In this ongoing project , Karl Otness and co-authors are evaluating a proof of concept multiscale approach to predicting subgrid forcings in climate models. They see encouraging preliminary results from first making a prediction in a fine-to-coarse direction which yields a lower resolution, but more confident prediction followed by a refinement to predict finer scale details.

This paper won “Best ML innovation” at the ICLR 2023 Workshop: Tackling Climate Change with Machine Learning.