Document Type : Research Paper
Authors
1
Mousa Abedini Professor Department of Geomorphology Faculty of Literature and Humanities University of Mohaghegh Ardabili- Ardabil,Iran
2
Ph.D Student, Department of Physical Geography, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardabil, Iran.
10.22034/gp.2026.72191.3538
Abstract
Objective: The objective of this study is to analyze and map landslide susceptibility in the Shalmanrud watershed, Guilan Province, Iran, using machine learning algorithms including Support Vector Machine (SVM), Random Forest (RF), and Gradient Tree Boosting (GTB).
Methods: First, effective factors influencing landslide occurrence such as elevation, lithology, slope, slope aspect, distance to stream, distance to fault, distance to road, slope curvature, precipitation, and vegetation cover were compiled using remote sensing data, the 30-meter ASTER Digital Elevation Model (DEM), and other authoritative sources, followed by processing in Geographic Information System (GIS) and Google Earth Engine (GEE) environments. For modeling purposes, 450 identified landslide points were utilized as dependent variables and, along with non-landslide points, were randomly divided into training (70%) and validation (30%) datasets. Subsequently, the SVM, RF, and GTB models were trained within GEE to generate continuous landslide susceptibility maps. These maps were categorized into five susceptibility classes (ranging from very low to very high) using the natural breaks classification method.
Results: The results demonstrated that the spatial pattern of landslide hazard in the study watershed exhibits a decreasing trend from south to north. High-risk zones are predominantly concentrated in the southern and southwestern areas, characterized by steep slopes, high elevation, and unstable lithology, whereas northern regions display lower risk and relative stability. Model comparison revealed that all three algorithms achieved excellent performance in predicting landslide-prone areas, with Area Under the Receiver Operating Characteristic Curve (AUC-ROC) values of 0.999, 0.997, and 0.998 for RF, SVM, and GTB, respectively.
Conclusions: Overall, the findings highlight the high accuracy and reliability of these models in landslide susceptibility mapping. Variable importance analysis indicated that slope angle plays the most critical role in landslide triggering. These findings can be effectively applied to mitigate landslide-induced damages in similar geographic settings.
Keywords
Main Subjects
)