National Assessment Synthesis Team
October 15, 1998

Introduction

This paper discusses the handling of socio-economic and ecological (SEE) assumptions in assessing the impacts of climate change and variability in the National Assessment. It is revised from the paper circulated before the Monterey meeting, and incorporates discussion and feedback from Monterey, Woods Hole, and other conversations.

The paper outlines a Framework to approach the problem of describing socio-economic and ecological (SEE) characteristics of the future world for which climate impacts are assessed. The framework is intended to provide some level of consistency in products of regional and sectoral teams, so their results can be used in preparing the National Synthesis Report. The framework is not intended to be a final product, but to be further developed through continuing discussions. Moreover, the framework cannot possibly seek to be a complete set of instructions, specifying how to make every required socio-economic or ecological (SEE) assumption for every analysis: each team will have to make many such decisions. The final section of this paper identifies some resources and contact people to help you find data and think through your input decisions.

The first section discusses a few general principles for making SEE assumptions. The second provides more specific, illustrative suggestions and examples for putting these principles into practice. The third discusses the specific SEE data that will be provided, and identifies sources of further information and support.

In addition to the common approach outlined here, your team may wish to perform various other types of analyses, perhaps using approaches very different from what is outlined here. If you have time, we encourage you to experiment with a variety of methods and approaches, and to report both what worked and what did not work, so that one of the products of this assessment is insight into how to do a better job next time.

General Principles:

In discussions in Woods Hole and elsewhere, it has been useful to make explicit a few principles that have helped to organize our thinking about SEE assumptions. This section summarizes these principles. The next section provides some suggestions for putting them into practice.

1. Climate Impact Assessment needs socio-economic and ecological assumptions.

   Simply describing how climate varies or changes is not sufficient to understand impacts or vulnerability. Climate impacts happen to societies, economies, and ecosystems, so describing impacts requires information about the current (and in some cases the cumulative past) state of the economic, social or ecological domain that bears the impacts. Consequently, assessing impacts and vulnerability requires assumptions (explicit or implicit) about the future state of society and ecosystems in which climate impacts occur. These will not be the same as today's society and ecosystems. In the future climate-changed world, patterns of population, economic activity, technology, resource use, and ecosystems will all likely differ from those of today, in ways that are relevant to thinking about impacts. Consequently, while precise and reliable prediction is impossible, assumptions about the future cannot be avoided. In most cases, it is more reasonable to take on the hard job of making the required SEE assumptions and making them explicit, than to assume implicitly that the future will be like the present. If on reflection you judge that the most reasonable assumption about some particular future characteristic is that it will be unchanged from the present, you should also make this assumption explicit.
    

2. The approach must accommodate each team's data needs and specialized expertise, while maintaining broad consistency of national assumptions.

   The national assessment includes many separate studies of impacts at finer-scale domains - impacts on particular activities, resources, or values (e.g., coastal buildings, winter wheat yields, or heat stress on human health) at particular spatial scales (e.g., a county, a metropolitan area, a watershed, a particular coastal zone, a state, a region). Each study will need different SEE data inputs, and each study team will know best what data they need, and often what values are plausible. The assessment process must take advantage of this specialized expertise as much as possible.

   But these fine-scale studies must ultimately be synthesized into an assessment of climate impacts and vulnerability for the nation as a whole. A few key variables are likely to be required inputs to many studies, and to this national synthesis. Population and GDP are obvious examples, although there could be others. For these, all studies should use consistent assumptions. Moreover, the aggregation across separate studies will be much easier if teams use a broadly parallel approach to their studies, even when their specific data needs are unconnected.
    

3. Studies should choose a few key impact domains and impact-determining variables.

   The complexity of impact assessment can be overwhelming. Multiple dimensions of climate potentially influence a vast number of coupled economic, societal, and ecological values, and these influences are mediated by a host of individual, organizational, and governmental choices. Developing scenarios for such studies is a much harder problem than developing emissions scenarios, because of the intensely fine-grained, detailed, and coupled information required to describe and understand impacts. Even within a single region or sector, no team can hope to grasp the entire, massively complex problem.

