Landslide susceptibility mapping using remote sensing and geographic information system: The Alakir River basin (Antalya, Turkey)

This study was conducted to map the sensitivity of remote sensing and Geographical Information Systems (GIS) landslide susceptibility of the Alakır Creek basin located within the boundaries of Antalya province. Landslides are among the natural disasters that are widespread in the world and our country (in special mountainous regions). In order to reduce the loss of life and property, the landslide susceptibility mapping work has become very important. The study area, being the Alakir catchment area, covers 212 km2 and is 77 km from the Antalya city center. Different parameters and indices were used in the study. Slope, aspect, TWI, LS factor, convergence index, plan curvature and profile curvature are the DEM derivatives produced. Distance parameters; distance to fault lines and distance to drainage network are used. Mineral ratios (Alunite, Kaolinite, Calcite) and decorrelation stretch vary among other parameters. Landslide inventory map and ASTER L3A satellite image are used as ready data. All the data to be used in the study were prepared for the Random Forest Algorithm from the machine learning algorithms together with remote sensing and image processing techniques. The area under the Receiver Operating Characteristic (ROC) curve (AUC) was used and the highest values were taken from the training-2 set. For the algorithm applied by using 10,100 and 1000 tree numbers, 0.860, 0.874 and 0.877 results were obtained, respectively. KEYWORDS: Antalya, Landslide susceptibility mapping, Machine learning, Random Forest, Support Vector Machine, Two-level random sampling