Analysis of the influence of dimensional unity in landslide susceptibility assessment

Abstract

Previous studies on the susceptibility assessment of regional landslides mainly focused on comparing and improving the results of different evaluation models, while neglected the preservation of information on selected disaster-causing factors and the issue of how to unify factor dimensions. To explore the correlation and dimensionality of disaster causing-factors and their impact on susceptibility assessment, this study selected 12 factors such as elevation, slope, aspect, and terrain undulation, and used new factors extracted from principal component analysis in susceptibility assessment in the northern Jingyuan County. Data standardization, landslide density, and information quantity substitution methods were used to unify the dimensionality of disaster-causing factors. The landslide susceptibility zoning map was drawn based on the GIS platform in the study area. The accuracy of the susceptibility assessment of each mode was evaluated by the receiver operating characteristic curve. The results show that among the information model, the logistic regression model, and the perceptron model, the accuracy of the model evaluation obtained by the factors processed by principal component analysis is the highest. Using the information value substitution method to unify the dimensions of factors can further improve the accuracy of the evaluation of the logistic regression model and the perceptron model. The perceptron model has the highest accuracy (AUC=0.936 7), which is superior to the information model (AUC=0.917 3) and the logistic regression model (AUC=0.927 2). This is an ideal model for the landslide susceptibility assessment in the study area and should be given priority. The results can provide the basic theoretical information for disaster prevention and mitigation in similar areas.

Keywords

• landslide susceptibility assessment
• disaster-causing factors
• principal component analysis
• information models
• logistic regression model
• perceptron model
• dimension