Figure — Reasons for concern about climate change impacts (source: IPCC WG 2 TAR, figure SPM-2). The left part of the figure displays the observed temperature increase up to 2000 and the range of projected increases after 2000 as estimated by IPCC, WG I (IPCC, 2001a) for scenarios from the Special Report on Emission Scenarios (SRES — see the Emission scenarios). The right panel displays conceptualizations of five reasons for concern regarding climate change risks evolving through 2100. White indicates neutral or small negative or positive impacts or risks, yellow indicates negative impacts for some systems, and red means negative impacts or risks that are more widespread or greater in magnitude. This figure shows that the most potentially dangerous impacts (the red colors on the figure) typically occur after a few degrees warming — thus, my later use of 3.5 ^oC as a tentative “threshold” for serious climate damages is very conservative (e.g., see Mastrandrea and Schneider, 2004 for a lower threshold value). The risks of adverse impacts from climate change increase disproportionately with the magnitude of climate change.

How Much Will It Warm Up?

The major challenge policymakers have given IPCC scientists is to assess the potential dangers of climatic warming. The UN Framework Convention on Climate Change (UNFCCC), fashioned at the Rio Earth Summit in 1990 and signed by 176 nations (including the US), agrees to avoid “dangerous anthropogenic interference with the climate system.” But what constitutes 'dangerous' is not defined. Of course, determining what is dangerous involves value judgments since it requires the consideration of many incommensurate amenities (see “Dangerous” Climate Impacts, below). Nevertheless, the risk of triggering abrupt nonlinear climatic changes or causing very high damages goes up rapidly as climate warms beyond a few degrees (IPCC WG 2). Thus, it is imperative that scientists estimate the warming possibilities over at least a century (see, for example, the figure, Three climate sensitivities, below).

Go to top

“Dangerous” Climate Impacts and the Five Numeraires

What are 'Dangerous' climate impacts? Here are the slides of an overview, in a presentation I made to the symposium, Avoiding Dangerous Climate Change, and here is an accompanying paper, Schneider and Lane, 2005.

Nordhaus, 1994a, conducted a survey of social and natural scientists and economists, similar to that undertaken by Morgan and Keith, but with a focus on the impacts of several scenarios of climatic change on world economic product, including both standard market sector categories (e.g., forestry, agriculture, heating and cooling demands) and so-called non-market sectors like biological conservation, international equity, and national security. Respondents were instructed to give estimates of aggregate losses for each warming scenario and to estimate what fraction of those losses were specified in market categories (i.e., in Standard National Accounts) and what fraction would be non-market losses. Respondents were not asked to specify exactly which metrics of change (known as numeraires) they used in each case, an issue I'll return to below.

Some of the explanation for the large spread (see the figure, Probability distributions (f(x)) of climate damages) of estimated climate damage costs in the Nordhaus survey was related to differing perceptions of the nature of climate change. Much economic analysis of potential climatic damage is derived from static situations — for example, comparing the land values in a hot place to a cold place to get an idea about how land values in the colder place could change as a result of global warming. This is problematic because the time evolution of changes may not mimic the static set of conditions assumed in such cross-sectional studies and because climatic variability and extreme events can be changed by changing climatic averages (see Schneider, Easterling, and Mearns, 2000, Climatic Change, section 3). Both effects would lessen our capacity to foresee future changes and thus lower our capacity to adapt, thereby increasing impact cost estimates. (See Hanemann, 2000 for a sharp critique of cross-sectional methods and Schlenker et al., 2003 for a more specific example). IPCC Working Group II addressed variability changes, and the Table below is a representation of some of the variability and extreme event factors influencing climate damage estimates that also need to be weighed in any analysis of what constitutes “dangerous” climate change.

Table — Projected effects of global warming during the 21^st Century (adapted from IPCC 2001b, table SPM-1).

Five Numeraires. Another reason for some of the large range of future climate scenarios projected in the Nordhaus survey is due to the differences in perception between economists, other social scientists, and natural scientists regarding the fraction of damages coming from non-market categories. Roughgarden and Schneider (1999) analyzed the Nordhaus (1994a) data set and found that most respondents who had estimated large damages were natural scientists who had placed the bulk of their estimated damages in the non-market basket. The economists, on the other hand, tended to have smaller overall damage estimates and fewer non-market-related costs. (See Probability distributions (f(x)) of climate damages.) This raises a major issue about the dimensions of damages, which, I have argued, need even finer subdivision than the binary characterization of market and non-market. Schneider, Kuntz-Duriseti and Azar (2000) have separated out “Five Numeraires” of climate impact (see the box below): monetary loss (market category), loss of life, quality of life (including coercion to migrate, conflict over resources, cultural diversity, loss of cultural heritage sites, etc.), species or biodiversity loss, and distribution/equity (e.g., longer growing seasons from warming increasing production in richer northern countries while crop losses occur in poorer 'overheated' tropical countries). There is little agreement on how to place a dollar value on the non-market impacts of climate change, such as an increase in the gap between rich and poor, the loss of human life, biodiversity, or ecosystem services, but this does not mean that these numeraires should be ignored.

