An Algorithm for Precipitation Correction in Flood Season Based on Dendritic Neural Network

Abstract

In recent years, the National Climate Center has developed a dynamic downscaling prediction technology based on the Climate-Weather Research and Forecasting (CWRF) regional climate model and used it for summer precipitation prediction, but there are certain deviations, and it is difficult to predict more accurately. The CWRF model simulates the summer precipitation forecast data from 1996 to 2019 and uses a combination of dendrite net (DD) and artificial neural networks (ANNs) to conduct a comparative analysis of summer precipitation correction techniques. While summarizing the characteristics and current situation of summer precipitation in the whole country, the meteorological elements related to precipitation are analyzed. CWRF is used to simulate summer precipitation and actual observation precipitation data to establish a model to correct the precipitation. By comparing with the measured data of the ground station after quality control, the relevant evaluation index analysis is used to determine the best revised model. The results show that the correction effect based on the dendritic neural network algorithm is better than the CWRF historical return, in which, the anomaly correlation coefficient (ACC) and the temporal correlation coefficient (TCC) both increased by 0.1, the mean square error (MSE) dropped by about 26%, and the overall trend anomaly (Ps) test score was also improved, showing that the machine learning algorithms can correct the summer precipitation in the CWRF regional climate model to a certain extent and improve the accuracy of weather forecasts.

Keywords: anomaly correlation coefficient (ACC); dendrite net (DD) and artificial neural networks (ANN); machine learning; mean square error (MSE); summer precipitation correction; temporal correlation coefficient (TCC).