Assessment of the susceptibility of loess geological hazards in Kangdian Town on the south bank of the Yellow River based on the I-PSO-BP model

Abstract

Geological disasters occur frequently in the loess distribution area of the Yellow River basin. Only during the "7·20" extremely heavy rainstorm in Zhengzhou, hundreds of geological disasters occurred in the loess distribution area of the Yellow River basin in western Henan, resulting in significant losses of personnel and property. Carrying out vulnerability assessment research is the key to prevent and control geological disasters of loess. 13 evaluation factors are selectd including slope, landform, distance from structure, distance from water system, land use type, and distance from buildings. Based on information, I, BP neural network, and particle swarm optimization(PSO), I model, I-BP model, and I-PSO-BP model are constructed to evaluate the vulnerability of geological disasters in Kangdian Town, Gongyi City, which suffered the most serious geological disasters during the "7·20" extremely heavy rainstorm in Zhengzhou on the south bank of the Yellow River basin. Compare and analyze the accuracy, predictive ability, and rationality of different models using AUC values, frequency ratios, and field validation. The result show that the accuracy of the three coupled models(I-PSO-BP) is higher than that of other models(I and I-BP). The AUC values of the three models are 0.968 3, 0.976 4, and 0.978 6, respectively, and the accuracy rates are 59.14%, 92.00%, and 92.60%, respectively. The single model has the lowest accuracy. I-PSO-BP has higher accuracy and stronger prediction ability, which is most consistent with the field verification practice. The proportion of high, medium, low, and extremely low prone areas in the I-PSO-BP model is 5.09%, 2.25%, 7.52%, and 85.15%, respectively. The medium and high prone areas are concentrated along the gullies where human engineering activities are frequent, and human engineering activities such as slope cutting are important factors in the occurrence of geological disasters in Kangdian Town.

Keywords

geological hazards of loess in the Yellow River Basin / susceptibility / amount of information / BP neural network model / particle swarm optimization algorithm / Kangdian Town