FENGSHUN AI-based forecasts initiated between 1 and 25 July 2025

The “FENGSHUN” S2S AI-based forecasting system, released by the China Meteorological Administration, establishes an efficient spatiotemporal modeling framework by integrating a Cascaded Swin Transformer into a U-shaped network, using moving-window attention to capture multi-scale dependencies for high-resolution S2S forecasting. To address climate prediction uncertainty, the system incorporates an intelligent perturbation module that jointly optimizes Kullback-Leibler divergence and the L1 loss function to learn low-rank Gaussian parameters, enabling the generation of ensemble forecasts that effectively mitigate autoregressive error accumulation. This dataset includes the "FENGSHUN" outputs of global rainfall amount initiated between 1 and 25 July 2025. Data structure (take Z_NAFP_C_BAQH_20250701000000_P_FENGSHUN-PR-GLB-SURF-20250701-060 for instance): netcdf Z_NAFP_C_BAQH_20250701000000_P_FENGSHUN-PR-GLB-SURF-20250701-060 { dimensions: time = UNLIMITED ; // (60 currently) lev = 100 ; lat = 121 ; lon = 240 ; variables: double time(time) ; time:long_name = "Time" ; time:units = "days since 2025-06-30 12:00:00" ; float lev(lev) ; lev:long_name = "Ensemble" ; float lat(lat) ; lat:long_name = "Latitude" ; lat:units = "degrees_north" ; float lon(lon) ; lon:long_name = "Longitude" ; lon:units = "degrees_east" ; float tp(time, lev, lat, lon) ; tp:units = "mm/h" ; tp:long_name = "tp" ; tp:standard_name = "Total Precipitation" ; // global attributes: :description = "Fengshun AI model tp Product" ; :creator = "Fengshun AI model Development Team @CMA." ; }