Precipitation by a regional climate model and bias correction in Europe and South Asia

Abstract

Because coarse-grid global circulation models do not allow for regional estimates of the water balance or trends of extreme precipitation, downscaling of global simulations is necessary to generate regional precipitation. This paper applies for downscaling the regional climate model CLM as a dynamical downscaling method (DDM) and two statistical downscaling methods (SDMs). Because the SDMs neglect information available to the DDM, and vice versa, a combination of the dynamical and statistical approaches is proposed here. In this combined approach, a simple statistical step is carried out to correct for the regional model biases in the dynamically downscaled simulations. To test the proposed methods, coarse-grid global re-analysis data (ERA40 with ∼ 1.125° grid spacing) is downscaled in two regions with different climatology and orography: one in South Asia and the other in Europe. All of the methods are tested on daily precipitation with 0.5° grid spacing. The SDMs are generally successful: the standardized root mean square error of rain day intensity is reduced from ERA40's 0.16 to 0.10 in a test area to the west of the European Alps. The CLM simulations perform less well (with a corresponding error of 0.14), but represent a promising approach if the user requires flexibility and independence from observational data. The proposed bias correction of the CLM simulations performs very well in European test areas (better than or at least comparable with the SDMs; i.e., with a corresponding error of 0.07), but fails in South Asia. An investigation of the observed and simulated precipitation climate in the test areas shows a strong dependence of the bias correction performance on sampling statistics (i.e., rain day frequency) and on the robustness of bias estimation.