Machine learning-assisted laser-induced breakdown spectroscopy method for the detection of Pb and Cr in soils from...

Laser-induced breakdown spectroscopy (LIBS) is an advanced elemental analysis technique that enables rapid, in situ, and simultaneous detection of multiple soil elements, thereby supporting environmental monitoring and data-driven agricultural management. Heavy metals such as chromium (Cr) and lead (Pb) typically occur in soil at trace levels (mg/kg), and their accurate detection by LIBS is a challenging issue due to complex soil composition and substantial matrix variations across regions. In this study, a machine learning-assisted LIBS method was developed to achieve highly sensitive and accurate detection of Pb and Cr in diverse soils. Following the optimization of instrumental parameters for pellet-based measurements, limits of detection of 9.08 and 4.15 mg/kg were achieved for Cr and Pb, respectively. To mitigate the effects of soil matrix variability on LIBS detection accuracy, the competitive adaptive reweighted sampling (CARS) and convolutional neural network (CNN) algorithms were employed for feature wavelength selection and modelling. Compared to signal-concentration calibration models, CARS-CNN models demonstrated higher detection accuracy for Cr and Pb, yielding R_v² values above 0.99 and reducing detection errors by over 5-fold. Finally, external validation was conducted using 51 real soil samples collected from 11 provinces of China, achieving R² values of 0.773 and 0.877 and relative root-mean square errors of 24.36 % and 29.35 % for Cr and Pb, respectively. The findings of this study demonstrated the machine learning-assisted LIBS method had high-sensitivity and robust detection performance for heavy metals in cross-regional soils.