Abstract:

For the high frequency of landslide occurrences in the watershed of Jinshajiang, certain factor (CF) and weight of evidence (WOE) were both adopted to evaluate landslide susceptibility in this area. In this study, ten variables were selected to combine diversely in the process of modeling, including slope, aspect, normalized difference vegetation index, altitude, lithology, profile curvature, distance to roads, distance to rivers, distance to faults and the types of Quaternary sediments. Validated by the area under the prediction rate curve (AUC), the results indicated that CF with the total 10 variables had the best performance among the two models with combinations of various variables. The succeed rate and predictive rate of CF with the 10 variables respectively reached to 83.40% and 74.43%, while the results of WOE merely reached to 74.52% and 69.89%. It is proved that CF has high accuracies in both experiment area and verification area. Thereafter, the landslide susceptibility index generated from CF was further categorized into 4 subdivisions by the natural-break classification, including extremely difficult, moderately difficult, moderately easy and extremely easy area to landslide occurrences. As shown in the result of classification, either the north-south extended zones of the Chenghai Fault or the areas along the middle segment of Jinshajiang are prone to triggering landslides. The result implied a special correlation among faults, rivers and landslide occurrences.

Key words: landslide, susceptibility evaluation, GIS, certain factor, weight of evidence, Jinshajiang