An ensemble risk assessment model for urban rainstorm disasters based on random forest and deep belief nets: a case study of Nanjing, China

Abstract

Effective management of rainstorm risk is essential for reducing regional rainstorm disaster risks and losses. In this paper, we discussed the influencing factors of urban rainstorm disaster (URSD) risk from four aspects and then constructed the index system of URSD risk assessment which includes 16 influencing factors. Furtherly, important indexes were extracted as the input of deep belief nets (DBN) model after analyzing the types and risk characteristics of URSD. As well as a coupling risk assessment model of URSD based on random forest and deep belief nets (RF–DBN) was established due to the capacity of high-dimensional data processing of RF and robustness of DBN. To test the validity of this risk assessment model, it was applied to evaluate the rainstorm disaster risk in 11 districts of Nanjing, China, from May to September during 2009 and 2017. Finally, the risk grade map of rainstorm disaster in Nanjing was drawn and the corresponding countermeasures for the regulation and control of URSD were put forward. The results show that the rainstorm risk in Nanjing is generally high during the period of rainy season and the risk of rainstorm disaster has egional features during the flood season.