FlowCast is the first end-to-end probabilistic model leveraging Conditional Flow Matching (CFM) as a direct noise-to-data generative framework for precipitation nowcasting. The approach outperforms state-of-the-art diffusion models while leveraging CFM's inherent efficiency, requiring significantly fewer sampling steps.

This is an extension of Bernardo's master's thesis. His thesis was about Deep Learning for Precipitation Nowcasting in Slovenia, where he identified the novel application of Conditional Flow Matching as the best approach, after investigating a better probabilistic approach than the previous State of the Art diffusion models.

After additional experiments with his supervisor, Asst. Prof. Dr Jana Faganeli Pucer, they converted the thesis results into a paper format and submitted it to the International Conference on Learning Representations (ICLR) 2026 in Rio de Janeiro (Brazil).

Congratulations, Bernardo!