SeaCast is an innovative high-resolution forecasting system for the Mediterranean that harnesses AI to deliver faster and more energy-efficient predictions than traditional models. Unlike existing global AI models, which operate at lower resolutions and primarily rely on ocean data, SeaCast integrates both ocean and atmospheric variables, capturing complex regional dynamics . Its graph-based neural network accounts for intricate coastlines and lateral boundary conditions, overcoming one of the major challenges in regional ocean forecasting.
The model operates at a high resolution of about 4 km (1/24°) , the same resolution as the CMCC Mediterranean operational forecasting system MedFS (which is coupled with a wave model and covers the full ocean depth), delivered through the Copernicus Marine Service , and produces forecasts down to a depth of 200 meters . This is made possible by training the model on CMCC Mediterranean reanalysis data , which are provided at the same resolution and are freely available through the Copernicus Marine website .
SeaCast consistently outperforms the Copernicus operational model over the standard 10-day forecast horizon and extends predictions to 15 days . The efficiency gains are striking: while the operational numerical system requires around 70 minutes on 89 CPUs ( central processing units , conventional processors used in most computers) to produce a 10-day forecast, SeaCast can generate a 15-day forecast in about 20 seconds using a single GPU , a highly efficient processor designed for parallel calculations and widely used in machine learning.
These advancements are crucial for ocean and climate research. For example, SeaCast’s improved computational speed enables rapid “ what-if scenario” testing and probabilistic ensemble forecasts, where multiple simulations are used to better estimate forecast uncertainty – scientific tools that are invaluable not only for research, but also for coastal management and decision-making.
Accurate ocean forecasts are crucial for practical applications such as shipping, aquaculture, environmental monitoring, and coastal risk management. SeaCast’s ability to provide fast, reliable, high-resolution predictions helps anticipate impacts, inform planning, and support proactive measures in the Mediterranean region.
“This achievement demonstrates how bringing together oceanography, atmospheric science, and AI expertise produces tangible results, and can unlock a new generation of regional ocean forecasts,” says Emanuela Clementi , CMCC researcher and co-author of the study. “Combining physical insight with advanced AI allows us to improve forecast accuracy while dramatically reducing computational costs. Interdisciplinary collaboration was essential to tackle problems that neither field could solve alone.”
“Working closely with CMCC was a key part of making SeaCast possible. I learned a great deal about the Mediterranean, and I had the chance to visit both Bologna and Lecce and interact with inspiring researchers at both offices,” says Daniel Holmberg , researcher at University of Helsinki and first author of the study. “Beyond the science, the warm hospitality and the many discussions over coffee made this collaboration especially memorable.”
A key innovation of SeaCast is the integration of atmospheric forcing data alongside ocean variables during the model training, validation process and forecasting. Results show that including atmospheric information substantially improves forecast accuracy, especially near the surface, where ocean dynamics are strongly influenced by the atmosphere. Sensitivity experiments, in which researchers tweak different input variables to see their impact on predictions, further demonstrate which atmospheric variables contribute most to improved marine predictions and how longer training periods – up to 35 years of historical data – enhance model skill.
Looking ahead, CMCC researchers are now working to integrate SeaCast into operational forecasting chains alongside traditional physics-based models, further enhancing the prediction and management of Mediterranean Sea conditions. This first regional, high-resolution AI ocean model sets a new benchmark for marine forecasting and opens a pathway for faster, smarter, and more reliable ocean predictions worldwide.
Scientific Reports
Accurate Mediterranean Sea forecasting via graph-based deep learning