Abstract：The multisource heterogeneous data generated by modern meteorological observation networks provide an important opportunity to improve the accuracy of shortterm forecasts.The effective integration of observation systems such as ground automatic stations,Doppler radar,wind profile radar,and satellite remote sensing with numerical model forecast fields can significantly improve the spatiotemporal accuracy of shortterm forecasts ranging from 2 to 12 hours.Based on the fusion algorithms such as optimal interpolation,threedimensional variation and ConvLSTM deep learning,a multisource fusion framework covering temperature,humidity and wind pressure fields,radar reflectance,satellite cloud images and convective parameters was constructed.Practical business applications have shown that the fusion system demonstrates obvious advantages over the single data source method in severe convective weather forecasting,providing a reliable technical approach for businessoriented precise forecasting.
2026, Vol. 43 
