Real-time urban rainstorm and waterlogging disasters detection by Weibo users

Abstract. With the process of urbanization in China, the urban waterlogging caused by rainstorm occurs frequently and often leads to considerable damage on natural environment, human life, and the city economy. Rapid detection of rainstorm and urban waterlogging is an essential step to minimize related losses. Weibo, a popular microblogs servicer in China, can provide many real-time Weibo posts for rapid detection. In this paper, we propose a method to identify microblogs with rainstorm and waterlogging information and apply them to waterlogging risk assessment. After pre-processing the microblog texts, we evaluate the performance of clustering (K-means) and classification (support vector machine, SVM) algorithms in the classification task. Apart from the word vector features, we also introduce the sentiment and publisher features for a more real-time and accurate results. Furthermore, we build a waterlogging dictionary to assess the waterlogging risk from the Weibo texts, and get a risk map with ArcGIS. To examine the efficacy, we collect Weibo data from two rainstorm and waterlogging disasters in Beijing city as examples. The results indicate that the SVM algorithm can be applied for real-time rainstorm and waterlogging information detection. Compared to the official authentication and personal certification users, the microblogs posted by general can better show the intensity and timing of rainstorm. The location of waterlogging points is consistent with the risk assessment results, which can be used as a reference for timely emergency response.