A Precipitation Prediction Method Based on UNet and Attention Mechanism

Weather forecasting accuracy provides a wide range of social and economic advantages. Traditional numerical weather models need a lot of computer resources in the forecasting process, and they're pretty painful at forecasting for short periods of time (5-10 minutes). In this paper, a new structure based on UNet has been proposed ,which uses a self-attentive mechanism. The dataset is based on a sample of the German Weather Service's RY product (DWD). The experimental findings suggest that the model presented in this research uses less computer resources in the prediction process and has a lower error rate than existing neural network-based models.