Machine learning weather prediction models have advanced rapidly over the past few years and are generating a lot of excitement in the weather prediction community and general public alike. Recent studies, however, have shown some of the weaknesses of these machine learning models for real-time weather forecasting.

In Geophysical Research Letters, Sofar Head of Weather Steve Penny offers commentary on a new study that shows how the underlying mathematical principles behind the weather forecasts that you see every day can be applied to merge 1) the expert knowledge of meteorologists with 2) new advances in AI, to produce new hybrid models that leverage the best of both.