An adaptive and evolvable deep reinforcement learning framework for weather prediction

A new deep reinforcement learning framework, Feitian Adaptive Ensemble Weather (FTAE-Weather), has been developed to improve weather prediction. It coordinates an open pool of pretrained forecasters by learning when and where to trust each model, significantly reducing prediction errors across various atmospheric variables and lead times. This framework transforms a fragmented inventory of specialist models into a unified system that benefits from new AI weather forecasting architectures.