Statistical downscaling of a climate simulation of the last glacial cycle: temperature and precipitation over Northern Europe

Abstract. Earth system models of intermediate complexity (EMICs) have proven to be able to simulate the large-scale features of glacial–interglacial climate evolution. For many climatic applications the spatial resolution of the EMICs' output is, however, too coarse, and downscaling methods are needed. In this study we introduce a way to use generalized additive models (GAMs) for downscaling the large-scale output of an EMIC in very different climatological conditions ranging from glacial periods to current relatively warm climates. GAMs are regression models in which a combination of explanatory variables is related to the response through a sum of spline functions. We calibrated the GAMs using observations of the recent past climate and the results of short time-slice simulations of glacial climate performed by the relatively high-resolution general circulation model CCSM (Community Climate System Model) and the regional climate model RCA3 (Rossby Centre regional Atmospheric climate model). As explanatory variables we used the output of a simulation by the CLIMBER-2 (CLIMate and BiosphERe model 2) EMIC of the last glacial cycle, coupled with the SICOPOLIS (SImulation COde for POLythermal Ice Sheets) ice sheet model, i.e. the large-scale temperature and precipitation data of CLIMBER-2, and the elevation, distance to ice sheet, slope direction and slope angle from SICOPOLIS. The fitted GAMs were able to explain more than 96% of the temperature response with a correlation of >0.98 and more than 59% of the precipitation response with a correlation of >0.72. The first comparison with two pollen-based reconstructions of temperature for Northern Europe showed that CLIMBER-2 data downscaled by GAMs corresponded better with the reconstructions than did the bilinearly interpolated CLIMBER-2 surface temperature.