Abstract
Video prediction, which aims to synthesize new consecutive frames subsequentto an existing video. However, its performance suffers from uncertainty of thefuture. As a potential weather application for video prediction, short timeprecipitation nowcasting is a more challenging task than other ones as itsuncertainty is highly influenced by temperature, atmospheric, wind, humidityand such like. To address this issue, we propose a star-bridge neural network(StarBriNet). Specifically, we first construct a simple yet effectivestar-shape information bridge for RNN to transfer features across time-steps.We also propose a novel loss function designed for precipitaion nowcastingtask. Furthermore, we utilize group normalization to refine the predictiveperformance of our network. Experiments in a Moving-Digital dataset and aweather predicting dataset demonstrate that our model outperforms thestate-of-the-art algorithms for video prediction and precipitation nowcasting,achieving satisfied weather forecasting performance.