Application of the hidden Markov model in a dynamic risk assessment of rainstorms in Dalian, China

Disaster risk evolves spatially and temporally due to the combined dynamics of hazards, exposure and vulnerability. However, most previous risk assessments of natural disasters were static and typically based on historical disaster events. Dynamic risk assessments are required to effectively reduce risks and prevent future losses. Based on rainstorm disaster data and meteorological information collected in Dalian, China, from 1976 to 2015, the hidden Markov model (HMM) was used to detect inter-annual changes in rainstorm disaster risks. An independent sample test was conducted to assess the reliability of the HMM in dynamic risk assessments. The dynamic rainstorm risk in Dalian was simulated based on the observation probability matrix, which characterized the relationship dependence between rainstorm hazard and risk, and the probability matrix of state transition, which reflected the probability of changes for the risk level. High rainstorm risk was associated with high-hazard rainstorms and continuously appeared with little probability in several successive years. The reliability applied the HMM to simulate the rainstorm disaster risk was approximately 67% in the dynamic risk assessment. Additionally, the rainstorm disaster risk in Dalian is predicted to be at a medium-risk level in 2017, with a probability of 0.685. Our findings suggest that the HMM can be effectively used in the dynamic risk assessment of natural disasters. Notably, future risk levels can be predicted using the current hazard level and the HMM.