Study of Slope Geological Hazard Susceptibility Valuation with Small Sample Based on CF and SVM

Abstract

The number of geological hazards is one of the important factors that influence the sensitivity evaluation result. Owing to the small number of developmental geological hazards as sample points in a certain year, the applicability of sensitivity evaluation model is unclear. For this problem, Wuzhai County in northwestern Shanxi Province was selected as the study area in this paper. The susceptibility valuation factor set was established on the basis of four hazard pregnant factors: topographic factor, geological factor, human disturbance and destruction factor, and natural factor. Certainty Factor (CF model), Support Vector Machine (SVM model), and a combined evaluation model (CF-SVM model) were adopted to evaluate their sensitivity by using 53 slope geological hazards in Wuzhai County in 2017. At the same time, the combination model (I-SVM model) constructed by Information Value (I model) and SVM model was compared. The results show that: 1) the AUC values of CF model, I model, SVM model, CF-SVM model, and I-SVM model are 0.906, 0.855, 0.844, 0.934, and 0.891, respectively, and the accuracy of these models is good. The accuracy of CF-SVM model is increase by 3.09% and 10.66%, respectively, compared with that of CF model and SVM model, and by 9.24% and 4.83%, respectively, compared with that of I model and I-SVM model. The CF-SVM model has the highest evaluation accuracy among the three single models and two combined models, and has the best applicability in the case of small samples. 2) The areas of low sensitive area (45.09%), medium sensitive area (30.98%), high sensitive area (15.43%), and extremely high sensitive area (8.50%) of slope geological hazard susceptibility valuation in the study area by CF-SVM model are 627.205, 430.873, 214.652, and 118.270 km2, respectively. The distribution of disaster points account for 3.77%, 13.21%, 26.42%, and 56.60% of the total disaster points, and the frequency ratio is 0.08, 0.43, 1.71, and 6.66, respectively. 3) The very high sensitive areas of slope geological hazards in the study area are mainly distributed in the western loess hillyIgully area and the southeastern earthen and rocky mountain area along the trend of river system, while the low sensitive area extends outward from the high and medium sensitive area to the whole area.

Keywords

• slope geological hazard
• spatial distribution characteristics
• certainty factor
• support vector machine
• information value
• susceptibility valuation