Selecting Optimal Conditioning Factors in Landslide Susceptibility Mapping Using Certainty Factor and A Multi-Class Weight of Evidence Approach

Abstract

One of the important steps in assessing landslide susceptibility is the selection of conditioning factors, of which many have been examined in the literature for related mapping; however, factors that produce a better result for an area under analysis have not yet been convincing. With the available landslide conditioning factors, assessing the best combination has become an important research issue. Those employed in this research were: elevation, slope, aspect, curvature, plan curvature, profile curvature, lithology, TPI (Topographical Position Index), TWI (Topographical Wetness Index), SPI (Stream Power Index), distance to river, drainage density, soil, land use cover, distance to fault, and precipitation. All maps were subdivided into different classes by value or feature, and were then converted to a raster format in the ArcGIS 9.3. The objective of this study was to investigate the application of weight-of-evidence (WoE) and certainty factor approaches for selecting corresponding conditioning factors to produce an effective landslide susceptibility map of a landslide-prone area. Eleven out of fifteen factors were selected by certainty factor. Meanwhile,a combination of seven factors were selected as independence factors by WoE, using the conditional independence test. The validation results showed that the CF model has a slightly greater area under the curve (AUC) at 90.3% and a predictive rate curve of 90.2%, which is better than WoE (AUC 90.1% and a predictive rate curve of 89.9%).