Landslide Susceptibility Assessment using Bivariate and Data Mining Methods

Aims: The primary objective of the study is to examine the simultaneous use of machine learning with complex modeling processes and to compare their accuracy with that of the frequency ratio method, a simple statistical method, in northern Tehran. Due to its milder climate compared to Tehran City, residential areas and gardens have developed, leading to increased road construction and, in turn, a rise in landslide incidents. Material & Methods: A landslide distribution map was prepared using Google Earth and field survey data. Twelve factors were selected as conditioning factors. Generalized linear models, multivariate adaptive regression splines, and frequency-ratio models were applied to generate susceptibility maps. The ROC curve was used for model validation. The areas of the susceptibility classes were also calculated for three models.Findings: The FR and GLM models achieved good accuracy, while the MARS model demonstrated very good accuracy. The areas under the ROC curves were ۰.۷۷۱, ۰.۷۶۷, and ۰.۸۲۲ for the FR, GLM, and MARS models, respectively. The susceptibility classes show that ۳۷%, ۴۴%, and ۴۴% of the study area have high and very high susceptibility in the FR, GLM, and MARS models, respectively. Conclusion: The calculated susceptibility area indicates that the region is very susceptible to landslides, warranting careful attention in regional planning and development. Geographical datasets and landslide susceptibility maps provide valuable resources for sustainable planning in the area, land-use planning, and identification of vulnerable regions.

کلیدواژه ها:

Capital of Iran ، Frequency Ratio ، GLM ، Machine learning ، MARS ، ROC Curve ، Susceptibility Mapping