Rapid urban flood risk mapping for data-scarce environments using social sensing and region-stable deep neural network

Urban flooding is one of the most widespread natural hazards in modern cities. Risk mapping provides critical information for flood risk management to reduce life and economic loss. As a widely used approach to support flood risk mapping, physical-based modeling suffers from model accuracy and computation complexity. Empirical methods rely on availability of rich disaster data and are not normally transferable for fast flood prediction to different cities. Both methods require high-quality hazard data and disaster information for reliable prediction, which are not always available. This paper presents an alternative near real-time flood risk mapping method for data scarce environments developed using social sensing and region-stable deep neural network (RS-DNN). By extracting disaster information in near real-time using social sensing techniques and considering risk distribution factors rather than flood influencing factors, this new method enables flood risk mapping and analysis cross a large domain in minutes, with all input data openly available. The proposed method can be adapted to different disaster process and different case study cities through timely social sensing.