A1373
Title: Incorporating climatological constraints into statistical models to improve the post-processing of extremes
Authors: Bastien Francois - KNMI (Netherlands) [presenting]
Harun Kivril - KNMI (Netherlands)
Maurice Schmeits - The Royal Netherlands Meteorological Institute (Netherlands)
Kirien Whan - KNMI (Netherlands)
Philippe Naveau - CNRS-IPSL (France)
Abstract: Extreme events, such as extreme wind gusts, can generate huge impacts on society, and anticipating them is essential for taking preventive measures. Ensemble forecasts exhibit biases and under-dispersion and have to be calibrated using observations before being used. Several statistical post-processing methods have, therefore, been developed and applied. Among them, some are non-parametric and are thus not able to predict beyond the largest value observed during training. To overcome this problem, parametric distributions such as those from extreme value theory (EVT) can be fitted on post-processed outputs to allow extrapolation. However, depending on the outputs, the extrapolation may sometimes be insufficient. A new method is proposed to enable the extrapolation of post-processed outputs by forcing the fitted EVT distributions to have an upper bound aligned with the maximum of the climatological records at the stations. The proposed method is applied to forecasts of 6-hourly maximum wind gusts from 2021 to 2024 over the Netherlands using the ECMWF-IFS ensemble data. Results are compared against several state-of-the-art post-processing methods. The proposed algorithm shows skill improvements for wind gust extremes depending on lead times, stations and thresholds while maintaining performances for intermediate values. It encourages further research on adding better constraints to methods for the post-processing of extremes.
