Change Detection of Amazonian Alluvial Gold Mining Using Deep Learning and Sentinel-2 Imagery

Monitoring changes within the land surface and open water bodies is critical for natural resource management, conservation, environmental policy. While use of satellite imagery these purposes common, fine-scale change detection can be a technical challenge. Difficulties arise from variable atmospheric conditions problem assigning pixels to individual objects. We examined degree which two machine learning approaches better characterize in context current conservation challenge, artisanal small-scale gold mining (ASGM). obtained Sentinel-2 consulted with domain experts construct an open-source labeled land-cover dataset. The focus this dataset Madre de Dios (MDD) region Peru, hotspot ASGM activity. also generated datasets active areas other countries (Venezuela, Indonesia, Myanmar) out-of-sample testing. With data, we utilized supervised (E-ReCNN) semi-supervised (SVM-STV) approach study binary multi-class ponds MDD region. Additionally, tested how inclusion multiple channels, histogram matching, La*b* color metrics improved performance models reduced influence effects. Empirical results show that E-ReCNN method on 6-Channel histogram-matched images most accurate not only focal (Kappa: 0.92 (± 0.04), Jaccard: 0.88 0.07), F1: 0.05)) but prediction regions 0.90 0.03), 0.84 0.77 0.04)). methods did perform as accurately 6- or 10-channel imagery, matching low memory costs. These capable detecting specific object-oriented related ASGM. scalable outside area extended forms land-use modification.