While many developing countries may not experience as much absolute dollar loss in the market category as developed countries, they are likely to experience more human suffering and mortality. Already, developing countries are more affected by natural disasters, both in terms of frequency and severity. In fact, Perrings (2003) states that in the 1990s, 94% of all natural disasters and 97% of deaths related to such disasters occurred in what he calls "low-income countries", with about two thirds of those deaths occurring in southeast Asia alone. These statistics support the use of alternative numeraires for calculating damages.

On a related subject, the last numeraire, the issue of equity in climate change is and will likely continue to be contentious. Climate change inequality will likely come in two forms. First, it will produce inequity in effects. Some countries or sectors within countries will benefit from a certain degree of warming, whereas others will be harmed by it. The developed countries, who are responsible for most of the greenhouse gases emitted into the atmosphere thus far, may not be affected as much as the developing countries for two reasons: First, there is usually higher adaptive capacity in richer, cooler countries than in poorer, warmer ones. Second, developing countries that have not yet experienced many of the economic fruits of the Victorian Industrial Revolution, and want their chance to emit and industrialize, fear that policies to restrict emissions will deny them their “fair share” of the atmospheric commons to use—quite literally—as a waste dump. One strategy to solve this problem is “technology leapfrogging,” the transfer or development of cleaner technologies to developing countries on a much-accelerated time schedule (relative to the slow developments that have emerged over the twentieth century in now-rich countries). Moreover, as there are disparities in countries’ abilities to pay for global warming-related problems, once again, the developing countries will be impacted more yet have less of an ability to pay than the rich nations. (See Regional Impacts.)

Box — The Five Numeraires (source: Schneider, Kuntz-Duriseti and Azar 2000)

Despite this, any comprehensive attempt to evaluate the societal value of damages (or benefits) from climate change should include such things as loss of species diversity, loss of coastline from increasing sea level, environmentally-induced displacement or death of persons (see “Scientists report global warming kills 160,000 annually”), change in income distributions, and agricultural losses. The environment also possesses intrinsic worth that does not carry a clear market value, such as its aesthetic appeal, which suggests that it should be treated as an independent variable in utility. We should consider the “existence value” of the environment (i.e., stewardship) and place a priority on preserving it, even if we don’t intend to personally experience it. This is in addition to the “option value” of the environment: we may want to preserve it for our — or our grandchildren’s — possible personal use in the future.

Thus far, our discussion of climate change effects has been largely anthropocentrically-focused. However, species other than humans are also affected by global warming, as the five numeraires suggest. In some ways, this is more concerning, as natural systems have limited adaptive capacity and are expected to be less resilient than social systems -- particularly for rapid changes for which all but micro-species generally have no evolutionary mechanism to adapt quickly. In addition, natural systems will likely be the most undeserving victims of future climate change, as they obviously have no influence on the policymaking process, except to the extent that humans value their services and survival.

Scientists have already detected that the approximately 0.7^oC of warming that has occurred on Earth since the mid-1800s is having a discernible effect on various ecosystems. Root and Schneider, 2002 reported that the ranges of the species they researched were moving poleward, up mountain slopes, or both. They also found a strong pattern of consistent shifting towards earlier spring activities occurring in many of the species investigated. Their early findings were supported by futher work by that team (Root et al., 2003) and by Parmesan and Yohe, 2003, who, after analyzing more than 1,700 species, detected habitat shifts towards the poles (or vertically upward) averaging about 6.1 km per decade, as well as the advancement of typical spring activities by 2.3 days per decade.

Furthermore, ecological systems are more likely to be affected than human ones if/when thresholds are exceeded. Scientists have identified various "geoboundaries", locations such as coastlines and mountaintops that are more prone than other areas to irreversible losses. If these areas become unsuitable for their present occupants of disappear altogether, many of the plant and animal species that dwell there will be unable to find suitable alternative habitats, making extinction much more probable (e.g., Thomas et al., 2004).

