Evaluating Arsenic and Lead Contamination in Itata Valley Agricultural Soils, Chile: Integration of Slurry-TXRF and Machine Learning Technique for Efficient Monitoring

Abstract A machine learning method was applied to improve the accuracy of the determination of Arsenic and Lead by Slurry - Total Reflection X-ray fluorescence (Slurry-TXRF) with the idea of being applied to the ecological assessment of agricultural soils. Due to TXRF's relatively low resolution, a particular and well-known overlapping of arsenic signal Kα at 10.55 keV with Lαsignal at 10.54 keV of the lead can compromise its determination. Applying a multivariate calibration method based on a machine learning algorithm, for example, Partial Least Squares, could reduce variations due to interference and, consequently, improve the selectivity and accuracy in arsenic and lead determination. In this work the X-Ray fluorescence emission signals was evaluated for a set of 26 different synthetic calibration mixtures and a significant accuracy improvement for arsenic and lead determination was observed, overcoming the problems associated with spectral interferences. Furthermore, with these models, arsenic and lead were quantified from soils of a viticultural subregion of Chile, allowing the estimation of ecological indices in a fast and reliable way. The results report that the level of contamination of these soils concerning arsenic and lead is moderate to considerable.