Landslide susceptibility mapping using remote sensing data and geographic information system-based algorithms

摘要: Whether they occur due to natural triggers or human activities, landslides lead loss of life and damages properties which impact infrastructures, road networks buildings. Landslide Susceptibility Map (LSM) provides the policy decision makers with some valuable information. This study aims detect landslide locations by using Sentinel-1 data, only freely available online Radar imagery, map areas prone a novel algorithm AB-ADTree in Cameron Highlands, Pahang, Malaysia. A total 152 were detected integration Interferometry Synthetic Aperture RADAR (InSAR) technique, Google Earth (GE) images extensive field survey. However, 80% data employed for training machine learning algorithms remaining 20% validation purposes. Seventeen triggering conditioning factors, namely slope, aspect, elevation, distance road, river, proximity fault, density, river Normalized Difference Vegetation Index (NDVI), rainfall, land cover, lithology, soil types, curvature, profile Stream Power (SPI) Topographic Wetness (TWI), extracted from satellite imageries, digital elevation model (DEM), geological maps. These factors utilized generate susceptibility maps Logistic Regression (LR) model, Model Tree (LMT), Random Forest (RF), Alternating Decision (ADTree), Adaptive Boosting (AdaBoost) hybrid ADTree AdaBoost models, model. The was based on area under ROC curve (AUC) statistical measurements Positive Predictive Value (PPV), Negative (NPV), sensitivity, specificity, accuracy Root Mean Square Error (RMSE). results showed that AUC 90%, 92%, 88%, 59%, 96% 94% LR, LMT, RF, ADTree, algorithms, respectively. Non-parametric evaluations Friedman Wilcoxon also applied assess models’ performance: findings revealed is inferior other models used this study. Using handheld Global Positioning System (GPS), performed almost (30 locations) correctly detected. In conclusion, can be applicable hazard mitigation purposes regional planning.