A new study employing stacked encoding and AutoML techniques has offered significant breakthroughs in accurately identifying lead-zinc open pit mining areas around the Rampura Agucha region of Rajasthan, India. Utilizing Sentinel-2 satellite imagery, the researchers implemented machine learning (ML) algorithms to tackle the challenges typically associated with traditional remote sensing methods, characterized by spectral signature class mixing.
The core objective of this study was to effectively detect and classify the regions of interest, namely the lead-zinc open pit mines, integrating various spectral indices and band ratios to improve analysis accuracy. The findings indicated remarkable performance, with the Extra Trees classifier achieving the highest overall accuracy of 94.28%, underlining the potential for such advanced technological applications to inform sustainable resource management practices.
Mineral extraction, particularly of lead and zinc ores, has escalated over the years. The Rampura Agucha mine is reputed for possessing one of the most significant lead-zinc deposits globally, amounting to 6.15 million metric tons of reserves. This area, characterized by rugged terrain and dense semi-arid forests, presents inherent monitoring difficulties.
Using pre-monsoon data, the study mitigated atmospheric errors typically caused by seasonal weather variations, enhancing the effectiveness of spectral data. Multiple machine learning classifiers, including contemporary algorithms such as LightGBM and Random Forest, were assessed alongside the Extra Trees model. Each of these classifiers demonstrated guarantees of separation between active mining zones and other land cover features.
This study highlights the increasing importance of integrating satellite imagery with automated machine learning techniques, especially as the complexity of environments grows. Conventional survey methods are often time-consuming and may miss identifying areas at risk of illegal mining practices. Through automated classification derived from comprehensive spectral analysis, authorities can achieve real-time monitoring of mining activities, aiding efforts to enforce environmental regulations and minimize ecological impacts.
Efforts like this empower decision-makers to implement informed policies, as current knowledge gaps hampering effective mineral resource management can be addressed systematically. The successful deployment of these machine learning models signifies both practical and scientific advancements, effectively bridging existing challenges of invisible mining operations across under-monitored regions.
Going forward, the enriched methodology established through AutoML and stacked encoding approaches will pave the way for future explorations and technological innovations, showcasing the power of integrating remote sensing capabilities with advanced digital analysis for broader applications.