, using abridged US LCI data in Figure 6-13. Flow
Type
Unit
Amount
ProductFlow
m3
1
Carbon dioxide, fossil
air/unspecified
kg
2
Heat, natural gas, in boiler
ProductFlow
MJ
30
Inputs Natural gas, processed, at plant Outputs
Figure 6-13: Process Data for Generating Heat (from US LCI Heat, natural gas, in boiler)
Using the hardwood mill boiler as the baseline, our goal would be to compare the two systems based on a common functional unit of 'producing 1 MJ of heat and 5E-05 kWh of electricity' (or, alternatively, to the scaled functional unit 'producing 1 kWh of electricity and 20,000 MJ of heat'). Tracking the CO2 emissions, the CHP process emits 78 grams of CO2 for the 1 MJ heat/5E-05 kWh electricity functional unit, of which 70 g are biogenic emissions and 8 g are fossil-based. For the system expanded electricity and heat processes, generating 5E-05 kWh of electricity would emit 0.6 kg CO2 / kWh * 5E-05 kWh = 0.03 g CO2, and producing 1 MJ of heat would emit 2 kg/30 MJ = 67 g CO2. However, all of the CO2 in this expanded system is fossil-based. In fact, the systems are roughly comparable in total CO2 emissions, but the CHP unit using wood input has 59 g, or about 90%, less fossil CO 2 emissions (8 g versus 67.03 g). Instead of adding the heat process to the electricity process, we could credit the CHP process for avoided production of heat, as in Figure 6-10. The 2 kg/30 MJ = 67 g of fossil CO 2 from Figure 6-13 would be subtracted from the CHP system. In this case, it means the CHP system has 8 g – 67 g = -59 g of fossil CO2 emissions per 5E-05 kWh of electricity! The relative difference between the systems is the same (59 g less fossil CO2). It is worth noting that by using the alternative process data for heat production, the CHP system effectively has the same emissions factor (2 kg fossil CO2 / 30 MJ) as the alternate process. While we have not 'allocated' the CHP system's emissions, we have assigned the same share of emissions as in the alternate process.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
175
The final result of 'negative emissions' when using the subtraction method, as shown in Example 6-1, has generally caused some stakeholders to suggest not using this method, but again, the relative difference is the important result, regardless of the sign convention used. This negative result is an odd outcome, especially as compared to allocation, which only leads to positive flows. But, the Standard suggests avoiding allocation. If, for the sake of discussion, an energy-based allocation was done instead in Example 6-1, then the CHP system produces 1 MJ of heat and 0.00018 MJ of electricity (at 3.6 MJ / kWh) for a total of 1.00018 MJ. Thus the fossil CO2 would have been allocated 99.98% (1 MJ/1.00018 MJ) to the 1 MJ heat and 0.02% (0.00018 MJ/1.00018 MJ) to the 5E-05 kWh of electricity produced. Comparing the allocated CHP process and its explicit estimate of emissions for electricity to the natural-gas fired electricity process would be 0.0016 g CO2 versus 0.03 g per 5E-05 kWh, or 95% less, a larger difference than before but not qualitatively different. The CHP example motivates the argument of which alternate production process is used in the expansion, and additionally whether the alternative production process chosen is representative of the average, some best or worst case, or merely based on the only data point available. While there is no explicit suggestion of which process should be chosen, the choice should be sufficiently documented. Further discussion on using average or other processes follows in the next section.
Expanding Systems in the Broader Context For the additional production added via system expansion to make the various product systems equivalent, so far discussions and examples have addressed what, on average, is an appropriate additional production. This is because, so far, only an attributional or descriptive approach has been discussed for LCA studies. Attributional LCAs seek to determine the effects now, or in the past, which inevitably means that our concerns are restricted to average effects. However, emerging practice and need in LCA often seeks to consider the consequences of product systems or changes to them. In consequential LCA studies, marginal, instead of average, effects are considered (Finnveden et al. 2009). Marginal effects are those effects that happen 'at the margin', and in economics refer to effects associated with the next additional unit of production. Furthermore, consequential analyses seek to determine what would change or need to change given the influence of changing product systems on markets. Using the CHP example, heat is currently, on average, produced by burning average domestic natural gas. But on the margin, it is likely that such gas may be shale gas from unconventional wells. In the future, perhaps some other alternative fuel would be the marginal source. Certain resources, products, and flows may be Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
176
Chapter 6: Analyzing Multifunctional Product Systems
quite scarce, and a significant demand for one resource in a product system could, on the margin, lead to an alternative that is radically different. Note that the average and marginal technologies used in an analysis could be identical – they do not need to be different – but when different, the effects associated with them could be substantially different. Consequential studies often use separate economic models to aid in assessing changes. For example, increased exploration and production of unconventional natural gas (e.g., shale gas) has been stated through attributional LCA studies as leading to 50% reductions in the carbon intensity of electricity generation, since natural gas on average would replace coalfired electricity generation which emits 50% less CO2 per kWh. Some of these studies were done when coal was the source of 50% of our electricity generation. But on the margin, increasing the supply of low-priced natural gas has the prospect of replacing not just coalfired generation but also electricity from nuclear power plants, which have very low carbon intensities. Using such marginal assumptions, and an economically driven electricity dispatch model, Venkatesh (2012) suggests that a consequence of cheap natural gas on regional electricity CO2 emissions could be reductions of only 7-15%, far lower than the 50% expected on average. Another consequential effect seen in LCA studies relates to how land is used and managed. Historically, studies related to activities such as biofuel production modeled only the effects of plowing and managing existing cropland for corn or other crops (known as direct land use). That means that the studies assumed that production would, on average, occur in places where the same, or similar, crop was already being grown, and thus, the impacts of continuing to do so are relatively modest. Recent studies (Searchinger 2008, Fargione 2008) highlighted the fact that increased use of land for crops used to make bio-based fuels in one part of the world can lead to conversion of other types of land, e.g., forests, into cropland in other parts of the world (a phenomenon called indirect land use change). In such cases, the carbon emissions and other effects of converting land into cropland are far higher. This consequence of increasing use of cropland for biofuels has been quantitatively estimated and added to the other LCA effects, and leads to results substantially different than those that do not consider this effect. In LCA, considering the market-based effects of a substitution typically leads to a discussion of displacement of products. Displacement occurs when the production of a co-product of a system displaces (offsets) production of another product in the market. The quantitative effect of displacement is that the flows from what would occur when producing the alternative product are 'credited' to the main product system because it is assumed that the displacement results in less production of the alternative product. A traditional example of displacement is for a system where electricity is produced as a co-product. In such situations, usually the electricity co-product is assumed to displace alternative production of electricity, typically assumed to be grid electricity. The effect of displacement is thus crediting (subtracting from) the inventory of the product system with the inventory of an Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 6: Analyzing Multifunctional Product Systems
177
equivalent amount of grid electricity. More practically, a co-product will often displace a different product. An often-stated displacement assumption in the biofuels domain is that the ethanol co-product dried distillers grains with solubles (DDGS) displaces animal feed. This is because DDGS are assumed to be a viable alternative food for livestock, and so the inventory flows of producing a functionally equivalent amount of animal feed are credited to the product system. Here that displacement is not one-to-one, instead DDGS typically displaces only about 60% of animal feeds on a mass basis given differences in protein content. Production of heat may not be functionally equivalent either, and a displacement ratio may be needed to make a relevant comparison in those cases. Both of these are examples of the displacement ratio (or displacement efficiency) that would be expected, and it may be greater than, less than, or equal to one based on cost, quality, or other factors. Displacement is a consequence of production and market availability of a co-product; however, even attributional LCAs can consider effects from displacement. Consideration of the price and quantity differences resulting from displacement would be an appropriate addition for a consequential LCA. These "ripple-through" market considerations are at the heart of marginal analysis and economics, and thus consequential LCA. Whether a study is attributional or consequential in nature should be specified along with other parameters of a study. Of course, a study could be interested in both average and marginal effects, so as to consider the relative implications of the introduction of a product system to the market. Considering consequences or marginal effects is not exclusively the domain of consequential LCA. It could be useful in an attributionally-scoped LCA, e.g., by assuming an offset associated with marginal electricity production, or one that requires system expansion to consider alternative and substitute production of both average and marginal products. In such a study, both the average and marginal results could be presented. The core of this textbook will continue to be aligned with attributional analysis and methods, and relevant notes and guides towards consequential methods will be included where applicable. Additional sources on consequential LCA and differences from attributional LCA are provided in the references section at the end of this chapter.
Sensitivity Analysis for Allocation and System Expansion Previous chapters have defined sensitivity analysis and discussed its place within the Standard. While the Standard clearly states that allocation should be avoided in favor of system expansion, as consistent with other LCA study issues, a primary concern is whether a particular assumption has a significant effect on the study results. In this case, we are concerned as to whether the qualitative and quantitative conclusions change based on our choice of allocation method, and/or whether we should perform system expansion instead Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
178
Chapter 6: Analyzing Multifunctional Product Systems
of allocation. The study results should explicitly note whether the specific choice of allocation method, system expansion, etc. has a significant effect on the results (i.e., whether other choices would have led to quantitatively and qualitatively different results). To perform a sensitivity analysis of allocation methods, the results are found using each of the alternative allocations. It is common to see a table or graph comparing the effect at the process or whole product system level. Figure 6-14 shows a graphical comparison of the three different allocation methods for the three different fruits in the delivery truck example at the beginning of the chapter.
Allocated Flow (liters)
0.12 0.1 0.08 0.06 0.04 0.02 0 Item
Mass Apples
Watermelon
Economic LeFunction" dialog box helper. As with the example shown above, as long as you first select the cell range of the expected result (with the appropriate m x n dimensions), enter the formula, and click CTRL-SHIFT-ENTER at the end, you will get the right results. You will see an error (or a result in only one cell) if you skip one of the steps. While a bit cumbersome, using array functions in Excel is straightforward and very useful for small vectors and matrices. Figure 8A-12 shows a screenshot where [I-A]-1 and [I+A] Y1 have been created. E-resource: A Microsoft Excel file solving Examples 8-1 through 8-3 is posted to the textbook website.
Figure 8A-12: Result of Array Formula in Microsoft Excel for Example 8-2
Note that you can perform vector and matrix math without using the Excel name feature. In this case, you would just continue using regular cell references (e.g., B2:C3 for the A matrix in the screenshot above). All of the remaining instructions are the same. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 8: LCA Screening via Economic Input-Output Models
233
Brief MATLAB tutorial for IO-LCA modeling – section by itself? This short primer on using Mathworks MATLAB is no substitute for a more complete lesson or lecture on the topic but will help you get up to speed quickly. It presumes you have MATLAB installed on a local computer with the standard set of toolboxes (no special ones required). MATLAB, unlike Microsoft Excel, is a high-end computation and programming environment that is often used when working with large datasets and matrices. It is typically available in academic and other research environments. When MATLAB is run, the screen is split into various customizable windows. Generally though, these windows show: •
the files within the current directory path,
•
the command window interface for entering and viewing results of analysis,
•
a workspace that shows a listing of all variables, vectors, and matrices defined in the current session, and
•
a history of commands entered during the current session.
In this tutorial, we focus on the command line interface and the workspace. Despite the brevity of the discussion included here, one could learn enough about MATLAB in an hour to replicate all of the Excel work above. MATLAB has many built-in commands, and given its scientific computing specialties, is designed to operate on very large (thousands of rows and columns) matrices when installed. Some of the most useful commands and operators for use with EIO models in MATLAB are shown in Figure 8A-11. Many commands have an (x,y) notation where x refers to rows and y refers to columns. Others operate on whole matrices. Working with EIO matrices in MATLAB involves defining matrices and using built-in operators much the same way as was done in the Excel examples above. Matrices are defined by choosing an unused name in the workspace and setting it equal to some other matrix or the result of an operation involving commands on existing matrices. MATLAB commands are entered at the command line prompt ( >> ) and executed by pressing ENTER, or placed all in a text file (called a .m file) and run as a script. If commands are entered without a semicolon at the end, then the results of each command are displayed on the screen in the command window when ENTER is pressed. If the semicolon is added before pressing ENTER, then the command is executed, but the results are not shown in the command window. One could look in the workspace window to see the results.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
234
Chapter 8: LCA Screening via Economic Input-Output Models
Command
Description of Result
zeros(x,y)
creates a matrix of zeros of size (x,y). This is also useful to "clear out" an existing matrix.
ones(x,y)
same as zeros, but creates a matrix of all ones.
eye(x)
creates an identity matrix of size (x,x). Note the command is not I(x), a common confusion.
inv(X)
returns the matrix inverse of X.
diag(X)
returns a diagonalized matrix from a vector X, i.e, where the elements of the vector are the diagonal entries of the matrix (like the identity matrix).
sum(X)
returns an array with the sum of each column of the input matrix. If X is a vector then the command returns the sum of the column.
size(X)
tells you the size of a matrix, returning (number of rows, number of columns). This is useful if you want to verify the row and column sizes of a matrix before performing a matrix operation.
A'
performs a matrix transpose on A, inverting the row and column indices of all elements of the matrix.
A*B
multiplies matrices A and B in left-to-right order and with usual linear algebra.
A.*B
element-wise multiplication instead of matrix multiplication, i.e., A11 is multiplied by B11 and the results put into cell11 of the new matrix (A and B must be the same size).
A,B
concatenates A and B horizontally.
A;B
concatenates A and B vertically.
clear all
empties out the workspace and removes all vectors, matrices, etc. Like a reset. Figure 8A-11: Summary of MATLAB Commands Relevant to EIO Modeling
In this section, courier font is used to show commands typed in to, or results returned from, MATLAB. For example, the following commands, entered consecutively, would "clear out" a matrix named "test_identity" and then populate its values as a 2x2 identity matrix: >> test_identity=zeros(2,2) >> test_identity=eye(2)
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 8: LCA Screening via Economic Input-Output Models
235
and the results consecutively displayed would be: test_identity = 0
0
0
0
test_identity = 1
0
0
1
The format of matrices displayed in MATLAB's command window is just as one would write them in row and column format. Matrices are populated with values by either importing data (not discussed here) or by entering values in rows and columns, where columns are separated by a space and rows by a semicolon. For example, the following command would create a 2x2 identity matrix: identity_2 = [1 0; 0 1] which would return the following result in the command window: identity_2 = 1
0
0
1
The workspace window has a list of all vectors or matrices created in the session. All are listed, and for small matrices individual values are shown. For larger matrices, only dimensions (m x n) are shown. Display of the dimensions is useful to ensure that you do not try to perform operations on matrices with the wrong number of rows and columns. Double clicking on a vector or matrix in the workspace opens a new window with a tabbed spreadsheet-like view of its elements (called the Variable Editor). It is far easier to diagnose problems in this editor window than in scrolling through the results in the command window, which can be overwhelming to read with many rows and columns. As discussed above, commands can be run from a text file containing a list of commands. Code is written into such files and saved to a filename with a .m extension. To run .m files, you navigate within the current directory path window until your .m file is visible. Then in the command window, you type in the name of the .m file (without the .m extension) and hit ENTER. MATLAB then treats the entire list of commands in the file as a script and runs it sequentially. Depending on your needs, you may or may not need semicolons at the end of lines (but usually you will include semicolons so the command window does not become cluttered as results speed by in the background). Any commands without semicolons will Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
236
Chapter 8: LCA Screening via Economic Input-Output Models
have their results shown in the command window. If semicolons are always included, the results can be viewed via the workspace. As a demonstration, one possible sequence of commands to complete Example 8-1 (either entered line by line or run as an entire .m file) is: Z=[150 500; 200 100]; X=[1000 2000; 1000 2000]; A=Z./X; A command sequence for Example 8-2 is (assuming commands above are already done): y1=[100; 0]; y2= [0; 100]; direct=eye(2)+A; L=inv(eye(2)-A); directreq1=direct*y1; directreq2=direct*y2; totalreq1=L*y1; totalreq2=L*y2; where the final 4 commands create the direct and total requirements for Y1 and Y2. A command sequence for Example 8-3 is (assuming commands above are already done): R=[50 5]; R_diag=diag(R); E_direct_Y1 = R_diag*directreq1; E_direct_Y2 = R_diag*directreq2; E_total_Y1=R_diag*totalreq1; E_total_Y2=R_diag*totalreq2; E_sum_Y1=sum(E_total_Y1); E_sum_Y2=sum(E_total_Y2);
EIO-LCA in MATLAB
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 8: LCA Screening via Economic Input-Output Models
237
The EIO-LCA model in a MATLAB environment is available as a free download from the website (www.eiolca.net). The full 1997 model in MATLAB is available directly for download, and a version of the 2002 model excluding energy and GHG data is available directly for download. The 2002 MATLAB model with energy and GHG data is available for free for non-commercial use via a clickable license agreement on the www.eiolca.net home page (teachers are encouraged to acquire this license and the MATLAB file for local distribution but to make non-commercial license terms clear to students). Within the downloaded material for each model are .mat files with the vectors and matrices needed to replicate the results available on the www.eiolca.net website, and MATLAB code to work with producer and purchaser models. MATLAB .m files named EIOLCA97.m and EIOLCA02.m are scripts for the 1997 and 2002 models, respectively, to generate results similar to what is available on the website. For example, running the EIOLCA97.m file in the 1997 MATLAB model will successively ask whether you want to use the producer or purchaser model, which vector (economic, GHG, etc.) to display, and how many sectors of results (e.g., all 491 or just the top 10). Before running this script, you need to enter a final demand into one or more of the 491 sectors in the SectorNumbers.xls spreadsheet file. Results will be saved into a file called EIOLCAout.xls in the 1997 MATLAB workspace directory. Note: to run the EIOLCA97.m script file, you must be running MATLAB directly in Windows or via Windows emulation software (e.g., Boot Camp or Parallels on a Mac) since it uses Microsoft Excel read and write routines only available on Windows. The vectors and matrices in the 1997 model though are accessible to MATLAB on any platform. Due to these limitations, and the age of the data in the 1997 model, this section focuses on the 2002 MATLAB model (but similar examples and matrices exist in the 1997 model). Similarly, running the EIOLCA02.m file in the 2002 MATLAB model files will successively ask whether you want to use the producer (industry by commodity basis default) or purchaser model, the name of the vector variable that contains your final demand (which you will need to set before running the .m file), and what you would like to name the output file. Note the 2002 MATLAB model can be run on any MATLAB platform (not just Windows). Before running this script, you need to create and enter a final demand into one or more of the 428 sectors. The following MATLAB session shows how to use the EIOLCA02.m script to model $1 million of final demand into the Oilseed farming sector. All lines beginning with >> show user commands (and as noted above the user also needs to choose between the producer and purchaser models, and give names for the final demand vector and a named txt file for output – highlighted in green). Before running this code, you will need to
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
238
Chapter 8: LCA Screening via Economic Input-Output Models
change the current MATLAB directory to point to where you have unzipped the MATLAB code. >> y=zeros(428,1); >> y(1,1)=1; >> EIOLCA02 Welcome to EIO-LCA This model can be run in 2002 $million producer or purchaser prices. For producer prices, select 1. prices, select 2. Producer or Purchaser prices?
For retail (purchaser) 1
Name of the 428 x 1 final demand vector
y
Output file name? (include a ".txt") Filename
xout.txt
Total production input is:
1$M2002, producer prices
The resulting xout.txt file shows the total supply chain results across all sectors for $1 million of final demand in all data vectors available in the MATLAB environment (which would match those on the website), all in one place. This imported as a text file into Microsoft Excel with semicolon delimiters for more readable output and for easier comparison to the results on the website. An excerpt of rows and columns from this file is shown in Figure 8A-12 (sorted by sector number):
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 8: LCA Screening via Economic Input-Output Models
239
Total econ, $M
Total Energy, TJ
GHG Emissions, mt CO2e
Total, All Sectors
2.1
16.1
3029.8
1111A0
Oilseed farming
1.1
8.4
2218.7
1111B0
Grain farming
0.0
0.2
91.1
111200
Vegetable and melon farming
0.0
0.0
0.2
111335
Tree nut farming
0.0
0.0
0.1
1113A0
0.0
0.0
0.3
111400
Fruit farming Greenhouse and nursery production
0.0
0.0
0.5
111910
Tobacco farming
0.0
0.0
0.4
111920
Cotton farming
0.0
0.2
43.7
1119A0
Sugarcane and sugar beet farming
0.0
0.0
0.3
1119B0
All other crop farming
0.0
0.0
3.5
Figure 8A-12: First 10 Sectors of Output from EIOLCA02.m script for $1M of oilseed farming
The script .m files have much useful information in them, should you care to follow the code. For example, you can see the specific matrix math combinations used to generate the producer and purchaser models and their direct and total requirements matrices used in EIO-LCA. Instead of using the provided .m script files, the MATLAB workspaces for 1997 and 2002 can be used on any MATLAB platform to do tailored modeling using the various vectors and matrices. For example you may want to quickly just generate the Total GHG emissions for $1 million of oilseed farming in the same EIO-LCA 2002 model. Total GHG emissions are in the matrix called EIvect, in row 7 (rows 1-6 are the various energy vector values and rows 7-12 are the various GHG emission vector values) >> clear all >> load EIO02.mat >> y=zeros(428,1); >> y(1,1)=1; >> x=L02ic*y; >> E=EIvect(7,:)*x which returns: E = 3.0298e+03 The same value as in the first row of the last column of Figure 8A-12. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
240
Chapter 8: LCA Screening via Economic Input-Output Models
Likewise, you might be interested in generating the total GHG emissions across the supply chain for $1 million into each of the 428 sectors: >> allsects=EIvect(7,:)*L02ic; which returns a 1x428 vector containing the requested 428 values (where the data in column 1 is the same as above for oilseed farming). This simple one-line MATLAB instruction works because the 1x428 row vector chosen from EIvect (total GHG emissions factors per $million for each of 428 sectors) is multiplied by the column entries in the total requirements matrix for each of the sectors, and the result is the same as finding the total GHG emissions across the supply chain as if done one at a time. The first four values in this vector (rounded) are: [3030 4470 1303 1329], representing the total GHG emissions for $1 million of final demand into the first 4 (of 428) sectors in EIO-LCA. Much more is possible given the available economic and environmental/energy flow matrices that is not possible on the website or with the included script file. For example, you could do a similar analysis as above for the purchaser-based model to find the results of $1 million in all sectors.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
241
Chapter 9 : Advanced Life Cycle Models In this chapter, we define alternative approaches for LCA using advanced methods such as process matrices and hybrid analysis. Process matrices organize process-specific data into linear systems of equations that can be solved with matrix algebra, and represent a significant improvement over traditional process flow diagram approaches. Hybrid LCA models, combining process and input-output based methods, offer ways to leverage the advantages of the two methods while minimizing disadvantages. Three approaches to hybrid LCA modeling are presented, with the common goal of combining types of LCA models to yield improved results. The approaches vary in their theoretical basis, the ways in which the submodels are combined, and how they have been used and tested.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. Define, build, and use a process matrix LCI model from available process flow data. 2. Describe the advantages of a process matrix model as compared to a process flow diagram based model and an input-output based model. 3. Describe the various advantages and disadvantages of process-based and IO-based LCA models. 4. Classify the various types of hybrid models for LCA, and how they combine advantages and disadvantages of process and IO-based LCA models. 5. Suggest an appropriate category of hybrid model to use for a given analysis, including the types of data and process-IO model interaction needed.
Process Matrix Based Approach to LCA In Chapters 5 and 8 we introduced process-based and IO-based methods as two approaches to performing life cycle assessment. The bottom-up process method presented in Chapter 5 (which is more widely referred to as the process flow diagram approach) is a fairly limited application of the process method. It requires time for iteratively finding each needed set of process data, including for following the connections between processes. We found results by summing effects from each included process in the diagram in a bottom-up method. On the other hand, IO-LCA methods presented a distinct benefit in terms of delivering quick Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
242
Chapter 9: Advanced Life Cycle Models
and easy top-down results by exploiting matrix math methods to invert and solve the entire upstream chain. We can merge the concepts of input-output analysis with the data from a process flow diagram approach to create linear systems of equations that represent a comprehensive set of process models known as a process matrix. Now that you have seen both process and IO methods, you might have already considered a process matrix-based model. Conceptually, a process matrix model incorporates all available process data (whether explicitly part of the process flow diagram or not) into the system. The process matrix approach yields results similar to what would be expected if we added more and more processes to the process flow diagram. However, as we will see, the process matrix approach is able to improve upon the bottom up process diagram approach, as it can model the interconnections of all processes, and as in IO methods, will be able to fully consider the environmental flows of all upstream interconnections. The process matrix approach thus gives us some of the benefits of an IO model system but with data from explicit (rather than average) processes. Before going further, we use the linear algebra introduced in Chapter 8 to re-define process data and models. Figure 9-1 shows a hypothetical system with two processes, one that makes fuel and one that makes electricity.15 This example is similar to the main example of Chapter 5 that discussed making electricity from coal.
Figure 9-1: Process Diagrams for Two-Process System
Focusing on the purely technical flow perspective, process 1 takes the raw input of 50 liters of crude oil, and process 2 takes an input of 2 liters of fuel. Likewise, the output flow arrows show production of 20 liters of fuel and 10 kWh of electricity, respectively, in the two processes (the emissions shown in the figure will be discussed later). In this scenario, the functional units are the outputs, 20 liters of fuel and 10 kWh of electricity, since all of the
15
Thanks to Vikas Khanna of the University of Pittsburgh for this example system.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
243
process data corresponds to those normalized values. Without any analysis we know that fuel (process 1's output) must be produced to produce electricity (process 2's output). As with any linear system, but especially for the types of analysis of interest in LCA, we need to consider alternative amounts of outputs needed, and thus create a general way of scaling our production higher or lower than the functional unit values above. That is, we do not merely have to be constrained to produce 20 liters of fuel or 10 kWh of electricity. Once a scaling factor is established, the output for any input, or the input for any output can be found. Within our process system, we initially consider only flows of product outputs through the processes, e.g., fuel and electricity, not elementary flows. Thus for now we ignore necessary crude oil input and the various emissions (again, we will consider these later). In such a linear system we define a scaling factor vector X with values for each of the two processes, X1 and X2, and the total net production across the system for each of the two outputs, Y1 and Y2. Here, Y1 is the total net amount of fuel produced, in liters. Y2 is the total net amount of electricity produced, in kWh. We can define a sign convention for inputs and outputs such that positive values are for outputs and negative values are for inputs (i.e., product output that is input to other processes in the system). Given this framework and notation, we define the following linear system of equations which act as a series of physical balances given our unit process data: 20 X1 - 2 X2 = Y1 0 X1 +10 X2 = Y2
(9-1)
where the first equation mathematically defines that the total amount of fuel produced is 20 liters for every scaled unit process 1, net of 2 liters needed for every scaled unit produced in unit process 2. Likewise the second equation defines that the total amount of electricity produced is zero per scaled unit of process 1 and 10 kWh per scaled unit process 2. To scale our functional unit-based processes (up or down), we would insert values for X1 and X2. In general, these values could be fractions or multiples of the unit. If X1 = 1 and X2 = 1, we would generate the identical outputs in the processes shown in Figure 9-1. If we wanted to make twice as much fuel, then X1 = 2, which, for example would require 100 liters of crude oil. If we wanted to make twice as much electricity as in the unit process equation (20 kWh), then X2 = 2, requiring 4 liters of fuel input. Similar to what was shown in Chapter 8 (and its Appendices) we use the generalized matrix notation AX = Y to describe the system of Equations 9-1, as demonstrated in Heijungs (1994) and Suh (2005). Now the matrix A, which in the process matrix domain is called the
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
244
Chapter 9: Advanced Life Cycle Models
technology matrix, represents the technical coefficients of the processes linking the product systems together. In the system describing Figure 9-1 above, 𝐀=
20 0
−2 10
Note the structure of the matrix. The functional units, representing the measured outputs of each of the processes, are along the diagonal of A. The use of outputs from other processes within the system are the off-diagonal entries, e.g., -2 shows the fuel (output 1) used in the process for making electricity (output 2). To solve for the required scaling factor to produce a certain net final production in the system, the linear system AX = Y is solved as in Chapter 8, by rearranging the linear system equation and finding the inverse of A: AX = Y ó X=A-1Y
(9-2)
In this example, the inverse of A is: 𝐀!𝟏 =
. 05 0
. 01 .1
thus if we want to produce Y2 = 1,000 kWh of electricity, we can use Equation 9-2 to determine what the total production in the system needs to be. In this case it is: 𝑋 = 𝐀!𝟏 𝑌 =
. 05 0
0 10 . 01 = 100 . 1 1000
which says that to make 1,000 kWh of electricity in our system of two processes, then from a purely technological standpoint, we would need to make 10 times the unit process 1 of fuel production (200 liters total) and we would need to scale the electricity generation process 2 by a factor of 100. Within the system, of course, we would be making 200 liters of fuel, all of which would be consumed as the necessary (sole) input into making 1,000 kWh of electricity. Figure 9-2 shows this scaled up sequence of processes, including the dotted line "connection" of the two processes. The processes are defined identically as those in Figure 9-1, but with all values scaled by 10 for process 1 and by 100 for process 2.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
245
Figure 9-2: Scaled-Up and Connected Two-Process System
We could also use AX = Y notation for the linear system if we instead wanted to determine the total net output given a set of scaling factors. For example, if X = [10 ; 100] (our result from the example above) then: 𝐀𝑋 = 𝑌 = !
