Time-Series Flood Risk Assessment Based on Time Information Loss Compensation: Fusing Remote Sensing and Social Media Data

Abstract

The fusion of remote sensing and social media data has shown great potential in urban flood risk assessment. However, most related studies seldom consider the loss of time information and leverage social media data gathered throughout the disaster period for spatial information enhancement. Meanwhile, existing models for correcting the spatial bias between remote sensing and social media data rely on prior flood information, which is generally unavailable in countries and regions lacking urban flood monitoring infrastructure. In this study, we first combined spatiotemporal fusion of remote sensing data and feature extraction to obtain the flood prior distribution, which was used to extend the geographic optimal transport (GOT) model for scenarios with limited prior information. We then developed a morphology-based spatiotemporal enhancement method to fuse remote sensing features-derived information and relocated social media data for time-series flood risk assessment. Taking the 2016 Wuhan flood event as a case, our study showed that the precision of the relocated Weibo data reaches 82.41%, which greatly reduces the location uncertainty of the original data. In addition, the relocated Weibo data can help identify the flooded areas that may be ignored by remote sensing-based results and significantly increase the estimated flooding probability of ground-truthing derived flooded points at a local scale. Furthermore, the time-series flood risk maps at a 2-day interval exhibit good consistency with the real flooding situation. Overall, the proposed method shows potential to compensate for the loss of time information and realize time-series flood risk assessment based on data fusion.