AI-informed model-analogs for understanding subseasonal-to-seasonal jet stream and North American temperature predictability

Subseasonal-to-seasonal (S2S) forecasting is crucial for public health, disaster preparedness, and agriculture, yet both forecasting and diagnosing sources of potential forecast skill on this timescale remains particularly challenging. We adapt an interpretable AI-informed analog forecasting approach, previously used for longer timescales, to improve S2S model-analog prediction and understanding of its climate drivers. Using an artificial neural network, we learn a mask of weights to optimize analog selection and showcase its versatility across two prediction tasks: (1) regional continuous prediction of Month 1 Midwestern U.S. summer temperatures and (2) classification of Month 1–2 North Atlantic wintertime upper atmospheric winds. The AI-informed analogs outperform traditional model-analog forecasting approaches, as well as climatology and persistence baselines, for deterministic and probabilistic skill metrics on both climate model and reanalysis data; moreover, this skill gap grows for extreme predictions. Moreover, our interpretable-AI framework allows analysis of learned masks of weights, yielding improved understanding of the role of underlying physical processes upon predictability. We find skin temperature and the Northern Hemisphere to be more important predictors of North Atlantic wintertime upper atmospheric winds than upper atmospheric winds 1–2 months prior and the Southern Hemisphere, respectively.