Spatial and Temporal Aware Graph Convolutional Network for Flood Forecasting

Abstract

Intelligent flood forecasting systems provide an effective means to predict flood disasters. Accurate flood flow value prediction is a challenging task as it is influenced by both spatial and temporal relationships among flood factors. Popular deep learning architectures like Long Short-Term Memory (LSTM) networks lack the ability to model the spatial correlations of hydrological data, which hinders achieving satisfactory prediction results. Additionally, not all temporal information is equally valuable for flood forecasting. This paper proposes a novel Spatial and Temporal Aware Graph Convolutional Network (ST-GCN) for flood prediction, which extracts spatial-temporal information from raw flood data. Furthermore, a temporal attention mechanism is introduced to weight the importance of different time steps, leveraging global temporal information to improve flood prediction accuracy. Experimental results on two self-collected datasets demonstrate that ST-GCN significantly enhances prediction performance compared to existing methods.

Yirui Wu

Young Professor, CCF Senior Member

My research interests include Computer Vision, Artifical Intelligence, Multimedia Computing and Intelligent Water Conservancy.