The severity of the effects of climate change will vary among species, implying a dismantling of existing plant and animal communities as individual species' responses to climate unfold, which could create disruptions in what Daily, 1997 terms ecosystem goods (i.e., seafood, fodder, fuel wood, timber, pharmaceutical products) and services (i.e., purification of air and water, detoxification and decomposition of wastes, climate moderation, soil fertility regeneration). Additionally, the more rapidly climate changes, the more likely there will be "no-analog" communities (e.g., see Overpeck, Webb, and Webb III, 1992 and Schneider and Root, 1998), a scenario that will likely lead to increases in endangerment and extinction.

These are a few of the reasons it is essential for any full analysis of the costs of climate change impacts or mitigation strategies to explicitly consider alternative numeraires. In a May 25, 2004 e-mail to colleagues (myself included), Stuart Pimm presented a striking analogy about the decision on which costs to include in damage estimates: "Your hotel burns down tonight, you survive, but it's a close call and you only have the chance to bring a couple of items with you as your leave. Do you take your wallet? Or the only photograph you have of you, your wife, your kids, and your now-deceased parents?" To continue the analogy, I worry that market-is-everything economists would just take hte money (i.e., fail to value non-market goods).

Any climate damage assessment must be as clear as possible about which are being used and which have been omitted. Moreover, before any aggregation is attempted — e.g., cost-benefit optimization strategies — authors should first disaggregate costs and benefits into several numeraires and then provide a “traceable account” (see Moss and Schneider) of how they were re-aggregated. Such transparency is essential given the normative nature of the valuation of various consequences, as characterized by the five numeraires.

We will not easily resolve the paradigm gulf between the relatively optimistic and pessimistic views of those specialists (stereotypical economists and natural scientists, respectively) with different training, traditions, and world views. But the one thing that is clear from both the Morgan and Keith and Nordhaus studies is that the vast bulk of knowledgeable experts from a variety of fields admit to a broad umbrella of possibilities in the area of climate change — including both mild and catastrophic eventualities (see Probability distributions (f(x)) of climate damages). The wide range of probabilities follows from recognition of the many uncertainties in the data and assumptions inherent in climate models, climatic impact models, economic models, and their synthesis via integrated assessment models (see Schneider, 1997b). However, this wide range of potential future outcomes creates a risk of misinterpretation by those who are unfamiliar with the underlying issues. While it is necessary that a wide range of possible outcomes be included, they should be accompanied by subjective probabilities that knowledgeable assessment groups like the IPCC believe accompany each of those possible outcomes. Both “the end of the world” and the “it is good for business” outcomes, often polarized in media and political debates, carry very low probabilities, and the vast bulk of knowledgeable scientists and economists consider there to be a significant chance of climatic damage to both natural and social systems somewhere between these two extremes. Under these conditions — and given the unlikelihood that any amount of research will soon eliminate the large uncertainties that still persist — it is not surprising that most formal climatic impact assessments have called for cautious but positive steps both to slow down the rate at which humans modify the climatic system, to make natural and social systems more resilient to whatever changes do eventually materialize, and to mitigate some of the emissions in order to reduce the likelihood of “dangerous anthropogenic interference with the climate system” (e.g, National Academy of Sciences, 1991).

Figure — Probability distributions (f(x)) of climate damages from an expert survey in which respondents were asked to estimate tenth, fiftieth, and ninetieth percentiles of damages for the two climate change scenarios shown. Source: from Roughgarden and Schneider, 1999. Data from Nordhaus, W.D., 1994.

The Earth's surface temperature figure already showed the projected temperature increases up to 2100, as estimated by IPCC using two categories of models. One set of projections comes from the SRES, which gives a total of six “story line” scenarios. The vertical bars on the right hand side of that figure estimate the range of temperatures that each scenario implies. Scenario A1FI results in the highest temperatures, and A1T nearly the lowest. The second category of models determines the width of the bars. The IPCC authors used 6 GCMs, each with different climate sensitivities and different thermal delay times due to oceanic heat capacity, to calculate the range in the bars for each SRES storyline driver. The total range of warming to 2100 spans from about 1.5^oC to a very large 5.8^oC. But as large as this range is — implying warming that would be relatively easy to adapt to by many human systems up to potentially catastrophic warming — it is still too small to incorporate all recent climate sensitivity estimates, like those on the Single probability density function, or in Forest et al. (their Figure 4).