20 0
−2 10 0 !! !=! ! 10 100 1000
So if we want to make 10 times the unit production of process 1 (200 liters of fuel), and 100 times the unit production of process 2 (1,000 kWh of electricity), net production is only Y2 = 1,000 kWh of electricity, since all of the 200 liters of fuel produced in process 1 are consumed in process 2 to make electricity, resulting in a net of Y1 = 0 liters of fuel. This result (i.e., Y = [0; 1000]) is the same as used in the previous example.
So far, we have motivated the purely technological aspects of the simple two-process system. However Figure 9-1 gives us additional information on the resource use and emissions performance of the two processes. We create an environmental matrix, B, analogous to Chapter 8's R matrix, to represent the direct per-functional unit resource use and emissions factors. The B matrix has a conceptually identical basis of flows per functional unit as in Chapter 8, except that instead of consistently being flows per million dollars, the units are flows per functional unit, which vary across the processes (e.g., per kg, per MJ, etc.). E = BX = BA−1Y
(9-2)
Again we use a sign convention where negative values are inputs and positive values are outputs. From Figure 9-1, process 1 uses 50 liters of crude oil as an input, and emits 2 kg SO2 and 10 kg CO2. Process 2 has no raw inputs (only the product input of fuel already represented in the A matrix), and emits 0.1 kg of SO2 and 1 kg of CO2 emissions, respectively. Thus the environmental matrix B for our system, where the rows represent the flows of crude oil, SO2, and CO2, and the columns represent the two processes, can be represented as: −50 0 𝐁= 2 0.1 10 1 Of course, the linear system behind B reminds us of the connection of inputs, outputs, and emissions shown in Figure 9-1: E crude = −50 X1 ; E SO2 = 2 X1 + 0.1 X2 ; E CO2 = 10 X1 + X2 Put another way, the elements of BA-1 in Equation 9-2 represent total resource and emissions factors across the process matrix system, analogous to total economy-wide emissions factors R[I -A]-1 of an IO system (and the A matrices are different in the systems). Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
246
Chapter 9: Advanced Life Cycle Models
Building on our example from above, we can estimate the environmental effects of producing the required amount of outputs by multiplying the B matrix by our previously found vector Y [0 ; 1,000]. The resulting E is: 𝐸=
−500 30 200
While we have motivated this initial example with an intentionally small (two process) model, it is easy to envision how adding additional processes would change the system. If we added a third process, which potentially had interconnections to the original two processes, we would merely be adding another dimension to the problem. The A matrix would be 3x3, and the X and Y vectors would have an additional row as well. If we added no environmental flows, B would also be 3x3. If we added flows (e.g., another fuel input or emission) then there would be additional rows. The linear algebra does not become significantly more difficult. Depending on the necessary scope, your system may end up with 5, 10, or 50 processes. We generally use integer scaling factors and achieve integer results in the chapter examples. However, the linear algebra method can derive any real input and output values. Note that some processes may in fact only be able to use integer inputs (or be able to produce integer levels of output), in which case your results would need to be rounded to a near integer. E-resource: "Chapter 9 Excel Matrix Math" shows all of the examples in this chapter in a Microsoft Excel spreadsheet.
Connection Between Process- and IO-Based Matrix Formulations The most important aspect of the process matrix approach to recognize is the similarity to how we solved EIO models. The matrix math (AX = Y ó X=A-1Y) is identical, differing only where in the EIO notation the "A matrix" is instead I-A. If you look at the elements of the technology matrix A in the process matrix domain, and think through its composition, a more distinct connection becomes clear. As noted above, the diagonal entries of the technology A matrix summarize the functional units of the processes collected within the system. If we were to think only about the inputs into the process system, and/or collect an A matrix consisting only of the values of our own technology matrix from available process data, the matrix may not have any of those functional unit values – it would just contain data on the required inputs from all of the processes in the system. We would have no need to specify a particular sign convention for inputs, so we could include them as positive values. In the example above, the adjusted A matrix with this perspective would be: Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
𝐀∗ =
0 0
247
2 0
which would summarize the case where process 1 had no technological inputs from other parts of the system (i.e., no input of fuel or electricity) and where process 2 had a requirement of 2 liters of fuel. If we wanted to make productive use of this different process, we would need to add in the functional unit basis of the system (otherwise we would have no way of knowing how many units of output can be created from the listed inputs). In doing that, we would need to create a diagonalized matrix containing the functional unit values of each of the processes, which in this case is: 𝐖=
20 0
0 10
and we would combine the information in these two matrices before inverting the result. W is a matrix of positive values of the process outputs while A* is made of positive values for the process inputs. The net flows are found as outputs minus inputs: W + -(A*) = W – A* And our modified matrix math system would be: [W – A*]X = Y ó X=[W – A*]-1Y Of course, combining W and A* in this way gives exactly the original A process matrix, which is then inverted to find the same results as above. The key part to understand is that this is exactly what is done in IO systems, but since the system is in general normalized by a constant unit of currency (e.g., "millions of dollars"), all of the functional units are already "1", and thus the identity matrix I is what is needed as the general W matrix above. Nonetheless, this exposition should help to reinforce the similarity in derivation of the process based and IO-based systems.
Linear Systems from Process Databases The simple two-process system above used hypothetical values for inputs and outputs. But we may also build up our linear system with data from processes in available databases. We could envision the process flow diagram from Chapter 5, where we made electricity from burning bituminous coal that had been mined and delivered to the power plant by train. This example used actual data from the US-LCI database. In Chapter 5, we already saw how to build simple LCI models from this process flow diagram. We could find the same answers by building a linear system. Using the same notation as in our two-process example Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
248
Chapter 9: Advanced Life Cycle Models
above, but using the US LCI database values from Chapter 5, and assuming that bituminousfired electricity generation is process 1, bituminous coal mining is process 2, and rail transport is process 3, we could define the linear system in 9-3:
1 X1+ 0 X2 + 0 X3 = Y1 -0.442 X1 + 1 X2 + 0 X3 = Y2 -0.461 X1 + 0 X2 + 1 X3 = Y3
(9-3)
For this system, 1 𝐀 = −0.442 −0.461
0 1 0
0 0 1
So if we want to produce 1 kWh of electricity, as found in Chapter 5, we would need to produce the following in each of the three processes: 1 𝑋 = 𝐀!𝟏 𝑌 = 0.442 0.461
0 1 0
0 1 1 0 0 = 0.442 1 0 0.461
and considering only fossil CO2 emissions, 𝐁 = 0.994 0 0.0189 so, using E = BX , E ~ 1.003 kg CO2, the same result reached in Chapter 5 (Equation 5-1). This example shows us that we could build linear system models based on process data, and could include as many processes as we have time or resources to consider. Of course we can use software tools like Microsoft Excel or MATLAB to manage the data and matrix math. From the example in Chapter 5, we could expand the boundary to add data from additional processes like refining petroleum, so as to capture effects of diesel fuel inputs. As we add processes (and flows) we are just adding rows and columns to the linear system above. Beyond adding rows and columns as we expand our boundary, we also generally add technical coefficients to the A matrix that were not previously present (e.g., if we had data showing use of electricity by the mine). We would thus be adding upstream effects that would likely not have been modeled in a simple process flow diagram approach. The threeprocess example above does not shed light on this potential, because there are no upstream Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
249
coefficients that are in the system that were not in our process flow diagram example in Chapter 5. If building linear system models from the bottom up, we would eventually decide that we were unlikely to add significantly more information by adding data from additional processes or flows. The dimensions of the technology matrix A of our linear system would be equal to the number of processes included in our boundary, and the dimensions of the environmental matrix B would be the number of processes and the number of flows. However, if we have access to electronic versions of all the process LCI modules from databases, we can use them to build large process matrix models. Since databases like US LCI and ELCD are publicly available, matrices and spreadsheet models can be built that by default encompass data for all of the many interlinked processes and flows provided in the database. Many researchers and software tools incorporate external databases with the process matrix approach (e.g., SimaPro). In the rest of this section, we explore construction and use of these comprehensive process matrix models to facilitate rapid construction of LCI models. While not publicly available, licensees of ecoinvent data can download or build complete matrices representing all processes and flows. Chapter 5 discussed the availability of unit process and system process level data in the ecoinvent database and software tools. System processes are aggregated views of processes with relatively little detail, and no connections to the other unit processes. Using them is like using a snapshot of the process (i.e., where the matrix math has already been done and saved). Using ecoinvent unit processes allows the full connection to all upstream unit processes, and calculations involving them will "redo" the matrix math). While the US LCI database as accessed on the LCA Digital Commons website does not directly provide A and B matrices, the matrices can be either built by hand using the downloadable spreadsheets (see Advanced Material for Chapter 5), or by exporting the entire matrix representation of the US LCI database from SimaPro (choose the library after launching the software, then choose "Export Matrix" from the File menu). The US LCI database, as exported from SimaPro as of 2014, provides LCI data for 746 products and 949 flows. These 746 products are the outputs of the various processes available in US LCI. Given that the US LCI database has information on 746 products, we could form an A matrix of coefficients analogous to the linear system above with 746 rows and 746 columns, where the elements of the matrix are the values represented as inputs of other product processes within the US LCI database to produce the functional unit of another product's process. For example, the coefficients of our three-process US LCI example above would be amongst the cells of the 746 x 746 matrix. Of course, the A matrix will be very sparse (i.e., have many blank or 0 elements) since many processes are only technologically connected to a few other processes. This matrix would be similar to the IO direct Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
250
Chapter 9: Advanced Life Cycle Models
requirements matrix considered in Chapter 8. Likewise B can be made from the nonproduct input and output flows of the US LCI database, resulting in a 949 x 746 matrix. This matrix again will be quite sparse since the number of flows listed in a process is generally on the order of 10 to 20. For small-scale LCA projects, a Microsoft Excel-based process matrix model could suffice and provide the same quantitative results, yet will not provide the graphical and other supporting interpretation afforded by other LCA software tools. The process matrix models will, in general, represent more technological activity than bottom-up process flow diagrams because of the addition of many upstream matrix coefficients, and thus will generally estimate higher environmental flows. E-resources: The Chapter 9 folder shows two forms of the US LCI database represented as a complete matrix of 746 processes in Excel. The main (larger) spreadsheet shows all of the intermediate steps of building the matrices, including the raw exported data from SimaPro, matrices normalized by functional unit, transposed, inverted, etc., on separate worksheets. It also has a "Model" worksheet where you can enter production values for processes and see the "direct" (just the top-level technical requirements for the product chosen – as would be shown in the LCI module of the chosen process), total (X), and E results. It is useful to look at this spreadsheet and its various cell formulas, especially those involving array formulas, to see how such models can be built from the databases. Amongst the many features of this larger spreadsheet are that the coefficients of the A and B matrices can be modified, and changes would ripple through the model. A second spreadsheet (filename appended with "smaller") is the same model, but with just the final resulting matrices, without intermediate matrix math steps. It has the same functionality, but is significantly smaller in size. It may be more appropriate to use for building models that will not modify any of the A or B matrix coefficients. These two spreadsheets work by entering a desired input of product Y into the blue column of the Model worksheet to then estimate the direct and total technical requirements X and the environmental effects E, shown in yellow cells. Be sure to enter values pertinent to the listed functional unit of the product (i.e., check whether energy products have units of MJ or kWh). The spreadsheet conditionally formats in red any results deemed "significant", in this case greater than 0.000001, since imperfect matrix inversion results in many very small values (less than 10-16) throughout the model which can be treated as negligible. Example: Estimating the life cycle fossil CO2 emissions of bituminous coal-fired electricity using a process matrix model of the US LCI database. We can estimate the total fossil CO2 emissions of making coal-fired electricity in the US using the Microsoft Excel US LCI e-resource spreadsheets. The first step is determining the appropriate model unit process and input value to use. The same product used in other US Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
251
LCI examples, Electricity, bituminous coal, at power plant/US, (process number 416 of 747 US LCI processes when sorted alphabetically) is chosen and an input value of 3.6 MJ (equal to 1 kWh) is used. Figure 9-3 shows the "direct" technical flows from this input, corresponding to the seven direct inputs needed for this product in the US LCI data module for this process. Flows prepended with "Dummy" in the US LCI database were not discussed in Chapter 5. In short, these are known technical flows for a process, but for which no LCI data are included within the system. They thus act only as tracking entries in the model. Product
Unit
Flow
Transport, train, diesel powered/US
tkm
0.461
kg
0.442
tkm
0.056
kg
0.044
Bituminous coal, at mine/US Transport, barge, average fuel mix/US Dummy_Disposal, solid waste, unspecified, to unspecified treatment/US Dummy_Disposal, ash and flue gas desulfurization sludge, to unspecified reuse/US
kg
0.014
Transport, combination truck, diesel powered/US
tkm
0.003
Dummy_Transport, pipeline, coal slurry/US
tkm
0.002
Figure 9-3: Direct technological flows from 3.6 MJ (1 kWh) of Electricity, bituminous coal, at power plant in US LCI Process Matrix
The values of X and E for all processes and flows are also shown in the spreadsheet. Figure 9-4 shows the elements of X with physical flows greater than 0.001 in magnitude. There are 18 more products upstream of electricity than in the "direct" needs. While, as expected, the process matrix values for the same physical products are larger in Figure 9-4 than in Figure 9-3, the differences are generally small. We can also see that the additional amount of bituminous coal-fired electricity needed across the upstream chain within the process matrix model is small (0.037 MJ). The Air, carbon dioxide, fossil, column (number 231 out of 949 flows) shows the estimate of total emissions of fossil CO2 across the entire process matrix, 1.0334 kg CO2 (with apologies for the abuse of significant figures). While larger, this result is only marginally more than the result from the process flow diagram approach. This is not surprising, though, because it is well known that the main contributor of CO2 emissions in fossil electricity generation is the combustion of fuels at the power plant, which was included in the process flow diagram. If we were to choose a different product for analysis, we may see substantially higher environmental flows as a result of having the greater boundary within the process matrix. Note that the US LCI data provides information on various other carbon dioxide flows which have not been included in our scope. There are two "Raw" flows of CO2 (as inputs),
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
252
Chapter 9: Advanced Life Cycle Models
as well as 4 other air emissions. The only other notable one of these is the Air, carbon dioxide, biogenic flow, which represents non-fossil emissions, such as from biomass management.
Process Electricity, bituminous coal, at power plant/US Transport, train, diesel powered/US
Functional Unit MJ
Output Value X 3.637
tkm
0.466
Bituminous coal, at mine/US
kg
0.447
Dummy_Disposal, solid waste, unspecified, to underground deposit/US
kg
0.105
Electricity, at grid, US/US
MJ
0.068
Transport, barge, average fuel mix/US
tkm
0.057
kg
0.044
Transport, barge, residual fuel oil powered/US
tkm
0.044
Transport, ocean freighter, average fuel mix/US
tkm
0.037
Transport, ocean freighter, residual fuel oil powered/US
tkm
0.033
Electricity, nuclear, at power plant/US
MJ
0.015
Dummy_Disposal, ash and flue gas desulfurization sludge, to unspecified reuse/US
kg
0.014
Transport, barge, diesel powered/US
tkm
0.012
Electricity, natural gas, at power plant/US
MJ
0.012
Crude oil, at production/RNA
kg
0.008
Dummy_Transport, pipeline, unspecified/US
tkm
0.007
Dummy_Electricity, hydropower, at power plant, unspecified/US
MJ
0.005
Transport, ocean freighter, diesel powered/US
tkm
0.004
Transport, combination truck, diesel powered/US
tkm
0.003
Dummy_Transport, pipeline, coal slurry/US
tkm
0.002
Electricity, residual fuel oil, at power plant/US
MJ
0.002
Electricity, lignite coal, at power plant/US
MJ
0.002
Natural gas, at extraction site/US
m3
0.001
Natural gas, processed, at plant/US
m3
0.001
Electricity, biomass, at power plant/US
MJ
0.001
Dummy_Disposal, solid waste, unspecified, to unspecified treatment/US
Figure 9-4: Total technological flows (X) from 3.6 MJ (1 kWh) of Electricity, bituminous coal, at power plant in US LCI Process Matrix, abridged
Figure 9-5 shows the top products that emit CO2 in the upstream process chain of bituminous coal-fired electricity. As previously discussed, the combustion of coal at the power plant results in 97% of the total estimated emissions. The emissions from rail transport by train are another 1%. Thus our original process flow diagram model from Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
253
Chapter 5, which we motivated as a simple example, ended up representing 98% of the CO2 emissions from coal-fired electricity found in the more complex process matrix model. Process Total Electricity, bituminous coal, at power plant/US
Emissions (kg) 1.033
Percent of Total
1.004
97.2%
Diesel, combusted in industrial boiler/US
0.011
1.0%
Transport, train, diesel powered/US
0.009
0.9%
Electricity, natural gas, at power plant/US
0.002
0.2%
Residual fuel oil, combusted in industrial boiler/US
0.002
0.2%
Transport, barge, residual fuel oil powered/US
0.001
0.1%
Natural gas, combusted in industrial boiler/US
0.001
0.1%
Gasoline, combusted in equipment/US
0.001
0.1%
Electricity, lignite coal, at power plant/US
0.001
0.1%
Transport, ocean freighter, residual fuel oil powered/US
0.001
0.1%
Bituminous coal, combusted in industrial boiler/US
0.001
0.0%
Figure 9-5: Top products contributing to emissions of fossil CO2 for 1kWh of bituminous coal-fired electricity. Those representing more than 1% are bolded.
One of the aspects of the ISO Standard that we did not discuss in previous chapters is the use of cut-off criteria, which define a threshold of relevance to be included or excluded in a study. For example, the cut-off may say to only include individual components that are 1% or more of the total emissions. The Standard motivates the use of cut-off criteria for mass, energy, and environmental significance, which would mean that we could define cut-off values for these aspects for which a process is included within the boundary (and can be excluded if not). As an example, if we set a cut-off criterion of 1% for environmental significance, and our only inventory concern was for fossil CO2 to air, then we could choose to only consider the first two processes (three if we rounded off conservatively) listed in Figure 9-5. Likewise with a cut-off criterion of 0.1%, we would only need to consider the first 10. Neither of these cut-off criteria significantly affects our overall estimate of CO2 emissions. Cut-off criteria could be similarly applied for energy via LCI results. Mass-based cut-off criteria are evaluated separately than from LCI results, e.g., by considering the mass of subcomponents of a product. In all cases, the cut-off criteria apply to the effects of the whole system boundary, not to each product or process in the system, so if the product system were fairly large, it is possible that many initially scoped parts of the system could be excluded based on the cut-off criteria. For example, if the system studied is the life cycle of an automobile, and the scope is fossil CO2 emissions, then electricity use in production of
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
254
Chapter 9: Advanced Life Cycle Models
the vehicle may not be large enough to matter. In that case all the results in the example above would be excluded. The other side of the cut-off criteria selection decision issue is that of truncation error. LCI models are inevitably truncated when arbitrary or small boundaries are used. Thus, studies showing the effects of truncation can compare the results from within a selected boundary as compared to those with a complete (e.g., process matrix or IO system) boundary. In the example above, the truncation error is very small as long as the "at power plant" effects are within the boundary, but such errors can sometimes be substantial as analysts define boundaries based on what they think are the important components without knowledge of which are the most important. In the end, the process matrix approach is yet another valuable screening tool, albeit one with substantial process-based detail, to be used in setting analysis boundaries. As you have seen in this chapter, the process matrix approach provides an innovative way to use our process data models. The results from using the matrix approach will generally be larger and more comprehensive compared to the simpler process diagram approach, in the same way as when we used IO models. This is because the process flow diagram approach is inherently limited by what you include in the boundary. To some extent, a process flow diagram approach assumes by default that everything excluded from the diagram doesn't matter. But as we have seen with the process matrix (and IO) approaches, it is difficult to determine what matters until you have considered these larger boundaries. We should generally expect IO models to estimate the largest amount of flows, as they comprise the whole economy within the boundary, including overhead and service sectors, which are rarely included in LCI databases. It is for this reason that IO models are often used to prescreen the needed effort for a more comprehensive process-based analysis.
Extending process matrix methods to post-production stages In our examples above, the boundaries have only included the cradle to gate (including upstream) effects. We can extend our linear systems methods to include downstream effects as well, such as use and end-of-life management, since again this will merely involve adding rows and columns, as well as additional coefficients, to the matrices. In this case, lets build on the example from Figure 9-1 but where we make alternative descriptions of the processes involved. Assume that we want to make a lamp product requiring fuel and electricity to manufacture it, and that it will be disposed of in a landfill by a truck. We had previously referred to fuel production as process 1, and electricity production as process 2. Now we add lamp production as process 3, and disposal as process 4. Specifically, lets assume that producing each lamp requires 10 litres of fuel and 300 kWh Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
255
of electricity (and emits 5 kg CO2), and finally that disposing of the lamp at the end of its life consumes 2 litres of fuel (and emits 1 kg CO2). Our linear system can be written as in system 9-4:
20 X1 - 2 X2 - 10 X3 - 2 X4 = Y1 0 X1 +10 X2 - 300 X3 + 0 X4 = Y2 0 X1 +0 X2 + 1 X3 + 0 X4 = Y3 0 X1 + 0 X2 + 0 X3 + 1 X4 = Y5
(9-4)
where 20 0 𝐀= 0 0
−2 10 0 0
−10 −300 1 0
−2 0 0 1
In this case when we have an input (Y) to the system of 1 produced lamp and one disposed lamp, the total production is: . 05 0 𝑋 = 𝐀!𝟏 𝑌 = 0 0
. 01 0.1 0 0
3.5 30 1 0
0.1 0 0 1
0 3.6 0 30 = 1 1 1 1
Thus, in order to produce 1 lamp (the output of 1 lamp unit process), 3.6 units of fuel and 30 units of electricity are also needed. If we only model CO2 emissions, then 𝐁 = 10 1 5 1 and R = 72 kg of CO2. While the system has few interrelationships, it may not be as easy to validate this result as before, but if we think through our production, we need to make 72 litres of fuel (12 litres from producing the lamp and disposing of it, and 60 litres from making electricity), 300 kWh of electricity (all for making the lamp), and 1 lamp for every lamp produced. The total emissions of 72 kg come from 36 kg of CO2 from fuel production, 30 kg CO2 from electricity production, 5 kg from lamp production, and 1 kg from disposal. So our results make sense. While not shown here, we could add another row and column to represent use of the lamp. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
256
Chapter 9: Advanced Life Cycle Models
Advantages and Disadvantages of Process and IO Life Cycle Methods Before considering other advanced LCA methods, we summarize the strengths and weaknesses of process and IO methods. Both IO-LCA and process modeling have their advantages and drawbacks. In principle, they can yield quite different results. If so, the analyst would need to make a closer examination to determine the reasons for the differences and which one or combination of the two gives the best estimate. Figure 9-6 compares the strengths and weaknesses of the two types of LCA models.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
Process Models
Input-Output
Detailed process-specific analyses
Economy-wide, comprehensive assessments (direct and indirect environmental effects included)
257
Advantages
Specific product comparisons Process improvements, weak point analyses
System LCA: industries, products, services, national economy Sensitivity analyses, scenario planning
Future product development assessments
Publicly available data, reproducible results Future product development assessments Information on every commodity in the economy
System boundary setting subjective
Some product assessments contain aggregate data
Tend to be time intensive and costly
Process assessments difficult
Disadvantages
Difficulty in linking dollar values to physical units New process design difficult Use of proprietary data Cannot be replicated if confidential data are used Uncertainty in data
Economic and environmental data may reflect past practices Imports treated as U.S. products Difficult to apply to an open economy (with substantial non-comparable imports) Non-U.S. data availability a problem
Uncertainty in data Figure 9-6: Advantages and Disadvantages of Process and IO-Based Approaches
The main advantage of a process model is its ability to examine, in whatever detail is desired, the inputs and discharges for a particular process or product. The main disadvantage is that gathering the necessary data for each unit process can be time consuming and expensive. In addition, process models require ongoing comparison of tradeoffs to ensure that sufficient process detail is provided while realizing that many types of relevant processes may not have available data. Even though process matrix methods are quick and comprehensive, their boundaries still do not include all relevant activities. Process models improve and extend the possibilities for analysis, but we often cannot rely wholly on process models. The main advantage of an IO approach is its comprehensiveness, it will by default include all production-based activities within the economy. Its main disadvantage is its aggregated and Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
258
Chapter 9: Advanced Life Cycle Models
average nature, where entire sectors are modeled as having equal impact (e.g., no differentiation between types of electricity). IO-based models simplify our modeling effort and avoid errors arising from the necessary truncation or boundary definition for the network of process models. An IO model's operation at this aggregate level fails to provide the detailed information required for some analyses.
