Remote Sensing-Based Detection and Statistical Characterization of Large Landslides in the Transylvanian Basin

Landslides are geomorphological processes that have a significant impact on environmental components. As the Transylvanian Basin is landslide-prone, in this study, we used remote sensing and spatial analysis to detect and statistically characterize large landslides. Therefore, only landslides with an area greater than 0.5 km2 were analyzed. This approach helps fill existing knowledge gaps regarding the number, spatial extent, state of preservation, and geomorphological activity of landslides. A compact patch-based CNN trained on DEM-derived morphometric predictors was additionally used as a data-driven screening layer to support landslide identification. Digital elevation models (DEMs), optical RGB imagery, and Sentinel-2 multispectral images were used to detect and delineate landslides; spectral indices such as the NDVI, NDMI, and BSI were calculated from these images. The statistical characterization consisted of an analysis of morphometric parameters and geological characteristics. We assessed the distribution of large landslides and correlations between geomorphological variables and triggering factors and identified 171 landslides larger than 0.5 km2 in the Transylvanian Basin, including 82 ranging between 0.5 and 1 km2, 87 between 1 and 10 km2, and 2 exceeding 10 km2. Beyond providing valuable information for stakeholders involved in the management of large landslides in the Transylvanian Basin, these results significantly update current knowledge regarding such geomorphological processes.