Probabilistic ocean modeling is essential for decisions that depend on the range of possible future ocean states, not just one forecast. OceanFlow is a new method that addresses this by learning the conditional distribution of future ocean states.

The model is built on flow matching, a generative approach that constructs a probability path from a simple prior to the target distribution. This lets OceanFlow generate coherent samples of future ocean conditions, providing a richer picture than a single deterministic prediction.

By outputting a distribution rather than a point estimate, OceanFlow is designed to support risk-aware marine operations, climate analysis, and scientific decision-making. The paper is available on arXiv under identifier 2610.15518.