Evaluation and adaptation of portable x-ray fluorescence and data fusion techniques for non-invasive characterisation of metal-contaminated environments

Abstract

Metal extraction and refinement has left a legacy of brownfield and post metallurgical sites, which present risks to health and the environment. However, these sites present an opportunity to provide the critical resources needed to meet the global energy transition to net zero. A key barrier to the resource recovery and remediation of these sites is the expensive and slow surveying methods currently in use. Addressing this barrier, this PhD aimed to develop an innovative rapid, non-invasive method for quantifying elemental and speciated metals in contaminated soils using portable X-ray fluorescence (pXRF) and visible-near infrared (Vis-NIR) spectroscopy. Research investigated the applicability of these methods and evaluated the influence of sample preparation and inter-sample interferences through four objectives. Firstly, when investigating the effects of sample pre-processing on ex-situ pXRF measurements of samples collected from a post-metallurgical site; it was identified that sieving and grinding improved pXRF precision (average relative standard deviation fell by 7.17% and 8.37% respectively); while drying and grinding enhanced pXRF accuracy (average r2 increased by 0.03 and 0.10 respectively). Secondly, the correlation of pXRF and geophysical measurements was investigated for the first time ever at a heterogeneous post-metallurgical site. No correlations (r² < 0.46) were observed between the two approaches, which was attributed to the differing sample volumes and depths measured by these methods. It was concluded that matrix heterogeneity and the scale disparity between geophysical and chemical sampling present significant challenges to rapid methodologies. Thirdly, to overcome the interference caused by moisture and soil organic matter (SOM) on in-situ pXRF measurements, a novel data fusion framework using pXRF and Vis-NIR was developed. During its development, it was identified, for the first time, that moisture had an elementally dependent non-linear trend with substantial underpredictions by pXRF at ~15% moisture. Furthermore, this research also showed, for the first time, that SOM caused an insubstantial but significant difference in performance of pXRF predictions. However, the data fusion framework successfully accounted for both moisture and SOM in all elements measured, with Ca showing the largest improvement (r² = 0.308). And fourth, to predict to determine the feasibility of predicting metal speciation, a simplified, artificial sample matrix containing multiple iron compounds was again measured with both pXRF and Vis-NIR. This novel approach demonstrated for the first time that speciated Fe concentration can be accurately predicted (r² = 0.96) in a heterogeneous solid sample. This PhD thesis proposes a best-practice methodology for ex-situ pXRF analysis, addressing the effects of sample preprocessing. The methodology can be immediately implemented to reduce financial cost of collecting data in contaminated land surveys. Additionally, through further development of the data fusion framework for in-situ measurements, significant time and cost savings can be achieved, allowing for increased data collection and the generation of qualitative models in the early stages of contaminated land assessments. Furthermore, this research demonstrates the potential to predict metal speciation, with the possibility of reducing the cost of speciation data acquisition. Such advancements would enable more targeted remediation and resource recovery efforts at contaminated sites.