Categories of Hybrid LCA Models An inevitable goal is thus to develop hybrid LCA methods that combine the best features of process and IO-based approaches. In general, hybrid approaches use either a processbased or IO model as the core model, but use elements of the other approach to extend the utility of the overall model. While Bullard (1978) was perhaps the first to discuss such hybrid methods, Suh (2004) categorizes the types of hybrid models in LCA as follows: tiered, input-output based, and integrated hybrid analysis. In a tiered hybrid analysis, specific process data are used to model several key components of the product system (such as direct and downstream effects like use phase and end of life), while input-output analysis is used for the remaining components. If the process and IO components of tiered hybrid analysis are not linked, the hybrid total results can be found by summing the LCI results of the process and IO components without further adjustment (but identified double counted results should be deducted). The point or boundary at which the system switches from process to IO-based methods is arbitrary but can be affected by resources and data available. This boundary should thus be selected carefully, to reduce model errors. Note that unlike many of the other methods discussed in this book, there are no standard rules for performing the various types of hybrid analysis. In tiered hybrid models, the input–output matrix and its coefficients are generally not modified. Thus, analysis can be performed rapidly, allowing integration with design procedures and consideration of a wide range of alternatives. Process models can be introduced wherever greater detail is needed or the IO model is inadequate. For example, process models may be used to estimate environmental impacts from imported goods or specialized production. The most basic type of tiered hybrid model is one where the process and IO components are not explicitly linked other than by the functional unit. An example of such a tiered hybrid model could be in estimating the life cycle of a consumer product. In this case, one could use an IO-based method, e.g., EIO-LCA, to estimate the production of the product (and depending on which kind of IO model used, the scope of this could be cradle to gate or
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
259
cradle to consumer). Process based methods could then be used to consider the use phase and disposal of the product. Example 9-1: Tiered separable hybrid LCA of a Washing Machine To estimate the flows of a washing machine over its life cycle, we could assume that an EIOLCA purchaser basis model is able to estimate the effects from cradle to the point at which the consumer purchases the appliance. Likewise we could assume that process data can be used to estimate the effects of powering the washing machine over its lifetime. We use the data shown in Chapter 3 for "Washing Machine 1". IO component: Assuming that the purchaser price of a new washing machine is $500, we could estimate the fossil CO2 emissions from cradle to consumer via the 2002 US purchaser price basis model in EIO-LCA (Household laundry equipment manufacturing sector) as 0.2 tons. Process components: (Using the US-LCI process matrix Excel spreadsheet, we can consider the production and upstream effects of 10 years' worth (8,320 kWh) of electricity. Since the functional unit is MJ, we convert by multiplying by 3.6 (30,000 MJ), and the resulting fossil CO2 emissions are 6,140 kg (6.1 tons). Note: The US-LCI process matrix has no data on water production and landfilling so these stages are excluded from this example. Total: 6.3 tons fossil CO2.
The description of tiered hybrid methods noted the possibility that the process and IO components could both be estimating some of the same parts of the product system diagram. In these cases, common elements should be dealt with by subtracting out results found in the other part of the model. In Williams et al (2004), the authors performed a hybrid LCA of a desktop computer. The three main subcomponents of the hybrid model are shown in Figure 9-7, where most of the major pieces of a desktop computer system were modeled via process-based methods, capital equipment and feedstocks were modeled with IO methods, and the net "remaining value" of the computer systems not otherwise considered in the two other pieces were also then modeled with IO methods. The overall result from Williams is that the total production energy for a desktop computer system was 6400 MJ, 3140 MJ from the process-sum components, 1100 MJ from the additive IO component, and 2130 MJ from the remaining value. Common elements were subtracted but not represented in the values listed above.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
260
Chapter 9: Advanced Life Cycle Models
Figure 9-7: Subcomponents of Hybrid LCA of Desktop Computer (Source: Williams et al 2004)
Others have used tiered hybrid methods to consider effects of goods and services produced in other countries when using IO models, which would otherwise assume that impacts of production would be equal to domestic production. In an input-output based hybrid analysis, sectors of an IO model are disaggregated into multiple sectors based on available process data. A frequently mentioned early discussion of such a method is Joshi (2000), where an IO sector was proposed to be disaggregated to model steel and plastic fuel tanks for vehicles. In this type of hybrid model, the process level data allows us to modify values in the rows or columns of existing sectoral matrices by allocating their values into an existing and a disaggregated sector. In Chapter 8 we discussed various aggregated sectors in the US input-output tables, such as Power generation and supply, where all electricity generation types (as well as transmission and distribution) activities are all in a single sector. If one could collect sufficient data, this single sector could be first disaggregated into generation, transmission, and distribution sectors, and then the generation sector further disaggregated into fossil and non-fossil, and then perhaps into specific generation types like coal, gas, or wind. Another example of disaggregation that could be accomplished with sufficient process data is the Oil and gas extraction sector (which could be disaggregated into oil extraction and gas extraction). Any of these would be possible with sufficient data, but only if the resulting models would be better than what process-based methods could achieve.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
261
Of course when disaggregating, all relevant matrices need to be disaggregated and updated, and to use the disaggregated results in a model, the A matrix and R matrix values need to be adjusted based on the process data. Since the A and R matrices are already normalized per unit of currency, it is usually easier to modify and disaggregate make, use, or transaction matrices for economic values (and then re-normalize them to A) and to disaggregate matrices with total un-normalized flows to subsequently make R matrices. Let us consider that we want to build a hybrid LCI model based on Example 8-3 in Chapter 8. At the time a two-sector economy was defined as follows (values in billions): 1 2 V X
1 150 200 650 1000
2 500 100 1400 2000
Y 350 1700
X 1000 2000
Assume that sector 1 is energy and sector 2 is manufacturing, and that we have process data (not shown) to disaggregate sector 1 into sectors for fuel production (1a) and electricity (1b). The data tells us that most of the $150 billion purchased by sector 1 from itself is for fuel to make electricity, and how the value added and final demand is split between the fuel and electricity subsectors. We verify that the X, V, and Y values for sector 1 in the original example are equal to the sum of the values across both sectors in the revised IO table. 1a: Fuel 1b: Elec 2: Manuf V X
1a: Fuel 15 10 100 400 525
1b: Elec 100 25 100 250 475
2: Manuf 300 200 100 1400 2000
Y 110 240 1700
X 525 475 2000
The direct requirements matrix for the updated system (rounded to 2 digits) is: . 03 . 21 . 15 𝐀 = . 02 . 05 . 1 . 19 . 21 . 05 Finding values for the disaggregated R requires more thought. Emissions of waste per $billion were 50 g in the original (aggregated) sector 1 and 5 g for sector 2. Thus, the total emissions of sector 1 were originally (50g/$billion)($1,000 billion) = 50,000 g. If our available process data suggests that emissions are 20% from fuel extraction and 80% from electricity, then there are 10,000 g and 40,000 g, respectively. Given disaggregated sectoral outputs of fuel and electricity of $525 and $475 billion, the waste factors for sectors 1 and 2 are 19 g and 84 g, respectively (sector 3's value is unchanged). The disaggregated R matrix is: Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
262
Chapter 9: Advanced Life Cycle Models
19 𝐑= 0 0
0 84 0
0 0 5
Following the same analysis as done in Example 8-3 ($100 billion into each of the sectors), the total waste generated by each of the (now 3) sectors are 2.5, 9.9, and 2.0 kg, respectively. The new emissions for the disaggregated energy sectors are both fairly different than the original aggregated sector 1's emissions of 6.4kg. The emissions from sector 2 are slightly less (2kg compared to the previous 2.2 kg), since the revised A matrix from our hybrid analysis splits the purchases of energy by sector 2 differently, with relatively less dependence on the more polluting electricity sector. This example is intended to demonstrate the method by which one could mathematically disaggregate an A matrix with process data. Note that complete process data is not required, and that even limited process data for some of the transactions, coupled with assumptions on how to adjust other values, can still lead to interesting and relevant hybrid models. For example, if disaggregating an electricity sector into generation, transmission, and distribution, purchases of various services by the three disaggregated sectors may not be available. Assumptions that purchases of services are equal (i.e., divide the original sector's value by 3), or proportional to the outputs of the disaggregated sectors (i.e., distribute the original value by weighted factors) are both reasonable. Given the ultimate purpose of estimating environmental impact, its unlikely any of these choices on how to redistribute effects of a service sector would have a significant effect on the final results of the hybrid model. In the final category are integrated hybrid models, where there is a technology matrix that represents physical flows between processes and an economic IO matrix representing monetary flows between sectors. The general rationale for wanting to build an integrated hybrid model is that IO data may be comprehensive but slightly dated, or too aggregated to be used by itself. Use of process data can attempt to overcome both of these shortcomings. Both the process and IO matrices in this kind of model use make-use frameworks (see Advanced Material for Chapter 8), and are linked via flows at the border of both systems. Unlike the tiered or IO-based approached above, the models are called integrated because the process level data is fully incorporated into the IO model. Double counting is avoided by subtracting process-based commodity flows out of the IO framework. Integrated models require substantially more effort than the other two types of hybrid models because of the need to manage multiple unit systems (physical and monetary) as well as the need to avoid double counting of flows through subtraction. They may also require sufficiently detailed estimates of physical commodity prices across sectors. In general, the goal of an integrated hybrid model is to form a complete linear system of equations, physical and monetary, that comprehensively describe the system of interest. The general structure of such a model, analogous to Equation 8-5 is: Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
263
m m m x1m = z11 + z12 +...+ z1jm +...+ z1n + y1m
(mass)
m m m x 2m = z 21 + z 22 +...+ z 2jm +...+ z 2n + y 2m
(mass)
x $3 = z $31 + z $32 +...+ z $3j +...+ z $3n + y $3
(dollar)
: x $n = z $n1 + z $n2 +...+ z $nj +...+ z $nn + y $n
(dollar)
which leads to an A matrix with mixed units which has four separate partitioned matrices representing fully physical flows, fully monetary flows, and two connecting matrices for the interacting flows from the physical to monetary and monetary to physical sectors. The models can be "run" with inputs (Y) of physical and/or monetary flows, and outputs are then physical and monetary. Hawkins et al (2007) built an integrated hybrid model to comprehensively consider the implications of lead and cadmium flows in the US. Data on material (physical) flows were from USGS process-based data. The economic model used was the US 1997 12-sector input-output model (the most aggregated version of the benchmark model). E-resource: The Microsoft Excel spreadsheet for the Hawkins (2007) lead and cadmium models is available on the course website as a direct demonstration of the data needs and calculation methods for an integrated hybrid model (see the paper for additional model details). From the Hawkins model, Figure 9-8 shows direct and indirect physical and monetary flows from an input of final demand of $10 million into the manufacturing sector in 1997. Since the focus of this model is on lead, the left side of the graph summarizes flows (kg) through the physical lead-relevant sectors of the mixed-unit model while the right side of the graph summarizes monetary flows ($ millions) through the economic model. While the manufacturing sector is highly aggregated in this model, it shows the significant flows of lead through various physical sectors needed in the manufacturing sector. While a more disaggregated IO model may provide higher resolution insights into the specific flows of lead through manufacturing sectors, even this highly aggregated model required a personyear of effort to complete, and pushed the limits of the available USGS physical flow data. Given the specific data needs of these models, it is likely that a more disaggregated model is not possible given available process data.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
million $
16.7 13.7
8
2.0
6
1.5
4
1.0
2
0.5
0
0.0
L Le ead ad P Le Se rim ad co ary Min nd S i n ar me g y Le Sm ltin ad e g S ltin Le hee g En N ad tin d ew g L N of at Li Sto ea sold ur fe ra d e al S ge ox r re to i so ra Ba des ur ge tte ce B rie s att s an er d i C m es on in Tr M st ing an an r u sp uf cti a c on or Pr ta tu tio of rin es n an Tr g si on d ad ut e Ed al F uc an ina Info iliti at d b nc rm es io u ia n si l a atio an ne c n Le d h ss tivit is ea se ies ur e lth rvic an se e d rv s O hos ice th p s er ita se lity rv ic e O s th er
Lead (tonnes)
tonnes
Monetary Requirements million (million $)
264
Figure 9-8: Physical Lead And Monetary Output Required For $10 Million Final Demand Of Manufacturing Sector Output, 1997 (Source: Hawkins 2007). Output Directly Required Is Represented By The Hashed Areas While Dotted Gray Regions Depict The Indirect Output.
Chapter Summary Mathematics allows us to organize generic process data into matrices that can be used to create process matrix models. These process matrix models share some of the desirable features of input-output based models such as fast computation and larger boundaries while preserving the desirable process-specific detail. As LCA needs evolve, process and IO Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
265
models can be combined in various ways resulting in hybrid LCA models that leverage advantages of the two model types while overcoming some of the disadvantages. Hybrid LCA models vary with respect to the amount of resources and data needed, integration, and model development involved. All will generally yield more useful results than a single model. Now that we have introduced all of the core quantitative models behind LCA, we can learn how to take the next step, impact assessment.
Note that the E-resources provided with this book do not provide spreadsheet forms of the ecoinvent database, as it is a commercial product. However, if you have a license for ecoinvent directly from the website you can request the files needed to construct a process matrix. If you have an ecoinvent sublicense through purchase of SimaPro, you can use the "Export Matrix" option mentioned above to create your own Microsoft Excel-based ecoinvent process matrix. Note that the dimensions of the A matrix for ecoinvent 2.0 will be roughly 4,000 x 4,000, and the spreadsheet files will quickly become large (the B matrix dimensions will be 1,600 x 4,000). If trying to use ecoinvent data in a process matrix form, it is better to use MATLAB or other robust tools given issues in working with matrices of that size in Excel (see Advanced Material at the end of this chapter).
References for this Chapter Bullard C. W., Penner P. S., Pilati D.A., "Net energy analysis: handbook for combining process and input-output analysis", Resources and Energy, 1978, Vol. 1, pp. 267-313. Hawkins, T., Hendrickson, C., Higgins, C., Matthews, H. and Suh, S., “A Mixed-Unit InputOutput Model for Environmental Life-Cycle Assessment and Material Flow Analysis”, Environmental Science and Technology, Vol. 41, No. 3, pp. 1024 - 1031, 2007. Heijungs, R., "A generic method for the identification of options for cleaner products", Ecological Economics, 1994, Vol. 10, pp. 69-81. Joshi, S., "Product environmental life-cycle assessment using input-output techniques", The Journal of Industrial Ecology, 2000, 3 (2, 3), pp. 95-120. Suh, S., and Huppes, G., Methods for Life Cycle Inventory of a Product, Journal of Cleaner Production, 13:7, June 2005, pp. 687–697. Suh, S., Lenzen, M., Treloar, G., Hondo, H., Horvath, A., Huppes, G., Jolliet, O., Klann, U., Krewitt, W., Moriguchi, Y., Munksgaard, J., and Norris, G., "System Boundary Selection in Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
266
Chapter 9: Advanced Life Cycle Models
Life-Cycle Inventories Using Hybrid Approaches", Environmental Science and Technology, 2004, 38 (3), pp 657–664. Williams, Eric, "Energy Intensity of Computer Manufacturing: Hybrid Assessment Combining Process and Economic Input−Output Methods", Environmental Science and Technology, 38 (22), pp. 6166–6174, 2004.
End of Chapter Questions
1. Modify the two-process example from the Chapter (equations 9-1), and estimate E, if process 1 requires 1 kWh of electricity as an input. 2. The un-excerpted list of inputs in the US LCI database for the Bituminous coal, at mine process is shown below. The electricity flow is currently outside of our three-process linear system since we do not have an "at grid" electricity process. Input Bituminous coal, combusted in industrial boiler Diesel, combusted in industrial boiler Electricity, at grid, US, 2000 Gasoline, combusted in equipment Natural gas, combusted in industrial boiler Residual fuel oil, combusted in industrial boiler Dummy, Disposal, solid waste, unspecified, to underground deposit
Unit kg L kWh L m3 L kg
Amount 0.00043 0.0088 0.039 0.00084 0.00016 0.00087 0.24
Update the three-process example by assuming that the given flow of electricity is from bituminous coal-fired electricity (not grid average). How different are X and E? 3. Redo question 2 by updating the US LCI process matrix (746x959) found on the book website with the same electricity assumption. How different are X and E from the values in Figures 9-3 through 9-5? 4. Expand the scope of the three-process example by including the Diesel, combusted in industrial boiler process (an input at the mine) based on the US LCI data. What is your updated estimate of total fossil CO2 emitted across the new four-process system? 5. What price of electricity is needed as a final demand in the 2002 EIO-LCA producer price model to yield results comparable to the Microsoft Excel US LCI process matrix spreadsheet for electricity? Discuss which of the methods is likely more relevant, and why each model type is limited.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
267
6. Compare the percentage contribution results of the 2002 EIO-LCA producer price model Power generation and supply sector with the US LCI process matrix Electricity, at grid US process. Generally describe the differences in results of these two models. 6. Estimate the fossil CO2 emissions of 1kg alumina, at plant/US using the US-LCI process matrix model. How does your estimate of fossil CO2 emissions change if you apply cut-off criterion of 1%, 5%, and 10% to remove processes from the estimate? Given these findings and continuing to only be concerned with fossil CO2, what might be an appropriate cut-off criterion here? Have a problem where I truncate by hand the process matrix and have them run the same input into each and discuss results and effects of truncation?
Advanced Material for Chapter 9 – Section 1 – Process Matrix Models in MATLAB In this section, we build on the material presented in the chapter about process matrix models, which have already been demonstrated in Microsoft Excel, by showing how to implement them in MATLAB. One of the benefits of using MATLAB, if available, is that it is well-suited for manipulating a series of consecutive operations quickly and in real-time, without needing to save versions of normalized or inverted matrices for later use (as required for the Excel version of the model introduced in the chapter, leading to its large file size). E-resource: In the online supplemental material for Chapter 9 is a zip file containing matrix files and a .m file for the US LCI database (746 processes, 949 flows) that can be used in MATLAB. The A and B matrices are identical to those used in the Excel version. A subset of the code in the .m file is discussed below, which shows the MATLAB implementation of the same US LCI process matrix model as in the US LCI process matrix spreadsheets discussed in the chapter. Since the models were developed with the same parameters exported from SimaPro and using the same algorithm, results are identical. % matrices assumed to be in workspace (USLCI.mat): % USLCI_Atech_raw -
technology matrix from exported matrix (A)
% env_factors - environmental coefficients in exported matrix (B) % funct_units - row of functional units from exported Matrix clear all load('USLCI.mat') Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
268
Chapter 9: Advanced Life Cycle Models
%makes a "repeated matrix" with funct_units down columns funct_units_mat=repmat(funct_units,746,1); norm_Atech_raw=USLCI_Atech_raw./funct_units_mat; % normalizes the A matrix by functional units L=inv(eye(746)-norm_Atech_raw);
% this is the I-A inverse matrix
y=zeros(746,1); funct_units_env=repmat(funct_units,949,1); except has 949 rows
%
same
as
above
env_factors_norm=env_factors./funct_units_env; co2fossil=env_factors_norm(231,:); % row vector for the fossil CO2 flow % as example, enter a value into the y vector to be run through the model % default example here is 1 kWh into the bituminous coal-fired electricity process y(416,1)=3.6; (so 1 kWh) out=L*y;
%% funct unit basis is in MJ, this is MJ per kWh
% equivalent to x = [I-A]inverse * y
co2out=co2fossil*out; co2outcols=diag(co2fossil)*out; sum(co2outcols) % result of running this script will be the sum of fossil CO2 emissions throughout the upstream process matrix
If we run the .m code either by double clicking it in the MATLAB window, or selecting it and choosing the "Run" menu option, the result is 1.0334, which matches the Microsoft Excel version of the model.
Additional HW Questions for this Section Ecoinvent Example here?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
269
Advanced Material for Chapter 9 – Section 2 – Process Matrix Models in SimaPro In the Advanced Material for Chapter 5, demonstrations were provided on how to find process data in SimaPro. Here we show how the process matrix LCI results of a particular process can be viewed in SimaPro. Recall that SimaPro uses the same process matrix-based approach as shown for Microsoft Excel in the chapter. Using the same steps as shown in Chapter 5, find and select the US LCI database process for Electricity, bituminous coal, at power plant. Click the analyze button (shown highlighted by the cursor in Figure 9-9 below).
Figure 9-9: Analyze feature used in SimaPro to view data
The resulting window allows you to set some analysis options, as shown in Figure 9-10.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com Figure 9-10: New Calculation Setup Window in SimaPro
270
Chapter 9: Advanced Life Cycle Models
If needed, change the amount from 1 to 3.6MJ in the calculation setup window. Click the Calculate button. In the resulting window (shown in Figure 9-11), click the "Process contribution" tab. This shows the total technical flows from other products / processes needed to make the chosen (3.6 MJ) amount of electricity. Note the default units checkbox ensures the normally used units (e.g., kg) are displayed, otherwise SimaPro will try to maintain 3 digits and move to a higher or lower order unit (e.g., grams).
Figure 9-11: Results of Process Contribution Analysis for Process in SimaPro
You will see that these values are the same as those presented by the Microsoft Excel spreadsheets for the US LCI database (shown in Figure 9-4), but SimaPro tends to maintain Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
271
three significant digits so may be slightly different due to rounding. Clicking on the inventory tab of the results window shows the E matrix results for all tracked flows (Figure 9-12). Substances, compartments, and units are presented. While not all are shown, the results from the US LCI process matrix Excel spreadsheet would match those here.
Figure 9-12: Inventory Results in SimaPro
The final part to explore is the Network tab view. This tool creates a visualization of flows for the entire network of connected processes (as summarized in the process contribution tab). By default, SimaPro will truncate the Network display so as to reasonably draw the network system without showing all flows. Figure 9-13 shows a default network diagram for fossil CO2, with an assumed cut-off of about 0.09%. This cut-off can be increased or decreased to see more or less of the network.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
272
Chapter 9: Advanced Life Cycle Models
Figure 9-13: SimaPro Network view of process outputs (excerpt)
Further discussion of modeling with SimaPro is in a later chapter, but the analysis of LCI and LCA results uses this same analyze feature.
Additional HW Questions for this Section?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
273
Advanced Material for Chapter 9 – Section 3 – Process Matrix Models in openLCA As with SimaPro, openLCA also uses a process matrix approach behind the scenes of the tool. Again the focus on this Section will be on how to view LCI results. To do this, start openLCA as described previously, and click on the "Product systems" folder under the US LCI data folder. Choose the "Create a new product system" option. You may name it whatever you like, and optionally give it a description. In the reference process window, drill down through Utilities and then Fossil Fuel Electric Power Generation to find the Electricity, bituminous coal, at power plant process. Keep both options at the bottom of the window selected (as shown in Figure 9-14).
Figure 9-14: Creation of a New Product System in openLCA
You may then choose the calculate button at the top of the window (the green arrow with x+y written on it) to do the analysis of the process, as shown in Figure 9-15.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
274
Chapter 9: Advanced Life Cycle Models
Figure 9-15: Product System Information in openLCA
In the calculation properties dialog box, choose the "Analysis" calculation type, then click the Calculate button as shown in Figure 9-16.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
275
Figure 9-16: Calculation Properties Window in openLCA
The resulting window (not shown) has various tabs to display tables and graphs of the process matrix-calculated upstream results for your product system (your selected process in this case). If you enable it in the openLCA preferences, you can also download a spreadsheet export of analysis results here. The "LCI – Total" tab summarizes the inputs and output from the process matrix calculation as shown in Figure 9-17. Again, these are very similar to the results from the US LCI Excel spreadsheet or SimaPro. For our usual observation of the CO2 results, openLCA seems to aggregate all carbon dioxide air emissions into a single value (the US LCI database tracks 4 separate air emissions of CO2, including biogenic emissions).
Figure 9-17: LCI - Total Results for Product System in openLCA
The "Process results" tab shows the additional detail of the direct inputs and outputs from the chosen process as well as the total upstream, as shown in Figure 9-18.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
276
Chapter 9: Advanced Life Cycle Models
Figure 9-18: Process results view in openLCA
Additional HW Questions for this Section?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 9: Advanced Life Cycle Models
Another Advanced Material Section - Advanced features of EIO website (custom tab)?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
277
278
Chapter 10: Life Cycle Impact Assessment
Chapter 10 : Life Cycle Impact Assessment In this chapter, we complete the discussion of the major phases of the LCA Standard by defining and describing life cycle impact assessment (LCIA). This is the part of the standard where we translate the inventory results already created into new information related to the impacts of those flows, in order to help to assess their significance. These impacts may be on ecosystems, humans, or resources. As with the previous discussions about quantitative methods, life cycle impact assessment involves applying a series of factors to inventory results to generate impact estimates. While many impact assessment models exist, we begin by assessing some of the more common and simpler impact categories, such as those used for energy use and greenhouse gases, and then move on to more comprehensive LCIA methods used around the world. As always, our focus is on understanding the quantitative fundamentals associated with these efforts. Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. Describe various impact categories of interest in LCA and the ways in which those impacts can be informed by inventory flow information. 2. Describe in words the cause-effect chain linking inventory flows to impacts and damages for various examples. 3. List and describe the various mandatory and optional elements of life cycle impact assessment. 4. Select and justify LCIA methods for a study, and perform a classification and characterization analysis using the cumulative energy demand (CED) and/or climate change (IPCC) methods for a given set of inventory flows.
Why Impact Assessment? To help motivate the general need to pursue LCIA, we create a hypothetical set of LCI results that we will revisit throughout the chapter. These LCI results for two alternative product systems, A and B, may have been generated either as part of a prior study intended to only be an LCI (as opposed to an LCA), or as the LCI results to be subsequently used in an LCA. Due to either data constraints, or explicitly chosen statements in the goal and scope of the study, only a few flows have been tracked. As shown in Figure 10-1, a life cycle interpretation analysis of these results based only on the inventory would be challenging. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
279
Option A has more fossil CO2 emissions (5 kg) and use of crude oil 100 MJ), but fewer emissions of SO2 (2 kg), than Option B (2 kg, 80 MJ, 5 kg, respectively). Aside from stating that obvious tradeoff, which is how much of a compromise we would need to achieve between the flows, it is not clear what else an interpretation may contribute towards the decision support for A versus B. Flow Carbon dioxide, fossil Sulfur dioxide Crude oil
Compartment air air
Units kg kg MJ
Option A 5 2 100
Option B 2 5 80
Figure 10-1: Hypothetical Study LCI Results
The ideal case, of course, for the interpretation of LCI results is vector dominance, where the flows of one option across all inventoried flows are lower for one option than the other. In such a case, we would always prefer the option with lower flows. In reality, vector dominance in LCI results is rare, even with a small number of inventoried flows. As inventory flows are added (i.e., more rows in Figure 10-1), the likelihood of vector dominance nears zero, because more tradeoffs in flows are likely to appear across options. It is the existence of tradeoffs, and the typical comparative use of LCA across multiple product systems, that makes us seek an improved method to allow us to choose between alternatives in LCA studies, and for that we need to use impact assessment.
Overview of Impacts and Impact Assessment In Chapter 1, we motivated the idea of thinking about impacts of product systems. We showed that we might have concern for various impacts, and seek indicators to help us to understand how to measure and assess those impacts. For example, we described how we might measure our concern for fossil duel depletion by tracking coal and natural gas use (in MJ or BTU). Similarly we described how we might measure our concern for climate change in terms of greenhouse gas emissions (in kg or tons). These indicators, which became our LCI results, were intended to be short-term placeholders on our path to being able to circle back and consider the eventual impacts. In this chapter, we take the next steps needed to achieve this goal. The idea of impact assessment is not new. Scientists have been performing impact assessments for decades. Entire career domains exist for those interested in environmental impact assessment, risk assessment, performance benchmarking, etc. A key difference between life cycle impact assessment and other frameworks is its link to a particular functional unit (and of course the entire life cycle as a boundary), which focuses our attention on impacts as a function of that specific normalized quantity. Typically, risk or environmental impact assessments are for entire projects or products, such as the environmental impact of a highway expansion or a new commercial development. That said, Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
280
Chapter 10: Life Cycle Impact Assessment
the methods we will use in life cycle impact assessment (LCIA) have in general been informed and derived from activities in these other domains. Impact assessment is about being able to consider the actual effects on humans, ecosystems, and resources, instead of merely tracking quantities like tons of emissions or gallons of fuel consumed as a result of production. This chapter cannot fully describe the various methods needed to perform LCIA. Our focus is on explaining what the ISO LCA Standard requires in terms of LCIA, and on the qualitative and quantitative skills needed to document and complete this phase. Before discussing the mandatory and optional elements for LCIA in the Standard, we reintroduce the notion that there are various impacts that one might be concerned about, and discuss in limited There are impact assessment detail how we might frame our concerns for such categories for energy use and impacts in an LCA study. climate change, as we will see later in the chapter. But before we get to the formal definitions of those methods, we will reuse energy and climate examples along the way, as these two concepts are likely already familiar to you. Many of the other impact categories and methods available are much more complex, and we will save all discussion of those for later.
