to
Online
About
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 coefficients that are only weakly constrained and tuned offline, often locking in persistent biases and limiting adaptability across regimes, resolutions, and climates.
In 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 observe the evolving state and update selected tunable components online during integration. We evaluate this approach across a hierarchy of idealised environments, from a simple single-parameter bias-correction setting to multi-parameter zonal energy balance models (EBMs), exploring both single-agent and federated multi-agent configurations, the latter mirroring the spatial decomposition used in general circulation models.
Across these settings, we find RL-assisted parameter updates consistently reduce area-weighted RMSE relative to static tuning, with the largest gains emerging in tropical and mid-latitude bands, while federated training accelerates convergence and enables geographically specialised control without sacrificing physically meaningful parameter adjustments. Overall, results from these idealised setups suggest that RL provides a viable pathway toward regime-aware, state-dependent parametrisations and a scalable framework for online learning within numerical weather and climate models.
About the speaker
Pritthijit Nath is a PhD student at the University of Cambridge, jointly working with the UK Met Office through the AI4ER CDT. His research focuses on integrating reinforcement learning into physical parametrisations to enhance weather and climate models, with a strong emphasis on maintaining physical consistency and interpretability. He has developed scalable testbeds for evaluating RL algorithms and is now applying them within the Unified Model framework.
With a background spanning machine learning, climate science, and high-performance computing, Pritthijit’s work aims to advance hybrid climate modelling using physically grounded AI for robust long-term predictions.
How to attend
About Journal Club
This talk is part of the ICCS Journal Club. Members of the VESRI teams, ICCS and the wider research community present research papers or concepts that are significant to their research and are of interest to the other teams.
Contact
ICCS team