Flood risk assessment by using an interpretative structural modeling based Bayesian network approach (ISM-BN): An urban-level analysis of Shenzhen, China

With increasingly uncertain environmental conditions under global change, it is rather important for water security management to evaluate the flood risk, which is influenced by the compound effect of severe weather events and strong anthropogenic activities. In this paper, a risk assessment model in the framework of Bayesian network (BN) was proposed through incorporating with the Interpretative Structural Modeling method (ISM), which would produce an integrated ISM-BN model for reliable flood assessments. The ISM is employed to identify the relations among multiple risk factors, and then helps to configure the BN structure to conduct a risk inference. The established model was further demonstrated in Shenzhen city of China to perform an urban-level risk analysis of the flood disaster, and the Enhanced Water Index (EWI) was introduced to derive model parameters for training and verification. The obtained results of risk assessment lead to an accuracy of 76% with the Area Under ROC Curve (AUC) of 0.82, and spatial distribution of risk levels also showed a satisfactory performance. In addition, it was found that the maximum daily rainfall among ten risk factors play a key part in flood occurrence, while the elevation and storm frequency are also sensitive indicators for the study area. Besides, the spatial flood risk map generated under various design rainfall scenarios would contribute to identifying potential areas that are worth paying particular attention. Thus, the developed assessment model would be a useful tool for supporting flood risk governance to achieve reliable urban water security.

Automated summary

This paper proposes a risk assessment model based on the Bayesian network and Interpretative Structural Modeling (ISM) method for reliable flood assessments. The model incorporates ISM to identify the relations among multiple risk factors and configure the BN structure for risk inference. The model was demonstrated in Shenzhen city, China, and achieved an accuracy of 76% with an Area Under ROC Curve (AUC) of 0.82. The spatial distribution of risk levels showed satisfactory performance, and the maximum daily rainfall, elevation, and storm frequency were identified as key indicators for flood occurrence in the study area. The spatial flood risk map generated under various design rainfall scenarios can help identify areas that require particular attention. Thus, the developed assessment model is a useful tool for supporting flood risk governance and achieving reliable urban water security.