Get Started
Home
Topics
Search
Library
Research questionHow can reinforcement-learning corrections adapt a global weather model without destabilizing its dynamics?A global weather model evolves a coupled atmospheric state, so corrections that reduce forecast error can also accumulate or destabilize the integration. The difficulty is adapting corrections as the model state changes without undermining the model’s dynamics.
AI
Machine Learning
Reinforcement Learning
Research Paper
Latest papersRecent research connected to this question, newest first.Online Reinforcement Learning in the Met Office Unified Model through Distributed Model-Agent CouplingThe evidence concerns the Met Office Unified Model coupled to distributed RL agents through rank-local tensors. A DDPG actor shared weights across 70 vertical levels per atmospheric column and applied bounded potential-temperature corrections to model tendencies. Training used ten nudged forecasts directed toward the UK Met Office operational analysis; a frozen policy was then evaluated during non-nudged inference. The workflow remained numerically stable in the evaluated case and improved selected six-hour error measures, but the evidence is limited to a single-case experiment.research paper · Sep 2, 2026
Related questions
How can historical base-model residuals robustly correct multivariate forecasts without overcorrection?How can learned world models prevent policies from exploiting prediction errors and failing in the real world?How can online time-series forecasters adapt to evolving, unseen patterns without costly continual parameter updates?How can high-altitude balloons maintain station-keeping under nonlinear dynamics and partially observed wind fields?