Effect of sample pretreatment on pelletization and performance of laser-induced breakdown spectroscopy for predicting key soil properties

Abstract

Laser-induced breakdown spectroscopy (LIBS) is a promising analytical technique for determining soil properties in a rapid, cost-effective, and environmentally friendly manner. However, an optimal sample pretreatment method that enhances sample homogeneity for improved prediction accuracy is needed. In this study, the effect of different sample pretreatments on pelletization of soil samples, and prediction of texture and soil organic carbon (SOC) content, was assessed. Four pretreatments, namely 2 mm, 200 μm, and 100 μm sieved, and milled pretreatment, were applied to 133 soil samples representing a wide variability in texture and SOC content range. Visual assessment of soil pellet surface quality was done where poor surfaces, deemed unsuitable for LIBS measurement, were separated from the good-surface pellets. Partial least square regression (PLSR) models for the pretreatments across the investigated soil properties were compared, followed by interval partial least square regression (iPLSR), to evaluate potential improvements in the accuracy of the prediction models. The ratio of performance to interquartile distance (RPIQ) was used to rank the PLSR models for each pretreatment across the soil properties. The 100 μm pretreatment had the highest number of good-surface pellets followed by the 2 mm pretreatment, and lastly the 200 μm and milled pretreatments, which were comparable. The milled pretreatment achieved the best model for sand prediction (RPIQ = 7), followed by the 2 mm pretreatment for clay prediction (RPIQ = 2.5), 100 μm pretreatment for silt prediction (RPIQ = 2), and the milled pretreatment for SOC prediction (RPIQ = 1). A reduction in RMSEP of 31%, 23%, and 11% for the prediction of sand, SOC, and silt content, respectively was achieved by the milled pretreatment while the 100 μm pretreatment recorded a 15% reduction in RMSEP for the prediction of silt content. Overall, PLSR performed better than iPLSR for the prediction of texture (apart from clay) and SOC. These results show that sample pretreatment that involves sieving, or milling, influences the pellet surface quality and the prediction accuracy of texture and SOC content.