A novel strategy for quantitative analysis of soil pH via laser-induced breakdown spectroscopy coupled with random forest

pH is one of the significant properties of soil, and is closely related to the decomposition of soil organic matter, anion-cation balance, growth of plants and many other soil processes. In the present work, laser-induced breakdown spectroscopy (LIBS) technique coupled with random forest (RF) was proposed to quantify the pH of soil. First, LIBS spectra of soil was collected, and some common elements in soil were identified based on the National Institute of Science and Technology database. Then, in order to obtain a better predictive result, the influence of different input variables (full spectrum, different spectral ranges, the intensity of characteristic bands and characteristic lines) on the predictive performance of RF calibration model was explored with the evaluation indicators of root mean square error (RMSE) and coefficient of determination (R2), the characteristic bands of four elements (Al, Ca, Mg and Si) were determined as the optimal input variables. Finally, the predictive performance of RF calibration model was compared with partial least squares calibration model with the optimal input variables and model parameters, and RF calibration model showed a better predictive performance, and the four evaluation indicators of Rp, 2 RMSEP, mean absolute error and mean relative error were 0.9687, 0.1285, 0.1114 and 0.0136, respectively. It indicates that LIBS technique coupled with RF algorithm is an effective method for pH determination of soil.