Landslide susceptibility mapping using ensemble learning integrating climate and seismic factors along the CPEC

Landslides represent a major global geohazard, and accurate landslide susceptibility mapping (LSM) is essential for disaster risk reduction, particularly under changing climatic and seismic conditions. This study develops an integrated machine learning framework to predict landslide susceptibility along the China-Pakistan Economic Corridor (CPEC). We evaluated and compared three individual models (Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) alongside an ensemble model, incorporating key climatic (e.g. rainfall indices, temperature) and seismic factors (e.g. fault density, epicenter density). Model performance was rigorously assessed using confusion matrix statistics, the area under the receiver operating characteristic curve (AUC), and landslide density (LD) validation. Results showed that the ensemble model outperformed all individual models across three distinct datasets (climate-inclusive, seismic-inclusive, and all-inclusive), achieving accuracies of 0.886, 0.865, and 0.850, respectively. The highest predictive accuracy (0.885) and AUC (0.95) were achieved using the all-inclusive dataset, underscoring the critical importance of data integration. The resulting susceptibility map classified 66.72% (626,038 km²) of the total study area (940,450 km²) as ‘Very Low’ risk and 10.51% (98,597 km²) as ‘Very High’ risk. This study provides a robust tool for informed land-use planning, infrastructure development, and proactive risk management along the CPEC.