Speaker
Description
In this work we introduce and investigate the zero degree of freedom non-central Chi Squared distribution for ensemble postprocessing. It has a point mass at zero by definition and is thus particularly suited for postprocessing weather variables naturally exhibiting large numbers of zeros, such as precipitation, solar radiation or lightnings. Due to the properties of the distribution, no additional truncation or censoring is required to obtain a positive probability at zero. The presented study investigates its performance compared to that of the censored generalized extreme value distribution (GEV0) and the censored and shifted gamma distribution (CSG0) for postprocessing 24h accumulated precipitation at 31 stations in Germany using an Ensemble Model Output Statistics (EMOS) approach with a rolling training period. The case study shows that the Chi Squared distribution is highly competitive to state-of-the-art distributions, specifically when predicting extreme precipitation events.
