Can bias correction methods improve the statistical characterization of extreme rainfall compound events in climate simulations?

Abstract. Compound events (CEs), commonly defined as the “combination of multiple drivers and/or hazards that contributes to societal or environmental risk”, often result in amplified impacts compared to individual hazards. To understand their evolution in terms of frequency under climate change, the outputs of climate simulations are used. Climate simulations are often statistically biased which can affect the representation of CEs. Hence, this study examines to what extends bias correction methods, including multivariate ones, improve the statistical characterization of CEs. It also aims at determining whether their evolution under climate change can be preserved by these methods. Two extreme rainfall events triggered by accumulated precipitation have been selected and analyzed either with a multivariate generalized Pareto modeling or a copula-based modeling. Two multivariate bias correction methods (dOTC and R2D2) and one univariate bias correction method (CDF-t) are applied to bias correct simulations from 10 global climate models. Bias corrected and raw data are compared in terms of return periods and in terms of extremal dependence structure. The results show that bias correction methods improve the representation of the two studied CEs and that the sign of their evolution is preserved in most cases.