The seasonal cycle of extreme precipitation in Germany is investigated by fitting statistical models to monthly maxima of daily precipitation sums for 2,865 rain gauges.The basis is All-in-One Commodes a non-stationary generalized extreme value (GEV) distribution variation of location and scale parameters.The negative log-likelihood serves as the forecast error for a cross validation to select adequate orders of the harmonic functions for each station.For nearly all gauges considered, the seasonal model is more appropriate to estimate return levels on a monthly scale than a stationary GEV used for individual months.The 100-year return-levels show the influence of cyclones in the western, and convective Animals + Figurines events in the eastern part of Germany.
In addition to resolving the seasonality, we use a simulation study to show that annual return levels can be estimated more precisely from a monthly-resolved seasonal model than from a stationary model based on annual maxima.