Analysis of Landslide Surface Deformation Using Geographically Weighted Regression

Abstract

Traditional regression analysis methods such as Ordinary Least Squares (OLS) are usually used to explore data relations, but they cannot reflect the spatial non-stationarity of the data. Geographically Weighted Regression (GWR) is an effective tool for dealing with this situation, whereas there has not any related studies about using GWR to analyze the landslide surface deformation. This paper tries to base on a typical reservoir-type landslide in Three Gorges Reservoir area of Yangtze River, China, and uses monitoring data, to build OLS and GWR model between landslide surface displacements and trigger factors by ArcGIS. Analysis showed that the GWR model has greater R2 and smaller Akaike information criterion (AIC) value, and the residuals spatial autocorrelation degree can be significantly reduced then the OLS model, what means the GWR model can capture the spatial non-stationarity of independent variables and is more reliable in analysis of landslide surface deformation.