Precipitation Nowcasting with Star-Bridge Networks

Yuan Cao, Qiuying Li, Hongming Shan, Zhizhong Huang, Lei Chen, Leiming Ma, Junping Zhang

arXiv preprint · 18 July 2019

In one sentence. StarBriNet, an RNN with sub-networks per rainfall intensity and a star-shaped bridge, improves radar-based precipitation nowcasting over prior methods.

Abstract

Precipitation nowcasting, which aims to precisely predict the short-term rainfall intensity of a local region, is gaining increasing attention in the artificial intelligence community. Existing deep learning-based algorithms use a single network to process various rainfall intensities together, compromising the predictive accuracy. Therefore, this paper proposes a novel recurrent neural network (RNN) based star-bridge network (StarBriNet) for precipitation nowcasting. The novelty of this work lies in the following three aspects. First, the proposed network comprises multiple sub-networks to deal with different rainfall intensities and duration separately, which can significantly improve the model performance. Second, we propose a star-shaped information bridge to enhance the information flow across RNN layers. Third, we introduce a multi-sigmoid loss function to take the precipitation nowcasting criterion into account. Experimental results demonstrate superior performance for precipitation nowcasting over existing algorithms, including the state-of-the-art one, on a natural radar echo dataset.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.