Document Type : Research Paper

Authors

• Aliakbar Nazari Samani ¹
• Reza fathiganji ²
• Sadat Feiznia ³
• Abdolreza Nourizdan ⁴

¹ Faculty of Natural Resources University of Tehran

² University of Tehran

³ Faculty of Natural Resources, University of Tehran,

⁴ General Department of Natural Resources and Watershed of Lorestan, Iran

Abstract

andslides, as a geo-hazard and disturbance, can exert significant direct and indirect impacts on human communities, the natural environment, and the landscape. Determining the factors influencing vulnerability to landslides is therefore essential for hazard management. This study aims to identify the key factors affecting land sensitivity to landslides and to create a landslide susceptibility map through the integration of machine learning methods and statistical modeling In this study, the random forest algorithm was used to assign weights and importance rankings to effective factors, including slope, slope aspect, TPI,TWI, Plan curvature, lithology, distance from faults, distance from streams, distance from roads, and landuse. Subsequently, artificial neural network and MaxEnt techniques were applied to model and predict landslide-prone areas.To assess the model accuracy, validation was conducted within the Kakashraf watershed boundary, and evaluation metrics were calculated. The results indicated that the distance from faults and slope are the most significant factors in determining land susceptibility to landslides. Based on the Area Under the Curve (AUC) metric, the neural network model with an AUC of 0.92 showed greater accuracy in predicting landslide-prone areas compared to the MaxEnt model, which had an AUC of 0.801.The highest vulnerability was found in lands near watercourses, within 200 meters of fault lines, and on slopes of 20-40%. Therefore, managing and monitoring land use in these areas is of high priority. These findings can be instrumental in improving land management and planning in landslide-prone areas,protecting natural resources, and facilitating effective risk management.

Keywords

• Sensitivity
• MaxEnt
• random forest
• neural network