by Judith Curry
At one level, analyzing climate risks is a matter of due diligence, given mounting scientific evidence. However, there is no consensus about the means for doing so nor about whether climate models are even fit for the purpose. An alternative to the scenario- led strategy, such as an approach based on a vulnerability analysis (“stress test”), may identify practical options for resource managers. – Brown and Wilby
At the recent Royal Society Workshop on Handling Uncertainty in Weather and Climate Prediction, among the interesting people that I met was Robert Wilby, who has an article in the latest issue of EOS.
Citation: Brown, C. and R. L. Wilby (2012), An alternate approach to assessing climate risks, Eos Trans. AGU,93(41), 401, doi:10.1029/2012EO410001. [link]
The recognized limitations of GCMs, including the lack of credibility on extremes, imply that GCM-based projections may have difficulty providing the information decision makers typically look for or even adding value to a risk analysis . One problem is the tendency for some stakeholders to perceive and treat projections as forecasts. Indeed, it is difficult to communicate exactly what climate projections mean from a decision standpoint— they simulate what might happen under some conditions but do not preclude other outcomes. In fact, climate analysts are often reluctant to say that one future is more or less likely than others.
In other disciplines such as decision analysis, scenarios are constructed to help decision makers explore the range of uncertainty in the key variables that affect their system or decision. However, climate change projections from GCMs are ill formed for doing so because of incomplete process representation, parameterization, and small effective sample sizes of models. As a result, the possible range of climate changes might not be fully explored if an analysis relies exclusively on climate projections. Changes beyond what current models project are possible. So if a decision maker wants to conduct a formal scenario analysis, restricting the analysis to this minimum range of uncertainty could result in a lack of consideration of possible climate outcomes.
Multimodel experiments such as the Coupled Model Intercomparison Project phase 3 (CMIP3), ENSEMBLES, Climateprediction .net, and others have helped to characterize aspects of climate uncertainty but not necessarily for those variables of greatest relevance to natural resource managers, such as variability statistics. Other climate modelers assert that the spread of uncertainty may be reduced by adjusting known model biases in simulating present climate. Some researchers are beginning to think that it is better to generate climate scenarios in such a way that one can control, by design, the range of climate changes in the specific variables of interest.
Given these concerns, climate risk analysis in a decision-making context should consider analyses other than climate projections. In some cases, a vulnerability analysis, or stress test, may provide greater insight. Like a sensitivity analysis, a vulnerability analysis provides information on how much a system of interest would respond (how sensitive it is) to changes in climate. Once risks are identified, model projections can be used to assess the plausibility, likelihood, or ranking of climate threats and opportunities based on the latest scientific evidence.
Climate scenarios can be generated parametrically or stochastically to explore uncertainty in climate variables that affect the system of interest. This allows sampling changes in climate that include but are not constrained by the range of GCM projections. The definition of scenarios can be developed as part of a stakeholder-driven, negotiated process, and climate projections can be used in this process. Alternatively, a very wide range of climate alterations can be developed independent of their plausibility and used to identify risks. For scenarios in which the climate consequences exceed coping thresholds, it is then fruitful to evaluate the plausibility of the scenarios. Climate projections, paleo- climate reconstructions, and subjective climate knowledge could all inform such discussions.
Hydrologists and engineers are developing methods based on sensitivity analysis that shift attention back on the water system of interest and use GCM projections to inform, rather than drive, the analysis. These include “scenario neutral” approaches and “decision scaling,” which uses decision analysis as a framework for incorporating climate information including GCM projections. These approaches might be termed “bottom-up meets top-down,” as they focus first on the issues of concern and then on how climate information might add value to the analysis. The basic steps in these methods are to (1) identify the problem, includ- ing defining objectives and performance measures; (2) use a stress test to identify the hazard and evaluate the performance of the system under a wide range of nonclimatic and climate variability and change; and (3) evaluate the risk using climate information including model projections.
An additional advantage of sensitivity approaches is that they may preclude the need for an expensive climate impact assessment and associated opportunity costs (i.e., time and money). For example, if a stress test is performed and no risks to operations emerge over a wide range of plausible climates, then a decision maker will have assessed climate risk, found little or none, and satisfied the review requirements without the large effort involved in typical GCM-led end-to-end uncertainty analysis. For instance, for many water systems, climate pressures may not be significant relative to other considerations [e.g. population increase], especially when economic discount rates in cost benefit analysis diminish the importance of the distant future.
JC comments: This article echoes many of the same themes developed in my presentation at the RS Workshop (.ppt; audio recording). Note, my interest in a follow on RS post has been stymied by only the audio recordings being available (no ppt, no podcast).