An approach of using social media data to detect the real time spatio-temporal variations of urban waterlogging

Urban waterlogging has occurred frequently in recent years due to the impact of climate change and human activities. Real-time waterlogging information is crucial for disaster emergency management, but how to quickly obtain such information remains challenging. Social media data has been widely used to derive damage information because of its high real-time response, low acquisition cost, and high content integration. In this study, we propose an approach to extract real-time waterlogging points from social media data (Sina Weibo). First, social media data is obtained through web crawler technology; Then, de-duplication and de-noising methods are used to filter the data; Finally, a waterlogging point extraction method based on deep learning BERT-BiLSTM-CRF model is proposed to extract waterlogging points. Taking the "7.20" rainstorm in Zhengzhou as an example, there was a rapid increase in the number of social media data during urban waterlogging. Social media data is highly sensitive to urban waterlogging disasters caused by extreme rainstorms. On the day with the heaviest rainfall (July 20), the number of Weibo waterlogging points (331) in the central city was 267 more than the official waterlogging points (64). There were many more Weibo-derived waterlogging points than the real-time official published waterlogging points. The waterlogging points obtained by this approach covered the official published real-time waterlogging points accounted for no less than 82% and they were mostly located around roads, especially in low-lying areas. In general, we demonstrate the feasibility and accuracy of social media data on rapid detection of real-time spatiotemporal variations of waterlogging caused by extreme rainstorms. Urban waterlogging disaster information extracted from social media data can rapidly reflect the real-time spatiotemporal variations of urban waterlogging disasters, and can effectively cover and supplement data reported by government agencies, and can provide data support for urban waterlogging disaster prevention.