Impact of bias correction of regional climate model boundary conditions on the simulation of precipitation extremes.

An accurate description of changes in extreme rainfall events requires high resolution simulations. Regional climate models (RCMs), where GCM data are used to provide input boundary conditions, are widely used as a way to resolve finer spatial scale phenomena. A problem with this, however, is that the inherent systematic biases within the GCM simulation are transferred to the RCM through the model boundaries. In this work we focus on the impact of bias correction of lateral and lower boundary conditions on simulated extreme rainfall events. Here three bias correction approaches are investigated. In increasing order of complexity, these are corrections for the mean, mean and variance, and the nested bias correction (NBC) approach that also corrects for lag-1 autocorrelations at nested timescales. These corrections are implemented on six-hourly GCM data taken from the GCM simulations which are used to drive the RCM along the RCM lateral boundaries. To evaluate the performance of bias correction on simulation of extreme rainfall events, daily precipitation extremes indices from the World Meteorological Organization (WMO) Expert Team on Climate Risk and Sectoral Climate Indicators (ET-CRSCI) are used. The results show that bias correction on the boundary conditions produce the results in significant improvement in extremes indices. It is clear that sea surface temperature (SST) plays an important role in driving the simulation. The results indicate that within the domain (far from boundaries) the errors in precipitation extremes are strongly dependent on the RCM, with a smaller effect coming from changes in the lateral boundary conditions.

Figure 3. Comparison of annual mean extremes indices (i.e., R95p, R99p, R95pTOT, and R99pTOT) using a bias map covering the Australian landmass comparing ERA-I-driven WRF outputs to RCM(CSIRO), RCM( M ), RCM ( M,s), RCM(NBC), and RCM(NBC-NoSST). Stippling indicates the bias at the 5% significance level