- 1Department of Environmental Engineering, University of Calabria, Rende (CS), Italy (alfonso.senatore@unical.it)
- 2IMK-IFU, Karlsruhe Institute of Technology (KIT), Campus Alpin, Garmisch-Partenkirchen, Germany
- 3Department of Water Resources Engineering, Lund University, Lund, Sweden
- 4National Center for Monitoring and Early Warning of Natural Disasters, Brazil
- 5Institute of Geography, Augsburg University, Augsburg, Germany
- 6Marine and Freshwater Solutions, Finnish Environment Institute (Syke), Helsinki, Finland
Water scarcity and drought, worsened by climate change, are growing threats to both the economy and society, particularly in hotspots like the Mediterranean Basin. Effective water resources management requires timely forecasting to implement appropriate countermeasures. Seasonal forecasts, based on knowledge-based models, are essential for managing water scarcity and drought risks. These forecasts have become more reliable with advancements in weather modeling systems. Currently, global seasonal forecasts are provided by various centers, such as the Copernicus Climate Change Service (C3S). However, the success of Artificial Intelligence encourages a shift towards a data-driven approach, which moves away from traditional knowledge-based models, relying instead on machine learning to improve forecast accuracy.
This study compares two advanced seasonal forecast models, one knowledge-based and the other data-driven, for the Calabrian peninsula in southern Italy. The knowledge-based, process-based model uses the SEAS5 ensemble forecasts from the ECMWF, with precipitation predictions disaggregated to a higher resolution (around 9 km) and bias-corrected according to the ERA5-Land product for improved accuracy. The data-driven approach predicts future precipitation using time series from 134 local rain gauges, employing methods like Gaussian process regression (GPR), support vector machines (SVM), and feed-forward neural networks (FFNN). The models’ performance is evaluated for the 2021-2023 period using indices such as bias, RMSE, and Pearson correlation coefficient across different spatial areas. Furthermore, the output of both approaches is used for further hydrological modeling.
The results show a high level of consistency between the two techniques and their respective reference datasets, emphasizing the significant potential of combining both approaches. This integration allows for the utilization of their individual strengths, such as probabilistic forecasting and physical consistency with other variables in knowledge-based methods, as well as flexibility and computational efficiency in data-driven models.
Acknowledgments: This study was funded by The Next Generation EU - Italian NRRP, Mission 4, Component 2, Investment 1.5, call for the creation and strengthening of ‘Innovation Ecosystems’, building ‘Territorial R&D Leaders’, Project Tech4You - Technologies for climate change adaptation and quality of life improvement, n. ECS0000009; The Next Generation EU - Italian NRRP, Mission 4 ‘Education and Research’ - Component C2, Investment 1.1, Research Project of National Interest (PRIN 2022 PNRR) - INnovative FOrecast-informed REServoir operations for sustainable use of water resources and climate change adaptation (INFORES, CUP H53D23001430006), Italian Ministry of University and Research; The Next Generation EU - Italian NRRP, Mission 4, Component 2, Investment 1.3, project WaterWISE - Water Management Strategies and Climate Change Adaptation in Southern Italy, n. PE00000005, CUP D43C22003030002.
How to cite: Senatore, A., Furnari, L., Nikravesh, G., Cortale, F., Lorenz, C., Naghibi, A., Cuartas, A., Kunstmann, H., Bertacchi Uvo, C., and Mendicino, G.: Performance comparison of knowledge-based and data-driven approaches for seasonal meteo-hydrological forecasts in the central Mediterranean, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7157, https://doi.org/10.5194/egusphere-egu25-7157, 2025.