PF-UNet: A Short-term Precipitation Correction Model Integrating Multiple Meteorological Elements

The short-time precipitation correction is to correct the forecast deviation of the numerical model by integrating the actual observation data. Previous precipitation correction methods based on deep learning mainly used a single precipitation element to correct the forecast data, and failed to fully consider the impact of other meteorological elements on short-term precipitation process. However, the precipitation process is coupled by many meteorological elements in its formation and evolution, resulting in its complexity and variability. Therefore, considering the influence of other meteorological factors in the correction process can better improve the accuracy of short-term rainfall forecast. In this paper, a deep learning network PF-Unet is proposed to correct the short-term precipitation of multiple factors. The PF-Unet model integrates the key features of various meteorological elements, strengthens the influence of meteorological elements on the generation and evolution of precipitation events, and improves the accuracy and reliability of short-term precipitation forecasting. The experimental results show that the precipitation area and precipitation size of the PF-Unet model with multi factors are closer to the global precipitation measurement (GPM) data. This study further improves the accuracy of the numerical prediction system in short-term precipitation prediction.