Risk assessment of flood disasters in the Poyang lake area

Abstract

Floods are one type of common and catastrophic disaster and cause extensive destruction areas, enormous casualties, and huge economic and environmental losses worldwide. Flood risk assessment is crucial to disaster mitigation and prevention, social stability and prosperity, and economic and environmental sustainability. This work establishes comprehensive risk indices of geoenvironmental, disaster-triggering, disaster-vulnerable, and disaster-preventive factors to improve the rationality of risk assessment. Moreover, a novel combined method of context-perception long short-term memory (LSTM), analytic hierarchy process, and entropy weight is suggested to perform relative accurate risk assessment in the Poyang lake region, one of the worst-hit areas in China. The new context-perception LSTM algorithm can adequately learn comprehensive and significant context characteristics and outperforms the context-based convolutional neural network and conventional LSTM algorithms. The test precision reaches 95.7%, 99.1%, 0.958, and 0.981 in Accuracy, TPR, F1-score, and AUC, respectively. Some significant insights into flood risk are suggested as follows. (1) High flood risk mainly occurs in the densely populated regions with gentle topography close to main streams. The concentrated population and intensive town construction have destroyed surface permeability and soil consolidation and caused high flood risk. (2) In the high-risk urban regions, the operational condition of old soil dikes should be continuously monitored, and the construction of high rock dikes are necessary to effective flood prevention in the future. In the densely-populated and high-risk urban regions, more emergency rescue infrastructures, for example hospitals and emergency shelters, are required to arrange or build. Rational planning of emergency shelters can not only protect people against floods but also reduce government investment. (3) Significant hydrological infrastructures require high-frequency or even near real-time monitoring to timely discover engineering diseases and to avoid huge losses caused by unstable prevention infrastructures during flash floods.