A pragmatic workflow towards the generation of pXRF datasets for large-scale soil monitoring programs

As an environmentally friendly and non-destructive tool, portable X-ray fluorescence (pXRF) has been increasingly utilised for the rapid estimation of total elemental concentrations in soil samples in the laboratory environment (Ravansari et al. 2020). With different manufacturer calibration modes (e.g. Soil and Geochem detection mode of the Olympus Vanta M Series) operating at varying energy levels, pXRF is widely applied for detection of macro- and trace elements (e.g. Stockmann et al. 2016), as well as soil contaminants (e.g. Weindorf et al. 2013). Beyond elemental analysis, pXRF has also been applied to predict soil properties through the development of empirical models related to mineral composition, such as cation exchange capacity (CEC), and soil organic matter (SOM), demonstrating adequate analytical accuracy across diverse studies (e.g. O’Rourke et al. 2016; Andrade et al. 2020). Additionally, pXRF-generated datasets also provided valuable proxies for the stability of SOM (Ravansari et al. 2020). As a result, pXRF presents a promising alternative or complementary tool alongside conventional laboratory methods for soil analysis.

For large-scale soil analysis, such as Australian national-scale soil organic carbon monitoring initiatives (Karunaratne et al. 2023) or European continental Land Use/Cover Area frame statistical Survey Soil program (Orgiazzi et al. 2018), optimising data collection protocols is essential to balance accuracy, efficiency and cost. Most studies and pXRF manufacturers recommend using thin film-packed sample cups to generate consistent pXRF elemental datasets, as supported by Towett et al. (2016). However, the time and cost associated with this method becomes significant when processing large sample numbers. Plastic bags, such as standard household sealable polypropylene bags, offer a more convenient and readily available alternative (Adams et al. 2017). Nonetheless, further evaluation of the accuracy of using a plastic bag compared to conventional pXRF cups is necessary to assess suitability of use for large scale soil analysis applications.

To identify the most time-efficient scanning protocol that achieves acceptable accuracy for rapid elemental concentration measurements across large sample numbers, a total of 130 national-scale soil samples were selected from the CSIRO National Soil Archive. Originally collected through the Soil Carbon Research Program (SCaRP; Baldock et al. 2013), these samples represent 4180 farmer paddock sites across Australian cropping and pasture regions, covering a wide range of soil types and landscapes. All samples were air-dried at 40°C and sieved to <2 mm to obtain the fine earth fraction. For more details, see the Supplementary materials.

Samples were prepared for pXRF scanning using two types of soil sample containers: (1) sample cups with 4 μm polypropylene thin film; and (2) standard polypropylene plastic bags with a thickness of 50 μm (50 mm × 75 mm). Sample cups were packed with fine earth soil (<2 mm) (at least 1 cm thick), filled with cling wrap, and sealed tightly with thin film (Fig. 1a). Each cup cost approximately AUD2 and required about 2 min of preparation, including assembly. For plastic bag containers, sample bags were filled with consistent quantities of 25–30 g of fine earth soil (Fig. 1a), folded and compacted for scanning (Fig. 1b). At approximately AUD0.02 and 1 min of preparation per bag, this method significantly reduced costs and packing time. As pXRF manufacturers (e.g. Olympus) generally recommend sample cups (e.g. as provided by Choice Analytical Ltd, Thornleigh, Australia) for optimal results across a wide range of elements, the sample cup dataset was used as the standard for comparison in this study.

Fig. 1.

(a) Soil packing using plastic cups within thin film (top) and bags (below). (b) Compacted soil in plastic bags for pXRF measurement operated in the field stand kit.

Data acquisition was carried out using an Olympus Vanta pXRF analyser (Vanta M Series, Olympus Scientific Solutions, Waltham, USA), with two manufacturer calibration modes that are commonly used in soil science: (1) Soil, which operates with three energy levels (i.e. beams) (50, 40 and 15 kV) with a total of 90 s standard analysis time (e.g. Ross et al. 2024); and (2) Geochem, which uses two beams (40 and 10 kV) with 60 s standard analysis time (e.g. Zhang and Hartemink 2019). These two calibration modes are designed to optimise elemental detection in soil-like materials. Their primary difference is their limit of detection (LOD) and elemental range that they capture (see Supplementary Table S1). In this study, we compared Mg, Al, and Si, which were only detectable in Geochem mode, other macro-nutrients (K and Ca), micro-nutrients (Fe and Cu), and elements important for pedological studies (Ti and Zr) (Solleiro-Rebolledo et al. 2019). The accuracy and agreement between datasets were evaluated using the coefficient of determination (R²), Lin’s concordance correlation coefficient (LCCC), and the root mean square error (RMSE).

