Spatial Proximity-Based Geographically Weighted Regression Model for Landslide Susceptibility Assessment: A Case Study of Qingchuan Area, China

• Main Authors: Yange Li, Xintong Liu, Zheng Han, Jie Dou
• Format: Article
• Language: English
• Published: MDPI AG 2020-02-01
• Series: Applied Sciences
• Subjects: landslide susceptibility assessment; geographically weighted regression; spatial non-stationary; spatial proximity; slope unit
• Online Access: https://www.mdpi.com/2076-3417/10/3/1107

Landslides pose a serious threat to the safety of human life and property in mountainous regions. Susceptibility assessment for landslides is critical in landslide management strategy. Recent studies indicate that the traditional assessment models in many previous studies commonly assume a fixed relationship between influencing factors and landslide occurrence within an area, resulting in an inadequate evaluation for the local landslides susceptibility. To address this issue, in this paper we propose a spatial proximity-based geographically weighted regression (S-GWR) model considering spatial non-stationarity of landslide data for assessing the landslide susceptibility. Spatial proximity is the basic input condition for the proposed S-GWR model. The challenge lies in defining the spatial proximity expression that shows the geographical features of landslides and therefore affects the model ability of S-GWR. Our solution chooses the slope unit as spatial adjacency, rather than the grid unit in DTM. The multicollinearity between landslide influencing factors is then eliminated through variance inflation factor (VIF) method and principal component analysis (PCA). The proposed model is subsequently validated by using data in Qingchuan County, southwestern China. Spatial non-stationary is identified for landslide data. A comparison with grid unit and four traditional evaluation models is conducted. Validation results using the area under the ROC (receiver operating characteristic) curve and success rate curve indicate that the spatial proximity-based GWR model with slope unit has the highest predictive accuracy (0.859 and 0.850 respectively).