Impact of land use change on landslide susceptibility in Pettimudi, Idukki district using GIS and Machine learning techniques

andslides are one of the most destructive natural disasters in the hilly terrains of Kerala, especially in the ecologically sensitive regions of the Western Ghats. The study area focuses on Pettimudi, a landslide-prone region within the Western Ghats. The increasing frequency and intensity of these events have underscored the importance of accurate and efficient landslide susceptibility mapping (LSM) for effective hazard mitigation and planning. This study focuses on the preparation of landslide susceptibility map using GIS-based composite Landslide Susceptibility Assessment Toolbox (LSAT)[2] in ArcGIS Pro 3.4, which integrates data preparation, model execution, and validation into an automated workflow. The study utilizes three widely accepted bivariate and multivariate statistical models, Frequency Ratio (FR), Logistic Regression (LR) and Multi - Layer Perceptron Model (MLP) to assess susceptibility. A total of nine causative factors were selected for analysis, including aspect, slope, soil, curvature, elevation, NDVI (Normalized Difference Vegetation Index), precipitation, Topographic Roughness Index (TRI), and Topographic Wetness Index (TWI) maps. These factors were reclassified and processed through the toolbox to generate susceptibility maps. In addition to the core LSM objective, the study also investigates the influence of LULC on landslide distribution, incorporating a change detection analysis between 1990 and 2020 to evaluate the effect of anthropogenic activities such as the expansion of tea plantations. LULC maps for 1990 and 2020 are created using Google Earth Engine (GEE), and an intersection matrix is developed to analyze which land use categories are predominant within high, moderate, and low susceptibility zones. This helped to understand the correlation between land use practices and landslide occurrences more clearly. The training and testing datasets are prepared using the landslide inventory of the region, ensuring robust model calibration and validation. Model performance was assessed using Receiver Operating Characteristic (ROC) curves and the Area Under the Curve (AUC) values, demonstrating reliable predictive capability for FR, LR and MLP models. The final susceptibility maps were categorized into high, moderate, and low zones. The toolbox developed through this study serves as a user-friendly, replicable platform for landslide susceptibility analysis, enabling faster decision-making and promoting sustainable land use practices in landslide-prone areas.