Figure 10-2 summarizes the different classes of issues of concern, called impact categories, which are commonly used in LCA studies. Also included is the scale of impact (e.g., local or global), and the typical kinds of LCI data results that can be used as inputs into methods created to quantitatively assess these impacts. This list of impact categories is not intended to be exhaustive in terms of listing all potential impact categories for which an individual or a party might have concern, or in terms of the potentially connected LCI results.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
Impact Category
Scale
281
Examples of LCI Data (i.e. classification) Carbon Dioxide (CO2), Nitrous Oxide (N2O), Methane (CH4), Chlorofluorocarbons (CFCs), Hydrochlorofluorocarbons (HCFCs), Methyl Bromide (CH3Br)
Global Warming
Global
Stratospheric Ozone Depletion
Global
Chlorofluorocarbons (CFCs), Hydrochlorofluorocarbons (HCFCs), Halons, Methyl Bromide (CH3Br)
Acidification
Regional, Local
Sulfur Oxides (SOx), Nitrogen Oxides (NOx), Hydrochloric Acid (HCl), Hydrofluoric Acid (HF), Ammonia (NH4)
Eutrophication
Local
Phosphate (PO4), Nitrogen Oxide (NO), Nitrogen Dioxide (NO2), Nitrates,Ammonia (NH4)
Photochemical Smog
Local
Non-methane hydrocarbon (NMHC)
Terrestrial Toxicity
Local
Toxic chemicals with a reported lethal concentration to rodents
Aquatic Toxicity
Local
Toxic chemicals with a reported lethal concentration to fish
Human Health Resource Depletion Land Use Water Use
Global, Regional, Local Global, Regional, Local Global, Regional, Local
Total releases to air, water, and soil. Quantity of minerals used, Quantity of fossil fuels used Quantity disposed of in a landfill or other land modifications
Regional, Local
Water used or consumed
Figure 10-2: Summary of Impact Categories (US EPA 2006)
Impact Assessment Models for LCA Figure 10-2 introduced various individual impact categories, but most of the attention and examples so far have related to climate change and energy. While these continue to be the most popular impact categories of interest in LCA (partly due to the relatively small amount of uncertainty regarding their application and thus the large degree of scientific consensus on their use), more comprehensive models of impacts exist that encompass multiple impact categories and have been incorporated into LCA studies and software tools. Some of the most frequently used LCIA methods are summarized and mapped to their available characterization models in Figure 10-3. Some of these may already be familiar to those that have reviewed existing studies. As Figure 10-3 shows, some LCIA methods are focused on a single category, e.g., cumulative energy demand (CED), while others broadly encompass all of the listed categories. Note that only the TRACI method is US-focused, with the remainder being mostly Europe-focused. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Resource consumption
Land use
X
CML2002
X
X
X
X
Eco-indicator 99
X
X
X
X
X
EDIP 2003/EDIP976
X
X
X
X
X
EPS 2000
X
X
X
X
Impact 2002+
X
X
X
X
IPCC
X
LIME
X
X
X
LUCAS
X
X
MEEuP
X
X
ReCiPe
X
Swiss Ecoscarcity 07 TRACI USEtox
Aquatic eutrophication.
Model CED
Terrest. eutrophication
Acidification
Ozone formation
Ecotoxicity
Ionising radiation
Human toxicity
Respiratory inorganics
Ozone depletion
Chapter 10: Life Cycle Impact Assessment
Climate change
282
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Figure 10-3: Summary of Impact Categories (Characterization Models) Available in Popular LCIA Methods (modified from ILCD 2010)
We will not be forced to choose a single impact category of concern. A study may set its study design parameters to include several, all, or none of the impact categories from the list in Figure 10-2, and thus may use one or more of the LCIA methods in Figure 10-3, with varying comprehensiveness. Using a diverse set of impact categories could allow us to make relevant comparisons across inventory flows so that we could credibly assess whether we should prefer a product system that releases 3 kg less of CO2 to air or one that releases 3 kg less SO2 to air (as in Figure 10-1). If our concerns are cross-media, such that some of our releases are to air and some are to water or soil, the challenge is even greater because we then need to balance concern for impacts in both ecosystems. Being able to take this next step in our LCA studies beyond merely providing LCI results is significant. The degree of difficulty and effort needed to successfully complete LCIA precludes some authors from even attempting it (which, as discussed above, is a big driver for why so many studies end at the LCI phase). As LCIA is perhaps most useful in support of comparative LCAs, it will not typically be very interesting or useful to know the LCIA results of a single product system. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
283
Beyond using multiple impact categories, multiple LCIA methods are often used in studies to assess whether different approaches agree on the severity of the chosen impacts. Of course, this is only useful when the LCIA methods use different underlying characterization models. The outputs of LCIA methods will be discussed below. In order to understand impact assessment, and thus LCIA, it is important to understand how LCI results may eventually connect to impacts. Figure 10-4 shows the cause-effect chain (also referred to as the environmental mechanism) for an emission category. Similar chains exist for impact categories like resource depletion and land use. At the top are emissions, sometimes referred to as stressors, so called because they are triggers for potential impacts (the 'causes' in the cause-effect chain). While shown as single chains in the figure, there may be various stressors all leading to the same potential impacts or damages. Likewise, the same emissions may be the first link in the chain for multiple effects (not shown).
Emissions ConcentraHons Impacts
Midpoints
Damages
Endpoints
Figure 10-4: General Cause-Effect Chain for Environmental Impacts (Adapted from Finnveden 1992)
Next in the chain are concentrations, which in the case of air emissions are the resulting contribution of increased emissions with respect to the rest of the natural and manmade molecules in the atmosphere. A relatively small emission would have a negligible effect on concentrations, while a large emission may have a noticeable effect on concentrations. In the case of climate change impacts, increased emissions of greenhouse gases lead to increased concentrations of greenhouse gases in the atmosphere. As concentrations are changed in the environment, we would expect to see intermediate impacts. For the case of climate change, increased concentrations of greenhouse gases are expected to lead to increased warming (actually radiative forcing). Emissions of conventional pollutant emissions lead to increased concentrations in the local atmosphere. These intermediate points of the chain are also called midpoints, which are quantifiable Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
284
Chapter 10: Life Cycle Impact Assessment
effects that can be linked back to the original emissions, but are not fully indicative of the eventual effects in the chain. Finally, damages arise from the impacts. These damages are also referred to as endpoints, since they are the final part of the chain and represent the inevitable ending point with respect to the original stressors. These damages or endpoints are the "effects" in the causeeffect chain. For global warming (or climate change), the damages/endpoints of concern may be destruction of coral reefs, rising sea levels, etc. For conventional pollutants, endpoints may be human health effects due to increased exposure to concentrations, like increases in asthma cases or hospital admissions. For ozone depletion, we may be concerned with increases in human cancer rates due to increased UV radiation. Note that LCIA will not actually quantify these damages (i.e., it will not give an estimate of the number of coral reefs destroyed or height of sea level change), but it will provide other useful and relevant information that could subsequently allow us to consider them. Fortunately, as we will learn below, the science behind impact assessment, while continuing to be developed, is available for us to use without needing to build it ourselves. But using the relevant methods still requires substantial understanding of how these methods work. Getting to the idea of an endpoint is hard, and again, that is partly why people stop at the inventory stage. Along the way, we have seen how LCAs can yield potentially large lists of inventory results. These are generally lists of inputs needed (e.g., fuels used) and outputs created (e.g., GHG emissions) by our product systems. The prospect of impact assessment may create an intimidating sense of "how will we pull together a coherent view of impact given this large list of effects?" However, in reality, impact assessment methods are created exactly to deal with using large inventories as inputs. Impact assessment methods will attempt to take the detailed information in those inventories and create summary indicators of impacts from them.
ISO Life Cycle Impact Assessment In Chapter 4, we began our summary discussion of the ISO Standard. Figure 10-5 repeats the original Figure 4-1 which overviews the major phases of the Standard. As we have already discussed, the various phases are all iterative. We remind you that the text in this chapter is not intended to replace a careful read of the ISO LCA Standard documents specific to LCIA, as here we only summarize the information, link it to previously discussed material, and show examples.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
285
Figure 10-5: Overview of ISO LCA Framework (Source: ISO 14040:2006)
In life cycle impact assessment (LCIA), we associate our LCI results with various impact categories and then apply other factors to these categorized results to give us information about the relevant impacts of our results. We also then iteratively return to the life cycle interpretation phase so that we can add to our interpretations made when only the LCI was complete. LCIA also connects iteratively back to the LCI phase, so that if the LCIA results do not help us in expected ways, we can refine the inventory analysis to try to improve our study. While not shown as a direct connection in Figure 10-5, we may also iteratively decide to adjust the study design parameters (i.e., goal and scope) if we interpret that our impact assessment results are unable to meet our objectives. As we will see below, some elements of LCIA may be subjective (i.e., influenced by our own value judgments). As stated in the Standard, it is important to be as transparent as possible about assumptions and intentions when documenting LCIA work so as to be clear about these subjective qualities. Figure 10-6 shows the various steps in LCIA, which includes several mandatory and several optional elements, each of which is discussed below. The steps in LCIA are commonly referred to by the shorthand name in parentheses in the figure.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
286
Chapter 10: Life Cycle Impact Assessment
Figure 10-6: Overview of Elements of LCIA Phase (Source: ISO 14040:2006)
Mandatory Elements of LCIA Selection The first mandatory element of LCIA is the selection of impact categories, their indicators, and the characterization models and LCIA methods to be used. In practice, this element also involves sufficiently documenting the rationale behind these choices, which need to be consistent with the stated goal and scope of the study. While we will discuss more of the various possible impact categories later in this chapter (as well as indicators and characterization models), we know from previous discussions that climate change is an impact category. If we wanted to include climate change as one of our study's impact categories, then we should justify why climate change is a relevant impact category given our choice of study design parameters (goal and scope) and/or given the product system itself – i.e., is it a product considered to be a major potential cause of climate change? The ISO Standard requires that the impact assessment performed encompass "a comprehensive set of environmental issues" so that the study is not narrowly focused, for example, on one particular hand picked impact that might be chosen since it can easily show Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
287
low impacts. Thus, our LCIA should use multiple impact categories. Our justification should include text for all of our chosen categories. A study's justification for selection of impact categories should not be subjective to the author's own personal wishes. They should consider those of the organization responsible for funding the study, or who is responsible for the product system. For example, if the organization requesting the study has long-term goals of mitigating climate change in their actions, that would be an appropriate justification for choosing climate change as an impact category when assessing their products. The LCIA methods selected should be relevant to the geographical area of the study. An LCA on a product manufactured in a US factory would not be well-served by using an LCIA method primarily developed in, and intended to be applied in, Europe. However, the majority of LCIA models have been created only for the US and Europe, and thus, it can be challenging to select a model if considering a product system in Asia. In such cases, it may make sense to use multiple models outside of the relevant geographic region to consider ranges of results and to try to generalize findings. This step should also document and reference the studies on impacts used, i.e., the specific scientific studies used to assess impacts of greenhouse gases. Of course, the vast majority of LCA studies will be using well-established LCIA methods. Beyond this initial LCIA element for justification of choices, the remaining mandatory elements involve the organization and application of indicator model values to your previously generated inventory results.
Classification Classification is the first quantitative element of LCIA, where the various inventory results are organized such that they map into the frameworks of the relevant impact category frameworks chosen for the study. Classification involves copying your inventory items into a number of different piles, where each pile is associated with one of the impact categories used by the selected LCIA methods. Consider again the hypothetical inventory from Figure 10-1. If a study has selected climate change as an impact category, then the carbon dioxide, fossil inventory flow would be classified into that pile since it is a greenhouse gas (and the other two flows would not). If you chose an impact category for energy, then the crude oil inventory flow would be classified there (and the other two would not). If you chose no other impact categories, then the sulfur dioxide flow would not be classified anywhere, and would have no effect on the impact assessment. To be able to perform classification, each LCIA method must have a list of inventory flows connected to that impact. As discussed in Chapter 5, LCI results can have hundreds or Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
288
Chapter 10: Life Cycle Impact Assessment
thousands of flows. Thus, the list of relevant connected flows for LCIA methods can likewise be substantial (hundreds or thousands of interconnections). Classification has no quantitative effect on the inventory flows other than arranging and creating piles. It is possible that the classified list of inventory flows relevant to a chosen LCIA method have different underlying units (e.g., kg, g, etc.). These differences will be managed in subsequent elements of the LCIA. Amongst the most widely used impact categories are those for climate change and energy use. Two specific underlying methods to support these are the Intergovernmental Panel for Climate Change (IPCC) 100-year global warming potential method and the cumulative energy demand (CED) method, respectively. We describe each below and use them to illustrate the mechanics of the various LCIA elements through examples. Since the IPCC and CED methods have all of the mandatory elements, they qualify as LCIA methods, but they are fairly simplistic and singularly focused compared to some of the more advanced LCIA methods in Figure 10-3. Studies that only consider energy and global warming impacts are sometimes viewed as being narrowly focused with respect to impact assessment, especially since energy and climate results tend to be very similar. Figure 10-7 provides an abridged list of substance names and chemical formulas (generally greenhouse gases) that are classified in the IPCC method introduced above. Thus, any of the substances in this list that are in an LCI would be copied into the pile of classified substances to be used in assessing the impacts of climate change.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
Name Carbon dioxide Methane Nitrous oxide CFC-11 CFC-12 CFC-13 CFC-113 CFC-114 CFC-115 Halon-1301 Halon-1211 Halon-2402 Carbon tetrachloride Methyl bromide Methyl chloroform HCFC-22 HCFC-123 HCFC-124 HCFC-141b HCFC-142b HCFC-225ca HCFC-225cb
289
Chemical Formula CO2 CH4 N2O CCl3F CCl2F2 CClF3 CCl2FCClF2 CClF2CClF2 CClF2CF3 CBrF3 CBrClF2 CBrF2CBrF2 CCl4 CH3Br CH3CCl3 CHClF2 CHCl2CF3 CHClFCF3 CH3CCl2F CH3CClF2 CHCl2CF2CF3 CHClFCF2CClF2
Figure 10-7: (Abridged) List of Substances Classified into IPCC (2007) LCIA Method
Likewise, Figure 10-8 provides an example list of energy flows that would be classified into the CED method. Note that the CED method further sub-classifies renewable and nonrenewable energy, as well as particular subcategories of energy (e.g., fossil, solar). Also note that the listings in Figure 10-8 are not specific to known flows in any of the databases. One database might have a flow for a particular kind of coal or wood that is named something different in another database. Category Non-renewable resources
Renewable resources
Subcategory fossil nuclear primary forest biomass wind solar geothermal water
Included Energy Sources hard coal, lignite, crude oil, natural gas, coal mining off-gas, peat uranium wood and biomass from primary forests wood, food products, biomass from agriculture, e.g. straw wind energy solar energy (used for heat & electricity) geothermal energy (shallow: 100-300 m) run-of-river hydro power, reservoir hydro power
Figure 10-8: Energy Sources Classified into Cumulative Energy Demand (CED) LCIA Method (Source: Hischier 2010)
If classification is done manually (which is rare), then various quality control problems could occur in creating the piles of classified inventory flows. For example, you would need to look at each of your inventory results and then check every LCIA method's list of classified substances to see whether it should be put into that pile, and to put it into the correct pile. It would be easy to make errors in such a process, either by not noticing that certain Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
290
Chapter 10: Life Cycle Impact Assessment
inventory flows are classified into a method, or by classifying the wrong flows (e.g., those in an adjacent row number). In practice, most classification is done via the use of software tools and/or matrix manipulation. Even so, making the classification process work efficiently is not easy. There are also potential problems associated with the computerized classification process (Hischier 2010). First, inventory flows reported in databases (or from primary data collection) may be named inconsistently with scientific practice, and cause mismatches or inability to match with LCIA methods. For example, one source may list CO2 and another carbon dioxide. Behind the scenes of the software tools, many of the "matches" are done by using CAS numbers to avoid such problems. CAS Numbers give unique identities to specific chemicals (e.g., formaldehyde is 50-00-0). Beyond naming problems, an LCIA method may have many listed flows that should be classified under it, but the inventory done may be so streamlined that none of the classified flows have been estimated. Conversely, a relatively substantial LCI may have no flows that can be classified into any of the selected LCIA methods. In short, the connection between flows in an LCI and the classification list of flows in the LCIA method is not one-to-one. Of course, should problems like these be identified during the study, then changes should be made to the study's goal, scope, or inventory results to ensure that relevant flows are identified so as to be able to make use of the selected LCIA method (or, of course, the LCIA method should be adjusted). This potential disconnect between available and quantified inventoried flows and the connections to the inputs of LCIA methods goes is critical to understand. Since most LCIA methods have a large list of classifiable LCI flows, it is critical that inventory efforts are sufficiently robust so as to make full use of the methods. Thus, inventories must track a sufficient number of flows needed for the classification step of the chosen method. A significant risk is posed when doing primary data collection of a new process. Imagine the case where the study author has chosen a climate change method for LCIA. If only CO2 is inventoried as part of the boundary set in the data collection effort, then the potential climate change effects due to non-CO2 GHGs (which are more potent) cannot be considered in the LCIA. One could use IO-LCA screening as a guide to help explicitly screen for inventory flows that should be measured or verified to be zero when using a particular method. For example, if another round of data collection could measure emissions of methane or other GHGs, it could have a substantial effect on the results. It is possible that you could have chosen an impact category (or categories) such that none of your quantified inventory flows are classified into the pile for that category, giving you a zero impact result. While unlikely, this again would be a situation where you would want to iterate back to the inventory stage and either redouble data collection efforts, or iterate all the way back to goal and scope to change the parameters of the study.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
291
It is likely that inventory flows will be classified into multiple impact category piles. For example, various types of air emissions may be classified into a climate change impact category, an acidification impact category, and others. In these cases, the entire flow is classified into each pile (not assigned to only one of the piles, and not having its flows allocated across the impact piles). Figure 10-9 shows the classification results for the hypothetical inventory example for the IPCC and CED methods. Once the classification is completed, the LCIA proceeds to the next required step, characterization. Classification: Climate Change Impact Category (IPCC) Compartment Units Option A Carbon dioxide, fossil air kg 5 Classification: Energy Impact Category (CED) Crude Oil kg 10 Flow
Option B 2 8
Figure 10-9: Classification of Hypothetical Inventory
Characterization The characterization element of LCIA quantitatively transforms each set of classified inventory flows via characterization factors (also called equivalency factors) to create impact category indicators relevant to resources, ecosystems, and human health. The purpose of characterization is to apply scientific knowledge of relative impacts such that all classified flows for an impact can be converted into common units for comparison. The characterization methods are pre-existing scientific studies that are leveraged in order to create the common units. For example, in the climate change impact example we have been using in this chapter, the characterization method is from IPCC (2007). This IPCC method is well known for creating the global warming potential equivalency values for greenhouse gases, where CO2 is by definition given a value of 1 and all other greenhouse gases have a factor in equivalent kg of CO2, also abbreviated as CO2-equiv or CO2e. Similar to other methods, this creates (in effect) a weighting factor adjustment for greenhouse gases. Furthermore, since all characterized values are in equivalent kg of CO2, the values can be aggregated and reported in the common unit of an impact category indicator. The IPCC report actually provides several sets of characterization factors, for different time horizons of greenhouse gases in the atmosphere. The factors typically used in LCA and other studies are the IPCC 100-year time horizon values, but values for 20 and 500 years are also available. Figure 10-10 shows the characterization (equivalency) factors for greenhouse gases in the IPCC Fourth Assessment Report (2007) 100-year method. Thus, 1 kg of methane has the warming potential of 25 kg of carbon dioxide. Any classified greenhouse gases (or other substances appearing in the list of characterized flows) would then be multiplied by the "kg CO2e/kg of substance" factors to create the characterized value for each inventory flow.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
292
Chapter 10: Life Cycle Impact Assessment
Name
Chemical Formula
Carbon dioxide Methane Nitrous oxide CFC-11 CFC-12 CFC-13 CFC-113 CFC-114 CFC-115 Halon-1301 Halon-1211 Halon-2402 Carbon tetrachloride Methyl bromide Methyl chloroform HCFC-22 HCFC-123 HCFC-124 HCFC-141b HCFC-142b HCFC-225ca HCFC-225cb
CO2 CH4 N2O CCl3F CCl2F2 CClF3 CCl2FCClF2 CClF2CClF2 CClF2CF3 CBrF3 CBrClF2 CBrF2CBrF2 CCl4 CH3Br CH3CCl3 CHClF2 CHCl2CF3 CHClFCF3 CH3CCl2F CH3CClF2 CHCl2CF2CF3 CHClFCF2CClF2
Characterization Factor (kg CO2-eq / kg of substance) 1 25 298 4,750 10,900 14,400 6,130 10,000 7,370 7,140 1,890 1,640 1,400 5 146 1,810 77 609 725 2,310 122 595
Figure 10-10: IPCC 2007 100-year Characterization Factors (abridged)
While similar in application, the Cumulative Energy Demand (CED) method introduced above has multiple subcategories for which inventory flows are classified, and thus an additional level of characterization factors by subcategory. The characterization factors transform original physical units of energy from each source into overall MJ-equivalent category indicator values. Category indicators for CED are typically reported by subcategory (e.g., fossil, nuclear, solar, or wind), aggregated into the categories (e.g., non-renewable and renewable), and then aggregated into a total (cumulative) energy demand. Figure 10-11 shows CED characterization values used in the ecoinvent model.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
293
CED characterization factors (MJ-equivalent per unit) Non-renewable primary forest fossil nuclear 9.90 19.10 38.29 560,000 45.80 9.9
Renewable wind geo(kinetic) solar thermal
water Source Unit biomass Coal, brown kg Coal, hard kg Natural gas Nm3 Uranium kg Crude oil kg Peat kg Energy, biomass, MJ 1 primary forest Energy, in MJ 1 biomass Energy, wind MJ 1 (kinetic) Energy, solar MJ 1 Energy, MJ 1 geothermal Energy, hydropower MJ 1 (potential) Figure 10-11: Cumulative Energy Demand Values Used In Ecoinvent Model (Abridged from Hieschier 2010). Nm3 means normal cubic foot (normal temperature and pressure)
Models may internally change their mappings between inventory flows. For example, ecoinvent maps hard and soft wood uses into the energy, biomass categories shown above. Likewise, energy values are often pre-converted and adjusted to appropriate categories when creating the processes in LCI databases (as in the kinetic and potential energy values in Figure 10-11). Due to differences in naming inventory flows in different systems, CED characterization factors often have to be tailored for different frameworks (i.e., CED values used for the US LCI database may be different than those above). All of these issues make comparisons of CED results across different databases and software tools problematic. While not discussed in this book, the science behind the development of characterization factors for use in LCIA methods is an extremely time consuming and comprehensive research task. Activities involved include finding impact pathways, relating flows to each other, and then inevitably the development of the equivalency factors. Such research takes many person-years of effort, yet the provision of convenient equivalency factors as shown above may make the level of rigor appear to be small. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
294
Chapter 10: Life Cycle Impact Assessment
Characterized flow = inventory flow (raw flow unit) * characterization factor (characterization factor units / unit inventory flow) Category Indicator Results or LCIA Results The summary of all category indicator values used is referred to as the LCIA profile. Using the IPCC and CED factors above, Figure 10-12 shows the LCIA profile associated with Figure 10-9. Note that the CED values are the product of the raw values for kg of crude oil (10 and 8 for Options A and B, respectively) with the CED characterization factor for crude oil of 45.8 MJ-eq/kg. Characterization: Climate Change (IPCC 2007) Indicator Units Option A Equivalent releases CO2 kg CO2 equiv. 5 Characterization: Energy (CED) Non-renewable fossil MJ-eq. 458 Non-renewable nuclear MJ-eq. 0 Non-renewable forest MJ-eq. 0 Non-renewable total MJ-eq. 458 Renewable total MJ-eq. 0
Option B 2 366 0 0 366 0
Figure 10-12: LCIA Profile of Hypothetical Example
Note from Cate: The sample question was originally posed as assessing the decision between an option with 3kg less CO2 or an option with 3 kg less SO2. But the SO2 was not addressed in either of the methods. Maybe there’s another example that can be tied up cleanly for the purpose of this chapter.
You presented research in class that a student had worked on about natural gas, diesel, and gasoline fueled buses and cars. I can’t find it now, but I thought it compated CO2 and SO2 emissions for the different fuel type. Not sure if it was and LCA . Maybe you could reference the results as a way to conclude the example, but without having to discuss an assessment method that addresses the sulfur dioxide. Characterization represents the last of the initial mandatory elements in LCIA, as the remaining elements are optional, and many LCA studies skip all optional elements. If it were to be the final step, one could next interpret the results in Figure 10-12. Given that only energy and greenhouse gas related impacts were chosen for the study, and that these impacts tend to be highly correlated, it is not a surprise to see that the characterized LCIA results suggest the same result, i.e., that Option B has lower impacts than Option A. The interpretation of course should still highlight the fact that this result occurs because of the chosen impact assessment methods. If other impact categories were selected, different answers could result, including tradeoffs between impacts.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
295
There is a final step, evaluation and reporting, which is not officially in the Standard but is described after the optional elements below. This step should be done after characterization regardless of whether the optional elements are pursued.
