Acid Mine Drainage Impoundments in the Nkangala District South Africa
Description
Hypothesis: Temporally stable mine-site impoundments (pit lakes, tailings, return, and seepage ponds) accumulate acid mine drainage (AMD) with higher rare-earth element (REE) concentrations than natural waters which are contaminated by AMD. These can be systematically mapped, screened, and quantified to prioritise recovery and remediation. What the data are: A spatial inventory of AMD impoundments across the Nkangala District, South Africa. The core product is a polygon layer of 2,886 AMD-containing waterbodies, with a detailed attribute table including geometry, morphometrics, monthly Sentinel-2 reflectance statistics (2024), and screening variables for AMD/REE potential. Key fields: Geometry; area (area_m2, area_km2); coordinates; reflectance statistics (mean, min, max, stdDev, selected indices); estimated depth and volume (m³, L) from DEM-based shoreline proxy; water type label (pit lake, tailings, seepage, return pond); monthly reflectance band stats (Jan–Dec 2024) for B1–B12 and B8A. How it was built: Mining boundaries were refined in QGIS by reconciling multiple datasets. In Google Earth Engine, Sentinel-2 L2A imagery (cloud/shadow masked) was processed; the AWEI index was selected after testing and applied with a threshold > −0.35. Features ≥ 4 pixels (~400 m²) and present in ≥ 50 % of 2024 monthly composites were retained. Polygons were vectorised, some were visually classified, and linked to morphometric data derived from Copernicus GLO30 DEM shoreline buffers. Temporal medians informed conservative depth proxies; volume = area × depth. Monthly reflectance values were computed per polygon across B1–B12 and B8A (± 7 days). 2023 field sampling inside vs outside mining boundaries confirmed higher sulfate, lower pH, and measurable REEs within mine sites compared to bodies sampled within a 10km radius of a mine. What the data show: AWEI combined with temporal persistence filters delineated 2,886 AMD impoundments across the study area. DEM benchmarking reproduced approximately 78–80 % of known capacities, supporting the reliability of the volumetric estimates. Indicative ∑REE concentrations derived from reflectance-based comparisons suggest an average of 419 mg/L in these AMD waters, with lows of 89 mg/L. How to interpret: Best used as a screening layer to: map AMD extent, target high-priority ponds for sampling, treatment, or REE recovery, and support techno-economic scoping. Volumes are lower-bound estimates (based on 30 m DEM); concentrations are indicative, not assay results. Users should pair this layer with site visits, water chemistry, and high-resolution elevation data. Format: GeoJSON (EPSG:4326) Software: QGIS, Google Earth Engine. Limitations: DEM resolution, empirical thresholds, sparse paired reflectance–chemistry samples, and no flow or residence-time data. Volumes are conservative; REE values are screening outputs.
Files
Steps to reproduce
This was developed through a remote sensing and GIS-based workflow designed to locate, delineate, and quantify AMD impoundments associated with coal mining in the Nkangala District, South Africa. The full workflow is reproducible in GEE and QGIS using publicly available datasets and the scripts referenced in the companion paper. Mining boundaries were extracted from multiple global and national datasets, including the FINEPRINT Global Mining Land Use dataset (Maus et al., 2020, 2022) and OpenStreetMap (OSM). These were cross-checked against high-resolution basemaps and manually corrected to eliminate false positives, merge fragmented polygons, and ensure all visible mine features were included. The corrected boundary dataset defined the “zone of influence” for AMD detection. Sentinel-2 Level-2A surface reflectance imagery (10–20 m spatial resolution) was accessed via the Copernicus Open Access Hub through GEE. Cloud and shadow contamination were removed using the Sentinel-2 Cloud Probability (s2cloudless) and Scene Classification (SCL) layers. All imagery was projected to WGS 84 / UTM Zone 35S (EPSG:32735) and clipped to the study area. To isolate AMD waters, three spectral indices (NDWI, MNDWI, and AWEI) were evaluated against field-verified AMD sites. The AWEI was selected as the most reliable and applied with a threshold of AWEI > −0.35. Raster outputs were vectorized and filtered to retain only features ≥4 connected pixels (~400 m²) and persistent in ≥50% of monthly composites (2024). This ensured temporal stability and excluded ephemeral waters. The Copernicus GLO30 Digital Elevation Model (DEM) was used to estimate depth and volume. Shoreline buffers were generated for each polygon and DEM elevations extracted. The temporal median of shoreline elevation differences served as a conservative depth proxy, and volume was calculated as area × depth. For each polygon and month, reflectance statistics (mean, min, max) were computed for Sentinel-2 bands B1–B12 and B8A within ±7-day windows. The resulting 12-month composite table formed the spectral dataset linked to each id. All computations were performed in GEE, QGIS or Python. Water samples were collected in 2023 from 30 sites inside and outside mining boundaries. Field measurements (pH, conductivity, DO, temperature) were taken using a YSI Professional Plus probe. Elemental concentrations were determined via ICP-OES (Agilent 5800 VDV) following AMD-optimized protocols. Results confirmed higher sulfate, lower pH, and measurable REE enrichment inside mining areas, validating the spectral-based mapping focus. All spatial analysis and visualization were performed using GEE, QGIS, and Python with the libraries pandas, SciPy, geopandas, numpy, scikit-learn, seaborn, and matplotlib. The workflow is fully reproducible with public datasets, provided GEE access and Sentinel-2 imagery for the specified dates.
Institutions
- Kobenhavns Universitet Institut for Plante og Miljovidenskab
- University of the Witwatersrand School of Geography Archaeology and Environmental Studies