Landslide identification and susceptibility assessment in high-consequence areas of the oilfield based on multi-band InSAR and machine learning

Loess landslides pose a serious threat to energy infrastructure on the Loess Plateau; however, studies on landslide identification and monitoring in high-consequence areas of oil and gas facilities remain limited. This study investigates landslide identification and susceptibility assessment in a high-consequence area of the Changqing Oilfield in Qingyang City, Gansu Province, China, using Lutan-1 and Sentinel-1 Synthetic Aperture Radar data and machine learning. Combining optical remote sensing images and applying the Minimum Covariance Determinant method and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to the InSAR derived deformation results, 65 potential landslides are identified. A random forest model, incorporating 13 conditioning factors, is constructed to generate a susceptibility map, with Shapley Additive explanations (SHAP) analysis used to interpret factor contributions. The results indicate that high susceptibility zones are predominantly controlled by surface roughness, distance to streams, and land surface temperature, and contain the majority of identified landslide points. 21.7% of oilfield facilities fall within these high-risk zones, highlighting the urgent need for continuous monitoring. This integrated framework provides a practical approach for geological hazard prevention in loess areas with critical energy infrastructure.