A new artificial intelligence-driven forecasting system is transforming how scientists predict conditions in the Mediterranean Sea, delivering detailed 15-day forecasts in just seconds a task that traditionally required more than an hour of intensive computing.
Developed by researchers at the Centro Euro-Mediterraneo sui Cambiamenti Climatici (CMCC), the high-resolution system known as SeaCast integrates ocean and atmospheric data to generate faster, more energy-efficient predictions. The breakthrough study detailing the model has been published in Scientific Reports.
SeaCast operates at a resolution of approximately 4 kilometres (1/24°), matching the scale of CMCC’s Mediterranean operational forecasting system, MedFS, which is delivered through the Copernicus Marine Service. While conventional numerical systems require around 70 minutes on 89 CPUs to produce a 10-day forecast, SeaCast can generate a 15-day projection in about 20 seconds using a single GPU.
The dramatic reduction in computing time marks a major advance for ocean and climate research. Faster simulations enable scientists to run multiple “what-if” scenarios and probabilistic ensemble forecasts, improving estimates of uncertainty and supporting more informed decision-making in coastal management.
Unlike global AI forecasting models that operate at lower resolutions and rely mainly on ocean data, SeaCast uses a graph-based neural network designed to capture the Mediterranean’s complex coastlines and boundary conditions. The model was trained on high-resolution CMCC Mediterranean reanalysis datasets spanning up to 35 years, allowing it to predict ocean conditions down to depths of 200 metres.
Why Atmosphere Ocean Coupling Matters
A key innovation of SeaCast is the integration of atmospheric forcing variables alongside ocean data during training and forecasting. Researchers found that incorporating atmospheric information significantly improved accuracy, particularly near the ocean surface where interactions between air and sea strongly influence marine dynamics.
Sensitivity tests showed that certain atmospheric inputs contributed substantially to forecast improvements, underscoring the importance of coupling atmospheric and oceanographic processes in regional models.
Real-World Applications Across the Mediterranean
Accurate and timely ocean forecasts are critical for industries and communities across the Mediterranean region, including shipping, aquaculture, environmental monitoring and coastal risk management. Faster predictions allow authorities to anticipate marine heatwaves, storm surges and other ocean-driven hazards, strengthening early warning systems and climate resilience planning.
Researchers at CMCC are now working to integrate SeaCast into operational forecasting chains alongside traditional physics-based models, combining the strengths of AI efficiency with established scientific frameworks.
By dramatically reducing computational costs while improving forecast accuracy, SeaCast sets a new benchmark for regional marine prediction. As climate variability intensifies, such AI-powered systems may play a crucial role in delivering faster, smarter and more reliable ocean forecasts worldwide.