*2.3. Chemical Analysis*

The electrical conductivity (EC; proportional to soluble salt content) of soil samples was determined on 1:5 solid: Deionised water suspensions using a calibrated conductivity cell electrode. The pH was measured on the same suspensions using a glass-reference pH electrode after a 2-point buffer calibration [18].

The near-total concentrations of 26 elements (Al, As, Ba, Ca, Cd, Ce, Cr, Cu, Fe, Gd, K, La, Mg, Mn, Mo, Na, Nd, Ni, P, Pb, S, Sr, Th, V, Y, and Zn) were measured on samples by inductively-coupled plasma optical emission spectrometry (ICP-OES) following digestion of soil in concentrated nitric and hydrochloric acids (i.e., *aqua regia*) at ca. 130 ◦C [19]. *Aqua regia* digestion is commonly used to determine environmentally significant concentrations, since it largely excludes elements within the matrices of silicates and other recalcitrant minerals. Before acid digestion, samples were ground to 50 μm using ceramic mortars and pestles. Reagent blanks, and grinding blanks composed of acid-washed silica sand, were included in analytical runs to check for contamination. The standard reference stream sediment material STSD-2 [20] was analysed identically to samples to assess analytical accuracy. Measurement precision was assessed using analytical duplicates on ca. 10% of samples.

The lower limits of analytical detection were calculated, where possible, from 3× the standard deviation of multiple reagen<sup>t</sup> blank concentrations [21]. Concentrations lower than mean blank values, or below calculated lower detection limits, or both, were deleted from the dataset.

#### *2.4. Statistical and Numerical Analysis*

Data managemen<sup>t</sup> and transformation of variables was conducted using Excel® (Version 2016, Microsoft, Redmond, WA, USA) Statistical and graphical analyses of data were performed in the statistical computing environment 'R' [22] and associated packages. Skewed variables (identified with the Shapiro-Wilk test for normality) were log10- transformed, or power-transformed based on the Box-Cox algorithm and re-checked for normality.

A general inability of variables to be transformed to yield normal distributions dictated the use of the non-parametric Spearman correlations, and Wilcoxon or Kruskal-Wallis tests for mean comparisons. If Kruskal-Wallis tests showed a significant difference, the R package 'PMCMR' [23] was used to apply the post-hoc Conover's test for pairwise comparisons of mean rank sums. Simple regression models were fitted using the log10-transformed variables. The potentially misleading effects of compositional closure were addressed using transformations to centred log-ratios [24], which were used for principal components analyses. Principal components analyses were conducted using only variables having minimal or no missing observations.

Distribution maps were constructed using the 'OpenStreetMap' package [25] with elevation contours interpolated from a dense grid of land elevations from Google [26] generated using the R package 'googleway' [27] and interpolated using the R package 'akima' [28]. Spatial autocorrelations were assessed using global and local Moran's I statistics, calculated using the R package 'lctools' [29]. Local Moran's I values showing significant association (*p* ≤ 0.05) were categorised using high-low notation, based on the point measurement relative to the median and the sign of the Local Moran's I statistic. Spatial interpolations were achieved using an inverse distance weighting method using the R packages 'sp' [30] and 'gstat' [31]. Preliminary analysis showed that inverse-distance interpolation gave similar results to simple kriging, but kriging interpolation was not used, based on the requirement of ≥100 observations to generate a reliable experimental variogram [32].

A composite estimate of soil contamination was calculated from the concentrations of As, Cu, Pb, and Zn as the Integrated Pollution Index, IPI [33], shown in Equation (1):

$$\text{IPI} = \left(\sum\_{i=1}^{n} \left(\frac{\mathbf{C}\_i}{\mathbf{S}\_i}\right)\right) / n\_\prime \tag{1}$$

In Equation (1), ∑ means the sum of terms 1 to *n*, C*i* = the measured concentration of th *i*-th element, S*i* = the background concentration of the *i*-th element, *n* = the number of elements. The S*i* values used (in mg/kg: As = 1.5, Cr = 10, Cu = 2, Pb = 5, Zn = 6) were published ambient background concentrations for the Perth region [34], but this report suggests a zero-background concentration for Ni. In this study 1 mg Ni/kg was used for background, which is the lowest (most conservative) 25th percentile concentration among similar datasets (e.g., [35]).
