Exploring coal mine-affected water chemistry data of the Fitzroy River Basin, Central Queensland Australia

Chapman, JM ORCiD: 0000-0002-9850-0403; Cozzolino, D ORCiD: 0000-0001-6247-8817; Jones, CE ORCiD: 0000-0002-1779-6231This study examined coal mine-affected river chemistry data from the Fitzroy River Basin, Queensland Australia. The raw dataset related to 106 de-identified sites, associated with 28 different coal mines, across 7 sub-basins. Data consisted of 377 samples concerning up to 50 different water quality parameters (each measured between 80 to 377 times), that were sampled during July 2016 to June 2017. Principal component analyses (PCA) is a useful exploratory data analysis technique for multi-parameter analytical/environmental chemistry datasets. Many statistical packages have built-in PCA capabilities. Data treatment, PCA results, and comparisons between two software programs (Sigma plot 13 and the Unscrambler X 10.3) were explored. After considering the percent of censored data, summary statistics and the distribution of individual observations, a reduced number of variables were included in the PCA. Data were log transformed, which increased normality and helped to deal with the different scales and units of measurement across the parameters. The Sigma Plot PCA model does not handle missing observations, hence, excluded all but 12 of the original samples. In contrast, the Unscrambler non-linear iterative partial least squares algorithm uses all available data. Turbidity (NTU), total suspended solids (mg/L), and total aluminium, chromium, copper, iron, nickel, zinc, cobalt and manganese (µg/L) were notable parameters (component loading > 0.75) in PC-1 of both models. In addition, SIGMA plot identified EC (µS/cm), pH, nitrate (as N mg/L), sulfate (mg/L) and fluoride (µg/L) (loadings < -0.7). The analyses indicated that concentrations of these metals increased during periods of elevated suspended particulate matter levels within the basin. Conversely, pH and some anions were negatively correlated with these total metal concentrations within the dataset