Influence of Bias Correcting Predictors on Statistical Downscaling Models

Abstract

AbstractStatistical downscaling models (SDMs) and bias correction (BC) methods are commonly used to provide regional or debiased climate projections. However, most SDMs are utilized in a “perfect prognosis” context, meaning that they are calibrated on reanalysis predictors before being applied to GCM simulations. If the latter are biased, SDMs might suffer from discrepancies with observations and therefore provide unrealistic projections. It is then necessary to study the influence of applying bias correcting to large-scale predictors for SDMs, since it can have impacts on the local-scale simulations: such an investigation for daily temperature and precipitation is the goal of this study. Hence, four temperature and three precipitation SDMs are calibrated over a historical period. First, the SDMs are forced by historical predictors from two GCMs, corrected or not corrected. The two types of simulations are compared with reanalysis-driven SDM outputs to characterize the quality of the simulations. Second, changes in basic statistical properties of the raw GCM projections and those of the SDM simulations—driven by bias-corrected or raw predictors from GCM future projections—are compared. Third, the stationarity of the SDM changes brought by the BC of the predictors is investigated. Changes are computed over a historical (1976–2005) and future (2071–2100) time period and compared to assess the nonstationarity. Overall, BC can have impacts on the SDM simulations, although its influence varies from one SDM to another and from one GCM to another, with different spatial structures, and depends on the considered statistical properties. Nevertheless, corrected predictors generally improve the historical projections and can impact future evolutions with potentially strong nonstationary behaviors.