Assessing hydrological change: deterministic general circulation models or spurious solar correlation?

Recent research has focused on providing impact assessments of climate changes, specifically those due to an enhanced greenhouse effect. Typically, general circulation models (GCMs) are used to predict future climate scenarios based on descriptions of atmospheric/ocean processes formulated on physical laws. GCM horizontal resolution is typically coarse (300–1000 km2), whereby each grid/pixel is assumed homogeneous. The various GCMs differ not only in their grid sizes and vertical layers, but also in the number of processes and relevant parameters that can be included. An obvious weak point of GCMs is their inability to model properly the physics of clouds; in this respect the various GCMs vary considerably, and often produce quite different results (van Dam, 1999). Other problems with GCMs include grossly uncertain land surface representations, the problem of sub-pixel heterogeneity, and, perhaps most seriously, the attendant problems of translating GCM output into hydrologically meaningful variables. Despite many shortcomings, GCMs are viewed as ‘best science’ in that they embody the sum total of known climate physics. However, as noted earlier, the understanding of the ‘physics’ of climate processes is currently deficient. The complexity of GCMs means that runtime is prohibitive, making uncertainties difficult to propagate. However, substantial uncertainties exist, especially in terms of the different model hypotheses that these models represent. In a strict Popperian view of science, the testing of GCMs is inadequate, as model structures (i.e. different sets of hypotheses) are not rigorously evaluated. At this point it may be worthwhile to note that these arguments have much in common with those made in earlier commentaries in HPToday. In particular, Beven (2000) has pointed out that, in the case of hydrological modelling, our interest should perhaps be best directed toward departures from our perceived (and in many cases, assumed) models of catchment processes. I will argue here that this is true also of our hydrological models of climate. For instance, the practice of flood frequency analysis, where observations are available, is primarily based on a static concept of climate—the future risk is defined by previously observed flood frequency. With notable exceptions, floods are assumed to be drawn from a single probability distribution model (although, in any given application, there are a number of models available!).