   Rather, the most we can hope for in the limited time available is that each team undertake a few narrowly targeted studies, providing thoughtful analysis of possible impacts, vulnerabilities, and key decisions in a few key domains (e.g., particular impacts, activities, resources, or locations) within their area. Both the regional and sectoral teams have begun this process with their scoping exercises and identification of key issues. But even studies of a few key domains or issues will require major simplifications to deal with the multiple factors that potentially bear on each domain. Attempting to describe fully the SEE factors that influence the climate vulnerability of, say, wheat cultivation in Kansas or heat-stress deaths in Chicago, will quickly become an impossibly complex - and arbitrary - exercise.

   It is futile to attempt to describe the future in all detail, even as it shapes climate impacts in one narrow domain. Instead, each study should seek to identify a small number of key SEE variables that are likely to have the strongest and most direct influence on climate impacts and vulnerabilities in their domain. The analysis can then proceed by assuming values for these few key variables, without specifying how (of the multitude of possible ways) they came to have those values. If a forest study decides that demand for commercial forest products is a key variable determining impacts, they can assume values for demand without having to specify what combination of demographic, market, ecosystem, and technological factors caused demand to take that value.
    

4. Studies should reflect uncertainty by choosing ranges, not points, for key variables.

   Because the future is uncertain, it is important not just to assume a single value for the key variables that influence impacts, but to consider alternate values. The simplest approach is to consider a high and a low value, so as to span a range. However confident you may be that the future is going in a particular direction (e.g., demand for forest products will increase), you should work hard to think through other possibilities that could broaden the range of plausible values for your impact variables. Considering a range of values for key inputs is necessary to investigate how impacts and vulnerability in your domain depend on these inputs. Moreover, when you estimate ranges, you must work hard to draw them wide enough to accurately reflect the applicable uncertainty. A host of research on decision-making has shown that you are likely to be too confident in your predictions. You should try to guard vigilantly against this bias.
    

5. Impacts in a domain may depend both on characteristics of that domain, and on characteristics of its broader socio-economic context.

   It is clear that climate impacts within a particular domain will depend on the SEE characteristics of that domain. But for many domains, impacts may also depend strongly on the broader context in which the domain is situated. Impacts and vulnerability in one county or state may, for example, depend on the condition of other counties or states nearby, or the nation as a whole; impacts in one economic sector may depend on the condition of other sectors. These contexts may affect impacts and vulnerability for several reasons. The context - other places or activities - may compete with the domain for some constrained resource (e.g., regional water supply may be shared among nearby regions, alternative forms of agriculture, or between agriculture and urban or industrial development.)

   Alternatively, the context may provide resources that shape or mitigate vulnerabilities: e.g., capital to support new investment or adaptation; knowledge and capacity for technological and social innovation; people to fill new jobs or places for people to move to; support for transition to different livelihoods; or a governmental capacity to make policies to manage vulnerabilities, adapt to impacts, or compensate those who suffer from them. Where such linkages are important, contextual conditions may either intensify or diminish the vulnerability of the domain to climate impacts, or the significance of these impacts.

Making the Principles Operational: Tentative Guidelines

Implementing these principles - getting SEE assumptions straight for a particular analysis - will pose serious practical problems, many of which will differ from study to study. Here, we provide suggestions and illustrations on how to proceed. Perhaps your study will fit the approach well; if not, you should feel free to adjust parts of the approach, or in extreme cases discard them.

In particular, the following discussion assumes that analyses and key SEE variables are quantified. Where possible this is a desirable approach, but where not possible it should not be forced. Where quantitative analysis or modeling is not feasible, a good qualitative treatment of possible impacts will be more useful, both to your team and to the national synthesis, than a seriously incomplete or biased attempt to perform a quantitative assessment.

1. Choose a few key impact domains to study, and a few key impact-determining variables:

   Most teams have already gone through a scoping exercise to identify a few key impact domains on which they will concentrate. We would only urge that all teams also establish a clear priority ranking of these domains, reflecting your collective best guess of their overall importance. With such a ranking, you can begin your analyses with the highest ranked domains, so when you run out of time (as we all will) you will at least be able to say something about the few most important impact domains in your region or sector.

   For each domain, the team should then list the most important SEE variables that are likely to influence impacts and vulnerability in that domain. You should choose a small number of such variables, that appear have the strongest and most direct influence. These variables should represent key SEE uncertainties, not objects of direct policy choice, since key policy decisions that influence vulnerabilities will be considered explicitly in your analysis.

