Study region: The predictive capabilities of seasonal forecasts (SFs) have been tested for Sicily Island (Italy), located in the central Mediterranean. Four case studies were considered, involving river basins feeding the artificial reservoirs supplying the metropolitan area of Palermo. Study focus: The study evaluates ECMWF SEAS5 Seasonal Forecasts (SFs) for temperature and precipitation and compares several bias correction (BC) techniques to enhance forecast accuracy. Traditional methods, including Quantile Mapping, Linear Regression, Mean Bias Correction, and Mean and Variance Adjustment, are compared with an Artificial Neural Network (ANN) approach applied via three strategies: Full Ensemble Median (FEM), Separated Monthly Median (SMM), and Individual Member Separated Monthly (IMSM) corrections. Retrospective analyses assess systematic biases and the performance of BC methods across different lead times. New forecasting insights for the region: Results indicate that raw SFs tend to overestimate low temperatures and underestimate high precipitation. Among BC methods, the ANN combined with IMSM achieves the best performance, substantially improving forecast accuracy while preserving ensemble variability. For precipitation, RMSE is reduced by up to 45 %, and probabilistic forecast skill, measured via the Continuous Ranked Probability Score, is significantly enhanced. These findings highlight the potential of advanced BC methods to generate reliable SFs, providing actionable information for proactive water resource management and climate adaptation in Sicily. • Raw SFs overestimate low temperatures and underestimate heavy precipitation. • ANN bias correction excels for temperature; simpler methods also work well. • ANN Bias Correction sharply reduces RMSE in precipitation forecasts. • Forecast skill drops with longer lead times, especially for precipitation.
Castaldo et al. (Tue,) studied this question.