Downscaling and uncertainty portrayal: a drought-projection example from Alaska

In many cases downscaling the output from general circulation models (GCMs) is a critical step in answering questions about the impacts of climate change. There may also be trade-offs. Practical limitations such as the amount of work necessary to downscale model output, limited high-resolution observational datasets, and large file sizes often mean that downscaling is applied to a smaller number of models, runs, and variables than are available from GCMs. Here we compare projected changes in potential evapotranspiration (PET) for the state of Alaska, based on high-resolution (2km), downscaled average monthly temperature from five GMCs (CCCMA-CGCM3.1 (t47), GFDL-CM2.1, MIROC3.2 (medres), MPI-ECHAM5, and UKMO-HadCM3) to a coarser resolution analysis based on monthly average temperature and radiation fluxes from the same set, as well as a broader suite of GCMs, using the Priestley-Taylor method. Downscaled radiation data were not available, as there are no high-resolution observed radiation datasets available as the basis for downscaling. Thus, components of the radiation-balance for the downscaled PET dataset were estimated from historical maximum and minimum temperature combined with changes in average temperature. PET based on downscaled data increases in all cases because the limited number of input variables reduces PET to a function of temperature, which increases in all five models. In contrast, PET calculated directly from simulated radiation fluxes increases in some models but decreases in others. Lack of long-term datasets for these variables makes their simulation difficult to evaluate, so it may be more challenging to understand the full range of uncertainty in projections derived from them. This could be quite important given the complexity of simulating clouds. High-resolution climate projections are desirable, as are projections based on our best and fullest understanding of the future, and it is important to understand how the choices we make about its use, as well as inherent uncertainties in climate model output, influence what we present to managers.