Examining the Performance of Statistical Downscaling Methods: Toward matching applications with the right tool

Keith Dixon

NOAA GFDL

Tuesday, Aug 08, 2017, 2:00 pm
DSRC Room 1D403

Abstract

Despite advancements in global climate model (GCM) spatial resolution and simulation quality, raw GCM output is often deemed to be inappropriate for direct use in many climate impacts studies. Statistical downscaling (SD) methods represent one approach to addressing GCM shortcomings via processing that gleans information from a combination of observations and the climate change response simulated by a dynamical model. All SD methods contain an underlying assumption that statistical relationships based on historical observations will apply equally well when used to refine historical model simulations and future climate projections. However, lacking observations of the future, there is no straightforward way to determine how a SD method’s skill might diminish when applied to future scenarios. For these reasons, we have developed a perfect model framework in which high resolution models serve as a proxy for observations, thereby allowing quantitative assessments of SD method skill both for the contemporary climate and for future projections. This presentation will describe the perfect model approach and present results illustrating how SD method performance can vary by location, season, downscaling method, climate variable of interest, amount of projected climate change, and whether one is interested in central tendencies or the tails of the distribution – a set of factors that are relevant when determining whether a particular SD data product is well matched to a climate impacts application.