Seismic landslide susceptibility assessment using principal component analysis and support vector machine

Ziyao Xu; Ailan Che; Hanxu Zhou

doi:10.1038/s41598-023-48196-0

Seismic landslide susceptibility assessment using principal component analysis and support vector machine

Sci Rep. 2024 Feb 14;14(1):3734. doi: 10.1038/s41598-023-48196-0.

Authors

Ziyao Xu¹, Ailan Che², Hanxu Zhou¹

Affiliations

¹ School of Naval Architecture, Ocean and Civil Engineering, Shanghai Jiao Tong University, 800 Dongchuan Road, Shanghai, 200240, China.
² School of Naval Architecture, Ocean and Civil Engineering, Shanghai Jiao Tong University, 800 Dongchuan Road, Shanghai, 200240, China. alche@sjtu.edu.cn.

Abstract

Seismic landslides are dangerous natural hazards that can cause immense damage to human lives and property. Susceptibility assessment of earthquake-triggered landslides provides the scientific basis and theoretical foundation for disaster emergency management in engineering projects. However, landslide susceptibility assessment requires a massive amount of historical landslide data. Evidence of past landslide activities may be lost due to changes in geographical conditions and human factors over time. The lack of landslide data poses difficulties in assessing landslide susceptibility. The aim of this study is to establish a generalized seismic landslide susceptibility assessment model for applying it to the Dayong highway in the Chenghai area, where earthquakes occur frequently but with a lack of landslide data. The landslide data used comes from the 2014 Ludian Ms (Surface wave magnitude) 6.5 earthquake in a region with geographical conditions similar to those in the Chenghai area. The influencing factors considered include elevation, slope, slope aspect, distance to streams, distance to faults, geology, terrain wetness index, normalized difference vegetation index, epicenter distance and peak ground acceleration. The frequency ratio method is used to eliminate influencing factors with poor statistical dispersion of landslides. Principal component analysis (PCA) is utilized to reduce the dimensionality of landslide conditioning factors and to improve the transferability of the assessment model to different regions. A support vector machine model is used to establish the susceptibility assessment model. The results show that the accuracy of the PCA-SVM model reaches 93.6%. The landslide susceptibility of the Chenghai area is classified into 5 classes, with the "Very high" landslide susceptibility class accounting for 0.63%. The 13-km section in the middle of the Dayong highway, which accounts for 8.9%, is identified as the high-risk area most obviously impacted by seismic landslides. This study provides a new approach for seismic landslide susceptibility assessment in areas lacking in landslide inventory data.