DERISC: Deep learning based Extreme Rainfall and flood warnIngs through Seamless foreCasting​

For the 4th edition of the pysteps hackathon, our team of dedicated developers and scientists gathered in the cozy setting of Chateau d’Avin in Belgium, joined by none other than the godfather of pysteps himself, Alan Seed (Weather Radar NZ). Together, we spent our days coding and our evenings cooking: the perfect recipe for a successful hackathon.

During this intense week of collaboration and scientific exchange, we pushed the boundaries of precipitation nowcasting with major progress in several key areas:

- pysteps v2.0 with xarray: A major leap for interoperability with the open-source radar and ML ecosystem.
- Novel nowcast-NWP blending: Implementing advanced methods like the reduced-space ensemble Kalman filter from Daniele Nerini and colleagues.
- A parametric pysteps nowcast: A new approach developed by Alan Seed
- AI forecast integration: Enabling blending of AI and traditional NWP models.
- Validation of pysteps: A variety of performance scores to monitor nowcasting performance.

Heartfelt thanks to all participants, both on-site and online, for their energy and expertise: Aitor Atencia, Alan Seed, Arthur Moraux, Avijit Majhi, Daniel Eduardo Villarreal Jaime, Daniele Nerini, Elise Verstraeten, Felix Erdmann, Joep Bosdijk, Lars Bonnefoy, Loic Faleu, Martin Rempel, Mats Veldhuizen, Ricardo Reinoso-Rondinel, Ruben Imhoff, zongkui guan, and especially Simon De Kock for the organisation!

We are grateful to BELSPO for their financial support through the DERISC project (Royal Meteorological Institute of Belgium (RMI) - Vrije Universiteit Brussel - KU Leuven). This unique format wouldn't be possible without it.