Extended Data Fig. 5: Case study of a mega dust storm event in northern China during April 16–20, 2023.
From: Advancing operational global aerosol forecasting with machine learning

a,b, Forecasts of DUAOD (a) and DUSMC (b) during the event, produced by operational AI-GAMFS, CAMS, CMA-CUACE/Dust, KMA-ADAM3, and FMI-SILAM models, with a 5-day lead time at 24-h intervals, compared against MERRA-2 reference data. Operational AI-GAMFS was initialized at 22:30 UTC on April 15, 2023, while the other models were initialized at 00:00 UTC on April 16, 2023. For comparison, the outputs of all models were interpolated to the same spatiotemporal resolution as AI-GAMFS. The overall accuracy metrics (i.e., spatial R and RMSE) for operational AI-GAMFS and the four dust models relative to MERRA-2 data are shown in the lower-left corner of each panel. Note that DUSMC is not included as a direct forecast variable for the CAMS model.