Climate scenarios of extreme precipitation using a combination of parametric and non-parametric bias correction methods in the province of Québec

Abstract

Realistic simulation of heavy precipitation in climate simulations is a major challenge for adaptation, as the grid resolution of most climate models is too coarse to explicitly resolve convective processes. When proper future extreme precipitation events are required, such as for adaptation to future flooding, users therefore rely on bias-corrected precipitation data. However, the commonly used quantile–quantile mapping procedure is not well suited to post-process distribution tails. As a response to a need expressed by the province of Québec for the purpose of the government’s INFO-Crue project, which aims in part to provide a better understanding of future floods and incorporate this information in flood mapping, a new, mixed method was proposed for post-processing the entire range of precipitation, including heavy precipitation. The method uses a non-parametric quantile–quantile mapping procedure for the bulk distribution and a parametric procedure based on extreme value theory for the right tail. The method’s performance is illustrated on a watershed in Québec (Canada), using external Generalized Extreme Value (GEV) parameters. Results show that the proposed method is able to keep the important characteristics of simulated distribution tails, such as the initial ranking and scaling between values, keep spatial coherence and provide robust estimates of high return levels. The proposed method represents a flexible framework that relies on the quantile–quantile mapping procedure that is trusted by the end users, while incorporating information from the statistical community where necessary to ensure that heavy precipitation that might drive flooding, such as the 20or 100-year 24h precipitation, is bias-corrected in a more robust manner. The method is available in the open-source package ClimateTools.jl written in Julia and Python’s xclim package.