Data-driven modeling of background and mine-related acidity and metals in river basins

• Published in: Environmental pollution (1987) 2014-01, Vol.184, p.530-539
• Main Author: Friedel, Michael J.
• Format: Article
• Language: English
• Subjects: Applied sciences; Basins; Cluster analysis; Colorado; Continental surface waters; Earth sciences; Earth, ocean, space; Engineering and environment geology. Geothermics; Environmental assessment; Environmental Monitoring - methods; Exact sciences and technology; Hydrothermal alteration; Mathematical models; Metals - analysis; Mineral-resource assessment; Mining; Mining activity; Models, Chemical; Natural water pollution; Neurons; Nonlinearity; Pollution; Pollution abatement; Pollution, environment geology; River basins; Rivers - chemistry; Self-organizing map; Statistics; Stochastic modeling; Uncertainty; Water Pollutants, Chemical; Water quality; Water treatment and pollution
• ISSN: 0269-7491; 1873-6424
• DOI: 10.1016/j.envpol.2013.09.036

A novel application of self-organizing map (SOM) and multivariate statistical techniques is used to model the nonlinear interaction among basin mineral-resources, mining activity, and surface-water quality. First, the SOM is trained using sparse measurements from 228 sample sites in the Animas River Basin, Colorado. The model performance is validated by comparing stochastic predictions of basin-alteration assemblages and mining activity at 104 independent sites. The SOM correctly predicts (>98%) the predominant type of basin hydrothermal alteration and presence (or absence) of mining activity. Second, application of the Davies–Bouldin criteria to k-means clustering of SOM neurons identified ten unique environmental groups. Median statistics of these groups define a nonlinear water-quality response along the spatiotemporal hydrothermal alteration-mining gradient. These results reveal that it is possible to differentiate among the continuum between inputs of background and mine-related acidity and metals, and it provides a basis for future research and empirical model development. The trained self-organizing map is used to determine upstream hydrothermal alteration (AS – acid sulfate; PROP – propylitic, PROP-V – propylitic veins, QSP – quartz-sericite-pyrite, WSP – weak-sericite-pyrite; Mining activity: MINES) from water-quality measurements in the Animas river basin, Colorado, USA. The white hexagons are sized proportional to the number of water-quality samples associated with that SOM neuron. [Display omitted] •We model surface-water quality response using a self-organizing map and multivariate statistics.•Applying Davies–Bouldin criteria to k-means clusters defines ten environmental response groups.•The approach differentiates between background and mine-related acidity and metals. These results reveal that it is possible to differentiate among the continuum between inputs of background and mine-related acidity and metals.