A Dynamic Risk Assessment Method of Waterlogging Points by Coupling Hydrology Model with Deep Neural Network

Abstract

With the intensification of global climate changes and rapid development of urbanization, waterlogging caused by torrential rain in urban areas has become potential risks that strongly affect the daily life and safety of urban residents. In order to comprehensively analyze the spatial-temporal dynamics of urban waterlogging points as well as their risk grades, in this paper, we propose a risk assessment model of waterlogging points by combining both hydrology model and deep neural network (DNN). In this method, we apply the hydrology model to calculate the dynamic water accumulating volume of the waterlogging points during rainfall. To measure the socio-economic effects of waterlogging, we use buffer analysis to extract Point of Interests (POI) surrounding the roads nearby the waterlogging points. Then, we employ a deep neural network to assess waterlogging points risk by using the calculated water accumulating volume and the statistics of POIs. To verify the reliability of the proposed model, we conducted assessment experiments using waterlogging survey data of central urban area of Wuhan city. The results showed that the model accurately assesses the risk of waterlogging and its spatial-temporal distribution in general, and can be used in decision making of early warning, planning, emergency response and research of urban waterlogging.