Item

Abstract

Extratropical storms are one of the major coastal hazards in the North Sea region and a substantial driver of coastal protection efforts. Skillful seasonal and decadal predictions of storm activity on a regional scale are still a challenging task, even with today’s state-of-the-art modeling capabilities. While recent studies have shown that predictions of high storm activity with large-ensemble models can be skillful for averaging periods of five years or more, the predictive skill for shorter lead times, such as the next winter, is still low and does not consistently outperform simple statistical prediction methods. In our study, we demonstrate how physical connection-based ensemble refinement can increase the prediction skill of a large-ensemble decadal prediction system based on the Max-Planck-Institute Earth System Model (MPI-ESM) for German Bight winter storm activity on seasonal-to-sub-decadal time scales. We identify physical atmospheric and oceanic predictors of German Bight winter storm activity from near-surface station observations and the ERA5 reanalysis. We then use these predictors to create first-guess predictions of winter storm activity and refine our model ensemble by discarding ensemble members that strongly disagree with the first-guess prediction. With this technique, we are able to significantly increase the prediction skill for winter storm activity at lead times of one to three years.