On the Use of Statistical Downscaling to Improve the Skill in Decadal Predictions of Temperature over the Continental United States

Abstract:

Increases in global temperature over recent decades and projected acceleration in warming trends over the 21^st century are established consequences of climate change. Possible manifestations of altered climate (e.g., more frequent, intense, and persistent temperature extremes) have resulted in a strong need to obtain information about future conditions. Under these circumstances, skillful decadal temperature predictions (DTPs) can have profound societal and economic benefits through informed planning and response. However, skillful and actionable DTPs are extremely challenging to achieve. Even though General Circulation Models (GCMs) provide decadal predictions of a number of climate variables, the direct use of GCM simulations is not encouraged because of the limited skill they exhibit. To address these shortcomings, we apply a statistical downscaling methodology to increase the GCMs’ skill in predicting decadal temperature over the continental United States. Here, we use kernel regression to establish statistical relationships between coarse resolution climate variables (predictors) and the fine resolution climate variable of interest (predictand). The climate variables used as predictors are obtained from National Centers for Environmental Prediction and the National Center for Atmospheric Research (NCEP/NCAR) reanalysis data, and ‘Parameter-elevation Relationships on Independent Slopes Model’ (PRISM) temperature data are used as predictand. Statistical relationships established over calibration period (1961-1990) are applied to retrospective decadal predictions by GCMs. The skill is quantified using a diagnostic skill score that allows the evaluation of the potential skill and conditional and unconditional biases associated with these predictions. Our approach has led to significant improvements in the prediction skill and to the removal of the biases. Results are discussed for predictions at different spatial and temporal scales.