A Guided Tour of GFDL's Statistical Downscaling Efforts

Downscaling is a means by which to translate climate information produced by large-scale dynamical models to the local scale for use by impacts communities. Downscaling also seeks to reduce systematic model biases. A major focus of GFDL's downscaling team is the evaluation and inter-comparison of various statistical downscaling methods. We expect that our efforts will provide guidance to both users of downscaled data as well as developers of statistical downscaling methods.

This talk introduces our philosophies, goals and approaches to such evaluations, and shows some of our early results. We employ some novel approaches, including "Perfect Model" and "Synthetic Data" frameworks. We are able to address the "stationarity assumption" that is implicit to all statistical downscaling methods. Namely, to what extent are statistical relationships derived from past data valid when applied to future climate change scenarios? The answer to this question proves elusive when employing traditional means.