The evaluation and the sensitivity analysis of GIS-based landslide susceptibility models

This paper evaluated two prevalent GIS-based models in landslide susceptibility; a deterministic-based approach and a bivariate-based approach. The bivariate-based approach defines the relationship between two variables individually, and this approach calculates the weights, which can also represent the contribution of factors. A deterministic-based approach is based on the geotechnical equation and calculates the quantitative values for landslide susceptibility. A number of GIS-based landslide projects and studies have been frequently conducted for identifying or estimating landslide prone areas using these two models. Results from those studies were easily implemented into disaster-related policies. Therefore, this paper highlighted discrepancies between these two models, and fulfilled the sensitivity analysis. After two models were run, the higher landslide susceptibility areas were quite different, and the reasons may come from the factors, the data availability, the data scale, and the data accuracy. The bivariate-based approach was sensitive to factors, and higher susceptible areas were new forest areas that have been presumably experienced by historical landslides. The deterministic-based approach was substantially affected by soil properties and slopes. The sensitivity analysis found that soil data should be complete and extensive for the accurate landslide susceptibility analysis.