Assessment of alternative forest road routes and landslide susceptibility mapping using machine learning

Abstract

Background: Forest roads are among the most basic infrastructure used for forestry activities and services. To facilitate the increased amount of biomass harvesting adequately, the existing road network may require modifications to allow forest transportation within harvesting units that are not yet accessed by the roads. The construction of a forest road can trigger landslides, so the necessary constraints should be considered when the road is being planned to preclude such problems. Landslide Susceptibility Mapping (LSM) has become an integral part of the growing process of machine learning (ML), providing a more effective platform for practitioners, planners, and decision-makers. This study aims to reveal the most suitable alternative routes for a forest road, especially in areas susceptible to landslides, and to provide an effective tool for decision-makers. Results: For this purpose, two models were developed through ML: Logistic Regression (LR) and Random Forest (RF). Elevation, slope, aspect, curvature, Topographic Wetness Index (TWI), Stream Power Index (SPI), distance from the fault, the road, and the stream, and lithology were considered as the main landslide susceptibility factors in these models. The best model was obtained by the RF approach with an Area Under ROC Curve (AUC) value of 81.9%, while the LR model was 78.2%. LSM data was used as a base, and alternative routes were obtained through CostPath analysis. Conclusion: It has been shown that the ML methods used in this study can positively contribute to decision-making by providing more effective LSM calculations in studies to determine alternative routes in a forest road network.