Research on regional short-term precipitation quantitative forecast revision technology integrating weather feature information

Abstract

This study proposes a method to improve the accuracy of heavy precipitation forecasts by incorporating weather system information into a random forest (RF) regression model. Based on meteorological elements from the European Centre for Medium Range Weather Forecasts (ECMWF) model, this method utilizes dimensionality reduction techniques to transform the four-dimensional training dataset into a three-dimensional one. Key weather systems, such as the 500 hPa trough line and the 850 hPa-925hPa shear line, are introduced as predictive factors into the RF model. Finally, the method corrects heavy precipitation forecasts using the China Meteorological Administration's Shanghai 3-km regional model and forecaster experience. The RF model demonstrated superior performance over the ECMWF model in predicting precipitation events, particularly excelling in forecasting heavy rain and above. From April to September 2022, the RF model achieved an 18 % increase in correct forecasts for heavy rain and more intense precipitation events, along with notably better TS scores across various intensities of rainstorms. Additionally, the RF model provided a more accurate representation of precipitation fall areas and intensities, closely aligning with actual rainfall patterns. This enhanced accuracy was particularly evident when the predicted rain band shapes and trends in the model corresponded with real-world observations, often resulting in a slightly stronger intensity of the forecasted rain. These findings demonstrate that the integration of weather system features into the RF model significantly improves the accuracy in forecasting heavy rain and more severe precipitation events. This advancement represents a more effective and reliable tool for meteorological applications, especially in predicting intense rainfall scenarios.