Fig. 4: Global atmospheric river (AR) forecast performance of the evaluated models.
From: Global performance benchmarking of artificial intelligence models in atmospheric river forecasting
Model skill scores for AR detection: a P, b R, and c F1 score. Solid lines represent the mean values across all days in 2023, and shaded areas denote the interquartile range from the 25th to 75th percentiles. d–i Spatial distribution of ARs in forecasts with an F1 score below 0.5, calculated against corresponding events in ERA5 (indicating unsatisfactory performance), at a 5-day lead time. d Results from Pangu; the top value indicates the ratio of detected ARs to all events detected in ERA5. e–i Corresponding results for FCN2, FuXi, NeuralGCM, GraphCast, and FGOALS.
