Design and evaluate a GIS-based model for landslide susceptibility mapping

Thesis Abstract

Landslides pose a significant threat to life, infrastructure, and economic development in hilly and mountainous regions, demanding effective risk assessment and management strategies. Despite the increasing availability of geospatial data and advancements in Geographic Information Systems (GIS), there remains a critical need for robust, spatially explicit models that accurately delineate landslide-prone areas to inform land use planning and disaster mitigation efforts. This study aims to design, implement, and evaluate a GIS-based landslide susceptibility model tailored to the mountainous region of the Central Highlands, utilizing an integrated approach of remote sensing, spatial analysis, and statistical modeling. The specific objectives include (i) identifying and mapping key landslide conditioning factors such as slope, lithology, land cover, and proximity to geological faults; (ii) developing a landslide susceptibility model using multi-criteria evaluation (MCE) and statistical techniques, particularly logistic regression analysis; (iii) validating the model through field verification and statistical accuracy assessments, including Receiver Operating Characteristic (ROC) curve analysis; and (iv) providing a comprehensive framework that can be adopted in regional disaster management planning. The study adopts a quantitative research design within a positivist paradigm, treating spatial and non-spatial data as primary sources for model development. The population for this research encompasses all landslide-affected zones and potential hazard areas within the study region, with a sample comprising 150 landslide locations confirmed through existing geological records and field reconnaissance. Data collection involved acquisition of high-resolution satellite imagery from Sentinel-2, Digital Elevation Models (DEMs), geological maps, land cover datasets, and field surveys that documented landslide occurrences and environmental parameters. Spatial data preprocessing included georeferencing, layer integration, and reclassification to ensure consistency. The validity and reliability of data were ensured through cross-verification with geological agency records and ground-truthing during field visits. Analytical methods employed include weighted overlay analysis for factor suitability and logistic regression for susceptibility classification, implemented within ArcGIS and R statistical software. The logistic regression model's performance is evaluated via ROC analysis, with model accuracy interpreted through Area Under Curve (AUC) scores. Additionally, the model's predictive capability is assessed through k-fold cross-validation to prevent overfitting. The findings are expected to reveal statistically significant relationships between selected conditioning factors and landslide occurrence, leading to a hierarchical susceptibility map categorizing zones into very high, high, moderate, low, and very low hazard levels. Anticipated results indicate that slope gradient, proximity to geological faults, and land cover type are the most influential variables. This research makes a notable contribution to advancements in spatial risk modeling by providing an empirically validated, GIS-based susceptibility framework that can be adapted and applied in similar geological contexts. It bridges a gap in localized landslide risk assessment models where prior studies relied heavily on qualitative or semi-quantitative approaches, thereby offering a more precise, data-driven tool for policymakers and land-use planners. The study also enriches the theoretical understanding of landslide susceptibility, supporting the applicability of the logistic regression model within a spatial multi-criteria evaluation framework, underpinned by Tobler's First Law of Geography and the Multiple Factors Theory. The main conclusion emphasizes that integrating remote sensing data with geostatistical analysis significantly enhances landslide prediction accuracy. The study recommends the adoption of the developed GIS-based model in regional land-use and disaster risk mitigation policies, alongside continuous updating with new spatial data for dynamic hazard assessment. Future research is suggested to incorporate climate variables and socio-economic factors to refine susceptibility models further and improve disaster preparedness strategies.