Production of landslide susceptibility maps by machine learning techniques: Hopa (Artvin) example

The region where landslides are most common in Turkey is the Eastern Black Sea Region. The landslides occurring in the provinces of Trabzon, Rize, Giresun and Artvin in the Eastern Black Sea Region constitute approximately 20% of the landslides occurring in Turkey. Landslides stand out as the primary natural disaster type in settlements with rugged topography such as Artvin. In the coastal areas of Artvin, landslides triggered by heavy rainfall cause loss of life, infrastructure and superstructure damage, destruction of the natural environment and serious economic damage. In this study, it is aimed to produce landslide susceptibility maps of Hopa district of Artvin by using Random Forest (RF) and Extreme Gradient Boosting (XGBoost) algorithms. In the study, 10 factors (lithology, slope, land cover, elevation, aspect, curvature, topographic wetness index, proximity to fault lines, road and drainage networks) were used, which were determined to be effective in the occurrence of landslides. The performance of the machine learning models used in the study was evaluated using the Receiver Operating Characteristic (ROC) curve and the area under the ROC Curve (AUC) approach. As a result of the evaluation, it was determined that the XGBoost algorithm performed better than the RF algorithm. In addition, it has been determined that the most effective factors in the occurrence of landslides in the study area are elevation, proximity to fault lines, slope, lithology and land cover, respectively. Curvature was determined as the least effective or least important factor. As a result, it was concluded that the landslide susceptibility map produced by the XGBoost algorithm can guide the decision makers to reduce the landslide damages in the region.