Testing the performances of change detection approaches in mapping post-event landslides

Reliable landslide inventory maps are required to create susceptibility, hazard, and risk maps for landslides. In geographies with heavy rainfall, such as in the Black Sea Region, many shallow landslides usually occur after an event. Owing to the climatic conditions of these shallow landslides, the vegetation becomes greener over time, and the landslide areas are closed. Therefore, methods that offer fast solutions are required to create accurate and reliable landslide inventories in these regions. In recent years, satellite technologies and the temporal and spatial high-resolution images obtained from these technologies have offered effective solutions for mapping landslides. In this thesis, the success of change detection approaches in post-event rapid landslide mapping is tested using PlanetScope satellite images, which provide a one-day temporal and spatially 3 m resolution image. In this context, the mapping of landslides triggered by heavy rains on August 22, 2020, in Eriçek village of Espiye district of Giresun province and its surroundings was carried out with change detection. In the change detection process, maps were produced according to Principal Component Analysis(PCA), Independent Component Analysis(ICA), Change Vector Analysis(CVA) and Normalized Difference Crop Index (NDVI) methods. The automatic threshold determination methods used in the literature, Kittler, Otsu, Tsai, and Kapur methods, and a manual threshold determination approach were applied to determine the landslide and non-landslide areas in the produced maps. The accuracy of the landslides was compared with that of the landslides produced by visual interpretation by an expert in the study area. As a result, he recorded the more accurate results of the NDVI approach and the Tsai determination method for mapping rain-induced landslides. In addition, the usability of the PlanetScope satellite images used within the scope of the study in rapid landslide mapping has been demonstrated.