Exploring the application of large-scale meteorological models in basin-scale extreme rainfall forecasting

ZHAO Tongtiegang; LI Qiang

doi:10.3969/j.issn.1000-1123.2025.09.005

Abstract：With the advancement of artificial intelligence technology,large-scale models have been developed and applied in operational meteorological forecasting.Focusing on the"23·7"extreme rainfall and flood event in the Haihe River basin,retrospective forecasting experiments are conducted using large-scale meteorological models.The precipitation forecasts are compared with the High-Resolution Forecast(HRES)from the European Centre for Medium-Range Weather Forecasts(ECMWF)to evaluate the applicability of large-scale meteorological models in flood disaster prevention.The results indicate that,compared to traditional numerical weather prediction,the three large-scale meteorological models—GraphCast,FuXi,and AIFS—provide more accurate forecasts regarding the rainfall process,spatial distribution,central location,and timing.For 6-hourly precipitation,these models demonstrate comparable forecast accuracy across different lead times.Regarding accumulated precipitation,GraphCast,AIFS,and HRES produce forecasts of precipitation intensity,rainfall process,and affected areas that closely match observations.When the forecast initialization time is set 1 day in advance,the average accumulated precipitation in the study area was 124.6 mm,with forecasted values of 132.7 mm,115.5 mm,and 140.0 mm,respectively.For maximum precipitation,GraphCast,FuXi,and AIFS exhibite larger errors in precipitation intensity compared to HRES but have smaller errors in timing and location.The observed maximum accumulated precipitation was 484.8 mm,while the forecasts from GraphCast,FuXi,AIFS,and HRES are 329.7 mm,190.1 mm,251.2 mm,and 415.3 mm,respectively,when the initialization time is set 1 day in advance.Overall,large-scale meteorological models can provide effective precipitation forecasts for flood disaster prevention operations.