مدلسازی حساسیت مکانی وقوع زمین‌لغزش با استفاده از الگوریتم یادگیری ماشین جنگل تصادفی در امتداد جاده هراز

Background and aim:

Landslides, as a major geomorphological hazard, threaten transportation safety along mountainous roads due to their severe human, economic, and environmental impacts. In this context, spatial landslide susceptibility modeling—considered a key step in landslide hazard and risk assessment—plays a crucial role in mitigating potential damages. In recent years, the rapid development of machine learning algorithms has significantly enhanced the capability of spatial modeling for landslide susceptibility. The main objective of this study is to apply the random forest machine learning algorithm to model the spatial susceptibility of landslide occurrence along the Haraz mountainous road. Haraz road is continuously exposed to various types of landslides due to its complex and diverse geological structures, landslide-prone lithological conditions, climatic factors, and intensive human activities, and is regarded as one of the most hazardous mountainous roads in Iran.

Materials and methods:

To model landslide susceptibility, the spatial distribution of landslides was first identified and compiled as a landslide inventory map. This inventory was prepared based on detailed field surveys, geological maps, and satellite images. Subsequently, a set of independent conditioning factors including geological, topographic, hydrological, vegetation, land-use and land-cover, and human-related factors, were generated within a one-kilometer buffer zone along the Haraz road. Due to the imbalance between landslide and non-landslide pixels in the study area, a balanced dataset with an equal number of landslide and non-landslide pixels was constructed. From this dataset, 70% of the pixels were randomly selected for model training, while the remaining 30% were used for validation. To evaluate the effect of random data partitioning on model performance, a 10-fold cross-validation approach was implemented within the random forest algorithm, and the training–testing process was repeated ten times. In addition, to improve the predictive performance of the random forest model, hyperparameter tuning was applied. Key algorithm parameters, including the number of decision trees and the number of variables selected at each node of split, were systematically tuned. The accuracy and performance of the ten training models and ten validation models were evaluated using confusion matrix, Kappa coefficient, and the Area Under Curve (AUC). Finally, the model with the highest AUC value in the validation stage was selected as the optimal model and used to generate spatial landslide susceptibility map.

Resulst and discussion:

The results related to the importance of conditioning factors affecting landslides in random forest algorithm, indicate that the geology is the most influential variable controlling landslide occurrence along the Haraz road, with a relative importance of 47.8%. In this regard, geological units consisting of scree deposits and the Shemshak formation play a dominant role in triggering landslides in the study area. Regarding the spatial modeling of landslide susceptibility using the random forest machine learning algorithm combined with a 10-fold cross-validation approach, the results demonstrate a high predictive accuracy and robust performance in identifying landslide-prone areas. The mean Area Under the Curve (AUC) values obtained from the ten training iterations and ten validation iterations were 0.93 and 0.85, respectively, indicating the strong capability of the proposed approach for reliable landslide susceptibility assessment along mountainous road corridors.

Conclusion:

The results indicate that landslide spatial susceptibility along the Haraz road is variable and heterogeneous. Within the one-kilometer buffer zone along the road corridor, approximately 130 km² of the study area is classified as high and very high prone landslide areas, encompassing nearly 36 km of the Haraz road length. The findings of this study provide valuable insights for geomorphological hazard management along this road and can be effectively used for prioritizing technical and engineering interventions. In particular, the generated landslide susceptibility map can support the assessment of unstable slopes, the implementation of slope stabilization measures in high- and very high-risk zones, and strategic planning aimed at enhancing road safety and reducing landslide-related risks.