GRSG 35th Conference 2024 Abstract
Title: Innovative Machine Learning Approach for Monitoring Acid Mine Drainage in the Rio Tinto Region of Spain Using Sentinel-2 Surface Reflectance Time Series Composites
Author: Veronika Strnadova
Organisation: Czech Geological Survey
Innovative Machine Learning Approach for Monitoring Acid Mine Drainage in the Rio Tinto Region of Spain Using Sentinel-2 Surface Reflectance Time Series Composites
Veronika Kopačková-Strnadová1,*
1Remote Sensing centre, Czech Geological Survey, Czech Republic
*Corresponding author: veronika.kopackova@seznam.cz
The Rio Tinto in Spain is known for its iron ore deposits as well as long-term mining activities. The region is named after the Rio Tinto river, which runs through the area and has a unique red color due to its high iron content. Mining activities in the region have caused environmental issues like air/soil/water pollution and Acid Mine Drainage (AMD) – a result of exposure of sulfide minerals during mining activities, which can lower the pH of water bodies, affect fish and other aquatic organisms, and degrade water quality. To monitor Acid Mine Drainage (AMD), a cutting-edge Machine Learning (ML) approach was utilized to analyze Sentinel-2 surface reflectance time series composites from 2018 to 2023, which were recently made available through the free S-2 Global Mosaic (S2GM-2) service. A new S-2 mineral index was developed to distinguish iron sulfate minerals from other mineral types, in conjunction with commonly used indexes like ferric iron and NDWI.
Two ML techniques, Random Forest (RF) and Radial Basis Function Support Vector Machine (RBF SVM), were evaluated and validated using two independent datasets: the Global Mining Land Use layer and Geochemical data from Rio Tinto, Spain, provided by USGS. The validation process showcased robust results, with RBF SVM surpassing RF in performance. RBF SVM effectively identified hazardous sulfate-bearing AMD substrates within all 215 mine footprint polygons located within the specified area of interest (10,000 km2).
Sites with low pH waters were accurately pinpointed within the S2 AMD classified pixels or within a 30-meter radius from detected AMD clusters. Furthermore, this approach successfully mapped numerous mine sites not accounted for in the Global Mining Land Use layer, as well as hazardous materials transported through the Rio Tinto fluvial system. The use of multi-temporal cloudless S-2 reflectance composites enabled the detection and quantification of seasonal and yearly changes in AMD between 2018 and 2023.This innovative approach showcases the efficient utilization of freely available global spectral datasets and ML techniques for novel monitoring systems at regional or global scales.
Keywords: Sentinel-2, S2GM-2, Acid Mine Drainage, Machine Learning, Mining, Copernicus Land monitoring Services