Original title: Neptune: An AI model for Global Ocean Subseasonal Prediction
Davide Donno, Italo Epicoco, Massimo Cafaro, Gabriele Accarino +7 · September 2026
Predicting ocean weather weeks to months ahead is vital for managing water, farming, energy, and disasters. Traditional computer models use physics equations to simulate ocean currents. However, they demand massive computing power and have complex code that is difficult to update. To solve this, researchers built Neptune, an artificial intelligence system that forecasts ocean and sea-ice changes up to 60 days ahead. Neptune works by pairing two machine learning tools: convolutional neural networks, which spot local details, and spherical Fourier neural operators, which track large-scale global patterns. Driven by daily atmospheric conditions, it predicts daily water temperature, salt levels, currents, sea surface height, and sea ice at the surface and deep underwater. The researchers found that Neptune successfully matches real ocean patterns over 60 days while staying stable over long periods. This shows that data-driven artificial intelligence can reliably track complex ocean behavior at fine detail.
You are packing for a two-week mountain hike. The trail crosses rivers and snowfields that change with the weather each day, so a plan made once at the trailhead will be wrong by the time you reach the high passes. In this paper, the ocean is the trail system, the daily weather is the input from the atmosphere, and Neptune is the guide who estimates the full ocean state — currents, temperature, ice — 60 days ahead without manually tracking every physical rule.
Decisions about water, farming, disaster response, energy, and insurance depend on forecasts weeks to months ahead. Traditional physics-based ocean models are computationally expensive and so complex that improving them is a struggle. Neptune offers a data-driven way to emulate the global ocean and sea ice daily, up to 60 days out, at high resolution — evidence that this approach can support next-generation forecasting.
The abstract doesn't say how Neptune compares head-to-head with traditional physics-based forecast models, what computational resources it needs, or where it makes mistakes. It only emulates the ocean and sea-ice component: because atmospheric fields must be prescribed as input, Neptune does not forecast the atmosphere itself.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Neptune: An AI model for Global Ocean Subseasonal Prediction — Davide Donno, Italo Epicoco, Massimo Cafaro et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.