Mapping Landslide Susceptibility in the Moldavian Plain, Romania

Shallow landslides represent a frequent geomorphological process in the study region, located in northeastern Romania. The area is characterized by gently undulating interfluves, fragmented slopes, and deeply incised valleys, developed predominantly on clayey substrates. These predominantly shallow slope failures have significant impacts on intensive agriculture, rural infrastructure, and slope stability. Recent climatic variability and anthropogenic modifications of land use amplify the vulnerability of this geomorphological unit. This study presents a detailed assessment of shallow landslide susceptibility through the integration of an extensive landslide inventory with conditioning factors derived from high-resolution geospatial data. The landslide inventory was developed predominantly using digital elevation models generated from LiDAR data (1–2 m resolution), complemented by current orthophotos, drone aerial imagery, slope maps, and selective field validation. The use of LiDAR data substantially improves the precision of delineating shallow unstable features and reduces propagation errors associated with conventional lower-resolution DEMs. This methodology enabled the precise delineation of hundreds of active and relict shallow landslide features, surpassing the limitations of traditional inventories based on photogrammetry or global DEMs.
Relevant conditioning factors for slope dynamics in this region included slope angle, aspect, plan and profile curvature, lithological units (predominantly Miocene-Pliocene clayey deposits), land use, and distance to the drainage network. The dataset was divided into 70% for calibration and 30% for independent validation. The Presence Only Model performance was evaluated through ROC curves and AUC metrics, with values consistently demonstrating excellent predictive performance of the hybrid approach employed.
Results highlight zones of high and very high susceptibility to shallow landslides concentrated along major valleys and their tributaries, and on slopes exceeding 12–15°, where favourable lithological conditions overlap with intensive agricultural land uses or reduced vegetation cover. Methodologically, this study aligns with established international approaches for landslide susceptibility assessment but distinguishes itself through the use of high-resolution LiDAR data (1–2 m), specifically adapted to the morphological context of the region—an area with gently rolling relief and deeply incised valleys. This choice enables substantial reduction of topographic uncertainties inherent in models based on medium or low-resolution DEMs, thereby improving the precision of shallow instability feature delineation and the robustness of local predictions. The result is a susceptibility model with high transferability potential to other similar geomorphological units in plain-to-hill transition zones affected by shallow landsliding.