Creating a landslide inventory in the Eastern Cape Province, South Africa: a pixel-based change detection method using fuzzy membership functions

Landslide mapping is essential for hazard assessment and mitigation efforts, with remote sensing techniques offering valuable tools for detecting and monitoring changes in land cover associated with landslide events. In this study, a pixel-based change detection approach with fuzzy membership functions was employed to map landslide scars in both the Pinetown and Port St. Johns areas of South Africa. Covariates including the Normalized Difference Vegetation Index (NDVI), slope, and Vertical Distance to Channel Network (VDCN) were selected to detect changes in land cover indicative of landslide occurrence. High-resolution satellite imagery from PlanetScope (3 m) was utilized to capture small-scale landslide features, enhancing the accuracy of the mapping process. Results from the calibration area (Pinetown) have a Kappa value of 0.60, F1 score of 0.93, and an overall accuracy of 0.88, whereas extrapolation areas (Port St. Johns) resulted in a Kappa value of 0.70, F1 score of 0.95, and an overall accuracy of 0.92%. Challenges such as spatial incoherence and over-detection were encountered, highlighting the need for careful interpretation and refinement of the mapping approach. Extrapolation of the pixel-based change detection method from the Pinetown to the Port St. Johns area revealed limitations associated with differences in land use practices, surface cover, and availability of satellite imagery. Despite these challenges, the approach detected additional landslide occurrences in both study areas. Overall, this study demonstrates the effectiveness of pixel-based change detection with fuzzy membership functions for mapping landslide scars in diverse settings.