Retrospective analysis of the accuracy of SEAS5 ECMWF seasonal forecasts and their bias correction: An application to Sicily

Abstract

Nowadays, seasonal forecasts (SFs) have become indispensable tools in many fields of life and science, including hydrology, hydraulic risk mitigation, and water resource management; their ability to forecast climate variables up to six months in advance is more valuable than ever since the presence of the effects of climate change. SFs play a crucial role not only in predicting floods and droughts but also in the water resource management, where they enhance decision-making processes, mitigate risks, optimize resource use, and secure water supplies, particularly in areas vulnerable to water scarcity and climate variability (Francipane et al. 2023). Despite their importance, the reliability of SFs remains a key concern. Biases inherent in climate models can compromise forecast accuracy, making retrospective analyses essential for evaluating forecasts by comparing them with observed values. When systematic errors are detected, various bias correction techniques, such as quantile mapping, linear regression, mean bias correction, the MOS-Analog method, and mean and variance adjustment, can be applied to adjust the forecasts (Marcos et al. 2018). In addition to the previous methods, this study also explores an innovative approach using Artificial Neural Networks (ANN), which may offer superior performance by capturing the nonlinear relationships (Moghim and Bras 2017). Here ECMWF SEAS5 SFs relative to monthly precipitation, P, and temperature, T, have been analyzed on different areas of Sicily, which include four river basins whose reservoirs serve the Palermo metropolitan area's water system, specifically Piana degli Albanesi, Rosamarina, Poma, and Scanzano; for these river basins the SFs have been compared with observed precipitation and temperature data provided by the Sicilian regional basin authority (AdB).