Precision forecasting of Atlantic blocking events: Utilizing explainable AI and transfer learning to discern critical precursors and achieve superior predictive models with limited observational data

Blocking events are an important cause of extreme weather, especially long-lasting blocking events that trap weather systems in place. However, the duration of blocking events is systematically underestimated in climate models. Explainable artificial intelligence (AI) is a novel data analysis method that may help to identify the physical causes of prolonged blocking events, as well as diagnose model deficiencies. We demonstrate such an approach on an idealized quasigeostrophic model due to Marshall and Molteni (1993). We build a sparse predictive model for Atlantic blocking long-time persistence conditioned on a temporary high-pressure anomaly. We find that high-pressure anomalies in the American Southeast and North Atlantic both contribute significantly to the prediction of sustained blocking events in the Atlantic region. This agrees with previous work that identified precursors in the same regions via wave train analysis. Moreover, we use Transfer Learning (TL) to leverage the large data set we produced with the Marshall-Molteni model on ERA5 reanalysis data to achieve much better predictions than direct training. Our work demonstrates the power of machine learning methods to extract meaningful precursors of extreme weather events and achieve better prediction using limited observational data.