Assessing Machine Learning Models for Surface Mining Detection Using Sentinel-2 Data

Remote sensing data, particularly from Sentinel-2 satellites, offers valuable insights into surface mining activities. This study evaluates the effectiveness of Sentinel-2 imagery in detecting and monitoring sand and gravel extraction sites across five mining areas in Schleswig-Flensburg, Germany, from 2015 to 2019. Three machine learning algorithms—Random Forest (RF), Support Vector Machines (SVM), and Artificial Neural Networks (ANN)—were compared to determine the most accurate classification approach. Models were trained using three different training data scenarios to assess their performance. While RF and SVM demonstrated greater robustness than ANN, SVM outperformed the others when validated against ground truth data. Consequently, an optimized SVM-based classification model was developed and implemented in R to analyze temporal changes in the study areas over five years. The findings highlight the potential of integrating Sentinel-2 imagery with machine learning techniques for accurate and efficient monitoring of surface mining activities, offering a scalable approach for environmental management and land-use planning.