Determination of GIS-Based Landslide Susceptibility and Ground Dynamics with Geophysical Measurements and Machine Learning Algorithms

Abstract

Landslide is one of the major natural disasters that threatens engineering structures as well as complicates the construction process. There has been a rapid increase in studies to identify ground dynamics in areas with the potential for landslides. Landslide susceptibility maps are created using Support Vector Machine (SVM) and Random Forest (RF) machine learning algorithms based on geographic information systems to identify possible failures in selected areas. The aim of this study is to train different spatial data with machine learning algorithms to determine susceptible landslide areas, so as to analyze soil properties with the Multi-channel Analysis of Surface Waves (MASW) method, which is a fundamental shallow surface seismic surveying method in geophysical engineering. Also Refraction Microtremor (Re-Mi) method applied in some stations to detect shear wave velocity (V s) up to engineering bedrock level. Obtained velocity values of soil layers from different seismic methods and historical records were used together to train the model. The seismic surveying results were used for the first time to train the machine learning algorithms to detect high susceptible areas for landslides. Some of the MASW applications were carried out in landslide areas and others in areas considered to be risky. Thus, with the contribution of the seismic method, the dynamic behavior that may occur was analyzed. All the measurements carried out in the Girne (Kyrenia) Mountains terrane. Consequently, it has been determined that the northeast-facing slopes of the Girne Mountains are the highest sensitivity for landslide, in other words, the most active in terms of ground dynamics.