A new method for flood disaster resilience evaluation: A hidden markov model based on Bayesian belief network optimization

A hidden Markov model based on Bayesian belief network optimization (BBN-HMM) is introduced to achieve accurate evaluations of flood disaster resilience without a training set. The 15 farms under the jurisdiction of the Jiansanjiang Branch of the China Beidahuang Land Reclamation Group Co., Ltd., are studied, and a sensitivity analysis is conducted. The results show that the weight design of the comprehensive flood control safety index displays obvious differences among different farms. An analysis of the spatiotemporal evolution of resilience shows that from 1996 to 2004, the resilience in the entire region remained at a relatively low level, and the level in the central region was slightly higher than that in the surrounding areas. From 2004 to 2008, the resilience level of central farms declined, and after that, the resilience levels of central and eastern farms increased significantly. However, by 2020, the Qindeli, Qinglongshan and Qianjin farms only reached level II, experiencing potential flood hazards. In 2008, the main driving force for the decline in the resilience of the Yalvhe, Honghe, Qianjin and Erdaohe farms was the proportion of water conservancy projects in the GDP and the ratio of tertiary industry to GDP. In 2020, the main driving force for the resilience level of the five farms with level comprehensive V resilience, Qianjin, Chuangye, Qianfeng, Yalvhe and Erdaohe, was GDP per capita. Based on the composition weight of the comprehensive flood control safety index, the construction indexes are selected, and a scenario simulation for the next 10 years is performed to guide the improvement of resilience. After conducting a consistency test and evaluations involving precision, rationality and stability, it is proven that the model has significant advantages in performance.