A Sensitivity Assessment Framework for Frequency Ratio-Based Landslide Susceptibility Models: Evaluating the Role of Data Classification and Zonation

Abstract

Landslide Susceptibility (LS) maps are essential tools for hazard assessment, yet the reliability of such maps is highly sensitive to methodological choices in data classification and discretisation. Although Frequency Ratio (FR) modelling is widely applied in LS assessments, limited research has systematically evaluated how the selection of classification techniques and classification schemes affects the accuracy and zoning of Landslide Susceptibility Index (LSI) outputs. This study proposes a sensitivity assessment framework that analyses the influence of four standard classification methods Natural Breaks, Quantile, Geometrical, and Equal Intervals under varying classification levels (4, 6, and 8 classes) on Frequency ratio-based LS modelling. Applied to two landslide-prone regions in Brunei Darussalam, the framework reveals that classification choices significantly alter predictive performance, map zoning, and model Receiver Operating Characteristic (ROC)-Area Under the Curve (AUC) rates (AUC=88%–97.7%). Of the 12 scheme–class-count combinations tested, a Geometrical-interval scheme with six classes gave the highest AUC (97.7 %) in the Kota Batu–Subok study areas, while a Quantile scheme with six classes was best for Jangsak–Tutong (AUC = 94.6 %). The framework is demonstrated on two rainfall-triggered hillslope study areas near Bandar Seri Begawan, Brunei Darussalam: Jalan Kota Batu–Subok (≈ 53 km²; 134 mapped debris-fall events) and Jalan Jangsak–Tutong (≈ 41 km²; 115 events), where steep sandstone–shale terrain and intense monsoon downpours routinely initiate shallow slides. We further show that varying classification can distort susceptibility zoning, potentially leading to misinformed risk decisions. These findings underscore the critical need for further investigation of classification methods in landslide modelling workflows and provide a reproducible method to reduce model uncertainty. Reproducibility is achieved through shareable ArcGIS ModelBuilder scripts and companion Excel templates that automate factor re-classification, frequency-ratio weighting, and fixed-seed validation, enabling any user with standard ArcGIS and Excel to recreate the maps and uncertainty metrics. The frequency-ratio probability maps allow planners to set evidence-based no-build buffers, rank slopes for stabilization works, and calibrate local rainfall-threshold warning systems, directly linking the research to real-world hazard mitigation.