Evaluating the Efficiency and Predictive Accuracy of Temporal Susceptibility Models for Co-Seismic Landslides Using Real-Time Validation: A Case Study from the NW Himalayas

The Himalayan region, characterized by its rugged terrain, distinctive geography, and active tectonics, ranks among the most landslide-prone zones globally. Landslide susceptibility and hazard mapping are critical tools to mitigate future risks and devise effective management strategies. This study uses data-driven statistical approaches to evaluate co-seismic landslide susceptibility in District Hattian, NW Himalayas, Pakistan. A comprehensive co-seismic landslide inventory comprising 349, 393, and 735 landslide events from 2005, 2007, and 2012, respectively, was utilized to train, test and validate predictive models. 
Thirteen landslide causative factors (LCFs), including topographic, environmental, geologic, and anthropogenic variables, were analyzed to determine their influence on landslide occurrence. Three data-driven statistical models i.e., Weight of Evidence (WoE), Information Value (IV), and Frequency Ratio (FR) were employed to develop landslide susceptibility maps (LSMs). Model training used 70% of the landslide inventory, while 30% was reserved for validation. Model performance was evaluated using Receiver Operating Characteristic-Area Under Curve (ROC-AUC) metrics and predictive accuracy assessments. Among the models, the WoE approach outperformed well among the other models as ROC-AUC SRC scores of 84.4, 84.2, and 85.3 for 2005, 90.4, 86.4, and 87.2 for 2007, and 81.9, 86.7, and 85.9 for 2012 for WoE, FR, and IV models, respectively. PRC scores of the WoE, FR, and IV models were recorded as 85.7, 89.4, and 82.5 for 2005, 87.5, 77.5, and 80.4 for 2007, and 80.7, 88.3, and 87.7 for 2012. For the validation of long-term predictivity, efficiency models are checked by comparing the generated LSMs with newly recorded landslide events. The 2005 model was validated using 2007 data, the 2007 model with 2012 data, and the 2012 model with 2024 data. Results revealed a gradual decline in the predictive accuracy of the LSMs model of all three approaches over time; however, WoE consistently outperformed from the IV and FR models, maintaining robust predictive capabilities even after 12 years.
This study highlights that landslide-prone zones in District Hattian exhibit persistent mass movement activity and underscores the urgent need for proactive landslide management to minimize life loss and economic damage in this tectonically active region. The integration of advanced susceptibility modelling techniques with real-time validation offers a reliable framework for hazard assessment and risk mitigation. Policymakers and stakeholders are encouraged to implement targeted interventions, such as optimized land-use planning, the establishment of early warning systems, and increased community awareness programs, to enhance resilience against landslide hazards in the NW Himalayas.