Multivariate analysis applied to X-ray fluorescence to assess soil contamination pathways: case studies of mass magnetic susceptibility in soils near abandoned coal and W/Sn mines

Abstract

Determining the origin and pathways of contaminants in the natural environment is key to informing any mitigation process. The mass magnetic susceptibility of soils allows a rapid method to measure the concentration of magnetic minerals, derived from anthropogenic activities such as mining or industrial processes, i.e., smelting metals (technogenic origin), or from the local bedrock (of geogenic origin). This is especially effective when combined with rapid geochemical analyses of soils. The use of multivariate analysis (MVA) elucidates complex multiple-component relationships between soil geochemistry and magnetic susceptibility. In the case of soil mining sites, X-ray fluorescence (XRF) spectroscopic data of soils contaminated by mine waste shows statistically significant relationships between magnetic susceptibility and some base metal species (e.g., Fe, Pb, Zn, etc.). Here, we show how qualitative and quantitative MVA methodologies can be used to assess soil contamination pathways using mass magnetic susceptibility and XRF spectra of soils near abandoned coal and W/Sn mines (NW Portugal). Principal component analysis (PCA) showed how the first two primary components (PC-1 + PC-2) explained 94% of the sample variability, grouped them according to their geochemistry and magnetic susceptibility in to geogenic and technogenic groups. Regression analyses showed a strong positive correlation (R² > 0.95) between soil geochemistry and magnetic properties at the local scale. These parameters provided an insight into the multi-element variables that control magnetic susceptibility and indicated the possibility of efficient assessment of potentially contaminated sites through mass-specific soil magnetism.

Keywords: Geochemistry; Magnetic minerals; Metal contamination; PCA and cluster analysis; Regression analyses; XRF spectra.

Fig. 1

Geologic setting of the studied pilot areas: São Pedro da Cova, Fojo, and Regoufe

Fig. 2

Maps of the sample locations in studied pilot areas: São Pedro da Cova (a), Fojo (b), and Regoufe (c)

Fig. 3

Mean-plot of spectra from studied soils (n = 90) around three mining areas: raw XRF (a) and transformed OSC-XRF spectra (b)

Fig. 4

PCA applied to: raw XRF (a) and OSC-XRF (b) spectra of studied soils (n = 90) around three mining areas (São Pedro da Cova, Fojo, and Regoufe). The scores plot of the first two principal components shows that ten self-burning subset of samples of Fojo (bounded by a dashed line) were separated into an independent group

Fig. 5

Cluster analysis using squared Euclidean distance, applied to samples (n = 50) of São Pedro da Cova complex area. After excluding one sample (19), as an outlier, a dendrogram illustrates the arrangement of two major clusters based on the relative distance of the corresponding XRF spectra: a cluster 1 in green (n = 30) and the cluster 2 in red (n = 19)

Fig. 6

PLS regression (calibration—blue, external validation—red) between OSC-XRF spectra and mass magnetic susceptibility (χ) content in soil samples from São Pedro da Cova obtained after cluster analysis: cluster 1 (a) and cluster 2 (b). The model used three latent variables (factor = 3) to optimize covariance between data sets. The inserted boxes contain statistical parameters that explain the goodness of regression with high values of R² for calibration (> 0.99) and validation (> 0.95), and low values of RMSE of calibration (equal to 8.1–8.7 × 10⁻⁸ m³ kg⁻¹) and validation (equal to 14–43 × 10⁻⁸ m³ kg⁻¹). Slope and offset (intercept) define the linear relationship between two variables of the regression line. The closer the slope is to 1, the data are better correlated. The offset is the intercept of the line when the X-axis is set to 0 and its value is nearly 0

Fig. 7

PLS regression (calibration—blue, external validation—red) between OSC-XRF spectra and mass magnetic susceptibility (χ) content in soil samples from: Fojo (a) and Regoufe (b). 3 latent variables (factor = 3) were used to maximize covariance between data sets. Inserted boxes contain the statistical parameters of the regression (the explanation of the relevant parameters used for statistical analysis of data is given in detail in Fig. 6)

Fig. 8

PLS regression (calibration—blue, external validation—red) between OSC-XRF and mass magnetic susceptibility (χ) content in soil samples from São Pedro da Cova (cluster 2) and Fojo (n = 44). 6 latent variables (factor = 6) were used to maximize covariance between data sets. Inserted box contains the statistical parameters of the regression (the explanation of the relevant parameters used for statistical analysis of data is given in detail in Fig. 6)

Fig. 9

The plots of weighted regression coefficients (B_w) obtained using 1 factor for the PLS regressions in soil samples from São Pedro da Cova: cluster 1 (a) and cluster 2 (b). The peaks with the highest regression coefficients (at 6.4 and 7.1 keV) corresponding to Fe, play the most important role for the mass magnetic susceptibility in the studied soils

Fig. 10

The plots of weighted regression coefficients (B_w) obtained using 1 factor for the validated PLS regressions in soil samples from: Fojo (a) and Regoufe (b). The peaks with the highest regression coefficients (at 6.4 and 7.1 keV) corresponding to Fe, play the most important role in the regression for the mass magnetic susceptibility in the studied soils from Fojo and Regoufe. In soils from Regoufe mining area, As (10.5 and 11.7 keV) was also very important variable for the PLS regression of the mass magnetic susceptibility