Influence of Environmental Factor Selection and Combination on Landslide Susceptibility Prediction Modeling

Different combinations of landslide environmental factors can be obtained under various selection methods,which are then used as input variables of landslide susceptibility prediction (LSP) models.Studying the modeling rules under different factors combinations can provide a theoretical and practical basis for more accurate LSP modelling.Taking Wanzhou District of the Three Gorges Reservoir Area as an example,23 environmental factors,such as topography,hydrology and lithology,are firstly selected.Then the correlation coefficient analysis (CA),linear regression (LR),principal component analysis (PCA),artificial neural network (ANN) and rough set (RS) selection methods are used to optimize the factor combinations.Next,the obtained factor combinations are used as input variables of support vector machine (SVM),Multi-layer perceptron (MLP) and other typical machine learning,to construct CA-SVM,CA-MLP and other coupled models.Meanwhile,these coupled models are compared with the All factors-machine learning models without considering environmental factors selection.Finally,the AUC accuracy,the mean value and standard deviation of predicted landslide susceptibility indexes are used to explore the modeling rules.Results show that :(1)landslide susceptibility predicted by All factors-machine learning is generally better than other models considering factors selection,indicating that factors selection is not ideal for improving the LSP performance;(2)The sensitivity of different factor selection methods to modeling performance is slightly lower than that of different machine learning models,suggesting that factors selection is unnecessary and may complicate the LSP modeling process.However,we still need to avoid using environmental factors with high correlation and similar mechanism to landslides.It can be concluded that,a satisfied selection and combination of landslide environmental factors can be constructed according to the principles of accurate data,rich types,clear significance,feasible operation and clear primary and secondary.