PXRF: PACKED WITH POTENTIAL - PROTOCOLS FOR BASALTIC SAMPLES

Portable XRF (pXRF) units are becoming more readily available and their usage within geoscience is increasing, but to what extent can we rely on these devices as research tools? This study investigates the best protocols for producing accurate datasets for volcanic rocks. Seven billets of basaltic lava flows of varying thickness were selected, and analyzed using best use practices for the device (an Olympus Vanta Innov-X, M Series). Controls on sample results came from previous XRF analysis, and associated LOI corrections and normalization. pXRF best use practices were developed in accordance with guidance from Olympus and previous studies.

A key aim for practical use of XRF datasets is to classify rock units, and for volcanic rocks this is done using a Total Alkali–Silica (TAS) Diagram. However, pXRF results are devoid of Na₂O (a key oxide), as this is below the detection spectrum of the instrument. To counter this lack of data, direct relationships were calculated using the relative relationship between Na₂O and K₂O, and regression models were then created for specific samples and for basaltic rocks with an SiO₂ content ranging from 41-57 weight percent. Calculated results were then normalized, and compared to previous XRF results. TAS-specific regression models (defined by TAS classification) produced more accurate results than the broader basaltic rock regression model (defined by samples with an SiO₂ content of 41-57 wt%), which in most cases correctly determined rock type.

Thickness of billets did not affect results, with the minimum tested at 1.15cm (just above the minimum suggested thickness). In comparison to calculated LOI from XRF, no relationship can be established between pXRF data and XRF-derived LOI, indicating that locked in water content remains stable during testing and does not influence data collection. To obtain accurate results, regression models specific to rock type are suggested to calibrate pXRF datasets. However, broader SiO₂-defined regression lines can be used to derive elements below detection limits.