Evaluating Statistical Downscaling Performance Under Changing Climatic Conditions

English Abstract:

A broad range of empirical statistical downscaling (ESD) methods of varying levels of complexity exist, all of which may be used to refine Global Climate Model (GCM) projections via processes that glean information from a combination of observations and GCM-simulated climate change responses. ESD output files are viewed as value-added products – deemed to be more suitable for downstream applications than the raw GCM results from which they are derived. Yet, different ESD methods have different performance characteristics, indicating that ESD processing introduces uncertainties of its own. Often, assumptions that may limit the suitability of statistically downscaled projections for specific decision-support applications are not well-conveyed to researchers interested in incorporating high resolution climate projections into regional climate change impacts studies. We are developing a framework to systematically evaluate performance characteristics of ESD methods. For example, there is an inherent stationarity assumption that presumes ESD techniques perform as well in the future as during the observed period – a difficult to evaluate assumption, given the lack of future observations. A “perfect model” experimental design quantifies aspects of ESD method performance both for a historical period and for late 21st century climate projections, thereby quantitatively assessing how well the stationarity assumption holds. Results illustrate how ESD performance varies geographically, by time of year, variable of interest, amount of climate change, and is dependent upon the specific ESD method used. By revealing sensitivities, strengths and weaknesses, these evaluations contribute to inform ESD method improvements and can provide guidance regarding aspects of the confidence one should attribute to different statistically downscaled climate projections used for regional climate change impacts studies.