Extreme precipitation shows non-stationarity, meaning that its distribution can change with time or other large-scale variables. For a classical frequency-intensity analysis this effect is often neglected. Here, we propose a model including the influence of North Atlantic Oscillation, time, surface temperature and a blocking index. The model features flexibility to use annual maxima as well as seasonal maxima to be fitted in a generalized extreme value setting. To further increase the efficiency of data usage, maxima from different accumulation durations are aggregated so that information for extremes on different time scales can be provided. Our model is trained to individual station data with temporal resolutions ranging from one minute to one day across Germany. Models are chosen with a stepwise BIC model selection and verified with a cross-validated quantile skill index. The verification shows that the new model performs better than a reference model without large-scale information. Also, the new model enables insights into the effect of large-scale variables on extreme precipitation. Results suggest that the probability of extreme precipitation increases with time since 1950 in all seasons. High probabilities of extremes are positively correlated with blocking situations in summer and with temperature in winter. However, they are negatively correlated with blocking situations in winter and temperature in summer.
