基于多元自适应回归样条的黄土区滑坡敏感性评价OA北大核心CSTPCD

[Objective]The purpose of this study is to explore and evaluate the influence of disaster factors on the susceptibility zoning of Loess landslide,so as to assist the monitoring,warning and prevention of landslide disaster.[Methods]Based on factor screening and independence test,multiple adaptive regression spline model(MARS)was used to independently segment the data,to fit several basis functions,to construct hinge functions to establish a connection mechanism between the basis functions,and then to reverse self-correct,to eliminate or modify part of the basis functions to achieve independent selection of disaster factors and give factor weights.At last,probability ratio(PR)model was introduced to analyze the accuracy.[Results](1)The physical meaning of the base function of MARS model is clear,which greatly reduces the complexity of the model and is easy to implement GIS spatial analysis.(2)The multicollinearity and correlation tests of assessment factors are helpful for MARS model optimization.(3)The classification of extremely sensitive and highly sensitive areas by MARS model is more rigorous and objective than that by PR model,and the result of landslide susceptibility classification is the result of the comprehensive effect of all the participating factors,and the influence of key factors on the assessment results is not obvious.[Conclusion]In the study on the susceptibility assessment of Loess landslide in Yaozhou District of Tongchuan City by MARS model,the AUC value of ROC curve is 0.879,the overall fitting effect is good,and the results are consistent with the field investigation results,and the results are reliable.