   Several teams went through this exercise at Monterey. Choosing a few key SEE input variables is difficult, but is necessary to keep the analysis to manageable complexity. The key uncertainties are likely to be different for each domain of analysis; they may even be different for the same domain at different times. For example, key inputs to a health study may include the size of particular vulnerablepopulations; for a water study, they may include the level of various categories of water demands; or for a coastal study they may include the intensity, character, and location of coastal development. Choosing the most important input variables for your study may require that you first do some scoping exercises or preliminary, exploratory analyses. For some domains, you may find that national economic growth, or some other characteristic of the national context, is a key input; for others, all important inputs might be characteristics of the domain itself.
    

2. Choose Ranges, not points, for key impact-determining variables.

   For each SEE variable you have chosen, estimate a high and a low value that span a plausible range. Existing data or projections may provide some guidance in estimating such a range; for certain demographic and economic data up to the year 2030, high and low values (as well as a middle, "baseline" value) will be provided. But in many cases, projecting values for SEE variables will depend on the collective judgement of team members.

   How wide a range should you choose? In qualitative terms, we suggest the range should include a substantial majority of uncertainty, but should not seek to include low-probability extreme events. In quantitative terms, the range should correspond - in the judgement of the team - to an 80 per cent confidence interval. In your judgement, there should be no more than 1 chance in 10 that the true value will be higher than your "high" value, and no more than 1 in 10 that the true value will be lower than your "low" value.

   The purpose of the range you construct is to provide a set of alternative SEE assumptions to construct baselines on which climate-related perturbations are imposed. Your assumptions will support a series of "what if" analyses, examining the impacts of various scenarios of climate change under various alternative sets of SEE assumptions, and decisions that might manage associated vulnerabilities. Consequently, these ranges should include all sources of uncertainty except climate itself and policy responses to it. Climate and climate responses should be examined explicitly in the analysis, not concealed in the range of alternative values assumed for SEE inputs.

   There is one important exception to this exclusion of climate-related uncertainties from SEE assumptions. For analyses where climate impacts or policies outside the US are important, the range of SEE assumptions should include these uncertainties. For example, climate and climate responses abroad might have strong influence on US immigration, or export demand for US agricultural products. If immigration or agricultural export demand are key assumptions for a particular analysis, their range should include this uncertainty.

   However you choose values for the key SEE inputs to your analysis, be careful to explain how you got them. When you scale national numbers to a region, please explain how you did it. It may be useful to discuss historical patterns of change, to illustrate what "high" and "low" have meant in the past.

   You should be keenly aware that you are likely to be too confident in your predictions, and to underestimate the uncertainty that applies to your estimates. Abundant research has shown that people are systematically over-confident in estimation and prediction, and fail to account adequately for uncertainty. In drawing uncertainty ranges for your key inputs, this means that you must be careful to draw the ranges wide enough, wider than you are initially likely to think is necessary.
    

3. Combine Assumed values of impact variables to construct high and low-impact scenarios.

   At this point you should have the following: an ordered list of a few key impact domains that are your priorities to study; for each of these domains, a list of a few key SEE factors that in your judgement most strongly influence impacts and vulnerabilities from a specified climate change; and for each of these factors, a range of values reflecting your collective judgement of its uncertainty.

   Your analysis will of course require many other data inputs and assumptions, in addition to these few key factors. To keep your analyses manageable, only the few key factors you have identified should be varied. Other inputs should be left at "best guess" or baseline values.

   From here, you should combine high and low values for various SEE inputs to generate a small, usefully diverse set of SEE scenarios. From this point, differences among particular analyses are likely to be of overwhelming importance in determining sensible ways to construct these scenarios. While we continue to offer suggestions on how to do this under certain ideal circumstances, you may find that minor or major departures from our suggested approach are necessary for your analysis to make sense.

   1. For each input variable, note which of its values (high or low) you expect to be associated with high versus low impacts and vulnerability.
       
   2. Cluster your key input variables, according to your judgement of how strongly they are likely to vary together. If possible, create two clusters, such that the members of each cluster are strongly correlated with each other (whether positively or negatively), but only weakly correlated with the members of the other cluster. If your important inputs include some local and some national, it might be most appropriate to gather all the local ones into one cluster, and all the national ones into the other.
       
