Developing Scalable Monitoring System for Acid Mine Drainage Detection

This study focuses on advancing the development of effective monitoring systems Acid Mine Drainage (AMD) by leveraging Machine Learning techniques on optical multi and hyperspectral data. More specifically, the research investigates the utilization of hyperspectral data (PIKA L) acquired through Unmanned Aerial Vehicles (UAV) and multi-temporal data sets from the Sentinel-2 satellite. The results of ML classifications have been validated using ground truth, and it has been determined that the Radial Basis Function Support Vector Machine (RBF SVM) and Random Forest (RF) performs better than other tested ML approaches demonstrating especially effectiveness in handling high-dimensional spaces, which is crucial for hyperspectral data. Future work will focus on testing machine learning techniques on extended multi-temporal data sets, expanding the training and validation data sets to validate results across all scales and evaluating the transferability of the model to other geographical locations.