Parametric post-processing of core and extreme values

Summary

Intense precipitation events are likely to increase in intensity and frequency in Canada. The study of the impacts of these changes is generally based on precipitation simulated by climate models. These are generally biased and must be post-processed before being used in impact studies.

The most common method of post-processing is to modify the simulated precipitation so that its distribution matches that of the observed precipitation. The empirical distribution function is often used but the accuracy of this estimate decreases in the tail of the distribution. It is therefore not suitable for post-processing extreme values. We propose to use the extended generalized Pareto distribution for the post-processing of simulated rainfall. This law allows modeling extreme values within the framework of extreme value theory while adding flexibility for the core of the distribution. Parametric post-processing using the extended Pareto law allows for correction of both the core and tail of the distribution. The proposed approach will be illustrated for the post-processing of the simulated precipitation of the ClimEX ensemble at grid points containing the cities of Montreal and Toronto.