   3. Finally, create four scenarios by setting all variables within a cluster to their high and low-impact values respectively; and combining high and low-impact settings for variables in the two clusters. If your two clusters consist of local and national variables respectively, the four scenarios will combine high vs. low impact values for the local variables, with high vs. low impact values for the national variables.

   Your problem may fail to fit this approach in various ways. All important SEE inputs may be local to your domain; the important inputs may span various spatial scales from the local to the national; all important variables may be strongly correlated; or several important SEE inputs may all appear to be independently variable.

   If these or other obstacles arise, you must rely on your judgment.

   Ideally, the goal should be four SEE scenarios based on combining important, independently variable SEE inputs when possible; where this is not possible or appropriate, two SEE scenarios should be constructed based on jointly varying one, or a few, key SEE inputs between their high-impact and low-impact values. If at all possible, you should strive to construct at least two distinct and plausible SEE cases.
    

4. Do the Analysis

   With decisions made about alternate values to assume for key SEE impact variables, you are ready to proceed with the analysis. As with constructing scenarios, the detailed form of analyses will vary greatly across individual study domains.

   A "canonical" analysis will examine the effect of a few distinct climate scenarios, under a few distinct assumptions regarding SEE conditions, to generate "what if" analyses that illustrate possible impacts of particular climates under particular SEE conditions. Differences between the baseline-climate world and the climate-scenario world will reveal illustrative impacts and vulnerabilities. Differences between impacts of the same climate scenario under alternative SEE conditions will illustrate how vulnerabilities depend on SEE factors, and may help to identify key decisions that strongly shape impacts and vulnerabilities.
    

5. Report, Revise, Iterate

   The national assessment is an ambitious and novel exercise, and we must all feel our way. We offer these detailed proposals for crafting SEE scenarios, even though they are certain to appear inapplicable or misguided for some - perhaps for many - analyses, in the spirit of pushing the discussion quickly to the level of practical decisions. If these suggestions are helpful, or if they are not, please let us know; in particular, please let us know if you can suggest specific improvements that would make this approach work better for your analysis.

   Since this is the first national assessment, it is crucial that we try to learn from the methods we employ, particularly including our failures. Each analysis should yield new substantive and methodological questions. Reporting both what worked and what did not work, and how and why, will help to build the capacity to do a better job in subsequent national assessments.

   Above all we stress the importance of being parsimonious, in the choice of both domains to study and SEE inputs to vary. If your team studies three high-priority domains under three or four climate scenarios, using three or four SEE scenarios, for two years (2030 and 2100), this already yields 54 to 96 separate analyses. Further enrichments will rapidly increase the number of combinations beyond all feasibility.
    

6. The Difference between 2030 and 2100

   The assessment focuses on two target years, 2030 and 2100. While these dates are both far in the future to be making projections, 2030 is within the scope of some projection models and strategic-planning tools, while 2100 is beyond almost all such projections. Consequently, the richness of estimates available from which to construct SEE scenarios differs greatly between the two target years.

   For 2030, population and economic projections for three different scenarios from a model developed by NPA Data Services, Inc, will be distributed to all teams on CD-ROM. County-level information for each year to 2050 will be available for population by sex and 5-year age cohorts; as well as income and employment by industry sector at the 1-digit SIC code level, plus farm and non-farm proprietors, federal and state/local government, and military. The information will also be presented aggregated to states and metropolitan areas; alternatively, your team can re-aggregate the county data to any other multi-county region that is useful for your analysis. The three scenarios will be: a baseline projection developed by NPA Data Services, Inc; and a low-growth and high-growth scenario developed by NPA in consultation with the NAST. The scenarios were developed to bound the most likely range of values for U.S. population growth and economic growth. If population, employment, and economic growth are among the most important variables likely to influence impacts and vulnerability in the domains that you are studying, we request that you use these projections.

   This model and data series are widely used as planning tools in various federal, state and local governments, and by the private sector. Staff at Oak Ridge National Laboratory (ORNL) supporting the national assessment are familiar with the model and data series, and are available to provide technical support to national assessment teams in their use.

