Evaluation of High-Resolution Ensemble Precipitation Forecasts for Early Flood Warnings in Small-Scale River Basins during the Heavy Rainfall Period 2016 in Germany

摘要

In recent years, ensemble-based precipitation forecasts from numerical weather models are increasingly used for operational flood warnings in Germany. However, studies which assess the quality of such forecasts for indicators of early flood warnings like the areal precipitation are still limited. Therefore, this paper presents an evaluation of the COSMO-DE-EPS precipitation forecasts for areal precipitation and its spatial variability of eleven small-scale river basins (42 km(2) to 746 km(2)) located in a mountainous region in Rhineland-Palatinate, Germany. The study period covers the heavy rainfall period in early summer of 2016, where extreme rainfall and floods events occurred in the study region. There was a good temporal agreement between the forecasted and observed areal precipitation (interpolated rain gauge data as well as weather radar), even on an hourly scale. Thus, COSMO-DE-EPS provided relatively reliable forecasts for this period, even for shorter lead times (< 15 hours). However, several shortcomings were also uncovered, which need to be further examined in future studies. The accuracy of the ensemble forecasts did not improve with decreasing lead times, although the ensemble spread decreased significantly. Furthermore, all the ensemble members often underestimated the observations (negative bias), especially for longer lead times. This is a critical point, since extreme events are either forecasted too late or, in the worst case, not at all. The investigation of the spatial precipitation variability showed that the ensemble members can reproduce the observed precipitation variability in the individual catchments much better than the ensemble mean. This is to be expected, as the averaging of the ensemble forecasts leads to a smoothing of the precipitations fields, especially for longer lead times for which the ensemble spread is usually larger. An ensemble-based flood forecasting system should therefore not rely on the ensemble mean of the precipitation forecast, but must incorporate all ensemble members for subsequent hydrological predictions.