X-ray Fluorescence Spectrometry (XRF) is undeniably a valuable asset for the simultaneous determination of mineral elements. This is a fast, non-destructive and inexpensive method in comparison with conventional analysis methods. The recent development of portable spectrometers (pXRF) further increases the potential of the XRF technique in environmental purposes by bringing the device to the field. This work focused on trace elements determination (Cu, Zn, Pb, Ni, Cr and As), most of which are subject to specific regulations, especially for sewage sludge (expect As) and contaminated soil management. In Wallonia, the reference method is based on aqua regia (HCl+HNO3, ISO 11466) digestion followed by Atomic Absorption Spectrometry (AAS) or Inductively-coupled plasma atomic emission- or mass- spectrometry (ICP-AES/ICP-MS). It is established that aqua regia digestion-based analysis underestimates the total content of elements because it does not completely digest silicates, while XRF is supposed to measure total content. To assess the performance of a pXRF (S1 Titan 600, Bruker), we compared the prediction values with the values from the aqua regia digestion for some reference values in soils.
2. Material and methods
Seventeen soils (mainly agricultural soil), all of which were already evaluated for concentration of some metallic trace elements, were analyzed by pXRF in desktop configuration with XRF cells (Ø 40 mm, Prolene film 4µm) according to a validation process and were compared to their current aqua regia digestion-AAS values. Soil selection was based on results of a principal component analysis (PCA) using metallic trace elements and major elements (Ca, Mg, K, P, Fe) aqua regia contents, followed by a hierarchical cluster analysis (Ward’s method) to extract samples as diversified as possible. In addition, three inter-laboratory reference materials from BIPEA were subjected to the same protocol to carry out checks on the laboratory experiments. All samples were air-dried, sieved and crushed to 200 µm. Time measurement was set to 30 seconds in dual phase (60 seconds total).
In order to assess the validity of the pXRF, the accuracy profile’s method  was chosen. Under intermediate precision conditions (5 days and 3 repetitions/day), results were calculated as the mean of 3 successive readings. The accuracy profile allows determining an interval which will contain 95% of the measurements. This interval was then compared to an acceptability interval, which was fixed at ± 20% of the reference value, to vouch for the validity. Reference values of each metallic trace element were calculated as the mean of 5 series of measurements according to an aqua regia digestion-AAS method.
For the purpose of improving the trueness, two types of regression were applied between XRF and reference values: a simple linear regression and a FREML regression . The advantage of the latter is that it can take into account errors on both X and Y variables. Dataset was split into one calibration set (2/3) and one validation set (1/3).
In addition, the performance of the pXRF was compared to a laboratory wavelength-dispersive X-ray fluorescence spectrometer (WDXRF) supposed to give more reliable results and total contents.
3. Results and discussion
Strong linear correlations were found in soils for Cu, Zn or Pb (R² > 0.99) between pXRF and aqua regia digestion-AAS. This linear correlation was very poor for Cr, probably due to internal calibration issues. Figure 1 shows the Zn accuracy profile, where the underestimation by pXRF can be seen. A simple slope and y-intercept correction of pXRF data could generally restore the trueness (bias) to improve the accuracy on a larger concentration range. However, concentration levels close to detection limits have a higher degree of random variability. This can be explained by the Horwitz curve where random variability increases with lower concentrations. This emphasizes the need of multiplying the number of measurements/readings.
The comparison made of the pXRF with the WDXRF showed that the pXRF underestimates the metallic trace elements content. Indeed, the pXRF results were lower than WDXRF results. But, in terms of prediction of the reference values, the pXRF seems to be only slightly worse than the WDXRF. This shows the power of the portable XRF to predict AAS reference values at a low cost.
Figure 1. Zinc Accuracy profile.
Red short dotted line: Acceptance limits. Orange long dotted line: Tolerance limits
S1 Titan XRF is an interesting tool and easy to use for the prediction of metallic trace elements content in soils. However, to predict reference values (aqua regia digestion-AAS) with sufficient accuracy, direct measurements are not suitable and a specific XRF calibration is recommended. A simple linear or FREML regression is adequate to improve the accuracy of the measured values in some cases, depending on the wanted future application.
 M Feinberg, M Laurentie 2010. Validation des méthodes d’analyse quantitative par le profil d’exactitude. Cah. Tech. l’INRA No Special, 139, 2010.
 Analytical Methods Committee (AMC). Fitting a linear functional relationship to data with error on both variables [technical brief no. 10]. R. Soc. Chem. 1(10), 2002.