Spatiotemporal assessment of urban flooding hazard using social media: A case study of Zhengzhou ‘7·20’

Abstract

Urban flooding is occurring more frequently and with more serious consequences. The importance of flood disaster management has been highlighted. Social media has become an important source of flood disaster information in flood hazard assessment, however, it suffers from a lack of geolocation information and many qualitative descriptions. To improve the application of social media in flood hazard assessment, this study proposed a flooding hazard assessment method. Machine learning (ML)-based named entity recognition (NER) technique was used to identify the location and time of inundation in tweets, which can complement and correct the spatiotemporal properties of tweets. The textual description of waterlogging level in social media was quantified by lexicon-based NER technique. Taking a heavy rainfall event that occurred in Zhengzhou in July 2021 as an example, the results show that social media-based flood hazard maps are effective in identifying waterlogging. This study provides a feasible way to obtain near real-time hazard information from social media to help rapid hazard response.