Weather prediction is one of the most challenging yet essential task, multitude of algorithms ranging from pure Auto-regressive architecture to pure GNN based architecture were formulated, algorithms like FourCastNet provided a deep insight into how the complex weather conditions could be modelled.
Adding on to that, very recently Google DeepMind came up with GraphCast, a GNN based encoder-process-decoder architecture that not only is able to perform high resolution predictions but also in significantly less time.
claim : 10 days prediction in 1 min with 0.25 degree resolution on T4 GPU. (roughly 28sq km).