Weather and climate models rely on parametrisation schemes to represent sub-grid processes that cannot be explicitly resolved on the computational grid. Many such traditional schemes depend on fixed coeffici ents that are only weakly constrained and tuned offline\, often locking in persistent biases and limiting adaptability across regimes\, resolutions\ , and climates. \n\nIn this talk\, I present a strategy that reframes part of parametrisation design as a sequential control problem by embedding a reinforcement learning (RL) agent within the model\, allowing it to observ e the evolving state and update selected tunable components online during integration. We evaluate this approach across a hierarchy of idealised env ironments\, from a simple single-parameter bias-correction setting to mult i-parameter zonal energy balance models (EBMs)\, exploring both single-age nt and federated multi-agent configurations\, the latter mirroring the spa tial decomposition used in general circulation models. \n\nAcross these se ttings\, we find RL-assisted parameter updates consistently reduce area-we ighted RMSE relative to static tuning\, with the largest gains emerging in tropical and mid-latitude bands\, while federated training accelerates co nvergence and enables geographically specialised control without sacrifici ng physically meaningful parameter adjustments. Overall\, results from the se idealised setups suggest that RL provides a viable pathway toward regim e-aware\, state-dependent parametrisations and a scalable framework for on line learning within numerical weather and climate models.\n\n
<
/p>
*NOTE - In person &\; ZOOM ONLINE TALK*\nMore information includi ng the Zoom link can be found on the "ICCS website":https://iccs.cam.ac.uk /events/journal-club-rethinking-weather-and-climate-model-parametrisations -reinforcement-learning\n\n
LOCATION:Centre for Mathematical Sciences\, MR5 END:VEVENT END:VCALENDAR