Assessing the influence of mining-induced subsidence on landslide susceptibility : a machine learning approach using remote sensing and geospatial data for the Grasberg-Ertsberg cave mining operations

In deep-cave mining operations, progressive cave propagation can induce widespread deformation, seismicity, and surface subsidence, which interact directly with natural slopes and the existing open-pit geometry. These coupled processes can increase the likelihood of slope instability and landslides, disrupting operations, threatening surface infrastructure, and endangering workers' safety. Despite the importance of understanding these cave-mine-slope interactions, practical and defensible frameworks that quantify where, when, and how far mining-induced subsidence contributes to landslide susceptibility remain limited, especially in complex mountainous terrain where natural dominant triggers such as rainfall coexist with mining-related drivers. This thesis develops an integrated spatial-temporal susceptibility framework to evaluate landslide susceptibility triggers in a complex mountainous area influenced by deep caving operations (the Grasberg Block Cave and Deep Mill Level Zone) at PT Freeport Indonesia, with a focus on the interactions between those mines and surface slope instabilities. Multi-source remote sensing and geotechnical datasets are compiled into a consistent grid-based inventory spanning 2014 to 2024, including landslide occurrences, topography, hydrology, environmental factors, geology, and mining-related factors. A suite of machine-learning classifiers is trained under severe class imbalance using cost-sensitive learning and resampling strategies. The model's robustness is assessed using year-based holdouts for out-of-sample susceptibility map testing. Results demonstrate that incorporating deformation- and proximity-derived predictors yields clearer discrimination of hazardous terrain and provides quantitative evidence of the relative contributions of mining-related drivers between pre- and post-subsidence periods. The proposed workflow provides a defensible, reproducible basis for separating mining-induced from natural drivers, supporting risk-informed monitoring, prioritization, and long-term hazard management.