Multitemporal Spatial Analysis for Monitoring and Classification of Coal Mining and Reclamation Using Satellite Imagery

Observing coal mining and reclamation activities using remote sensing avoids the need for physical site visits, which is important for environmental and land management. This study utilizes deep learning techniques with a U-Net and ResNet architecture to analyze Sentinel imagery in order to track changes in coal mining and reclamation over time in Tapin Regency, Kalimantan, Indonesia. After gathering Sentinel 1 and 2 satellite imagery of Kalimantan Island, manually label coal mining areas are used to train a deep learning model. These labelled areas included open cuts, tailings dams, waste rock dumps, and water ponds associated with coal mining. Applying the deep learning model to multitemporal Sentinel 1 and 2 imagery allowed us to track the annual changes in coal mining areas from 2016 to 2021, while identifying reclamation sites where former coal mines had been restored to non-coal-mining use. An accuracy assessment resulted in an overall accuracy of 97.4%, with a Kappa value of 0.91, through a confusion matrix analysis. The results indicate that the reclamation effort increased more than twice in 2020 compared with previous years’ reclamation. This phenomenon was mainly affected by the massive increase in coal mining areas by over 40% in 2019. The proposed method provides a practical solution for detecting and monitoring open-pit coal mines while leveraging freely available data for consistent long-term observation. The primary limitation of this approach lies in the use of medium-resolution satellite imagery, which may result in lower precision compared to direct field measurements; however, the ability to integrate historical data with consistent temporal coverage makes it a viable alternative for large-scale and long-term monitoring.