Addressing the mischaracterization of extreme rainfall in regional climate model simulations – A synoptic pattern based bias correction approach.

Addressing systematic biases in regional climate model simulations of extreme rainfall is a necessary first step before assessing changes in future rainfall extremes. Commonly used bias correction methods are designed to match statistics of the overall simulated rainfall with observations. This assumes that change in the mix of different types of extreme rainfall events (i.e. convective and non-convective) in a warmer climate is of little relevance in the estimation of overall change, an assumption that is not supported by empirical or physical evidence. This study proposes an alternative approach to account for the potential change of alternate rainfall types, characterized here by synoptic weather patterns (SPs) using self-organizing maps classification. The objective of this study is to evaluate the added influence of SPs on the bias correction, which is achieved by comparing the corrected distribution of future extreme rainfall with that using conventional quantile mapping. A comprehensive synthetic experiment is first defined to investigate the conditions under which the additional information of SPs makes a significant difference to the bias correction. Using over 600,000 synthetic cases, statistically significant differences are found to be present in 46% cases. This is followed by a case study over the Sydney region using a high-resolution run of the Weather Research and Forecasting (WRF) regional climate model, which indicates a small change in the proportions of the SPs and a statistically significant change in the extreme rainfall over the region, although the differences between the changes obtained from the two bias correction methods are not statistically significant.

Fig. 3. Classification tree used to separate significant different results (red) from insignificant ones (blue). The tree is pruned to focus on the important details. xi is the criteria used to splitting data at each node. x1 is the absolute change in the proportion of each cluster; x2 is the absolute difference of the correction factors between two clusters, x3 and x4 are the absolute change in quantiles of the corrected AMS for cluster 1 and cluster 2, respectively; and x5 is the KS test decision (i.e. 0 for accepting the null hypothesis, and 1 for rejecting the null hypothesis) on whether the corrected distributions of the two clusters for the future climate are significantly different from each other. The bias corrected distributions using the two methods are plotted for the cases falling into the left (right) branch of the first level of the tree.