ABSTRACT Landslides are significant geological hazards in mountainous regions, posing severe risks to human life and property. In the Darjeeling District of the Eastern Himalaya, frequent recurrence of landslides highlights the need for accurate prediction models and effective landslide susceptibility mapping. This study develops a GIS-based Artificial Neural Network (ANN) model to generate landslide susceptibility maps using thirteen conditioning factors, including soil, Land Use/Land Cover (LULC), Topographic Wetness Index (TWI), geology, rainfall, slope, elevation, Drainage Density (DD), curvature, Normalized Differential Vegetation Index (NDVI), Stream Power Index (SPI), aspect, and Lineament Density (LD). A total of 172 landslide locations were used to construct the landslide inventory map. The ANN model consists of one input and one output layer, each with 13 and 11 neurons, respectively. Factor importance is evaluated by sensitivity analysis using an ANN model with the integration of multicollinearity analysis. Model performance is assessed using Receiver Operating Characteristic (ROC)-based Area Under the Curve (AUC) and additional statistical measures. The final susceptibility map delineates five classes of landslide-prone areas ranging from very low to very high, covering a significant (95%) area within the moderate to high class. Sensitivity analysis reveals soil, LULC, and TWI as the dominant contributors to landslide occurrence. The ANN model exhibits a higher predictive capability in validation (AUC = 0.991) compared to testing (AUC = 0.713) on the dataset. These outcomes provide critical guidance for hazard mitigation, infrastructure development, and land-use planning, offering policymakers and disaster management authorities of this district a reliable tool to minimise landslide risks by developing an early warning system.
Shaw et al. (Sun,) studied this question.