Statistical Downscaling of the Bulk and Tail of Simulated Precipitations

Abstract

Intense precipitation events are projected to become more frequent and severe in the future. Impact studies analyzing these changes typically rely on simulated precipitation data generated by climate models. However, these simulated datasets often exhibit biases and require post-processing before they can be effectively used in impact studies. Standard post-processing techniques aim to align the statistical distribution of simulated precipitation with that of the observed data. This involves making necessary adjustments to the simulated precipitation to ensure a consistent match between the two datasets. While the empirical cumulative distribution function (CDF) is frequently employed for this purpose, its accuracy diminishes in the tail of the distribution, making it unsuitable for extreme values. In this study, we propose an approach for post-processing simulated precipitation based on the use of the extended generalized Pareto (EGP) distribution for modeling the precipitation. Unlike the empirical CDF, the EGP distribution is capable of consistently modeling both the bulk and the tail of the distribution in accordance with the extreme value theory. To demonstrate the efficacy of our method, we apply it to post-process daily simulated precipitation data from two Canadian cities. Additionally, we have made the proposed method, along with other relevant techniques from existing literature, accessible through the open-source Julia package QuantileMatching.jl.