Caveat Venditor, Lector, et Emptor (Invited Presentation)

Climate is defined by the Intergovernmental Panel on Climate Change “as the statistical description in terms of the mean and variability of relevant quantities over a period of time ranging from months to thousands or millions of years” and thus statistical methods are central for climate studies, as they are for many other disciplines. These tools are used to describe the effects of variability and uncertainty on myriad measures that have been developed to characterize the climate, and provide a context within which to evaluate differences in climate in space and time, and to infer information about the climate at locations or times that have not been sampled. That context for evaluation and inference includes the sampling, structural and distributional assumptions that are required to develop effective diagnostics and quantify their expected performance in repeated application under well understood, albeit necessarily idealized, conditions. Without such information to describe the reliability and sensitivity of evaluation and inference methods, it is impossible to assess the level of confidence that we can place in their results. Unfortunately, however, the routine application of statistical methods in climate research implies that users are often not fully aware of the limitations of the context that is implicit in their use. I will briefly discuss three examples illustrating topics where the statistical context appears to be understood and appreciated to different extents. The examples will draw from the body of research on climate change detection and attribution, the evaluation of extremes, and the characterization of the large-scale modes of climate variability. It is argued that the first two areas have a relatively well-understood statistical framing and a reasonable understanding of limitations, while the latter remains substantially more problematic.