CMCC has developed MedFormer, a novel AI assisted model that predicts Mediterranean Sea conditions up to nine days in advance with more accuracy and faster than traditional models. The new AI ocean emulator can make ocean forecasting more precise and efficient, providing information and data that might support marine safety, environmental protection, and coastal management by aiding in route planning, risk prevention, and response to storms, pollution, or heatwaves.

Accurate and quick ocean forecasts support many practical needs, from increased marine safety through improved route planning and risk prevention for ships and coastal operations, to environmental protection and coastal management that helps authorities respond to storms, pollution, or heatwaves.

Research by CMCC has led to the development of MedFormer, a fully data-driven deep learning model specifically designed for medium-range ocean forecasting in the Mediterranean Sea.

Consistently outperforming the state-of-the-art Mediterranean Forecasting System (MedFS), MedFormer highlights the potential of data-driven approaches to complement, or even surpass, traditional numerical ocean forecasting systems in both accuracy and efficiency.

On average, MedFormer reduced forecast errors by 10–20% compared to MedFS, depending on the variable and forecast duration. Moreover, while traditional models’ accuracy worsens quickly after a few days, MedFormer’s errors grew more slowly, meaning its forecasts remained reliable even after a week. Finally, because it’s AI-based, MedFormer can produce forecasts much faster and with less computing power than physics-based models, which normally need large supercomputers.

Traditional forecasting systems rely on complex physics-based simulations that require a lot of computer power and time. MedFormer, on the other hand, “learns” directly from past data and can run predictions in a fraction of the time. “This advancement matters because accurate and quick ocean forecasts support many practical needs,” says Epicoco. These include: Marine safety, improving route planning and risk prevention for ships and coastal operations; environmental protection, helping monitor sea temperature and salinity, which are linked to marine ecosystems and climate change; and coastal management, supporting authorities in responding to storms, pollution, or heatwaves in the sea.

The color blue dominates the charts, revealing how MedFormer outperforms MedFS for nearly every combination of depth and forecast day. Source: Epicoco et al., 2025

MedFormer operates at a very high horizontal resolution, much finer than most AI-based ocean models developed so far. This allows it to capture small-scale processes, like local currents and eddies, which strongly influence regional climate, marine life, and coastal conditions.

Moreover, the model doesn’t just forecast one day at a time but it learns to predict multiple days ahead by using its own previous outputs as new inputs. This makes the model more stable and realistic for longer forecasts (up to 9 days), reducing the error accumulation seen in other AI systems.

“We’ve taken something that used to require enormous computing power and time, and made it faster and more accurate. It’s a concrete example of how data and AI can help us understand and protect the Mediterranean Sea more efficiently,” concludes Epicoco.

For more information:

Epicoco, Italo, Davide Donno, Gabriele Accarino, Simone Norberti, Alessandro Grandi, Ronan McAdam, Donatello Elia et al. “MedFormer: a data-driven model for forecasting the Mediterranean Sea.” (2025). https://doi.org/10.48550/arXiv.2509.00015