PNNL

Abstract

Global weather forecast models are vital tools used for numerous applications, including public safety, agriculture, and transportation. Recent advancements in artificial intelligence (AI) and deep learning (DL) have shown potential to enhance weather forecasting accuracy and speed. In this study, we developed a spatiotemporal DL model, PredRNN, to generate mid-range hourly forecasts for five surface atmospheric variables, including wind speed and direction, mean sea level pressure, temperature, and precipitation. By incorporating a wavelet transform function for data pre-processing, our model enhances the forecasting of extreme weather events. The PredRNN model demonstrates promising results. It achieves high-resolution (25 km) forecasts with a 1-day lead time RMSE of 1.8 m/s for wind components, 180 Pa for mean sea level pressure, 1.8 K for temperature, and 4x10-4 m for precipitation. While our model may not yet outperform state-of-the-art AI weather forecast models in all aspects, its ability to provide accurate hourly forecasts and potential for future enhancements contribute to the advancement of DL weather forecasting methods. Our work highlights the importance of integrating sophisticated temporal components and data transformation techniques to improve the predictability and accuracy of weather forecasts, particularly for extreme events.