Low atomic number elements of Mg, Al, and Si, which are only available in Geochem detection mode, showed the largest discrepancies in concentrations (%) between sample cups vs sample bags. Fig. 2a shows no significant correlation for Mg, with a very weak R² (0.05), an LCCC close to 0, and an RMSE of 0.38%. This is likely because most samples fell below the LOD (0.25%). In contrast, for Al (Fig. 2b) and Si (Fig. 2c), there was a tendency towards a linear relationship, with R² values of 0.65 and 0.75, respectively. However, the agreements along the 1:1 line for both elements were poor with low LCCC (<0.1), indicating an underestimation of Al and Si concentrations when using sample bags. For macro-nutrients, K and Ca (Fig. 2d, e), displayed high R² values (>0.95), and similarly high LCCC values close to 1 for both modes. However, only when using bags, K and Ca were not detected in 24 and 35 samples, respectively. Thus, the plastic film thickness of the containers appears to influence the detection of low atomic number elements. This can be explained by the Beer-Lambert law (Evans 1955), which suggests that signal attenuation occurs due to material thickness. The thickness of the plastic bags caused significant attenuation of low-energy X-rays compared to the thinner film of the sample cups. Nevertheless, the linear relationship for Al and Si suggests that varying concentrations of these elements in soil samples remain detectable in the sample bag scanning mode.

Fig. 2.

Measured element concentrations (%) packed in bags versus cups for (a) Mg, (b) Al, and (c) Si in Geochem mode, and (d) K, (e) Ca, (f) Ti, (g) Fe, (h) Cu, and (i) Zr in Geochem and Soil modes. LOD, limit of detection; N, number of R²: coefficient of determination; LCCC, Lin’s concordance correlation coefficient; RMSE, root mean squared error.

Higher atomic number elements showed strong linear relationships for both Geochem and Soil modes. For micronutrients Ti, Fe, Cu, and Zr (Fig. 2f–i), also showed high R² values over 0.9 and LCCC over 0.85. Notably, Soil mode generally resulted in lower RMSE than Geochem mode for these elements, except for K, Ca, and Fe. This is likely because Soil mode with 50 kV energy level provides a broader detection range for these elements compared to Geochem mode.

To better illustrate the correlation between Geochem and Soil modes, Fig. 3a–f shows the measured element concentrations (%) for K, Ca, Ti, Fe, Cu, and Zr in samples packed in cups. Most R² values were above 0.9, indicating a strong correlation, except for Ti. In addition, most of LCCC were also high as over 0.85, except for Fe of 0.74. The RMSE were also low for most elements.

Fig. 3.

Measured element concentrations (%) packed in cups for (a) K, (b) Ca, (c) Ti, (d) Fe, (e) Cu, and (f) Zr in Soil vs Geochem modes. R², coefficient of determination; LCCC, Lin’s concordance correlation coefficient; RMSE, root mean squared error.

Based on our findings, we recommend using the Geochem mode due to its consistent ability to measure a broad range of macro- and microelements, including Mg, Al, and Si, which are not detectable in Soil mode (Table S1). In addition to this, the Geochem mode provides a quicker 60-s standard analysis, compared to the 90-s standard analysis duration required for Soil mode. Fedeli et al. (2024) evaluated the performance of pXRF in both Soil and Geochem mode across various certified reference materials of soil and plant origin and recommended Geochem mode because it provided more reliable results for a broader range of elements.

In this study, we used pXRF and the Olympus pXRF range manufacturer calibration modes and their standard measurement time of 30 s per energy beam to measure elemental concentrations of soil samples. The impact of differing energy beam scanning time settings was not investigated here but could be explored in further research. Recent studies using the raw pXRF spectral responses of each energy beam and a scanning time of only 2 s found that this fast pXRF measurement can be used effectively to predict a range of soil properties (Tavares et al. 2023).

Furthermore, we only investigated the use of a standard single plastic bag thickness (50 μm). Future research should investigate also the impact of differing plastic bag thickness (25 and 75 μm, for example) on measurement accuracy and assess the associated costs in order to refine and possibly optimise the protocol further. Also, regarding those elements under LOD, it is also necessary to explore and the results from complementary chemical analytical approaches (e.g. Microwave assisted digestion with HF).

Current pXRF applications in soil research have been limited to relatively small sample sizes at field or regional scales, such as 180 samples on the campus of the Federal University of Lavras, Brazil (Bócoli et al. 2025), and 300 samples in the Afzar region, Iran (Naimi et al. 2022). This study provides an optimised protocol for ongoing large soil monitoring projects at national and global scales (e.g. >3000 samples). Using the Geochem mode, combined with standard 50-μm plastic bags has the potential to significantly reduce both time and costs. Rapid and cost-effective acquisition of XRF measurements for large-scale studies is desired, as soil elemental data can provide valuable insights into the role of soil mineralogy in determining carbon storage capacity (Georgiou et al. 2022) and the biological-chemical stabilisation of SOM (McNally et al. 2017) across different environments.

Supplementary material

Supplementary material is available online.