Research on weather radar nowcasting extrapolation

The purpose of heavy rainfall forecast is to predict the distribution of local rainfall intensity in the next 0-2 hours, and the accurate extrapolation image can provide accurate spatial-temporal data reference for the nowcasting of heavy rainfall forecast. Although the accuracy of the radar extrapolation results of the deep learning model based on the recurrent network has been greatly improved compared with the traditional extrapolation model in recent two years, it still needs to be further improved in many aspects of the model. Based on the analysis of the existing ConvSLTM model and TrajGRU model of radar extrapolation, this paper improves them by increasing the number of radar layers and the weight of prediction results. Experiments are carried out with open competition data and real radar data as samples. The experimental results show that the modified network model can better capture spatiotemporal correlation and has more accurate extrapolation effect.