A comparative study among ensemble and independent statistical models: a high-accuracy and conservative framework for landslide susceptibility mapping in east sikkim himalaya

The Sikkim Himalaya, situated in the North-East Himalaya, is susceptible to landslides induced by precipitation, anthropogenic activities, and various natural calamities. This study aims to evaluate landslide potential in East Sikkim by developing a precise Landslide Susceptibility Map (LSM) with high conservativeness and examining the individual and combined effects of two statistical methods, such as, Evidence Belief Function (EBF) Method and the Frequency Ratio (FR) method, on landslide susceptibility. A further objective was to evaluate whether high model accuracy necessarily corresponds with a conservative prediction. Here, conservativeness is defined as a model’s inclination to err on the safer side by prioritizing the avoidance of failing to detect landslide-prone zones. This study considers twenty causative factors for landslides, including: Elevation, Slope, Aspect, Profile Curvature, Relief Amplitude, Topographic Wetness Index (TWI), Topographic Roughness Index (TRI), Roughness, Sediment Transport Index (STI), Stream Power Index (SPI), Topographic Position Index (TPI), Distance to Drainage (DTD), Distance to Lineament (DTL), Rainfall, Lithology, soil Geomorphology, Land Use Land Cover (LULC), Normalized Difference Vegetation Index (NDVI), Distance to Road (DTR) have been created utilizing diverse data from multiple sources in ArcGIS environment. Eight scenarios corresponding to individual EBF, FR, and the ensemble of these methods have been developed. After developing these scenarios, LSMs have been developed for each scenario. The accuracy of each LSM has been examined by the Receiver Operating Characteristic (ROC) curve. It has been demonstrated that an ensemble of different methods may sometimes help in increasing LSM accuracy. This study indicated that high accuracy does not necessarily equate to a conservative LSM; hence, LSM conservativeness should be evaluated in conjunction with accuracy prior to making recommendations. High accuracy and conservative LSM for East Sikkim has been attained in this study.