   These data may provide many of the inputs you need for your impact analysis in 2030, but probably not all. If you need economic data at finer sectoral resolution, you will have to extend the NPA data with additional assumptions; if you need information about specific activities or technologies, land use or land cover, or ecosystems, the NPA data will not provide it. In this case, you must find other data sources or generate your own assumptions. Staff at ORNL are also available to help you identify existing data sources or projections relevant to your needs. Some team leaders have also offered to help other groups with certain specific kinds of data problems.

   Contact information for these resources is provided below. 2100 is much more difficult: uncertainties are wider, fewer projections are available, and the problems of developing believable and useful sets of assumptions are much harder. The only projections available for this time horizon come from integrated-assessment models used to study alternative time-paths of greenhouse gas emissions. Some of these models provide values for US population, GDP, and emissions, but little or no data at finer scale relevant to assessing impacts and vulnerabilities.

   To get a sense of how challenging the problem of developing useful SEE scenarios for 2100 is, you might imagine trying to predict today's world from the perspective of 1898. At that time, American's fertility rates were higher; the social mix was sharply different from today's; many more people lived on farms; aircraft, electronics and nuclear technologies had not been invented; aluminum was a semi-precious metal; and automobiles existed only as primitive novelties.

   With such a daunting challenge and such sparse and suspect data, teams might choose two broad directions to proceed. You might choose to do the analysis for 2100 using essentially the same method as for 2030, generating your own estimates for much more of the input data: identify key domains and key SEE variables for a hypothetical USA in 2100; choose a range of plausible values for each key SEE variable; and subject a few alternate SEE futures to a few alternative climate scenarios.

   Uncertainties will be much wider and the entire approach more speculative and qualitative, but the results may still be useful and interesting. You might, for example, judge that the key SEE variables determining impacts and vulnerability in a particular domain change over time: e.g., economic activity levels might be most important in the near term, but some specific aspect of technological change most important in the long term.

   Alternatively, you might examine impacts in 2100 by using an inverse approach. This approach would be parallel to the use of inverse climate scenarios for 2030, where you assumed some specific SEE picture for 2030 and asked how much climate would have to change, in what ways, to yield large impacts.

   To use an inverse approach for SEE assumptions, you would not fix specific SEE assumptions in the year 2100. Rather, you would impose a specific climate scenario - either a GCM output or a historical analog climate - and ask what SEE characteristics would strongly alter vulnerabilities to this particular climate. Under what plausible SEE conditions would this climate bring large impacts, discontinuities, or qualitative changes? (e.g., rendering important activities non-viable; exhausting a key natural resource such as freshwater or a key economic resource such as insurance reserves; pushing major enterprises into bankruptcy or specialized communities below thresholds of viability.) And conversely, under what plausible SEE conditions would this climate impose essentially insignificant impacts?

Resources and Support

1. Tom Wilbanks and his colleagues at Oak Ridge are available to provide general support for assessment teams in identifying relevant sources of data and projections for socio-economic variables, land use and land cover, and in working with the NPA data set.

   They prefer to receive inquiries by e-mail, so they have a written record of the request to pass on to the relevant expert.

   For general data inquiries:

   Tom Wilbanks

   Phone: 865-574-5515

   Fax: 865-576-2943

   E-mail: twz@ornl.gov

   or

   Sherry B. Wright

   Phone: 865-574-8651

   Fax: Same as above

   E-mail: sbw@ornl.gov

   For inquiries specific to understanding and using the NPA data set (for these inquiries, it would be most efficient, where possible, to a team member with some knowledge of econometrics or regional economic modeling):

   David P. Vogt

   Phone: 865-574-5192

   Fax: 865-573-8884

   E-mail: dpv@ornl.gov
    

2. Several leaders and members of particular teams have expressed willingness to help with specific inquiries in their areas of expertise, subject to the constraints imposed by their own projects. Lynne Carter at the National Assessment Coordination Office has offered to field inquiries and direct them to people willing to help. She will also circulate a list of people who have offered to be resources for specific types of questions.

   Lynne Carter,

   National Assessment Coordination Office

   Phone: 202-314-2230

   Fax: 202-488-8681

   E-mail: lcarter@usgcrp.gov
    

3. Outputs from runs of ecosystem models under the VEMAP project will be available to national assessment teams. Contact Ben Felzer, National Center for Atmospheric Research.)

   Ben Felzer,

   National Center for Atmospheric Research

   Phone: 303-497-1703

   Fax: 303 497 1348

   E-mail: felzer@ucar.edu