Objective: One of the destructive natural hazards that causes great is the phenomenon of landslides. Due to its geographical conditions, Iran witnesses this hazard every year. Fereydounshahr County in the west of Isfahan Province is one of the high-risk areas for landslides. Therefore, it is necessary to investigate, identify susceptible areas, identify factors affecting this hazard, and prepare spatial maps more than ever before. In this regard, the purpose of this study is to evaluate the susceptibility of landslides in Fereydounshahr County using support vector machine and random forest algorithms.
Methods: One of the most important and destructive natural hazards that causes great loss of life and property is the phenomenon of landslides. Due to its geographical conditions, Iran witnesses this hazard every year. Fereydounshahr County in the west of Isfahan Province is one of the high-risk areas for landslides. Therefore, it is necessary to investigate, identify susceptible areas, identify factors affecting this hazard, and prepare spatial maps more than ever before. In this regard, the purpose of this study is to evaluate the susceptibility of landslides in Fereydounshahr County using support vector machine and random forest algorithms.
Results: In this study, first, a landslide distribution map of Fereydounshahr County was prepared with 151 points. 70% of the randomly selected points were used for training the models and the other 30% were used for evaluating and validating the models. Based on the study of various sources and the conditions of the study area, 9 effective criteria in landslide occurrence were identified using random forest and support vector machine algorithms to prepare a landslide susceptibility map, including height, slope, slope direction, vegetation, geology, distance from fault, distance from river, distance from road, and precipitation, and their maps were prepared and prepared in the ArcGis10.8.2 environment. Then, a landslide risk susceptibility map in Fereydounshahr County was prepared using the two aforementioned models in 5 classes, and the relationship between the criteria and the results of the models was statistically examined. Finally, the area under the ROC curve method was used to evaluate the models.
Conclusions: The analyses and visual analysis of the results show that high-risk areas (high and very high) are mainly located in steeply sloping areas, close to faults. Among the factors affecting the occurrence of landslides based on the prioritization of the slope factor, precipitation, vegetation parameter, and slope direction have the greatest impact.