Urban Flood Inundation Probability Assessment Based on an Improved Bayesian Model

Abstract

Urban flood inundation is spatially uncertain. To quantify this uncertainty, it is necessary to explore the spatial probability of urban flood inundation in different return periods. In this study, an urban flood spatial inundation probability assessment method based on an improved Bayesian model is proposed, which comprises three parts: data reconstruction based on undersampling; optimal Bayesian sample planning; and spatial inundation probability assessment. A case study of the central urban area of Jingdezhen City, China, is presented in this paper. The results indicate that (1) the inundation probabilities generated based on various return periods (20-, 50-, and 100-year return periods) are accurately determined and can provide more detailed inundation information. (2) The adoption of the random undersampling data reconstruction method solves the problem of an imbalanced number of inundations/noninundations during Bayesian modeling and substantially enhances the prediction accuracy compared with the traditional Bayesian modeling approach. (3) A sensitivity analysis reveals that inundation probability is sensitive to the drainage network and elevation rather than soil water retention and distance to river. With an increase in the return period, the inundation probability gradually increases. As the proposed method can quantify flood inundation uncertainty, it is valuable in supporting specific flood risk assessments.