Optional Elements of LCIA The remaining text in this chapter discusses the optional elements of LCIA. Note that each of the elements below is independently optional, i.e., one could extend the characterized result by doing none, some, or all, of these optional elements. The underlying concepts of the optional elements are far simpler than in the mandatory elements, and thus the explanations are more concise. Part of the reason that they are optional is that they build on the relatively objective results from the mandatory elements and may introduce subjective components (even if not perceived as subjective to the study authors) into the LCA. They also modify the "pure" results that end with characterization, which all use known and established scientific factors used throughout the community. It is in passing over the threshold between the mandatory and optional elements that two parties provided with the same characterized LCIA results could subsequently generate different LCIA results. Beyond the mere subjectivity issues, taking the additional optional steps can lead to results that are hard to validate or compare against in future LCA studies. Because of these issues, as noted above, many studies end the LCIA phase of the study at characterization. Normalization Normalization of LCIA results involves dividing by a selected reference value or information. A separate reference value is chosen for each impact. The rationale of normalization is to both provide perspective or context to LCIA results and also to help to validate results. There is no specified set of reference values to be used in all LCIA studies. The Standard provides suggestions on useful reference values, such as dividing by total (or total per-capita) known indicators in a region or system, total consumption of resources, etc. Another useful normalization factor is an LCIA indicator result for one of the alternatives studied (or the indicator value from a previously completed study of a similar product system) as a baseline. The chosen reference value might be the largest or smallest of the results. In this type of normalization, a key benefit is the creation of a ratio-like normalized indicator for which alternatives can be compared. For example, any normalized result greater than 1 has higher impact than the baseline, and any less than 1 has lower impact. A potential downside of normalization is that the vast majority of product systems studied will have negligible impacts compared to the total or even the per-capita values to be used as the reference value. Thus, normalized values tend to be extremely small and thus their effect can be viewed as irrelevant or negligible. This can be partially solved by choosing similar but Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
296
Chapter 10: Life Cycle Impact Assessment
normalized reference values; for instance, instead of using the total annual impact or resource consumption, one may choose a daily value. It can also be solved by assuming a level of production for the product system and scaling up the functional unit of the study so that the normalized values are larger. As an example, in a study considering the life cycle of gasoline, the functional unit could be 100 billion gallons per year instead of 1 gallon. Given the potential for comparability issues, it is often useful to develop multiple normalization factors, and to perform sensitivity analyses on the normalization results. Various LCA communities around the world have invested time and research effort in the development and dissemination of normalization databases and factors to be used in support of LCA studies. Such efforts are extremely valuable as they serve to provide a common set of factors that can be cited and used broadly in studies of impacts in the relevant country. It also removes the need for practitioners to independently create their own normalization values, which can cause problems in comparing results across studies. In the US, the EPA published a set of total and per-capita normalization factors to be used as relevant for the year 1999 (Bare et al. 2006) in support of the TRACI US LCIA model as shown in Figure 10-13 (yearly) and Figure 10-14 (yearly, per capita). The "NA" values in the tables represent normalization factors that are unnecessary, such as for greenhouse gas emissions to water, or fossil fuel depletion from air or water. Note that while these values were created as being specific to the year 1999, various practitioners have continued to use them as is for the past decade. The assumed population in deriving the per-capita estimates was 280 million, so, one could update the per-capita factors with the current population if desired while retaining the 1999 baselines.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
Water
Total normalized Value
Normalized value Impact category
Air
297
Normalized unit
Acidification
2.08 E+12
NA
2.08 E+12
H+ equiv/yr
Ecotoxicity
2.03 E+10
2.58 E+08
2.06 E+10
2,4-D equiv/yr
Eutrophication
1.44 E+09
3.58 E+09
5.02 E+09
N equiv/yr
Global warming
6.85 E+12
NA
6.85 E+12
CO2 equiv/yr
Human health cancer
7.03 E+07
1.76 E+06
7.21 E+07
benzene equiv/yr
Human health noncancer
3.69 E+11
4.24 E+10
4.11 E+11
toluene equiv/yr
Human health criteria
2.13 E+10
NA
2.13 E+10
PM2.5 equiv/yr
Ozone depletion
8.69 E+07
NA
8.69 E+07
CFC-11 equiv/yr
Photochemical smog
3.38 E+10
NA
3.38 E+10
NOx equiv/yr
Fossil fuel depletion
NA
NA
1.14 E+07
surplus MJ of energy/yr
Figure 10-13: Summary of Total Annual Normalization Factors for US, 1999 (Source: Bare et al 2006)
Air
Water
Total normalized Value percapita
Acidification
7.44 E+03
NA
7.44 E+03
H+ equiv/yr / capita
Ecotoxicity
7.29 E+01
9.24 E-01
7.38 E+01
2,4-D equiv/yr / capita
Eutrophication
5.15 E+00
1.28 E+01
1.80 E+01
N equiv/yr / capita
Global warming
2.45 E+04
NA
2.45 E+04
CO2 equiv/yr / capita
Human health cancer
2.52 E-01
6.30 E-03
2.58 E-01
benzene equiv/yr / capita
Human health noncancer
1.32 E+03
1.52 E+02
1.47 E+03
toluene equiv/yr / capita
Human health criteria
7.63 E+01
NA
7.63 E+01
PM2.5 equiv/yr / capita
Ozone depletion
3.11 E-01
NA
3.11 E-01
CFC-11 equiv/yr / capita
Photochemical smog
1.21 E+02
NA
1.21 E+02
Fossil fuel depletion
NA
NA
4.08 E-02
NOx equiv/yr / capita surplus MJ of energy/yr/ capita
Normalized value per-capita Impact category
Normalized unit percapita
Figure 10-14: Summary of Per-Capita Normalization Factors for US, 1999 (Source: Bare et al 2006)
Chris notes there is a new version of this paper/table to include Something to add for usetox? Given our example above, we could create normalized values for Figure 10-12 by dividing the equivalent CO2 releases by the total factor of 6.85x1012 and/or the per-capita value of 2.45x104, and the energy values by 1.14x107 and 4.08x10-2, respectively. Grouping Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
298
Chapter 10: Life Cycle Impact Assessment
Grouping of LCIA results is achieved by combining LCIA results to meet objectives stated in the goal and scope. If a study includes only one or two preselected impacts, then the results of grouping are not apparent. However, if more than a handful of impacts have been selected, then grouping them together for reporting and presentation can help to guide the reader through the results.
Note from Cate: It’s not critical for understanding, but it might be nice to show an excerpt of a study that presented its prioritized results in a color-coded heat map to give the reader an idea of one way that the results could be communicated. Grouping is accomplished by sorting and/or ranking the characterized or normalized LCIA results. The Standard allows sorting of the results along dimensions of the values, the spatial scales, etc. Ranking, on the other hand, is done by creating a hierarchy, such as a subjectively-defined impact priority hierarchy of high-medium-low, to place the impacts into context with each other. Since it involves an assessment of how to prioritize impacts, grouping should be done carefully, but should also acknowledge that other parties might create different rankings based on different priorities and ranked impacts.
Weighting Weighting of LCIA results is the most subjective of the optional elements. In weighting, a set of factors are developed, one for each of the chosen impact categories, such that the results are multiplied by the weighting factors to create a set of weighted impacts. Weighting factors may be derived with stakeholder involvement. As with grouping, the practice of weighting is subjective and could lead to different results for different authors or parties. The weights chosen in the study may be different than what may be chosen by the reader. Regardless, the method used to generate the weighting factors, and the weighting factors themselves, needs to be documented. Results should be shown with and without weighting factors applied. Beyond subjectivity concerns, it is possible that a consideration in doing the study was a particular set of potential impacts, such as local emissions of hazardous substances at a factory. In such a case, weighting such impacts greater than other impacts can be deemed credible and also fit well within the goal and scope considerations. It also means that a separate study of the same product system but with a different production location (or a different set of weights) could lead to a different perceived impact.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
299
Last Step - Evaluation and Reporting While not listed as an element in the Standard, a final step of LCIA is to evaluate and report on results from the various elements. It is important that intermediate LCIA profile results from the individual mandatory (and optional) elements be shown. This prevents the study from, for example, providing only final results that have been normalized and/or grouped and/or weighted, at the expense of not showing what the characterized results would have been. Showing the ‘pure’ impact assessment results also gives your study greater utility as it will be relevant for comparison to a larger number of other studies.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
300
Chapter 10: Life Cycle Impact Assessment
Chapter Summary As first introduced in Chapter 4, life cycle impact assessment (LCIA) is the final quantitative phase of LCA. LCIA allows us to transform the basic inventory flows created from the inventory phase of the LCA and to attempt to draw conclusions related to the expected impacts of these flows for product systems. While climate change and cumulative energy demand tend to dominate LCA studies, other impact categories of broad interest have characterization models that are scientifically credible and available for use. Despite the availability of these credible models and tools, many LCA studies continue to focus just on generating inventory results, or at most, use only climate and energy impact models. Now that we have reviewed all of the important phases of LCA, in the next few chapters we focus on ways in which we can create robust analyses that will serve our intended goals of building quantitatively sound and rigorous methods.
References for this Chapter Bare, Jane, Gloria, Thomas, and Norris, Gregory, "Development of the Method and U.S. Normalization Database for Life Cycle Impact Assessment and Sustainability Metrics", Environmental Science and Technology, 2006, Vol. 40, pp. 5108-5115. Finnveden, G., Andersson-Sköld, Y., Samuelsson, M-O., Zetterberg, L., Lindfors, L-G. "Classification (impact analysis) in connection with life cycle assessments—a preliminary study." In Product Life Cycle Assessment—Principles and Methodology, Nord 1992:9, Nordic Council of Ministers, Copenhagen. 1992. "ILCD Handbook: Analysing of existing Environmental Impact Assessment methodologies for use in Life Cycle Assessment", First edition, European Union, 2010. IPCC Fourth Assessment Report: Climate Change 2007. Available at www.ipcc.ch, last accessed October 30, 2013. "Life Cycle Assessment: Principles And Practice", United States Environmental Protection Agency, EPA/600/R-06/060, May 2006. Hischier, Roland and Weidema, Bo (Editors), "Implementation of Life Cycle Impact Assessment Methods Data v2.2 (2010)", ecoinvent report No. 3 St. Gallen, July 2010. Homework Questions for Chapter 10 e. TBA
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 10: Life Cycle Impact Assessment
Advanced Material – LCIA in SimaPro Follow along with lecture notes
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
301
302
Chapter 11: Uncertainty in LCA
Chapter 11 : Uncertainty and Variability Assessment in LCA Every number we measure or estimate is uncertain. In this chapter, we discuss issues related to uncertainty and variability in life cycle data, as well as in LCA and LCIA results. These issues continue to We also discuss the implications of uncertainty and variability in terms of interpreting study results. These issues are perhaps most critical when doing comparative assessments where our qualitative conclusions may be dependent upon the quantitative strength of our data and results. As already motivated in several chapters, uncertainty and variability play a big role in the use of data and models in LCA, and recognition of their issues should be addressed when using data, creating models, and interpreting results. These activities range from qualitative identification of uncertainty and variability, to sensitivity analysis and use of quantitative ranges, up to probabilistic definitions of data and results.
Chapter Quote: “A decision made without taking uncertainty into account is barely worth calling a decision.” Wilson 1985. Also Finkel 1990 on good decisions and good outcomes?
Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. Describe why uncertainty and variability affect LCA model results 2. Describe the various sources and types of uncertainty and variability for data and methods 3. Develop methods that incorporate uncertainty into LCA Models 4. Select and justify LCIA methods for a study, and perform a classification and characterization analysis using the CED and/or IPCC methods for a given set of inventory flows
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
303
Comments/placeholders/to do Maybe re-look at various places in book to here where uncertainty, variability mentioned. Chap 2, 3, 5, etc. Review what to keep at end of Chap 5 (cth Jun3 2014 comments). (CW - better framing example than paper v plastic? Lots of SDP choices that lead to the differences)
Why Uncertainty Matters (keep this here?) To help us frame where our work so far in the book has led us, Figure XX shows results from the actual Hocking (XXX) study, which continue to be the typical result of a generic comparative LCA. In typical LCA studies, a long sequence of assumptions and citations leads to a "one off" LCA model (and results) that can be expressed either in table or figure form. Such results are typically the total LCI (or LCIA, if we are lucky) results expressed as a certain value. When "A is less than B" for a particular LCI result or impact, we say that in comparison, A is better. Our threshold for making such an observation is generally not stated, but typically is a simple less than comparison. It doesn't matter whether A is only less than B in the third or fourth significant digit (as shown in Figure XX), its lower, so it wins the comparison test and is concluded to have better performance than B.
Chapter 1 introduced several of the landmark studies in the field of LCA. Notable amongst these were those associated with the "paper versus plastic" debate of the 1990s. These debates raged in terms of trying to promote paper or plastic as the material of choice for items such as cups and shopping bags. As summarized then, the general answer back then, and for many years since, in relation to a question of "which is better for the environment, paper or plastic?" has been a resounding "it depends", i.e., the results have been inconclusive. Similar "it depends" conclusions have resulted for comparisons of cloth and disposable diapers, internal combustion and hybrid-electric engine vehicles, as well as petroleum-based or bio-based fuels. While the particular reasons these comparisons failed to make a specific conclusion "depend" on many things, in this chapter we begin by more substantively – and quantitatively - discussing how the broad issues of uncertainty and variability (and various methods to appreciate them) raised in earlier chapters affect our studies, in the hopes that we can both better appreciate the causes to make better models, and also to better interpret our results.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
304
Chapter 11: Uncertainty in LCA
Before diving deeper into uncertainty and variability, we refocus on the practical reasons why they are important in LCA. Studies of many different product systems, not just those mentioned above, have led to inconclusive results. In other cases results that should have been deemed inconclusive have been touted as showing a significant difference. In a field that seeks to perform accounting of impacts, inconclusive results can be seen as a failure of a method. Put bluntly, the fact that the answer presented in LCA studies continues to be "it depends" in so many cases has led to observations that the domain of LCA is either unable to answer significant questions, or more seriously, that when faced with significant questions, the methods are not strong enough to help support these important decisions. But even the relatively important question of "paper vs. plastic", while important from a scale perspective given the massive quantities of each material used in the technosphere, pales in comparison to some of the more important and timely questions for which LCA has been sought to offer advice. These questions have been related to biofuels (as in "gasoline versus ethanol"), hybrid-electric vehicles (as compared to internal combustion engine vehicles), and others whose motivations lie beyond merely the environmental issues. These latter examples are policy questions for which society needs answers so as to more effectively decide how to allocate resources to incentivize investments that we believe will have farreaching benefits. At these levels, the importance is far greater than just promoting an ecofriendly drinking cup. These are the "decisions that matter" implied in the title of this book. In reality, it is rarely a failure of the LCA Standard when studies are not able to produce conclusive answers. The underlying failure is more often a lack of substantive and quantitative effort by the study authors to details provided in the data and methods. The main goal of this chapter is to better understand how we can leverage existing practices and methods to comfortably inform hard decisions -- the so-called decisions that matter. We seek more robust methods where we can feel comfortable with our stated conclusions about the performance of a product system, especially as when comparing it to other systems. We seek methods to show more specifically what our conclusions might depend on. Doing so will require that we reopen our introductory mentions of uncertainty and variability from earlier in the book, and also look at the wealth of available data so that our results can be informed by 'all of the data' rather than just data from a single known source that we choose for our study. In the remainder of this chapter, we will simply refer to 'uncertainty' as it pertains to both uncertainty and variability, and will explicitly call out issues only related to variability. Note that in many domains, uncertainties are also referred to as sources of error. (keep this here instead of earlier chapter? Needed here or repetitive? Put it below?) Sadly, in the field of LCA there are many practitioners who actively or passively ignore the effects of uncertainty or variability in their studies. They treat all model inputs as single values and generate only a single result. The prospect of uncertainty or variability is lost in their model, and typically then that means those effects are lost on the reader of the study. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
305
How can we support a big decision (e.g., paper vs. plastic?) if there is much uncertainty in the data but we have completely ignored it? We are likely to end up supporting poor decisions if we do so.
Measurement vs. Accounting One of the challenges in performing the accounting task of an LCI is that the input and output flows for a product system cannot be independently measured in the same way as scientists are accustomed to in other fields. Before discussing the specific underlying issues relevant to uncertainty and variability in LCA, lets consider how scientific methods are used in other domains to produce useful quantitative results to help support analyses and comparisons. (reviewers – skeleton in next para – but struggling to find good example XX – ideas or suggestions?)
Consider the case of XX, where scientific instrumentation is available to measure the XX by using a machine called an X. This machine leverages the known science of X to quantitatively measure X and produces a value with several significant figures and with an uncertainty range given in the machine's technical specifications. When considering the overall amount of X in the product, a standard could be developed for repeatedly testing the X in the X. Each measurement would represent an independent attempt. In the end the test procedure could dictate that the test result is the average of N repeated measurements. If we wanted to compare the X in multiple products, we might use the test procedure and compare the scientifically and quantitatively derived averages to decide which one was the best. In LCA, ideal measurement devices do not exist - in many cases, primary data for key underlying processes do not even exist. Our own internal goals quickly revert to merely producing the best possible study given the realities of data availability.
In LCA, an array of techniques may be used to create data used in databases or modules. Transportation process data may use no directly measured data. For example, the data for a truck transportation process may use similar methods as used in the hypothetical fruit delivery truck in Chapter 6. That example used average load and fuel economy assumptions to derive an estimate of the fuel needed to deliver a certain amount of product over a certain distance. The result was an estimate of diesel use (xx) per ton-km. By applying an emissions factor of approximately 20 pounds CO2 per gallon, an estimate of CO2 emissions (xx) per ton-km was estimated. When looking at the metadata for process data, it is often difficult to tell whether any input or output flows were directly measured versus estimated, but if the sources provided are all reports, then typically estimates were created using methods as described in Chapter 2, as shown above for trucks. For the sake of typical LCI studies, such Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
306
Chapter 11: Uncertainty in LCA
effort is sufficient. However it is not the same as attaching a measuring device in a truck's fuel tank to see the exact amount of diesel used, or measuring emissions rate through the truck's tailpipe. This inevitably means that all LCA data, especially data not produced via measurements, has an appreciable uncertainty factor that should be considered when looking at the results of models. Given that the complexity of product systems in LCA could include tens, hundreds, or more processes, there could be substantial uncertainty associated with the final result. This is a much different modeling outcome than might be typical in environmental studies based on measurement. Repeated text: It may not be clear generally from the metadata provided with a data module what type of measurement was used. Many data modules do not use actual measurement technologies. Many estimate values, and then allocate them on a functional unit basis. For example, the process data module for a delivery truck (see example in Chapter 6) may use no measured data in its reporting of flows. Such a data module could use assumptions on fuel economy and vehicle capacity to determine flows of fuel and emissions per ton-kilometer. It is unlikely that one ton of product was driven one kilometer in the truck while the diesel fuel input and air emission outputs were measured – which is the type of scientific measurement that might be done in different circumstances and may be presumed to be behind such data. Even though well-developed and appropriate scientific measurement technology exists, in these cases they still may not be used. Problems that could arise from using such methods would not be about measurement! They would be related to imperfectly applying assumptions or calculations in the method used to create the data module values. Sidebar about Railroad ads on NPR – "we can ship XX on a gallon of diesel fuel"? Before developing the uncertain nature of LCA further, consider the emerging development of a breakthrough product that rapidly provides measurements relevant to LCA practitioners.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
307
Emerging Development: The Carbon Calorimeter Imagine a scientific device – a "carbon calorimeter" - that can measure the embodied fossil CO2 emissions of a manufactured product to four significant figures. The user opens the lid of this device and puts in a computer, or a smartphone, and after several minutes it returns a precise measure of the fossil CO2 needed to produce all of its subcomponents, raw material and energy inputs, and transportation of all of these components through the global supply chain until received by the customer. Imagine that additional value-added features were that it could quantify specific greenhouse gases emitted (e.g., methane and non-fossil CO2), and could measure emissions of use phase and disposal. This would be an amazing device, and even with some modest measurement error rates, it would render the LCA Standard and all of its uncertainties obsolete. Of course, you probably realize that such a device does not exist, and further, is impossible to create. It is impossible to measure embodied CO2, as there is insufficient residual carbon left in the product to link that to the carbon needed to make it. Likewise it is not possible to know a product's journey through the global supply chain by analyzing just the final product in its current location. Inevitably, the challenge faced in the LCA world is that various stakeholders assume that such a device not only exists (or at least, the underlying science needed to create it), but can be regularly used to inform a range of questions. Some of these stakeholders think LCA is such a device. In the absence of a calorimeter, we instead use LCA to 'measure' values like embodied fossil CO2 in a product. Of course, our methods are far more primitive than what the idealized calorimeter might do. As such, uninformed stakeholders and critics of LCA have unrealistically high expectations on the scientific and quantitative results of our studies– they expect perfect numbers with little or no uncertainty. And every time an LCA is done that does not sufficiently deal with issues of uncertainty and variability and does not report such factors, for example by ignoring uncertainty and providing only point estimate results, a disservice is done to the LCA world and opportunities are lost to educate the various stakeholders about the practical reality of LCA specifically and environmental system modeling in general.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
308
Chapter 11: Uncertainty in LCA
This "lack of true measurement tools" and the thought example above is an important aspect to consider regarding how rigorously to compare results across multiple product systems, and for how an audience may interpret your results. The calorimeter example hopefully inspires reflection on the feasibility of meeting goals of an LCA, as well as highlighting the inevitable limitations of such a Standard. Results comparable to those available from measurement systems are impossible. An appropriate second-best goal is to seek results that are robust enough to overcome known sources of uncertainty and variability. That is also the goal of this chapter. Now that a distinction has been drawn between measurement methods and approaches, and accounting methods, such as LCA, the next section discusses the topics of uncertainty and variability to be addressed in this context.
Types of Uncertainty and Variability Relevant to LCA Prior chapters have in several places already introduced the concepts of uncertainty and variability and also discussed examples where uncertainty and variability issues affect results, whether through issues associated with LCA data or models. These prior discussions were intentionally terse, seeking only to introduce and motivate important connections when discussing other concepts. In this section, more substantive discussion of uncertainty and variability as related to data and modeling are discussed. In Chapter 2, variability was defined as related to diversity and heterogeneity of systems, and uncertainty as resulting from a lack of information or an inability to measure. But having seen examples of data and methods used in LCA, we can more specifically define these terms in the context of LCA. Thoughts on these definitions? Uncertainty in LCA occurs as a result of using data or methods that imperfectly capture the effects in the product system. Variability in LCA occurs as a result of the data or methods used yielding results being unable to produce consistent results. As noted in Chapter 2, uncertainty can generally be reduced by performing additional work or research, while variability cannot be similarly reduced. These statements remain true in the context of LCA, but we should seek ways to either manage or represent uncertainty and variability in our work.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
309
Figure 11-1 shows a general representation of a model, where data are the inputs into methods, and the combination of data and methods produce results. All three of these components can be uncertain or variable, and the forms of uncertainty can be similar across the three. Heijungs and Huijbregts (2004), Williams, et al (2009) and Lloyd and Ries (2007) provide excellent descriptions of the various uncertainties relevant to LCA. From these sources the following generalizations are made in the remainder of this section. Uncertaintyrelated issues are always relevant in LCA studies, but perhaps most important when building comparative models (especially those that will lead to comparative assertions). The discussion begins with uncertainty related to data.
Figure 11-1: General Model Flow Diagram
Data Uncertainty We start with several general definitions that refer largely to data uncertainty. Measurement uncertainty generally refers to the case where a 'ground truth' or perfect measurement is possible using a particular technology, and measurement using an alternative technology will lead to differing degrees of imperfect results. This is analogous to the graduated cylinder example in Chapter 2 – if it were possible to produce a cylinder with more gradual lines on it, we would expect to be able to produce measurements that were less uncertain. In LCA terms, this might more specifically refer to… The problems may be more than just determining the appropriate number of significant figures to report. But in the context of LCA, the flows reported are emissions or releases, or quantities of energy and resource use. Even though it was noted above that LCA data may not be from measured sources, Nonetheless, measurement problems can still have an effect on LCA…
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
310
Chapter 11: Uncertainty in LCA
Add: Williams – measurement versus complex system. Which is an LCI? Connect to "idealized LCI" in Williams et al JIE paper? Parameter uncertainty exists when the parameters used in a model are uncertain. Typically all parameters in a model have some degree of uncertainty, except for physical constants. In LCA… Beyond these two underlying definitional relations about uncertainty, more specific types of uncertainty are relevant in LCA, as summarized in Figure 11-2. Each of these types are discussed in more detail below.
Uncertainty Type
Brief Description
Data
Due to errors or imperfections in model inputs
Cutoff
Due to choices in modeled product system boundaries
Aggregation
Due to similar higher- or lower-level process data being used as a proxy for desired process.
Geographical
Due to variations in where processes occur as related to (potentially uncertain) data available and used to model the processes
Temporal
Due to technological progress not being fully able to be represented in (potentially old) data Figure 11-2: Categorization of Sources of Uncertainty in LCA (Modified from Williams et al 2009)
Just because table says process doesn't mean it only applies to PLCA. Unfortunately, many of these types of errors are inherent in any LCA study, including process or IO-based methods. The remainder of this section describes the other uncertainty types shown in Figure X. More on Sources of Data Uncertainty Specific to LCA Survey Errors: Separate from measurement uncertainty, uncertainties in basic source data can result from sampling and reporting errors. Data sources used in government reports (as inputs into process data modules or IO models) often come from surveys of firms or individual facilities. Surveys are not sent to (and responses are not required of) all relevant firms or facilities in an industry – statistical sampling methods are used. These sampling methods may lead to unrepresentative sets of facilities asked. Survey questions can be misinterpreted, and data requested can be incorrectly provided. Minimizing survey errors Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
311
depends upon the actions of the firms surveyed and the data compilers in the government, so users cannot reduce this uncertainty. Incomplete and missing data: The databases used as sources are limited by their accuracy and completeness of coverage (aside from survey errors). For example, US EPA's Toxic Release Inventory is not collected for some industrial sectors or for plants below a specified size or threshold level of releases. As a result, estimates of toxic emissions tend to be underestimated. The largest problem with missing data comes from data that are not collected. Incomplete and missing data certainly contribute to uncertainty in results, but reducing this uncertainty requires considerable effort.
CW #3) price uncertainty in EIO. sector average price vs. indivual product price. You know I like to harp on this one :)
Cutoff Uncertainty The process model approach is particularly vulnerable to missing data resulting from choices of analytical boundaries, sometimes called the truncation error. Lenzen (2000) reports that truncation errors on the boundaries will vary with the type of product or process considered, but can be on the order of 50%. To make process-oriented LCA possible at all, truncation of some sort is essential. CW: 2) link with goal and scope: Can to some extent define away uncertainty by narrowing goal and scope. Role of PCRs, particularly related to cutoff error--if goal and scope is set to PCR and PCR defines allocation schemes, you're basically left with parameter uncertainty
Aggregation Uncertainty Aggregation errors arise primarily in IO-based models and occur because of heterogeneous producers and products within any one input–output sector. Aggregation: Even the nearly 500 sectors do not give us detailed information on particular products or processes. For example, we might like data on lead-acid or nickel-metal hydride batteries, but have to be content with a single rechargeable battery sector. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
312
Chapter 11: Uncertainty in LCA
For example, the battery sector contains tiny hearing aid batteries as well as massive ones used to protect against electricity blackouts in telephone exchanges. $1 million spent on hearing aid batteries will use quite different materials and create different environmental discharges than $1 million spent on huge batteries. But the EIO-LCA model assumes that the products within a sector are uniform. Some sectors group a large number of products together, such as all industrial inorganic and organic chemicals in one sector. Again, the environmental impact of producing different types of chemicals may vary. The best means to compensate for these types of uncertainty is to use detailed information about particular products. Formally, this can be accomplished with the hybrid approach, as described in Chapter 2. Less formally, the environmental impacts or input requirements calculated by the EIO-LCA model may be adjusted to reflect actual information about specific products.
Geographic Uncertainty
Imports: EIO-LCA represents typical U.S. production within each sector, even though some products and services might be produced outside the United States and imported. Leather shoes imported from China were probably produced with different processes, chemicals, and environmental discharges than leather shoes produced in the United States. Lacking data on the production of goods in other countries, the Department of Commerce assumes that the production is the same as in the United States. The magnitude of this problem is diminished by the fact that many imports are actually produced by processes comparable to those in the United States and with similar environmental discharge standards. As a general note, the uncertainty associated with imports is larger as the fraction of imported goods increases within the economy generally or within a particular sector. International trade as a percentage of Gross Domestic Production ranges from nearly 200% (Singapore) to 2% (Russia) (Economist 1990). The U.S. ratio is 12%. Smaller countries tend to have higher ratios. Sectors that have very little domestic production deserve particular attention since the EIO-LCA tables may be inaccurate.