In fact, estimating the likelihood of the warming in 2100 requires doing more than the IPCC did in either Working Group I or the SRES. Neither provided estimates of the subjective probabilities of climate sensitivity (Working Group I), nor that of emissions (Working Group III - SRES). Thus, very different estimates of risk could be obtained by invoking different assumptions of the likelihood of various climate sensitivities and emissions. I have challenged both Working Groups to remedy this lack of probabilistic estimation in their next IPCC report, the Fourth Assessment Report — currently being called the AR4 (see “What is 'Dangerous' Climate Change?” and “Can We Estimate the Likelihood of Climatic Changes at 2100?"). But not everybody in the process agrees that we can — or even that we should try in the AR4 — to make such highly subjective probabilistic estimates of future events.

Grübler and Nakicenovic (2001), key authors of the SRES effort, responded to my call (Schneider, 2001) for probabilistic assessment of emissions and climate scenarios with a cautious “no thanks.” Indeed, these authors suggest that we are better off, as IPCC did, assigning each representative scenario the label “equally sound,” implying no distinguishing probabilities. Grübler and Nakicenovic have thus implied that a “consequences alone” definition of risk is all we can deal with, as assigning wrong probabilities would be worse than nothing (i.e., a Type I error):

Risk Equals Probability Times Consequence

Indeed, I agree that any such estimates will be highly subjective and often carry a fairly low confidence (e.g., Moss and Schneider, 2000), and this should be freely admitted and kept in mind. Moreover, these estimates will likely change over time as knowledge is gathered and reassessed. But to duck the call to produce probabilistic estimates for such scenarios is to circumvent the completion of the risk equation, where risk = probability x consequence. It is very difficult for policymakers to have a “consequences alone” definition of risk such as that seemingly advocated by Grübler and Nakicenovic, in which only ranges of plausible scenarios are given “independent of the question of how likely they are to occur.” Let me suggest that if a range of possibilities with no estimates of probabilities attached was the sole basis of the scientific input for decision-making, then we should all focus on studying and preventing the next massive asteroid from colliding with the Earth, an event whose consequences would undoubtedly be orders of magnitude worse than any other environmental event we could imagine, anthropogenic or natural. The reason we do not all redirect our efforts towards the problem of avoiding an asteroid collision is that its probability is typically given on the order of one in ten million or less per year, whereas the probability of serious climate change is obviously many orders of magnitude higher than that – likely to be in the first decimal place (many tens of percent).

Probability matters greatly in a resource-constrained world struggling to determine the relative priorities for various investments (e.g., preventing dangerous climate changes, improving health care, or providing cleaner water). Nevertheless, I agree with Grübler and Nakicenovic that the assignment of such probabilities is fraught with deep uncertainty. While risk is still probability x consequence, the probability of some level of future climate change is not directly determinable by any set of frequency experiments – so-called frequentist probabilities – so we can only rely on scientific judgments based upon as much empirical observation as is possible (see 'Mediarology'). But these empirical observations will not be of the future climate – which would be impossible in principle before the fact – but rather of the behavior of the subcomponents of a complex systems model, which are then used as our “hazy crystal balls” to make the future projections. Thus, subjective judgments – essentially Bayesian judgments about the plausibility of the assumptions and structure of the systems model — are made on the basis of empirical validation of the subcomponents of the systems model (see Climate Modeling), not the performance of the overall system in the future (as that cannot be done empirically before the fact). Empirical data can and should be used as much as possible in the construction of the elements of such systems models used to estimate future changes, and empirical testing of the whole systems model on events that have already occurred (e.g., paleoclimatic events for a climate model or oil price shocks for an economic model) is desirable when possible, although these empirical tests will rarely be perfectly analogous to the future events climatologists are asked by society to assess.

The very nature of such subjective probabilities makes them controversial, and some scientists believe it is better to offer no probabilities than subjective ones — as implied in the above quotation, for example. I strongly disagree, using the asteroid collision argument made earlier as case in point. For us simply to redefine the classical definition of risk to “consequences alone” because subjective probabilistic analysis is fraught with deep uncertainties, is in essence to offer no advice to the policy community about how it should prioritize climate change actions or policies, for without probabilities, it is very difficult to engage in risk management. If we in the scientific assessment business do not offer some explicit notions of the likelihood of projected events, then the users of our products — policy analysts and policymakers — must guess about what we think these likelihood estimates are. That is hardly preferable, in my view, to a carefully-worded and explicitly-offered set of subjective probabilistic estimates accompanied by our (often low) confidence ratings of such estimates.

Go to top