Figure 1.3 Confidence in attribution of observed trends in impacts related to extreme weather. Graphical interpretation of the table in Cramer et al. (2014) documenting the synthesis of evidence of an effect of historical trends in extreme weather on various natural, managed, and human systems.

In the figure, confidence in the impact is necessarily no higher than confidence in the relevant climate driver, because the latter is a component of the former. Note that no assessment was made about whether the climate trends were driven by human activities or represent some natural fluctuation. Hansen and Stone (2016) did examine the role of humans in trends in climate averages that they considered relevant for the extreme weather, and they provided some indication of the robustness of some assessments that included attribution to human activities. In general, the snowmelt flood and coral bleaching assessments ought to be unaffected, whereas the effect on the Arctic coastal erosion assessment depends on the balance between the importance of thermofrost degradation (unaffected) versus regional sea ice retreat (strongly affected). Hansen and Stone (2016) did not examine the human role in other climate trends listed in this figure.

      There are three main observations one may make from this illustration. The most obvious is that not that many impacts were covered and many included were limited to very specific statements (for instance, the distinction between erosion of Arctic versus non‐Arctic coasts). The synthesis was conducted for two types of impacts: broad synthesis statements of general interest (e.g., monetary losses) or assessments of a more narrow set of impacts selected on the basis of whether strong evidence existed one way or the other (e.g., Arctic coastal erosion). In this sense, the assessment fell short of a full global synthesis across all systems, at least in part because it was conducted under the framework of detection and attribution.

      The second observation is that the figure is an amalgam of trends in impacts related to extreme weather, but these trends are not necessarily due to trends in the extreme weather itself. For instance, the evidence of increased erosion of Arctic coasts is based on understanding that storms can now erode the coast more easily because the summer permafrost has disappeared and is no longer providing structural strength and because there is a much longer distance for waves to grow in the space vacated from retreating sea ice. In other words, the erosion occurs during the storms, but the storms themselves are not changing, only the way they interact with the coast is because of more gradual changes.

      The third, more arguable, observation is that there are two types of conclusions present. The assessments for coral bleaching, snowmelt floods, and Arctic coastal erosion are all of at least medium confidence of a major role of climate change (which is mostly unaffected when extended to a major role of anthropogenic climate change). The other assessments are of lower confidence and apply only to the existence of a role of climate change. The former group arise because large‐scale warming is a simple direct driver, warming is the most visible manifestation of recent climate change, the warming and impacts have been fairly well monitored, and the systems are relatively sensitive to temperature (e.g., the snow line on mountains or the sea ice edge). One or more of these factors is lacking in the second group.

1.4. IMPLICATIONS FOR THE FUTURE

      This chapter has focused mainly on the past, specifically about detection and attribution of changes. This places heavy burdens on the evidence base that has the advantage of producing coherent, strongly supported conclusions, but it also has the disadvantage of being unable to provide information on some types of impacts. Does this matter when predicting future risk? After all, predictions concerning risks related to the extreme RFC were made many years before the first assessments of changes in past risks.

      First, as time elapses further from the initiation of the UNFCCC process in 1992, we need to know whether we are meeting the UNFCCC’s objective of preventing “dangerous anthropogenic interference with the climate system.” In other words, we will need to continually update our documentation of how anthropogenic emissions are affecting various aspects of human, managed, and natural systems around the world. This is fundamentally the detection and attribution problem, and hence not only requires understanding of how the world works but also monitoring how everything is (or is not) changing.

      As for the relevance for predicting the future, it helps to consider conditions under which detection and attribution analysis provides inconclusive results and to consider those conditions in the context of understanding future risks. There are three possible reasons for detection and attribution analysis to provide inconclusive results: poor monitoring, poor understanding of how the system operates, or bad luck (the observations and understanding do not match because of a statistical fluke). Poor understanding will be just as relevant for errors in predicting the future as they are for the past, in fact, perhaps more so because those errors are likely to be amplified as the climate change signal and other signals become stronger. Statistical flukes occur because the analysis is inherently probabilistic in nature but ought to happen rarely. It does remind us that specific aspects of the predicted future may not materialize in the end simply because the climate and various impact systems are inherently chaotic. Poor monitoring is also relevant, though, because if we do not have a reliably observed baseline and if we do not obtain reliable observations of future states, then we will lack an important input in the process of refining later predictions. The ability to calibrate predictions by evaluating against past behavior, that is, through detection and attribution analysis, will be especially important for our assessment of risk in cases where understanding remains poor in the future.

This chapter has focused on types of synthesis assessments that might be useful for informing the UNFCCC process or some similar global, multisectoral interest. Of course, synthesis assessments might be useful for other audiences too. At the national or a subnational administrative level, synthesis assessments may have a similar purpose, that is, informing the development and monitoring of the effectiveness of government policy, and so such syntheses might take a similar form to a UNFCCC‐motivated synthesis. However, syntheses relevant for industries, whether for large company, industry organizations, or government ministries, may have a more restricted remit in terms of types of impacts. That may mean that a single quantitative metric, such as insured monetary losses, is applicable. In these cases, there may be a clear and obvious method for performing a synthesis, too.

      Given the diversity in what is required of synthesis assessments, this chapter has refrained from specific recommendations that might be relevant only for a very particular class of assessment. Instead, there are some broad general guidelines that should considered in the