Temporal Issues Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
313
(move NREL LCI graph from chapter 5 in here instead?) Old data: The latest economic input–output table developed by the Department of Commerce may be up to seven years old. Similarly, there is a lag in receiving environmental data. The lag matters little for some data, such as the economic input–output coefficients for existing industries, but will be more important for other data, such as air emissions from vehicles. Fortunately, modern information technology is speeding up the process of compiling input–output tables so the lag in data is getting shorter, particularly for annual models, which are based upon aggregate updates to the benchmark models estimated every five years. It takes considerable time to assemble the data required for national input–output tables and the various environmental impacts. During this time, many elements of the economy may change. There may be technological change in some sectors, as new techniques of production are introduced; replacing human labor with robotics is a typical example. There may be changes in the demand for certain sectors, resulting in capacity constraints and changes in the production mix. New products may be invented and introduced. Relative price changes may occur which lead manufacturers to change their production process. LCA analysts compound the problems of change over time by extrapolating into the future. LCA users are usually most interested in impacts in the future, after the introduction of new designs and products. While the national economy is dynamic, there is considerable consistency over time. For example, an electricity generation plant lasts 30 to 50 years and variations from year to year are small. Input–output coefficients are relatively stable over time. Carter (1970) calculated the total intermediate output in the economy to satisfy 1961 final demand using five different U.S. national input–output tables. The results varied by only 3.8% over the preceding 22 years:
Using 1939 Coefficients: $324, 288 Using 1947 Coefficients: $336,296 Using 1958 Coefficients: $336,941 Actual 1961 Output: $334,160
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
314
Chapter 11: Uncertainty in LCA
Environmental discharges change more rapidly. Table 4-1 shows several impacts for the generation of $1 million of electricity as calculated by EIO-LCA using the 1992 and 1997 and 2002? benchmark models. The economic transactions expected in the supply chain are comparable for the two periods, with only a 3% difference, even though the sector definitions changed over time. The 1997 benchmark separated out corporate headquarters operations into its own economic sector; energy use and greenhouse gas emissions each declined from 1992 to 1997 by about 30%, suggesting that the sector and its supply chain became somewhat more efficient and cleaner over time. However, the generation of hazardous wastes and the emission of toxic materials increased. Both of these effects may be due to changes in reporting requirements rather than changes in the performance of different sectors. In particular, the electric utility industry was not required to report toxic releases in 1992. Users of the EIO-LCA model can use the different model dates to assess these types of changes over time for sectors of interest.
[Table 4-1] – INSERT from EIOLCA Book?
Model or Method Uncertainty CW: 1) model vs. parameter uncertainty. Many existing chapters mostly discuss parameter uncertainty, which is well handled by MC methods, but less on model uncertainty, which requires alternative model structures Beyond issues associated with uncertainty in data, LCA can also be affected by uncertainties related to the methods used. Input-output based models assume proportionality in production (e.g., the effects of producing $1000 from a sector are exactly 10 times more than producing $100 from the same sector), implying that there are no capacity constraints or scale economies. Likewise, the environmental impact vectors use average impacts per dollar of output, even though the incremental or marginal effects of a production change might be different. For example, a new product might be produced in a plant with advanced pollution control equipment. The linear assumptions of the input–output model may be thought of as providing a firstorder, linear approximation to more complicated nonlinear functions. If these underlying nonlinear functions are relatively flat and continuous over the relevant range, then the firstorder approximation is relatively good. If the functions are changing rapidly, are discontinuous, or are used over large changes, then the first order approximations will be relatively poor. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
315
Reducing errors of this kind requires more effort on the part of the LCA practitioner. A simple approach is to alter the parameters of the EIO-LCA model to reflect the user's beliefs about their actual values. Thus, estimates of marginal changes in environmental impact vectors may be substituted for the average values provided in the standard EIO-LCA model. This requires substitution of the average emissions in particular sectors with the marginal emissions. An analytically elegant approach to such updating is available through Bayes' Theorem, in which posterior parameter estimates are obtained by combining the existing EIO-LCA parameters with a user's beliefs about the actual parameter values (Morgan 1990). Unfortunately the formal use of Bayes' Theorem requires information about the distribution of the existing parameter estimates, which is not generally available from the underlying sources (as described in Chapter 5). A second approach is to combine detailed analysis of processes not expected to follow the EIO-LCA model format with EIO-LCA for other inputs. For example, suppose a new product will be assembled in a new manufacturing plant. The expected air emissions from a new manufacturing plant may be used to replace the average emissions for the entire sector. This is more easily accomplished using a hybrid approach in which the manufacturing, use, and disposal of the new product is considered on its own, and the EIO-LCA model is used to estimate the environmental impacts of the inputs to manufacturing, use, and disposal. Relevant methods are described in Chapter 2.
Result Uncertainty The final main area where uncertainty affects LCA studies is through the results generated (the interaction of data and method). Estimates of effects, or results, developed in LCA studies inevitably have considerable uncertainty. Computers know nothing about uncertainty and will print estimates to as many significant figures as desired. We generally limit reports of impacts to two significant digits, and rarely believe that the estimates are this accurate. As one indication of the degree of uncertainty, Lenzen (2000) estimates that the average total relative standard error of input–output coefficients is about 85%. However, because numerous individual errors in input–output coefficients cancel out in the process of calculating economic requirements and environmental impacts, the overall relative standard errors of economic requirements are only about 10–20%. Point to advanced material?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
316
Chapter 11: Uncertainty in LCA
Fortunately, deciding which of two products or processes is more environmentally desirable will be less uncertain than characterizing the impact of any single product. Because two competing alternatives share many characteristics, their associated uncertainty is usefully positively correlated (varying together), so that the impact differences between the two alternatives will be known with greater certainty than either impact separately. Suppose one product uses slightly less electricity than another. There is considerable uncertainty in the environmental impact of producing the electricity for either product. However, since the electricity use is less for the more energy efficient product, we can more confidently predict that the environmental impact due to electricity use is better as well. As a numerical example, Cano-Ruiz (2000) compared the environmental impact of chloralkali processes using a mercury cell, a membrane cell, and a diaphragm cell. If correlations among errors were ignored, then the difference in estimated impacts for the three methods was not statistically significant. However, if the positive correlations were considered, two alternatives were still similar in performance, but the mercury cell alternatives had only an 8% chance of being better than the others. As LCA study results are typically the most featured component of a study, with prominent graphics and references in the summary, much of the focus on managing uncertainty in LCA in this chapter is centered around the effects of uncertainty on study results, and how to manage them. Transition.. Now that the types of uncertainty have been categorized, qualitative and quantitative methods to manage them are discussed.
Methods to Address Uncertainty and Variability As introduced at the beginning of the chapter, our primary reason for caring about the implications of uncertainty is because we need to be aware of whether the uncertainties affect our results, and more specifically, the conclusions written as interpretation of the results. Two prominent motivations for LCA studies are 1) studies that seek to identify 'hot spots' of a single product system in support of improvements, and 2) studies that week to compare alternate technologies, processes, or approaches. These two prominent types of studies are also critically connected to consideration of uncertainty. We would want to ensure our interpretation of hot spots is focused on the appropriate components, or that we are able to confidently assess which of multiple systems is expected to have the lowest impact. If uncertainty is substantial, a particular process could be wrongly tagged a hot spot (or not tagged when it should be). Likewise, in comparative studies we are concerned about the robustness of the model or the level of confidence possible for a conclusion of whether A can be said to be better than B in the face of the uncertainty. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
317
There are a variety of alternative approaches, ranging from qualitative to quantitative, that can be used in studies.
Qualitative Methods As defined in Chapter 2, qualitative methods are those that qualify, rather than quantify, results. They do not generally use numerical values as part of the uncertainty analysis, but instead focus on discussion and text-based representations of uncertainty.
Discussion of Sources of Uncertainty Specific to a Study An initial example of a qualitative assessment of uncertainty in an LCA study would be to describe in words the expected effect of the various kinds of uncertainty (e.g., those in Figure X). This could include separate discussions pertaining to data, geographical, and other uncertainties. A specific description might look like Figure X, which builds on the concept of data quality indicators from Chapter X. Figure here showing DQI example.. While such a description cannot give specific quantitative support to uncertainty assessment, it is useful in ensuring that the reader is aware that the study was done with knowledge of the stated uncertainties (as opposed to being ignorant of them). As mentioned earlier (where?), any study should discuss data quality issues, including aspects of uncertainty and variability of data.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
318
Chapter 11: Uncertainty in LCA
Figure 11-3: Comparison of LCI Effects for Two Options
Semi-quantitative methods As listed here, semi-quantitative methods are those that use numerical values in support of uncertainty assessment, but do not incorporate the quantified values into the LCA modeling.
Pedigree matrix – A development over time relevant .. don't do it! Why? Arbitrary, etc. Pedigree matrix approach values?
Significance Heuristics Note to readers of this draft – the next few paragraphs are verbatim copies of text appearing in earlier chapters. I am still deciding where to put this material, and at what level of detail. Any suggestions or comments will be appreciated. First set here are copied from Chapter 2: Thus you will see many studies create internally consistent rules that define "significance" in the context of comparing alternatives. These rules of thumb are rooted in the types of significance testing done for statistical analyses, but which are generally not usable given the small number of data points used in such studies. Often used rules will suggest that the uncertainty of values such as energy and carbon emissions are at least 20%, with even higher percentages for other metrics. When implemented, that means our values for Alternatives A and B would need to be at least 20% different for one to consider the difference as being meaningful or significant. The comparative results would be "inconclusive" for energy use using such a study's rules of thumb. In the absence of study rules of thumb for significance, what would we recommend? Returning to our discussion above an LCA practitioner should seek to minimize the use of significant digits. We generally recommend reporting no more than 3 digits (and, ideally, only 2 given the potential for a 20% consideration of uncertainty). In the example of the previous paragraph that would mean comparing two alternatives with identical energy use – i.e., 7.6 kWh. The comparison would thus have the appropriate outcome – that the alternatives are equivalent. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
319
Copied from Chapter 5 While on the subject of assessing comparative differences, it is becoming common for practitioners in LCA to use a "25% rule" when testing for significant differences. The 25% rule means that the difference between two LCI results, such as for two competing products, must be more than 25% different for the results to be deemed significantly different, and thus for one to be declared as lower than the other. While there is not a large quantitative framework behind the choice of 25% specifically, this heuristic is common because it roughly expresses the fact that all data used in such studies is inherently uncertain, and by forcing 25% differences, then relatively small differences would be deemed too small to be noted in study conclusions. Helpful to use EPA Warm project like classification? Signs or orders of magnitude different, values >20%?
Examples from studies – is this pasted from somewhere? Now that we have come so far in our discussions about LCA, hopefully this simple less than test is able to constructively and effectively demonstrate how short-sighted is such a comparison criteria. Our first suggestion for an improved metric might be to require that any pronouncement of improved performance must first pass through a simple difference threshold. Many LCA consulting firms, who perform LCA work under contract for sponsors, now have a series of such thresholds pre-set and documented so that comparative results must prove to be greater than this threshold so as to be considered sufficiently different to merit a distinction of one being lower than the other. For example, Figure X shows a hypothetical summary of threshold differences documented in a study.. meaning that the study requirements are such that the comparative emissions between A and B must be more than 20% different for the study to conclude that one is better than the other. Any cases where the difference is less than 20% causes an in-conclusive result, which translates into "the uncertainty or variability in the underlying model or data is too great so as to prevent a clear conclusion on which is better". Armed with such a list going into a study is a useful practice, and is in line with the goals we have in this Chapter of making sure that we are sufficiently aware of the uncertainties in our work when preparing our results. Conceptually.. We would read such a graph differently, i.e., if we were to put +- 20% error bars on each of our results shown above, then we would require that the top of one error bar does not touch the bottom of another bar – even with uncertainties considered, we would feel comfortable than one was better than the other. Where can we get summaries of the threshold values to use? Where do they come from? Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
320
Chapter 11: Uncertainty in LCA
Sources? This minimum difference in values idea is becoming mainstream..
Alternative Model Approaches Another useful approach to assessing uncertainty in LCA studies is to compare alternative approaches and assumptions. For example, in estimating the impact of nanotechnology on automobile catalysts (Chapter 8), we compare EIO-LCA results to those available from a commercial, process-based LCA tool, GaBi. The results are fairly close for each scenario analyzed, increasing our confidence in the conclusion. But it is still not sufficiently robust for being the quantitative support we need or want when trying to use LCA for our goal – supporting big decisions (or decisions that matter) While the qualitative approaches above can help convey considerations of uncertainty, quantitative methods are needed to fully represent the uncertainty in studies.
Quantitative Methods to Address Uncertainty and Variability In each of the quantitative approaches discussed in this section, attention is paid to developing visual aids that serve to express the quantified uncertainty in inputs, model, or results, so that the audience is better able to quickly appreciate it. Keep eyes on qualitative goal – how robust is model – does answer change, etc?
Figure 11-4: General Flow for Model with Parameter Ranges
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
321
Ranges
Figure 11-4 shows the general flow for a model with ranges, which is otherwise similar to Figure 11-1 except that the inputs and results are expressed with ranges (here shown with a box-whisker plot like representation). When introduced in Chapter 2, ranges were suggested as a simple way of representing multiple estimates or sources instead of reporting only a single value. Ranges can also be used in the LCA context to express values from different sources or data modules. In cases where multiple data points or value exist, ranges can be useful. They can pertain to just inputs or inputs and outputs of LCA models.
Figure 11-5: Comparison of Results with Ranges
xx
Process Flow Diagram-based Example with Ranges
Pasted From chap 5: As noted in Chapter 2, ideally you would identify multiple data sources (i.e., multiple LCI data modules) for a given task. This is especially useful when using secondary data because you are not collecting data from your own controlled processes. Since the data is secondary, it is likely that there are slight differences in assumptions or boundaries than what you would have used if collecting primary data. By using multiple Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
322
Chapter 11: Uncertainty in LCA
sources, and finding averages and/or standard deviations, you could build a more robust quantitative model of the LCI results. We will discuss such uncertainty analysis for inventories in Chapter 10. Considering again the initial process flow diagram example (as shown in Figure 5-5), the study boundary included mining coal, shipping it by rail, and burning it at a coal-fired power plant. That initial example provided only single sources from the NREL US LCI database, and since the GWP characterization factors for greenhouse gases had not been discussed, methane emissions were excluded. As a first demonstration of using ranges in LCA, Figure XX.. shows CO2e values from the NREL US LCI database including only fossil CO2 (as in Figure XX) as well as including previously excluded methane emissions using the IPCC 2007 method (with a 100-year CO2e characterization factor of 25). The columns show.. explain LCI factor use (see Equation 5-X).. Process (functional unit)
LCI factor
Fossil CO2e / funct unit
Fossil CO2 and Methane
Coal-fired electricity generation (kWh)
1 kWh / kWh
0.994
0.994+8.31e-06*25 = 0.994
Coal mining (kg)
0.442 kg coal / kWh
0
0.00399*25 = 0.1
Rail transport (ton-km)
0.461 ton-km / kWh
0.0189
0.0189 + 9.05e07*25 = 0.0189
Total
1.003 Figure 11-6: INSERT TITLE HERE
1.04
Thus if our study scope were broadened to alternately include the effect of methane emissions, our model would estimate CO2e emissions using the IPCC 2007 method of (1.003-1.04) kg CO2 / kWh. This is not a very dramatic example of how ranges could be used, both because the difference is only 4% and because we have just expanded the scope to include more CO2e emissions, but demonstrates .. (results could be A more complex example could consider multiple data sources for the three main processes, as summarized in Figure 11-7. The benefit of using ranges in this way is that the (relatively basic) process flow diagram model can be used to structure .. The individual values shown are real, but not attributed directly to specific studies. Do this instead with Pauli's original NG/LNG paper, etc?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
Process (CO2-e per functional unit)
Fossil CO2 and Methane
Flow factor
0.994+8.31e-06*25 = 0.994
1 kWh / kWh
Coal mining (kg /
0.00399*25 = 0.1
0.442 kg coal / kWh
Rail transport (kg / ton-km)
0.0189 + 9.05e-07*25 = 0.0189
0.461 ton-km / kWh
Coal-fired electricity generation (kg / kWh)
Total
323
Range
1.04 Figure 11-7: INSERT TITLE HERE
Process Matrix-based Example with Ranges
Despite demonstrating how to use ranges, the examples above have been limited by using only process flow diagrams as opposed to process matrix approaches, which were shown to be more comprehensive in Chapter 9.
IO-LCA-based Example with Ranges
HERE.. Considering more advanced petrurbations in either process or IO-based matrix approaches, see Advanced Material for Chapter 11, Section 1.
Visualizing Ranges The results of the range-based assessments above were tabular. The appropriate graphical representation of ranges uses error bars. Chris H's comments: -best reporting means (sig figures, bars, etc.)
Sensitivity Analysis
As mentioned earlier in the book, sensitivity analysis is a means of assessing the effect on model outputs (results) from a percentage change in a single input variable. Chapter X Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
324
Chapter 11: Uncertainty in LCA
noted that sensitivity analysis is explicitly called for in the LCA Standard as a means of considering uncertainty. Beyond LCA models, sensitivity analysis is generally used (and automated by software) to automatically check the sensitivity of model outputs on all modeled input variables. Software such as DecisionTools Suite has Microsoft Excel plug-ins to automate sensitivity analysis in spreadsheet-based models. The key feature of these .. Sens anal is one input at a time. Example? Building on which one from earlier in book – something where we can look up multiple electricity grid factors?
Beyond just input parameters, sensitivity analysis can also be done for different assumptions. These assumptions could relate to choices amongst alternative allocation or system expansion schemes, assumptions about the use or availability of renewable electricity in a process, etc. If the sensitivity of all inputs or assumptions can not be assessed in an automated fashion, then various key parameters should be selected from the study, and tested for the quantitative effect.
Look at variability of results for some common processes (in US LCI, look at electricity) Electricity, eastern, at grid 0.2 kg / MJ Texas – similar Western – 0.15 Similar example for a transport process – train, barge? Do uncertainty analysis of using AR5 uncertainty ranges to see if GWP values change?
Figure of such values? Grid emissions factors/ tehnology emissions factors / etc Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
325
Probabilistic Methods and Simulation
Finally, in this section, we discuss the most complex approach to managing uncertainty. In this section, we use data from various sources to generate probability distributions for inputs, and then use spreadsheets or other techniques to track the effect of these probabilities through the model to generate results that also have probability distributions.
If we are trying to convince a stakeholder that the results of our LCA are "good enough to support their decision" then the final possible step discussed relates to considering the available data by creating probabilistic distributions of data instead of point estimates or ranges. By doing this, we explicitly aim to create models where the output could support a probabilistic assessment of the magnitude of a hot spot, or the percentage likelihood that Product A has less impact than Product B. Relatively Basic Type 1 vs type 2 error, etc. Aranya's comments: for case 2) 1. When the uncertainty is really large, and spans across an order of magnitude like in the case of Kim's biofuel work. Uncertainty becomes important while comparing LCA results to other baselines, and you need to get a good understanding of what drives the underlying uncertainties and variabilities, and whether this can be reduced either through technical improvements or policy decisions. 2. When the uncertainty isn't as significant, but the difference between life cycle results and a baseline is much smaller - like in the case of our fossil fuel uncertainty results. The difference between CNG and gasoline life cycle emissions for transport is small on average, ~5%, but this isn't a robust enough result since the difference is not statistically significant. So, it isn't a good idea Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
326
Chapter 11: Uncertainty in LCA
to make any decisions based on those comparisions.
The most commonly used approach is a Monte Carlo simulation to include key uncertainties in inputs parameters to the LCA model. I think this method can be extended to decision-support frameworks as well. For example, if you used an optimization model and one of your objectives or constraints was an environmental life cycle metric, you could still incorporate these uncertainties to identify robust 'optimal' solutions wherever possible. i.e. solutions that work across all or most of the scenarios that characterize uncertainty in the LCA 'space'.
Figure 11-8: General Diagram for Model with Probabilistic Inputs and Results
To numerically estimate the uncertainty of environmental impacts, Monte Carlo simulation is usually employed. Several steps are involved:
The underlying distributions, correlations, and distribution parameters are estimated for each input–output coefficient, environmental impact vector, and required sector output. Correlations refer to the interaction of the uncertainty for the various coefficients. Random draws are made for each of the coefficients in the EIO-LCA model. The environmental impacts are calculated based on the random draws. Steps 2 and 3 are repeated numerous times. Each repetition represents another observation of a realized environmental impact. Eventually, the distribution of environmental impacts can be reasonably characterized. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
327
Steps 2–4 in this process rely heavily upon modern information technology, as they involve considerable computation. Still, the computations are easily accomplished on modern personal computers. Appendix IV illustrates some simulations of this type on a small model for interested readers. Step 1 is the difficulty in applying this method. We simply do not have good information about distributions, correlations, and the associated parameters for input–output matrices or environmental impact vectors. Over time, one hopes that research results will accumulate in this area. A simpler approach to the analysis of uncertainty in comparing two alternatives is to conduct a sensitivity analysis; that is, to consider what level some parameter would have to attain in order to change the preference for one alternative. Mullins curves / -probabilities of drawing incorrect conclusions about signs (pos or neg) or comparisons.
Troy: Finally, I would say the field seems to be moving toward more detailed models based on stock datasets built up over time. These will reduce some sources of variability within LCA results and facilitate the calculation of end point metrics practitioners tend not to use very often at present, ie DALYs.
Add text on meta-analyses (eg JIE special issue?)
Still deciding - Move stochastic text from Chapter 3 back here too? Pretty big jump for chap 3 as is..
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
328
Chapter 11: Uncertainty in LCA
Unfortunately, the software tools for LCA do not make it easy to practice uncertainty assessment for students. The course license version of SimaPro does not include uncertainty analysis features (although a "student research" version of the program does). OpenLCA does include uncertainty assessment tools. The Advanced Material Section X shows..
Copied from Chapter 3..
Deterministic and Probabilistic LCCA Our examples so far, as well as many LCCAs (and LCAs, as we will see later) are deterministic. That means they are based on single, fixed values of assumptions and parameters but more importantly it suggests that there is no chance of risk or uncertainty that the result might be different. Of course it is very rare that there would be any big decision we might want to make that lacks risk or uncertainty. Probabilistic or stochastic models are built based on some expected uncertainty, variability, or chance. Let us first consider a hypothetical example of a deterministic LCCA as done in DOT (2002). The example considers two project alternatives (A and B) over a 35-year timeline. Included in the timeline are cost estimates for the life cycle stages of initial construction, rehabilitation, and end of use. An important difference between the two alternatives is that Alternative B has more work zones, which have a shorter duration but that cause inconvenience for users, leading to higher user costs as valued by their productive time lost. Following the five-step method outlined above, DOT showed these values: Without discounting, we could scan the data and see that Alternative A has fewer periods of disruption and fairly compact project costs in three time periods. Alternative B's cost structure (for both agency and user costs) is distributed across the analysis period of 35 years. Given the time value of money, however, it is not obvious which might be preferred. At a 4% rate, the discounting factors using Equation 3-1 for years 12, 20, 28, and 35 are 0.6246, 0.4564, 0.3335, and 0.2534, respectively. Thus for Alternative A the discounted life cycle agency costs would be $31.9 million and user costs would be $22.8 million. For Alternative B they would be $28.3 million and $30.0 million, respectively. As DOT (2002) noted in their analysis, "Alternative A has the lowest combined agency and user costs, whereas Alternative B has the lowest initial construction and total agency costs. Based on this information alone, the decision-maker could lean toward either Alternative A (based on overall cost) or Alternative B (due to its lower initial and total agency costs). However, more analysis might prove beneficial. For instance, Alternative B might be revised to see if user Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
329
costs could be reduced through improved traffic management during construction and rehabilitation." For big decisions like that in the DOT example, one would want to consider the ranges of uncertainty possible to ensure against a poor decision. Building on DOT's recommendation, we could consider various values of users' time, the lengths of time of work zone closures, etc. If we had ranges of plausible values instead of simple deterministic values, that too could be useful. Construction costs and work zone closure times, for example, are rarely much below estimates (due to contracting issues) but in large projects have the potential to go significantly higher. Thus, an asymmetric range of input values may be relevant for a model. We could also use probability distributions to represent the various cost and other assumptions in our models. By doing this, and using tools like Monte Carlo simulation, we could create output distributions of expected life cycle cost for use in LCCA studies. The use of such methods to aid in uncertainty assessment are discussed in the Advanced Material at the end of this chapter.
Chapter Summary The practice of Life Cycle Assessment inevitably involves considerable uncertainty. This uncertainty can be reduced with careful analysis of the underlying data and production processes. Fortunately, the relative desirability of design alternatives can be assessed with greater confidence than the overall impact of a single alternative because of the typical positive correlation between the impacts of alternatives. In general, we recommend that users of the EIO-LCA model use no more than two significant digits in considering results from the model.
As first introduced in Chapter 4, life cycle impact assessment (LCIA) is the final quantitative phase of LCA. LCIA allows us to transform the basic inventory flows created from the inventory phase of the LCA and to attempt to draw conclusions related to the expected impacts of these flows in our product systems. While climate change and cumulative energy demand tend to dominate LCA studies, various other impact categories of broad interest have had characterization models created that are scientifically credible and available for use. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
330
Chapter 11: Uncertainty in LCA
Despite the availability of these credible models and tools, many LCA studies continue to focus just on generating inventory results, or at most, use only climate and energy impact models. Now that we have reviewed all of the important phases of LCA, in the next few chapters we focus on ways in which we can create robust analyses that will serve our intended goals of building quantitatively sound and rigorous methods.
References for this Chapter Williams, Hawkins, et al (xxx) and Lloyd (xxxx) Morgan and Henrion Homework Questions for Chapter 11 here
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
331
Advanced Material for Chapter 11 – Section 1 Uncertainty in Leontief Input–Output Equations: Some Numerical Examples (need to fix vector/matrix notation throughout..) In this Advanced Material section, issues associated with propagation of uncertainty through matrix-based methods is demonstrated via perturbing values in an IO transactions matrix. Such perturbations would be useful if considering the effect of structured uncertainty ranges in a product system to assess whether changes have significant 'ripple through' effects. In general, this methods shows that making small changes in these matrices leads to varying levels of effects. (More trans here?) Tables of direct requirements and total requirements are often reported to six significant figures. It is said that these tables allow one to calculate to a single dollar the effect on a sector of a one million dollar demand. It should go without saying that we do not believe that we have information that permits us to do this kind of arithmetic. Little advice is given to the novice EIO analyst about how many significant figures merit attention. We have advised our colleagues and students to be careful when going beyond two significant figures, and in this book we restrict virtually all of our impact estimates to two significant digits. This paper explores some ways to investigate systematically the uncertainty in Leontief input– output analysis. We can judge the results of using the input–output estimates by considering the effects of errors or uncertainties in the requirements matrix on the solution of the Leontief system. Here, we will generally assume that we know the elements of the final demand vector Y without error. Uncertainty in the values of the elements of the total output vector X will result from uncertainty in the elements of the requirements matrix A. Uncertainties in X will result from the propagation of errors through the nonlinear operation of calculating the Leontief inverse matrix. The Hawkins-Simon conditions require that all elements in the A matrix are positive and less than one, and at least one column sum in the A matrix must be less than 1. The determinant of the A matrix must be greater than zero for a solution to exist. For empirical requirements matrices, such as are reported by the Bureau of Economic Services (BES) for the United States, values of the determinants of A are very small (on the order of 10–12). Values of the determinant of the associated Leontief matrices are greater than 1. We will use the U.S. 1998 BES 9×9 tables for the direct requirements matrix A and the calculated Leontief inverse matrix to illustrate the meaning of the sector elements (see Table IV-1). The two tables are given below. Of the 81 terms in A, five are zero. The numbers in Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
332
Chapter 11: Uncertainty in LCA
any column of the A matrix represent the amount, or direct requirement, that is purchased in cents from the sector-named rows to produce one dollar of output from the sector-named column. For example, for a one dollar output from the construction sector, 29.8 cents of input is required from the manufacturing sector. The column sum for the construction sector is 53.8 cents for all nine input sectors; this says that 1.00 – 0.538 = 46.2 cents is the value added by the construction sector for one dollar of output. The values in the [I-A]-1 matrix are the total requirements or the sum of the direct and the indirect requirements. Hence, from [I-A]-1 we see that a one dollar direct demand from the construction sector requires a total of 52.4 cents to be purchased from the manufacturing sector to cover both the direct and indirect requirements. The sum of the A values in the construction column shows that one dollar demand for construction results in $2.09 of economic transactions for the whole economy. [Table IV-1] - INSERT
Deterministic Changes in the [I–A] and [I-A]-1 matrices Sherman and Morrison (1950) provide a simple algebraic method for calculating the adjustment of the inverse matrix corresponding to a change in one element in a given matrix. Their method shows how the change in a single element in A will result in the change for all the elements of the Leontief inverse, and that there is a limit to changing an element in [A] that requires that the change does not lead to A becoming singular. We use the 1998 9×9 A and Leontief inverse matrices to demonstrate the numerical effect of changing the value in A on the Leontief inverse matrix. We do not use the Sherman-Morrison equation for our calculation, but instead use the functional term for calculating the inverse matrix in the spreadsheet program Excel.
