Evaluation of Landslide Susceptibility by Multiple Adaptive Regression Spline Method

  Objectives  According to general landslide susceptibility evaluation methods, landslide condition factors cannot be effectively selected.

  Methods  The prediction model of landslide susceptibility index (LSI) is constructed by multiple adaptive regression spline (MARS), and the landslide susceptibility condition factors are automatically selected, and the landslide susceptibility map is produced by 15 landslide susceptibility factors. In addition, the accuracy of the model is compared between logistic regression (LR) and MARS.

  Results  The results show that the accuracy of landslide susceptibility model constructed by MARS is better than LR. The accuracy of MARS success curve is 0.945 4, and the accuracy of MARS prediction rate curve is 0.923 8. At the same time, the model also selects the important influencing factors of landslide (elevation, slope angle, rainfall, distance to faults, NDVI, plan curvature, geological petrofabric).

  Conclusions  Research suggests that the MARS is an effective method for landslide prediction in study area and can provide decision support for reducing nature disaster.