Scientific challenges in the attribution of harm to human influence on climate.

• Author: Allen, Myles

The authors of this Article review the current state of the science of attribution of anthropogenic climate change, with particular emphasis on the methodological challenges that are likely to confront any attempt to establish a direct causal link between greenhouse gas emissions and specific damaging weather events. Standard "detection and attribution" analyses, such as those cited by the Intergovernmental Panel on Climate Change (IPCC), are generally sufficient to establish the strength of human influence on large-scale, long-term-average climate, but fall short of quantifying the role of greenhouse gas emissions in almost any conceivable case of actual harm, since nobody is directly exposed to a change in global average temperature alone. The authors argue that it should be possible to agree on a relatively objective approach to quantifying the role of human influence on climate in cases of actual harm. There are, however, a number of questions to be resolved, including: can we apply the concept of Fraction Attributable Risk, developed for population studies in epidemiology, to the analysis of an unprecedented change in a single system such as the world's climate? Can we rely on computer simulation to address counter-factual questions such as "what would the climate have been like in the absence of twentieth century greenhouse gas emissions, "given that we are working with imperfect simulation models? Due to multiple anthropogenic and natural contributions to changing weather risks, it will always be necessary to apply some kind of principle of ceteris paribus to quantify the role of any particular causal agent, such as greenhouse gas emissions. How is this principle to be applied? These questions are not, in themselves, scientific issues, although how they are to be resolved will have a direct bearing on how and whether climate science can inform specific causal attribution claims. In summary, we need the legal community to ask the scientific community the right questions. It is imperative that these issues be resolved as soon as possible, to avoid having them become entwined in the outcomes of specific cases. Thus, this Article serves as a kind of tutorial, going over some material that many will find familiar in order to place it in the context of attribution.

1. INTRODUCTION AND MOTIVATION: WHY LAWYERS NEED TO UNDERSTAND THIS STUFF

   In addressing a crowd of lawyers, the first and overarching challenge for a physicist is to keep the audience awake. This is particularly pertinent when the topic is climate change, because many in the legal community may feel that the methodological details of causal attribution are of academic interest only. When and if cases ever come to court claiming that actual or imminent harm is attributable to human influence on climate, this argument runs, competing sides will invite experts to testify for and against some assertion like "past greenhouse gas emissions contributed substantially to harmful weather event X," and cases will be decided on the basis of the credibility and qualifications of the relevant experts. Hence lawyers do not need to trouble themselves with the details; they simply need to line up authoritative expert witnesses.

   We would argue that such heavy reliance on expert testimony is both unnecessary and counterproductive. It is unnecessary because it should be possible to agree on a standard method for the attribution of harm to human influence on climate, such that results do not depend, at least as a first approximation, on the prior opinions of the scientists responsible for making the assessment. If the legal community demands it, the meteorological "raw material" for any assessment of harm could be provided by a straightforward extension of routine weather forecasting services, generated by the same neutral government agencies that currently provide forecasts. Expert judgment will naturally still be required for the interpretation of this material in any particular case, but, except in the most marginal situations, this reliance on expert judgment can be minimized if we can agree on an "industry standard" operational approach to the attribution problem, such as the one we advocate here.

   Heavy reliance on expert testimony in causal attribution claims is also likely to be counterproductive. First, there is considerable room for confusion over what expert testimony actually means. For example, we would all subscribe to the following two statements: "human influence on climate played a substantial role in causing the European heat wave of 2003"; and "it is impossible in principle to attribute any single weather event to human influence on climate." To a layperson, these statements appear contradictory. They only make sense if it is understood that "causing" an event in the first statement must be interpreted as "contributing to the risk of" that event occurring, while the second statement is only true if we interpret "attribute" narrowly as "were it not for human influence, this event would not have occurred." Hence the legal community cannot expect the scientific community to do its job for it: it will always be possible for both sides in a case to persuade experts (and possibly even the same expert) to subscribe to statements that sound as if they support their cause. Expert testimony thus will only be useful if the scientific context is understood.

   Overreliance on expert testimony is also likely to have a damaging impact on the scientific community. We do not have to look very far for an example of the damage that can be caused by personalizing scientific debate. Some critics of the Mann et al. reconstruction of Northern Hemisphere temperatures over the past millennium, (1) as cited by the 2001 Scientific Assessment of the Intergovernmental Panel on Climate Change (IPCC), (2) have made a sustained effort to divert the debate from methodological issues of paleoclimate reconstruction to procedural issues and ad hominem attacks over what source code was available when, how the reconstruction came to be cited by the IPCC, and so forth. (3) Scientific research will rapidly become impossible if individual researchers have to operate under the threat of their research software being subpoenaed or if scientific statements must be individually scrutinized for their potential legal implications.

   The only way to avoid this (scientists') nightmare scenario is to operationalize the attribution problem now, which must include the development of an agreed approach to incorporating new research into the attribution procedure. Any operational attribution procedure will always tend to be somewhat behind front-line research, whereas the "latest" research results tend, by their very nature, to be relatively unreliable until they have been fully digested by the scientific community. For this reason, weather forecasters are typically conservative about incorporating new research into their forecast models; they generally require some demonstration that an innovation will demonstrably improve forecast skill before making such a change. Demonstrating that an innovation--such as an improved model or a new reconstruction of solar variability--will improve the accuracy of attribution statements is clearly more problematic, however, because we do not have a sequence of cases on which to test out the impact of the innovation as the weather forecasters do.

   Popular discussions of the climate issue often give the impression that the whole science of attribution has to return to square one whenever a new feedback process is identified or a new solar reconstruction is published. This would obviously make life impossible for the legal community because new research papers that might potentially have some bearing on attribution claims will always be coming out. Nevertheless, we believe it should be possible to agree on a methodology for deciding to what extent new research results refute past attribution claims, and by applying the same standards, to agree on whether or not the current version of the operational attribution system is good enough to be applied to specific questions.

   To conclude this introduction, we stress that this review is focusing exclusively on the meteorological attribution problem: to what extent human influence on climate can be "blamed" for observed weather trends and specific weather events such as floods, storms, or heat waves. We will not be addressing the equally important problem of how other factors, such as decisions to build in vulnerable locations or to maintain coastal defenses, as well as offering health services, can contribute to or mitigate actual losses, nor how these losses are themselves related to meteorological variables. We believe that these questions can be considered separately because only in very exceptional cases (such as the carbon released by large-scale regional deforestation) will there be any significant feedback between local decisions affecting vulnerability to weather and global changes affecting the weather itself.

2. "CLASSICAL" DETECTION AND ATTRIBUTION: THE EVIDENCE FOR HUMAN INFLUENCE ON GLOBAL CLIMATE

   The aim of this section is to outline the principles behind the standard approach to detecting and attributing human influence on global and regional climate as they are used by the scientific community to provide the basis for successive IPCC Scientific Assessments. (4) We will avoid methodological details, and instead focus on clarifying the precise question that is being asked in conventional detection and attribution studies and the assumptions underlying their answers in order to highlight issues that might prove contentious if these studies are ever used in any legal action.

   Detection and attribution studies have traditionally focused on large-scale, long-term changes in variables such as seasonal or annual mean surface temperature. (5) Changes in such metrics do relatively little damage in and of...