Abstract
Humidity cell tests (HCTs) are commonly used in material characterization projects to reduce uncertainty in ARD potential, to estimate lag time to acidity, and to predict long-term water quality, especially under acidic conditions. Despite the wealth of information provided by HCTs, operators often struggle to predict when actual onset of acid generation will occur at the mine scale. However, the multivariate nature of HCT results provides an opportunity to leverage unsupervised machine learning techniques to predict degradation of mine waste facility effluent with greater accuracy.
Here we use Principal Component Analysis (PCA) and k-mean cluster analysis conducted on results from fifty-three HCTs from a proposed mine site to assess the multivariate evolution in water quality as materials transition from acid buffering to acid generating. PCA provides useful insights into the chemical evolution of materials that become acid generating while k-mean cluster analysis can be used to classify HCTs into discrete chemical domains. The dataset can be best described as five main domains: 1) silica, 2) alkalinity, 3) transition 4) acid generating, and 5) extreme acid generating. This type of analysis, when combined with leachate chemistry from mine facilities such as waste rock dumps and tailings storage facilities can be used to provide early warning that a facility is becoming acidic, even ahead of a change in pH. Such chemical indicators can be used to apply proactive and preventative measures, potentially saving operators significant expense.