Example 1. The effect of changing one element in the A matrix on the elements in the Leontief matrix. If element A4,3 is increased by 25%, we want to know the magnitude of the changes in the Leontief inverse matrix. The original A4,3 = 0.298 in the 1998 table and the increased value is A4,3 = 1.25×0.298 = 0.3725. All other elements in A are unchanged. All calculated elements in the new Leontief matrix show some change, but the amount of change varies. For an increase of 25% in A4,3, the inverse element [I – A]-14,3 increases by nearly the same amount. The manufacturing column and the construction row elements all change by the same percentage. In the Leontief matrix system, the sum of the column is called the backward linkage and the sum of the row is called the forward linkage. As a result of the 25% increase Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
333
in the direct requirement of the manufacturing sector on the construction sector, each backward element in the construction sector increases by 0.17%. There is a similar increase in each element in the forward linkage of the manufacturing sector. Other sectors change by different amounts. In percentage terms, the changes for the new Leontief matrix compared to the original 1998 matrix are shown in Table IV-2. [Table IV-2]- INSERT Example 2. The effect of a small change in a single cell of A on the Leontief matrix. The value of A4,3 is rounded from three decimal places to two, from 0.298 to 0.30. The relative change in the new Leontief matrix is small for all cells, and no cell has a positive change (see Table IV-3). The largest change is in [I-A]-14,3 of –0.6%.
[Table IV-3]- INSERT
Example 3. The effect of rounding all cells of A from three decimal places to two. The changes in the new Leontief matrix are both positive and negative, and are larger than for the single cell rounding change illustrated in the previous example (see Table IV-4). [I– A]–14,3 changes by –1.7% when all cells in A are rounded down. Rounding all the cells in A to two decimal places results in large changes in many cells of the Leontief matrix. The largest negative change is –71.1% in [I–A]–17,1, and the largest positive change is 54.1% in [I– A]–19,2.
[Table IV-4]- INSERT
Modeling Changes in the [I–A] and Leontief matrices with Probabilistic Methods
The literature dealing with uncertainty in the Leontief equations is not extensive. The earliest work that we have found that deals with probabilistic errors was from the PhD thesis of R.E. Quandt (1956). Quandt's (1958, 1959) analysis of probabilistic errors in the Leontief system was limited by the computing facilities of the late 1950s. His numerical experiments Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
334
Chapter 11: Uncertainty in LCA
were confined to small (3×3) matrices. He developed equations to calculate expected values for means and variances of the Leontief matrix based on estimates of these parameters for the [A] matrix. Quandt investigated changes in the Leontief equations by examining them in this form:
[I – A – E]–1 y = x
(IV-3)
Quandt specified conditions on his errors E that each element [Aij + Eij] > 0 and that column j, sum of all elements [Aij + Eij], must be less than 1, that is, the uncertain A elements satisfied the Hawkins-Simon conditions. His work examined eleven discrete distributions for E, (eight were centered at the origin and two were skewed about the origin.) The probabilities of these errors were also modeled discretely with choices of uniform, symmetric, and asymmetric distributions. For each distribution Quandt selected a sample of 100 3×3 matrices. Each sample set represented about 0.5% of the total population of 39 = 19,683 matrices. From this set of experiments he calculated the variance and the third and fourth moments of the error distributions and the resulting vector x. Quandt used a constant demand vector y for all his experiments. The mean values of x were little changed from the deterministic values and the variance of A had little effect on the mean values of x. Quandt concluded the following: 1. the skewness of the errors in A are transmitted to the skewness of the errors in the x vector. 2. The lognormal distribution provides a fairly adequate description of the distribution of the x vector elements irrespective of the distribution of the errors in A. 3. One can use the approximate lognormal distribution to establish confidence limits for the elements in the solution x.
West (1986) has performed a stochastic analysis of the Leontief equations with the assumption that elements in A could be represented by continuous normal distributions. He presents equations for calculating the moments of the elements in A. West's work is critically examined by Raa and Steel (1994) who point out shortcomings in his choice of Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
335
normality for the A elements—mainly that elements in A cannot be less than zero—and suggest using a beta distribution limited to the interval for A elements between 0 and 1 to keep elements in A positive.
Some Numerical Experiments with Stochastic Input A + E matrices The examples presented in this section are constructed using Microsoft Excel spreadsheets and @Risk software. They illustrate the ease by which we can study the effects of changes in the form of elements of A on the Leontief matrix and some of the multipliers and linkages commonly used in Leontief analysis. Numerical simulation is easy, and results from more than a thousand iterations are obtained quickly. Still, the critical issues are the formulation of good questions and the interpretation of the results of numerical experiments. As we have pointed out before, the lack of a detailed empirical database to support our assumptions about the statistical properties of the elements in the direct requirements A matrix is the most important limitation of this analysis. For each of our numerical experiments, we compare the properties of the input A + E direct requirements matrix and the output [I – A – E]-1 total requirements matrix, where E is an introduced perturbation. The results of stochastic simulations for the [I – A –E] inverses are compared to the deterministic calculation of [I – A] inverse. We report some representative results for four scenarios. 1) In each scenario, the means of A are the 1998 values reported to three decimal places. 2) Four types of input distributions are examined: a uniform distribution and two triangular distributions with both positive and negative skewness, and a symmetric triangular distribution. An Excel spreadsheet is constructed with the 1998 9×9 matrix A, and used to calculate the Leontief inverse. @Risk uses the Excel spreadsheet as a basis for defining input and output cells. The chosen probabilistic distribution functions can be selected from a menu in @Risk; the number of iterations can be set, or one may let the program automatically choose the number of iterations to reach closure. We expect changes from the results of the deterministic calculation of the Leontief matrix from A; Simonovits (1975) showed that
Exp (I – A)–1 > (I – Exp (A))–1
(IV-4)
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
336
Chapter 11: Uncertainty in LCA
where Exp is the expected value operator. We use the software to numerically simulate the values of the Leontief matrix given an assumed distribution for A. Table IV-5 shows the results of our simulations for the manufacturing:construction element in A, namely A 4,3. Each of the 81 values in A is iterated over l000 times for each simulation. Here only the distribution of values for one cell, the manufacturing:construction intersection, is reported.
[Table IV-5]- INSERT
The numerical simulations show that the mean value for [I-A]-14,3 for the symmetric uniform distribution is identical to the deterministic value for this sector pair of manufacturing:construction. The mean values for [I-A]-14,3 for the two skewed distributions are both lower than the deterministic value for [I-A]-1 4,3, and so is the mean value for [A 4,3] for the symmetrical triangular distribution. The coefficient of variation (COV: the ratio of the standard deviation to the mean) of [I-A]-14,3 is smaller than the COV for [A 4,3] for all simulations except for the uniform input distribution. Consistent with Quandt's conjecture, the skewness for every [I-A]-1 4,3 increases except for the uniform distribution input. Additional work remains to show the patterns for the entire distribution of cells in [A].
Energy Analysis with Stochastic Leontief Equations The following table is representative of the 9×9 U.S. economy with nearly $15 trillion of total transactions and the total value added, or GDP, of more than $8 trillion. We have included a column called the percentage value added for this economy. If we think in terms of $100 million of value added, or final demand, for the U.S. economy, we can also think of this demand in disaggregated sector demands of nearly $18 million for manufactured products, more than $5 million of construction, etc. For this energy analysis, we show three columns of energy data in mixed units, energy per $ million of total sector output. Hence, the manufactured products sector uses nearly 3.5 TJ of total energy per $ million of output, 0.24 million kWh of electricity per $ million of output, and 0.44 TJ of coal per $ million of output.
[Table IV-6] Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
337
The total energy use and the direct energy use for each sector are given by:
r = [R diagonal] x = [r diagonal] [I – A]–1 y
(IV-5)
and r direct = [R diagonal] [I + A] y
(IV-6)
where r is a vector of energy use by sector, [R diagonal] is a matrix with diagonal cells equal to the energy use per dollar of sector output and off-diagonal terms equal to zero, and A is the requirements matrix.
Example 1. For this example, we use Excel and @Risk to build a model to calculate the uncertainty in the physical units both the direct and the total energy use for $ 100 million of final demand for the U.S. economy. The demand distributed among the nine sectors proportionally to the distributions of value added for the economy. We assume we know the demand vector with certainty, and that we know the physical energy use with certainty. All uncertainty for this example is in A. Assume that the entries in A may be represented by a symmetric triangular distribution with a low limit of zero, a mode equal to the three decimal place value reported by BES, and a high limit of two times the mode. The mean of this distribution is equal to the mode. This is equivalent to saying that the coefficient of variation is constant for all entries with a value of 0.41. Previously, we presented the results of simulations for this triangular distribution on the Leontief matrix In this example, we examine the distribution of the r direct and the total r. @Risk performed 5000 iterations to calculate the mean value and the standard deviation of the energy use for each of the nine sectors for a $100 million increment of GDP proportionally distributed across the economy. The uncertainty in the energy output for each sector is shown by the standard deviation and the COV. The sum of the total energy use for the whole economy is 730 TJ and the direct energy use is 563 TJ. The sector values for r direct are smaller than the total r values, and the r values have more uncertainty than the r direct values as shown by the COVs. The COVs for all sectors are smaller than the constant COV of 0.41 assumed for A. Direct energy use is lowest as a percentage of the total energy Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
338
Chapter 11: Uncertainty in LCA
use for the agricultural products and minerals sectors. For all other sectors, the direct energy use is more than 70% of the total energy use. [Table IV-7]
Summary Uncertain values in the cells of the requirements matrix generate uncertain values in the cells of the total requirements or Leontief matrix. Three cases have studied. Two deterministic cases are presented; in one case only a single value in A is modified, and in the second case all the values in A are changed. For a set of probabilistic examples we used Excel and @Risk to calculate the Leontief matrix for a uniform and three triangular distributions of A as input. The simulations show small effects on the mean values of the Leontief matrix and larger changes in the second and third moments. An example of an energy analysis for a 9×9 sector model of the U.S. economy shows the effect of uncertainty from A on the total energy use r and the direct energy use r direct for each sector.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 11: Uncertainty in LCA
Advanced Material – how does SimaPro do Uncertainty?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
339
340
Chapter 12: Advanced Hotspot and Path Analysis
EXTRA PAGE FOR FORMATTING
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
341
Chapter 12 : Advanced Hybrid Hotspot and Path Analysis In this chapter, we describe advanced LCA methods that consider all potential paths of a modeled system as separate entities, instead of summarizing aggregated results. For process matrix-based methods, these methods are often referred to as network analyses, and for IObased methods, as structural path analyses. These methods are helpful in support of screening level analyses, as well as in helping to identify specific processes and sources of interest. In addition, we discuss hybrid methods that allow us to consider changes to the parameters used in the default network or structural path analysis in order to estimate the effects of alternative designs, use of processes, or other assumptions.
Learning Objectives for the Chapter At the end of this chapter, you should be able to: 1. Explain the limitations of aggregated LCA results in terms of creating detailed assessments of product systems. 2. Express and describe interdependent systems as a hierarchical tree with nodes and paths through a tree. 3. Describe how structural path analysis methods provide disaggregated LCA estimates of nodes, paths, and trees. 4. Interpret the results of a structural path analysis to support improved LCA decision making. 5. Explain and apply the path-exchange method to update SPA results for a scenario of interest, and describe how the results can support changes in design or procurement.
Results of Aggregated LCA Methods While the analytical methods described in the previous chapters are useful in terms of providing results in LCA studies, these results have generally been aggregated. By aggregated results, we mean those that have been 'rolled up' in a way so as to ignore additional detail that may be available within the system. In process-based methods, aggregated results express totals across all processes of the same name that are modeled within the system boundary (or within the boundary of the process matrix). For example, Figure 9-5 (revisited here as Figure 12-1) showed aggregated fossil Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
342
Chapter 12: Advanced Hotspot and Path Analysis
CO2 emissions to air across the entire inverted US LCI process matrix system by process in the production of bituminous coal-fired electricity. Process Total Electricity, bituminous coal, at power plant/US
Emissions (kg) 1.033
Percent of Total
1.004
97.2%
Diesel, combusted in industrial boiler/US
0.011
1.0%
Transport, train, diesel powered/US
0.009
0.9%
Electricity, natural gas, at power plant/US
0.002
0.2%
Residual fuel oil, combusted in industrial boiler/US
0.002
0.2%
Transport, barge, residual fuel oil powered/US
0.001
0.1%
Natural gas, combusted in industrial boiler/US
0.001
0.1%
Gasoline, combusted in equipment/US
0.001
0.1%
Electricity, lignite coal, at power plant/US
0.001
0.1%
Transport, ocean freighter, residual fuel oil powered/US
0.001
0.1%
Bituminous coal, combusted in industrial boiler/US
0.001
0.0%
Figure 12-1: Top products contributing to emissions of fossil CO2 for 1kWh of bituminous coal-fired electricity. Those representing more than 1% are bolded.
The total emissions across all processes are 1.033 kg per kWh of electricity, with the top processes contributing 1.004 kg CO2 to that value from producing coal-fired electricity and 0.011 kg CO2 from producing diesel fuel that is combusted in an industrial boiler. These two values generally include all use of coal-fired electricity and diesel in boilers though the system, not just direct use. Similarly, in IO-based methods, the aggregated results expressed as the output of using an IO model with the Leontief equation provides totals across all of the economic sectors within the IO model. Figure 8-5 (revisited here as Figure 12-5) showed total and sectoral energy use for producing $100,000 in the Paint and coatings sector. The total CO2 emissions across the supply chain are 107 tons, with the top sectoral sources being electricity (25 tons) and dyes and pigments (17 tons).
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
343
Total ($ thousand)
CO2 equivalents (tons)
Total across all 428 sectors
266
107
Paint and coatings
100
3
Materials and resins
13
5
Organic chemicals
12
5
Wholesale Trade
10
1
Management of companies
10
Soft drink and ice manufacturing'. This nomenclature will also be used in this chapter.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
353
Figure 12-9: Hierarchical Tree View of Structural Path Analysis for Soft Drink and Ice Manufacturing
The PowerPoint template used to make Figure 12-9 is in the Chapter 12 folder. Despite providing significant detail, SPA results expressed in tabular form can be difficult to interpret. Figure 12-9 shows a truncated graphical representation of most of the SPA results shown in Figure 12-8 (several of the values come from the underlying SPA, not shown). The values in the blue rectangles represent the sector names of the activities and the GHG emissions at the site, similar to Figure 12-3(b), but streamlined to only include a subset of them at each tier. The numerical values above the rectangles are the GHG LCI values from Figure 12-8. For example, at the very top of the hierarchy is the rectangle representing the initial (Tier 0) effects of Soft drink and ice manufacturing, which Figure 12-8 says has a value of 36 tons CO2e, and the LCI emissions, rounded off to 940 tons. Likewise, the Tier 1 site emissions of Wet corn milling are 52 tons, and the LCI of 119 tons CO2e. As promised, SPA, unlike aggregate methods, shows a far richer view of where flows occur in the product system. By using SPA, we could improve our understanding of our product, or the design of our own LCA study. For example, in the soft drink example above, we could ensure that key processes such as wet corn milling and other high emissions nodes are within our system boundary. If such nodes were excluded, we would be ignoring significant sources of emissions.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
354
Chapter 12: Advanced Hotspot and Path Analysis
Web-based Tool for SPA Visual structural path analyses similar to those shown above can be generated online via the EIO-LCA SPA tool (accessed at http://www.eiolca.net/abhoopat/componentviz/ ). The SPA tool has four elements, which display results of the 2002 benchmark model: 1) A search or browse interface to find the input-output sector you want to analyze 2) A pull down menu for selecting the effect you want to analyze (e.g., energy, carbon emissions, water use, etc.) 3) A sector hierarchy with categories of products you want to browse amongst to choose a sector to analyze 4) A Structured Path Analysis graphic displaying the results (shown after the three above are chosen) A key component of the structural path visualization is the ability to 'cut off' the many small paths in the supply chain (e.g., all paths with impacts less than 0.1%) in order to more effectively and efficiently focus on visualizing the results of the larger paths for decision making. The SPA tool allows the user to select any of the 428 IO sectors in the 2002 model, and then visualize the hierarchy or chain of effects that lead to significant impacts. Figure 12-10 shows the initial screen displayed when using the SPA tool to select a sector.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
355
Figure 12-10: Home Screen of the Online SPA Tool
Once a sector is chosen in the online interface, either by searching (i.e., by starting to type it in words and then selecting from a set of auto-completed options) or browsing (i.e., using the categorical drill down + symbols), the SPA display begins. Figure 12-11 shows the initial SPA screen for energy use (by default) associated with Electronic computer manufacturing18. The user can click the 'Change Metric' button to instead display SPA results for greenhouse gas emissions or other indicators of interest.
Figure 12-11: Initial SPA Interface Screen for Electronic Computer Manufacturing (Showing Energy, by default)
The tool also provides in-line help when the cursor is moved over screen elements. For example if the metric is changed to 'Greenhouse Gases', and the cursor is hovered over any of the elements in the top row (the top 5 sources), Figure 12-12 shows how the tool summarizes why those values are shown, i.e., they are the top sectors across the supply chain of producing computers that emit greenhouse gases.
18
Note: For consistency, a revision of this web tool example will be updated to show the same soft drink example described above.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
356
Chapter 12: Advanced Hotspot and Path Analysis
Figure 12-12: SPA Interface for Greenhouse Gas Emissions, with Help Tip Displayed
Likewise, Figure 12-13 shows how moving the cursor over elements in the first row of the structural pathway display will spell out the acronyms of sectors chosen (and which are abbreviated to fit on the display). It also explains the concept of depth introduced above which is relevant to how deep in the supply chain the path is being displayed. At the first level (or depth) of the visualization, all of the top-level activities that go into final assembly of a computer are represented as boxes in the large horizontal rectangle of activities. On the left hand side of this top level is always the sector chosen (in this case, computers), and then sorted to the right of that choice are the top sectors that result in emissions associated with the highest level of the supply chain. These include computer storage devices, semiconductors (shown below), etc.
Figure 12-13: SPA Interface Showing Detail of Elements in First Tier of Display
Each of the boxes in the lower (pathway) portion of the SPA tool shows the respective percentage of effects from that sector in the overall path. Of all of the highest level processes in the path of a computer, 17% of the emissions come from computer storage device manufacturing (CSDM), 16% comes from semiconductors (S&RDM), 13% from printed circuit assembly (PCAAM), etc. But each of those activities themselves also has an upstream supply chain.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
357
The red bar at the bottom of each of those grey process boxes is denoting that it has a further upstream process that may contribute effects in the overall SPA. By clicking on any of these boxes in the top level, the visualization drills down to all of the activities associated with that specific pathway. Figure 12-14 shows the SPA visual that would result from choosing computers, and then subsequently choosing the semiconductor manufacturing process at the first level.
Figure 12-14: SPA Interface with Second Tier Detail Shown
In this case, the largest emitting activity in the upstream supply chain of semiconductors is power generation, showing 20% of the relevant upstream emissions from that process. All other basic inorganic chemicals (AOBICM) would be next highest at 14%. Selecting the power generation box at this level would again drill down to the next level of the SPA, resulting in the display in Figure 12-15. At the third path level, the result is that almost all of the emissions come from the generation of electric power itself, with a few smaller upstream processes like coal mines, etc., to the right.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
358
Chapter 12: Advanced Hotspot and Path Analysis
Figure 12-15: SPA Interface with Third Tier Detail Shown
Finally, the SPA visual can connect the top impacts with the results shown in the SPA display at the bottom. By moving the mouse over any of the top 5 sectors on the top of the screen, the SPA will highlight in the same color all of the processes in your selected structural path that are associated with that sector. In Figure 12-16, all power generation boxes are shown. The point of this feature is to visually reinforce the importance of these top 5 sectors.
Figure 12-16: SPA Interface with top sources highlighted as nodes in levels
Note that the example drill down of the SPA shown above is just one of thousands of combinations that could be done for any particular top-level sector. For example, instead of choosing semiconductors in the first row, the SPA could have elected to follow computer storage devices, etc. The resulting visualizations would be different and are not shown here. The discussion and demonstration above hopefully provide further motivation as to why one might be interested in path specific results in LCA. In the next section, methods are presented that incorporate additional data to update the structural path results available from IO models. These methods represent the most detailed hybrid methods available to support detailed questions at the level of specific nodes in the supply chain.
The Structural Path Exchange Method While the results of an SPA may be inherently tied to sectoral average data and methods, the path and node-specific information may be useful when considering the effects of alternative design or procurement decisions on the relative impacts of product systems. For example, Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
359
we may use generic IO-based SPA results to develop a baseline for a product system, and replace various baseline results with our own data or assumptions as they relate to alternative designs or purchases. At the simplest level, path exchange (PXC) is an advanced hybrid LCA method conceived by Treloar and summarized theoretically in Lenzen (2009) that 'exchanges' path values from a baseline SPA, e.g., from a national level model, with data related to alternate processes, i.e., that differ from those modeled by average and aggregate IO data. The alternate data may be from primary data, supplier information, assumptions, or locally-available data on specific paths. The values exchanged may be only for a specific node, an entire subtree of the node, or more. The main purpose of path exchange is to create an improved estimate of effects across a network / supply chain, with the alternate exchanged data exploiting the comprehensiveness of the SPA. The PXC method targets specific nodes in the supply chain. Baboulet (2010) provides an excellent practical demonstration of path exchange in support of decision making and policy setting for a university seeking to reduce its carbon footprint. The steps of path exchange can thus be summarized as: (1) Perform a general national IO-based SPA for a sector and effect of interest to develop a baseline estimate. (2) Identify paths where alternate process data would be used (e.g., paths with relatively high values in the baseline, or where process values are significantly different than averages), and where data is available to replace the baseline path values in the SPA. For each of these exchanged paths, do the following steps: •
Develop a quantitative connection between the alternative process data and the nature of the relationship of the chosen paths, including potential unit change differences (e.g., mass to dollars).
•
Normalize available process data to replace information in the default path.
•
Update the path value.
(3) Re-calculate the SPA results with path exchanges and compare the new results to the baseline SPA. As a motivating example, consider trying to estimate the carbon footprint of a formulation of soda where renewable electricity has been purchased in key places of the supply chain (instead of using national average grid electricity everywhere). You would (1) run a baseline SPA on the soda manufacturing sector, (2) look for nodes in the SPA where electricity is used and has large impact, and use alternate data on renewables, derive alternate path values, Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
360
Chapter 12: Advanced Hotspot and Path Analysis
and (3) change the path values and recalculate and compare to the baseline SPA to see the overall effect of the green power. Example 12-1 shows a brief example to inspire how PXC works before we dive into more details and scenarios.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
361
Example 12-1: Use the path exchange method to estimate the GHG emissions reductions of using renewable electricity on site for soft drink and ice manufacturing (SDM) in the US. Answer: Consider that an estimate is needed for the total GHG emissions of a physical amount of soda that, when converted to dollars, is $1 million (maybe this is one month's worth of physical production from a facility). Further, the facility making the soda buys wind power. The results from Figure 12-8 can be used as the baseline since they were generated for $1 million of soda manufacturing. Figure 12-17 shows an abridged version of Figure 12-8 with the top three paths, and excludes several unused columns, sorted by site CO2e emissions. Recall that the path in the third row (path length 0) shows that the site emissions of the soda manufacturing facility are 36 tons CO2e and the total LCI (for the whole supply chain below it including the site) are 940 tons CO2e. Row 1 shows that the path for electricity directly used by the soda factory (path length 1), represents 79.6 tons of CO2e (83.6 tons considering the whole supply chain below this node). Baseline SPA Results LCI @ Site Path Description (tons) 83.6 Power generation and supply > SDM
1
GWP Site (tons) 79.6
1
52.1
119.4
Wet corn milling > SDM
0
36.0
940.4
SDM
Length
Total
940.4
Path Exchanges GWP Site (tons) 0
LCI @ Site (tons) 4.0
Reasons
Share Exch'd
Green Power
-100% site
860.8
Figure 12-17: Abridged PXC Worksheet (Top Three Site Values only) for $1 million of Soda Manufacturing
The right hand side of Figure 12-17 shows a worksheet for annotating path exchanges. If we assume that the green power purchased by the soda factory has no direct greenhouse gas emissions, we note that 100% of site emissions would be reduced, and record a path-exchanged value of 0 for the site CO2e emissions. The upstream (LCI) portion of the renewable energy system may or may not have 0 emissions. The existing upstream LCI value of 4 tons CO2e is for average US electricity, involving a weighted average for the upstream emissions of extracting and delivering fuel to power plants, equipment manufacture, etc. If we did not have specific information on the generation type and/or upstream CO 2e value for our green power, we could choose to maintain the 4 tons CO2e LCI value from the baseline SPA. Of course, if we did have specific information we could justify an alternate value, like an assumption for 0 in the LCI category as well. If we made no other changes to the baseline SPA results, our path-exchanged total system would be 860.8 (940.4 – 79.6) tons CO2e emissions – the extra significant figures shown to ensure the math is clear. This is a fairly significant effect for only one path exchange – an 8% reduction from the baseline SPA results.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
362
Chapter 12: Advanced Hotspot and Path Analysis
The basic path exchange in Example 12-1 also lays the foundation for the general PXC method. PXC does not manipulate the underlying A or R matrices of the IO model used for the SPA, and thus does not make broad and consistent changes to the entire economic system. Following Example 12-1, if PXC changed the R matrix for GHG emissions of electricity (in this case, made them 0), then all purchases of electricity by all facilities in the entire economy would be exchanged. Such a change of course would overestimate the effect of the decision by a single facility. Instead, PXC adjusts specific uses of A or R matrix values for nodes of a specific path (e.g., those used in Equation 12-2). This is a benefit of SPA and path exchanges – we can target very specific places in the network. Equation 12-2 helps to motivate that there are only two general kinds of exchanges – to the transaction coefficients (e.g., Aij) or intensity coefficients (Ri) underlying the SPA results used to generate the site and LCI values for specific paths. Transaction coefficient-based exchanges are those rooted in a change in the level of purchases made. If we remember what the values in an IO Use Table look like that eventually become elements of an A matrix, then we can consider that the 'production recipe' for a particular path can be presented as a value in units like cents per dollar. In the drill-down generated by an SPA, we might be able to assess that the economic value of a particular node is 10 cents per dollar. If we make a decision to change consumption of this particular node in a future design or decision, then we would edit the underlying 10 cents per dollar value to something more or less than 10. Buying 50% less would change this transaction coefficient to 5 units. On the other hand, changing intensity coefficients is done to represent different decisions or opportunities where the degree of effect is different. The waste example at the beginning of the chapter had intensities of 50 and 5 grams per $billion. Again, a path exchange could increase or decrease these values. Finally, an exchange can involve both transaction and intensity changes. Regardless of the type of exchange, and depending on the depth of the path you are trying to exchange, you may need to perform significant conversions so that you can determine the appropriate coefficients to use in the exchange. This could take the form of estimation problems (see Chapter 2), dealing with several physical to monetary (or vice versa) unit conversions, or other issues. In the end, what you will be exchanging is the path value from the baseline to the exchanged value (e.g., from 79.6 to 0 in Figure 12-17). You may be able to determine the appropriate exchanged path value without describing all of the transaction or intensity conversions (Example 12-1 exemplifies this in showing the exchange to 0). Building on the prior examples in the chapter about soda manufacturing, Example 12-2 shows how to use SPA to consider the effects of reducing the amount of corn syrup used in soda, in support of a more natural product. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
363
Example 12-2: Use the path exchange method to estimate the GHG emissions reductions of using 50% less corn syrup on site for $1 million of soft drink and ice manufacturing (SDM) in the US. Answer: Corn syrup (e.g., high fructose corn syrup) is one of the primary ingredients of soda and is the product of wet corn milling processes. The results from Figure 12-8 can again be used as the SPA baseline. Figure 12-18 shows our PXC worksheet that includes separate columns to track transaction or intensity coefficient changes. If we assume that the second row of the table represents all of our direct purchases of corn syrup, then the values we choose to exchange will fully represent the effect. Behind the scenes, this would be equivalent to reducing our Aij cell value by 50%. A 50% reduction would reduce the GHG site and LCI values by 50%. Of course this would have the equivalent effect as finding an alternative corn syrup supplier with 50% less site and LCI emissions. Baseline SPA Results
1
GWP Site (tons) 79.6
1
52.1
119.4
Wet corn milling > SDM
0
36.0
940.4
SDM
Length
Total
LCI @ Site Path Description (tons) 83.6 Power generation and supply > SDM
940.4
GWP Site (tons)
LCI @ Site (tons)
26.0
59.7
Path Exchanges Trans Reasons Share Exch'd Reduce syrup
Intensity Share Exch'd
-50% site, LCI
880.7
Figure 12-18: Abridged PXC Worksheet (Top Three Site Values Only) for $1 million of Soda Manufacturing
If we made no other changes to the baseline SPA results, our path-exchanged total system would be 880.7 (940.4 – 59.7) tons CO2e emissions – the extra significant figures shown to ensure the math is clear. This is a fairly significant effect for only one path exchange – a 6% reduction from the baseline SPA results. The same result occurs if the syrup is purchased by a supplier able to make the same amount of syrup with 50% lower emissions. The exchange would instead be entered in the intensity share exchange column.
