Forecasting extreme precipitation in the central Mediterranean: Changes in predictors' strength with prediction lead time
Nikolaos Mastrantonas, Luca Furnari, Linus Magnusson, Alfonso Senatore, Giuseppe Mendicino, Florian Pappenberger, Jörg Matschullat
Abstract
Abstract This work compares forecasting strategies for improving the prediction of extreme precipitation events at different timescales. Our study focuses on forecasting extreme precipitation in Calabria, southern Italy. This region with complex and abruptly varying topography poses additional challenges in forecasting precipitation. Here, we use gridded observational precipitation data from the E‐OBS dataset, reanalysis data from ERA5, and forecasts from ECMWF reforecasts. We show that different forecasting horizons require different forecasting strategies. For short to medium‐range lead times, post‐processing the forecasted extreme precipitation probabilities provides the most informative outputs for end users. For extended‐range forecasts though, it is beneficial to use the large‐scale weather variability, as depicted based on nine Mediterranean patterns in combination with moisture‐related information, to infer reliable information about extreme precipitation. We present benefits and limitations of the methods based on long‐term statistical analysis using a range of indicators, such as the Brier skill score and its decomposition, and the cost–loss model.