Example 12-2 shows a comparable GHG reduction as shifting our soda factory to 100% green power. An important part of a decision of pursuing one or the other alternative would be the relative costs (not included here). This further demonstrates why SPA and PXC are such powerful tools – the ability to do these kinds of 'what if' analyses to compare alternative strategies to reduce impact. When performing PXC, it is important to be careful in tracking the site and LCI values. While the examples above show both site and LCI values for the abridged baseline SPA for soda, recall that the full SPA has about 1,100 paths, including separate row entries for nodes upstream of some of the nodes with large LCI values. Tracking and managing effects in upstream nodes may be more difficult than these examples imply. In Examples 12-1 and 12Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
364
Chapter 12: Advanced Hotspot and Path Analysis
2, this was done by showing the resulting exchanged values for site and LCI in the same row. It may be easier to track site and LCI changes separately. For example, Figure 12-8 shows the top 15 paths of the soda SPA. Row 2 shows the path 'Wet corn milling > SDM', and Row 6 shows the path 'Grain farming > wet corn milling > SDM'. The LCI value for row 2 is 119 tons, while the site value for row 6 (which falls under the tree of the node in row 2) is 30 tons. Row 6 represents a significant share of row 2's LCI value. When Example 12-2 reduced the purchase of syrup by 50%, we also reduced the LCI value by the same amount, which makes sense given the transactional nature of the choice. However, there may be other path exchanges where we want to independently adjust these connected site and LCI values (i.e., separately edit site values in the PXC worksheet for rows 2 and 6). Example 123 shows how to represent multiple, offsetting exchanges via PXC. Example 12-3: Use the path exchange method to estimate the GHG emissions reductions of shifting 50% of direct truck delivery of soda to rail. Answer: Results from Figure 12-8 can again be used as the SPA baseline. Figure 12-19 shows our PXC worksheet for path length 0 (the entire LCI of system), and the paths of length 1 for truck and rail transportation (the latter not previously shown in Figure 12-8 but available in the supplemental resources). To reduce 50% of direct deliveries by truck, we exchange a value in Row 2 of the worksheet. This 50% transactional reduction would reduce the site and LCI values, or about 10.9 tons total. The offsetting increase in rail may not be simple, as the baseline amount of soda shipped by rail is not given, and the underlying physical units are not known (e.g., tons/$). Physical or monetary unit factors for the two transport modes are needed to adjust the rail value. If we assume that truck and rail emit 0.17 kg and 0.1 kg of CO2 per ton-mile, respectively, then the original 15 tons of CO2 from delivery by truck (row 2) equates to (15,000 kg / 0.17 kg CO2 per ton-mile), or 88,200 ton-miles. A 50% diversion is 44,100 tonmiles, which at 0.1 kg CO2 per ton-mile of rail emits 4.4 more tons CO2 than what is already shipped by rail in the baseline. Relative to the SPA site baseline of 2.5 tons (row 3), this is a factor of 2.76 (176%) increase, which we could apply to both the site and LCI values for rail. Baseline SPA Results LCI @ Site Path Description (tons) 940.4 SDM
Path Exchanges
0
GWP Site (tons) 36.0
1
15.0
21.7
Truck Transportation > SDM
7.5
10.9
1
2.5
3.2
Rail Transportation > SDM
6.9
8.8
Length
Total
940.4
GWP Site (tons)
LCI @ Site (tons)
Reasons
Divert truck to rail
Trans Share Exch'd
Intensity Share Exch'd
-50% site, LCI +176% site, LCI
935.1
Figure 12-19: Abridged PXC Worksheet for $1 million of Soda Manufacturing
If we made no other exchanges, our path-exchanged total system would have 935.1 tons CO2e emissions – a fairly insignificant effect for what is likely a large amount of logistical planning. From an economic perspective, though, it is likely much cheaper. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
365
Software and code exists to help with PXC activities, for example the University of Sydney's BottomLine. These provide detailed interfaces showing path summaries, transaction and intensity coefficients, etc., to be edited for path exchanges. Without such software, PXC must be done with exchange worksheets (potentially done in Microsoft Excel) as in Figure 1217. Exchanges will sometimes involve more than one path. A substitution in a design may involve a reduction of transaction or intensity from one path and an increase in another. For example, if our company elected not to use direct truck transportation for soda, we could not reasonably deliver our product, and could not capture the full effects of such an exchange by zeroing out truck transportation. We would need to increase the use of some other mode of transportation (e.g., rail, which was not shown in Figure 12-8). While this chapter has focused on IO-based structural path analysis, network analysis of process matrix models is analogous. The same matrix definitions and techniques are used, and the main inputs needed are the raw matrices used for the process model. See the Advanced Material for additional help on network analysis of process matrices.
Chapter Summary Structural Path Analysis (SPA) is a rigorous quantitative method that provides a way to disaggregate IO-LCA results to provide insights that are otherwise not possible. These disaggregated results can be very useful in terms of helping to set our study design parameters to ensure a high quality result. Path exchange is a hybrid method that allows replacement of results from specific paths in an SPA based on available monetary or physical data. These advanced hot spot analysis methods provide significant power, but remain critically dependent on our data sources.
References for this Chapter Baboulet, O., and Lenzen, M. Evaluating the environmental performance of a university, Journal of Cleaner Production, Volume 18, Issue 12, August 2010, Pages 1134–1141. DOI: http://dx.doi.org/10.1016/j.jclepro.2010.04.006 Crama, Y.; Defourny, J.; Gazon, J. Structural decomposition of multipliers in input-output or social accounting matrix analysis. Econ. Appl. 1984, 37 (1), 215–222. Defourny, J.; Thorbecke, E. Structural path analysis and multiplier decomposition within a social accounting matrix framework. Econ. J. 1984, 94 (373), 111–136. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
366
Chapter 12: Advanced Hotspot and Path Analysis
Lenzen, M. and R. Crawford, The Path Exchange Method for Hybrid LCA, Environ. Sci. Technol. 2009, 43, 8251–8256. Peters, G. P. & Hertwich, E. G. Structural analysis of international trade: Environmental impacts of Norway. Economic Systems Research 18, 155-181, (2006). Treloar, G. Extracting embodied energy paths from input-output tables: towards an inputoutput-based hybrid energy analysis method. Economic Systems Research, 1997, 9 (4), 375–391.
Homework Questions for Chapter 12 For questions 1-4, use the Microsoft Excel file 'SPA_Automobiles_1million_GHG.xls' posted in the Homework files folder for Chapter 12 to answer the questions. This file shows the results of a baseline SPA for $1 million of Automobile manufacturing in the 2002 EIO-LCA producer model with respect to total GHG emissions (units of tons CO2e). 1. Draw a hierarchical tree (either by hand or by modifying the posted PowerPoint template for soft drinks) for the top 10 paths that is similar in layout to Figure 12-9. 2. Find the percent of the total emissions in the system specified by the path analysis, and describe in words what the path analysis results tell you about the GHG hot spots in the supply chain for automobiles. 3. Use the path exchange method to estimate the net CO2e effects of each of the following adjustments to the 2002 baseline SPA. Without doing a cost analysis, discuss the relative feasibility of each of the alternatives. a. Use renewable electricity at automobile assembly factory b. Use renewable electricity at all factories producing motor vehicle parts c. Reduce use of carbon-based fuels by 50% at the automobile assembly factor (assume all site GHG emissions are from use of fuels) d. As done in Chapter 3, consider substituting aluminum (top path 17) for steel (top path 1) in 50% of all motor vehicle parts. Assume $17,000 of steel and $1,000 of aluminum per $million in parts currently, and that prices are $450 per ton of steel and $1,000 per ton of aluminum. Aluminum can substitute for steel at 80% rate. 4. Discuss the limitations of using SPA and path exchanges to model the life cycle of an automobile. Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
367
Advanced Material for Chapter 12 – Section 1 - MATLAB Code for SPA (Put the theoretical stuff – Figs 12-1 and 12-2 here instead?) The MATLAB code used to generate structural path analysis (SPA) results throughout Chapter 12 is available in the Web-based resources for Chapter 12 on the lcatextbook.com website (SPAEIO.zip). The core code was originally developed by Glen Peters and is provided with his permission. Use of alternative SPA tools or code could lead to different path analysis results than those presented in the chapter. To use the code, unzip the file into a local directory. The specific .m file in SPAEIO.zip that is used to generate the results for the waste example in Chapter 12 is called RunSPAChap12Waste.m, and uses the code below to generate the values for Figure 12-5 and Figure 12-6: clear all F = [1 1]; % for econ SPA paths ('1' values just return L matrix) %F = [50 5]; % for waste paths A = [0.15 0.25; 0.2 0.05]; filename = 'chap12example'; % code to make default sector names if needed (comment out if not) [rows, cols] = size(A); sectornames=cell(rows,1); for i=1:rows sectornames{i}=['Sector' num2str(i)]; end L = inv([eye(2)-A]); F_total = F*L; y = zeros(2,1); y(1,1) = 100;
% The $100 billion of final demand
percent = 0.01; T_max = 4;
% 'cut-off' of upstream LCI (as % of total emissions) % Max tiers to search
% this prints the T_max, percent, etc. params in the file % change to 0 or comment it out if not needed thresh_banner=1; % this last command runs a function in another .m file in the zip file % parameters of function are the data matrices and threshold parameters SPAEIO02(F, A, y, F_total, T_max, percent, filename, sectornames, thresh_banner); Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
368
Chapter 12: Advanced Hotspot and Path Analysis
The rest of the .m files in the ZIP folder provided build the hierarchical tree, sort it, traverse it across the various paths, and return results, printing only those that meet the threshold criteria (e.g., if T_max=4, it will not output any paths at or below that tier). The other .m files should not generally need to be modified19. To use the code, you use or edit the matrices and parameters in RunSPAChap12Waste.m, and then run it in MATLAB. It will generate a CSV text file (named chap12example here) with the economic path results below, where the intermediate calculations are summarized in Figure 12-5. You may want to check the math for some of the paths below and ensure you see which of the nodes they correspond to. Paths = 15, T_max = 4, percent 1: 0:100.0000:151.8152 2: 1:20.0000:29.0429 : 3: 1:15.0000:22.7723 : 4: 2:5.0000:7.5908 : 1 5: 2:3.0000:4.3564 : 1 6: 2:2.2500:3.4158 : 1 7: 3:1.0000:1.4521 : 1 2 ; Sector2 8: 2:1.0000:1.4521 : 1 9: 3:0.7500:1.1386 : 1 1 ; Sector1 10: 3:0.7500:1.1386 : 1 1 ; Sector1 11: 3:0.4500:0.6535 : 1 2 ; Sector2 12: 3:0.3375:0.5124 : 1 1 ; Sector1 13: 3:0.2500:0.3795 : 1 1 ; Sector1 14: 3:0.1500:0.2178 : 1 2 ; Sector2 15: 3:0.0500:0.0726 : 1 2 ; Sector2
= : 1 1 ; ; ; ;
0.01000, Total Effects = 1.518152e+02 1 ; Sector1 ; Sector1 : 2 ; Sector2 ; Sector1 : 1 ; Sector1 Sector1 : 2 ; Sector2 : 1 ; Sector1 Sector1 : 1 ; Sector1 : 2 ; Sector2 Sector1 : 1 ; Sector1 : 1 ; Sector1 Sector1 : 2 ; Sector2 : 1 ; Sector1 :
; Sector1 : 2 ; Sector2 : 2 ; Sector2 ; Sector1 : 2 ; Sector2 : 1 ; Sector1 : ; Sector1 : 1 ; Sector1 : 2 ; Sector2 : ; Sector1 : 1 ; Sector1 : 1 ; Sector1 : ; Sector1 : 1 ; Sector1 : 1 ; Sector1 : ; Sector1 : 2 ; Sector2 : 2 ; Sector2 : ; Sector1 : 1 ; Sector1 : 2 ; Sector2 : ; Sector1 : 2 ; Sector2 : 2 ; Sector2 :
The format of this output is as follows: the first row displays all of the threshold parameters, total paths given the thresholds, and the total effects - in this case, economic results in billions. The rows below this row show results for each path (sorted by the site effect value): the path number (here 1-15), the path length, the site and LCI effects (here, $billions), then the ordered path, e.g., path #1 is the top level purchases from sector 1 of the final demand, and path #15 is the purchases in the path Sector 2 > Sector 2 > Sector 2 > Sector 1). The Print_sorted_EIO2.m optionally displays the threshold criteria in the output file, printing of the sector names, and number of significant digits to display. These could all be edited if desired. 19
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
369
extraneous digits in the site and LCI values come from the SPA code (which, by default, are 4 post-decimal digits). The CSV text file results generated by MATLAB can be imported into Microsoft Excel by opening the text file in Excel and using the import wizard with colon and semicolons provided as delimeters (a colon needs to be typed into the 'other' field). The economic SPA results represent 99% of all economic effects throughout the supply chain in only 15 paths. Since the variable percent is 0.01, the SPA code searches for paths up to T_max where the LCI values are greater than 0.01% of $151.8 billion, or $0.0152 billion. If there were a 16th path that had been identified, it was ignored because its LCI value was less than that amount (but path 15 was not ignored). If you comment out the second line of code in RunSPAChap12Waste.m (F = [1 1];) and un-comment the third line (F = [50 5];) and re-run the .m file, it will instead return the waste path results (as summarized in Figure 12-6): Paths = 15, T_max = 4, percent = 0.01000, Total Effects = 6.402640e+03 1: 0:5000.0000:6402.6403 : 1 ; Sector1 2: 1:750.0000:960.3960 : 1 ; Sector1 : 1 ; Sector1 3: 2:250.0000:320.1320 : 1 ; Sector1 : 2 ; Sector2 : 1 ; Sector1 4: 1:100.0000:442.2442 : 1 ; Sector1 : 2 ; Sector2 5: 2:112.5000:144.0594 : 1 ; Sector1 : 1 ; Sector1 : 1 ; Sector1 6: 2:15.0000:66.3366 : 1 ; Sector1 : 1 ; Sector1 : 2 ; Sector2 7: 3:37.5000:48.0198 : 1 ; Sector1 : 1 ; Sector1 : 2 ; Sector2 : 1 ; Sector1 8: 3:37.5000:48.0198 : 1 ; Sector1 : 2 ; Sector2 : 1 ; Sector1 : 1 ; Sector1 9: 3:16.8750:21.6089 : 1 ; Sector1 : 1 ; Sector1 : 1 ; Sector1 : 1 ; Sector1 10: 3:5.0000:22.1122 : 1 ; Sector1 : 2 ; Sector2 : 1 ; Sector1 : 2 ; Sector2 11: 2:5.0000:22.1122 : 1 ; Sector1 : 2 ; Sector2 : 2 ; Sector2 12: 3:12.5000:16.0066 : 1 ; Sector1 : 2 ; Sector2 : 2 ; Sector2 : 1 ; Sector1 13: 3:2.2500:9.9505 : 1 ; Sector1 : 1 ; Sector1 : 1 ; Sector1 : 2 ; Sector2 14: 3:0.7500:3.3168 : 1 ; Sector1 : 1 ; Sector1 : 2 ; Sector2 : 2 ; Sector2 15: 3:0.2500:1.1056 : 1 ; Sector1 : 2 ; Sector2 : 2 ; Sector2 : 2 ; Sector2
Note that the path numbers in the economic and waste path results are different, as they are sorted based on total economic and waste effects, respectively. Path #1 in both happens to Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
370
Chapter 12: Advanced Hotspot and Path Analysis
be the path of length 0. But Paths #2-15 do not refer to the same paths. Economic path #2 (Sector 2 > Sector 1) corresponds to waste path #4. Connecting back to the chapter discussion on the coefficients to be changed in the path exchange method, the economic and waste path results above provide the node values, which are products of the underlying coefficients (Equations 12-1 and 12-2). The path results do not show the various individual coefficients. For example, the node value for economic path #3 is $15 billion (second row of Figure 12-5) and the effect node value for the corresponding waste path #2 is 750 g (second row of Figure 12-6).
Using the SPA Code for Other Models In terms of edits to the code shown in RunSPAChap12Waste.m, the y vector and/or F, A, and L matrix assignments can be modified. For example, to use the SPA code in conjunction with the 2002 US benchmark EIO-LCA model (described in the Advanced Material for Chapter 8 Section 5), the load command can be added to load the .mat file containing the EIO-LCA F, A, and L matrices, and then edit the lines of code defining F, A, and L to point to specific matrices in that model. RunEIO02SPA.m, also included in the SPAEIO.zip file, does a path analysis of sector #70, Soft drink and ice manufacturing in the 2002 EIO-LCA producer model (code different than RunSPAChap12Waste.m is in bold): clear all load('Web-030613/EIO02.mat')
% relative path to 2002 EIOLCA .mat file
F = EIvect(7,:); % matrix of energy & GHG results, 7 is total GHGs A = A02ic; % from the industry by commodity matrix filename = 'softdrinks_1million'; % sector names in the external .mat file sectornames = EIOsecnames; L = L02ic;
% industry by commodity total reqs
F_total = F*L; y = zeros(428,1); y(70,1) = 1;
% sector 70 is soft drink mfg (soda), $1 million
percent = 0.01; T_max = 4;
% 'cut-off' of upstream LCI (% of total effects) % Max tiers to search
% this prints the T_max, percent, etc. params in the file % change to 0 or comment it out if not needed thresh_banner=1; Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
371
% this last command runs two other .m files in the zip file % the parameters on the right hand side are the threshold parameters SPAEIO02(F, A, y, F_total, T_max, percent, filename, sectornames, thresh_banner);
The load command looks for the path to the EIO02.mat file, which in the code above is in a directory within the same directory as the .m file. You would need to edit this to point to where you put it. The next few lines of code set the inputs to various components of the EIO02.mat file. The F vector is set to a column in the matrix EIvect of the EIO02.mat file, which, as stated in Chapter 8, contains all of the energy and GHG multipliers for the R matrix (row 7 is the total GHG emissions factors across 428 columns and is thus transposed), A points to A02ic (the 2002 industry by commodity direct requirements matrix), L points to the already inverted L02ic, and sectornames is set to the vector of sector names (EIOsecnames), and y has a 1 in row 70 and 0's in all other 427 rows. The RunEIO02SPA.m code is run the same way as the RunSPAChap12Waste.m code, and yield the following excerpted results (only first 10 paths shown, used to make Figure 12-8): Paths = 1095, T_max = 4, percent = 0.01000, Total Effects = 9.403932e+02 1: 1:79.6233:83.5978: 70 ; SDM : 31 ; Power generation and supply 2: 1:52.0585:119.4311: 70 ; SDM : 44 ; Wet corn milling 3: 0:36.0017:940.3932: 70 ; SDM 4: 2:32.4451:48.0935: 70 ; SDM : 174 ; Aluminum product manufacturing from purchased aluminum : 173 ; Alumina refining and primary aluminum production 5: 2:32.0096:33.6074: 70 ; SDM : 148 ; Plastics bottle manufacturing : 31 ; Power generation and supply 6: 2:29.7694:42.8620: 70 ; SDM : 44 ; Wet corn milling : 2 ; Grain farming 7: 1:19.9944:166.0256: 70 ; SDM : 174 ; Aluminum product manufacturing from purchased aluminum 8: 1:15.1703:22.8200: 70 ; SDM : 11 ; Milk Production 9: 1:14.9953:21.7487: 70 ; SDM : 324 ; Truck transportation 10: 2:14.7441:15.4801: 70 ; SDM : 174 ; Aluminum product manufacturing from purchased aluminum : 31 ; Power generation and supply
Since the format was discussed above, we note only that the first line of results shows that the total LCI for $1million of soft drink manufacturing is 940.4 tons CO2e (as shown in Figure 12-7 or Figure 12-8). The other 10 rows show the path-specific results, which were reformatted and rounded off to one decimal digit for Figure 12-8). Soft drink and ice manufacturing has again been abbreviated SDM to conserve space. Summing all of the site values in the 1,095 paths would give a value of 684 tons CO2e, which is 73% of the total Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
372
Chapter 12: Advanced Hotspot and Path Analysis
940.4 tons CO2e. As motivated earlier, this is an expected outcome when using threshold parameters to limit the runtime of the code and the number of paths produced. Increasing T_max and/or reducing the percent parameters in the SPA code will always increase the number of paths and the total site emissions in the paths, and consequently, the percentage coverage of the SPA compared to the total will increase. Homework Questions for this Section: 1. Using the provided RunSPAChap12Waste.m file for the 2-sector economy example from the chapter, fill in the table below for the total waste effects across paths as T_max ranges from 2 to 5 and as percent ranges across 0.01, 0.1, and 1. The total waste effect found above is already entered into the table. Describe in words what the results in the table tell you about the SPA of this 2-sector economy. percent 0.01 0.1 1
2
3
T_max 4
5 6,403 g
2. Perform the same analysis as in question 1, but using the RunEIO02SPA.m file and the same soft drink and ice manufacturing sector, and tracking total CO2e emissions. Describe in words what the results in the table tell you about the SPA for this sector. percent 0.01 0.1 1
2
3
T_max 4 940.4
5
3. Change the provided RunEIO02SPA.m code to that it uses the 2002 commodity by commodity A and L matrices (keeping all other values the same). How different are the GHG results as compared to the 940.4 tons CO2e with the industry by commodity values? Discuss why they change. 4. Use the 2002 EIO-LCA producer model to create an expanded SPA for $1 million of automobiles that includes lifetime gasoline purchases with respect to total GHG emissions. Assume year 2002 cars cost $25,000, have fuel economy of 25 mpg and are driven 100,000 miles. Assume 2002 gasoline price was $1.30 per gallon. Discuss how this SPA differs from the SPA for $1 million of automobile manufacturing only. Also discuss the limitations of this Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
373
model for the life cycle emissions of gasoline-powered vehicles. (Hint: all of the MATLAB code for SPA has discussed entering just a single value into Y). 5. Modify the provided MATLAB code to use the US LCI process matrix (from Chapter 9) and present the processes with the top 10 site LCI values for fossil CO2.
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
374
Chapter 12: Advanced Hotspot and Path Analysis
: Reporting, Peer Review, Claims, etc. Role of chair, members Typical questions asked of a peer reviewer How to do the review, what to look at, etc. ISO 14071 From LCA framework slides: Goal: ¨ (1) Intended application, (2) audience ¤ Are the intended applications or audience relevant (or the most relevant) possible? Is it missing the point? ¤ Does chosen goal/audience preclude use by others? ¤ Example from actual study: "..The findings of the study are intended to be used as a basis for educated external communication and marketing aimed at the American Christmas tree consumer." ¤ Is it useful for supporting actual purchasing decisions? Is it useful for non-consumers (retailers?) ¤ When reviewing studies- who did the work, who provided data, and who paid for it are critically important ¨ Everyone has biases, conflicts of interest. whether they affect the work done
Question is
¨ Always useful to be skeptical and force study to convince you ¨ ACTA is a trade association of artificial tree manufacturers
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
Chapter 12: Advanced Hotspot and Path Analysis
375
¨ Haven't shown final results – but what do you predict they show? ¨ Don't panic. Sometimes sponsors want single credible study out there when result is well known (i.e., maybe artificial is better!) ¤ Not required to stipulate any of this in goal statement ¨ (3) Purpose, (4) whether it will make comparisons ¤ a.k.a. "why we did it and what will we do with it?") ¤ Can we expect the audience to use it to make these decisions? (I don't choose Xmas trees based on this) ¤ Example from actual study: "The goal of this LCA is to understand the environmental impacts of both the most common artificial Christmas tree and the most common natural Christmas tree.." ¤ Inevitably results will be shortened / generalized in secondary sources (e.g., "artificial trees are better") ¨ Are those the study author's fault? What could be done to ensure best possible attribution of results?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
376
Chapter 12: Advanced Hotspot and Path Analysis
Glossary?
Life Cycle Assessment: Quantitative Approaches for Decisions That Matter – lcatextbook.com
