The spectral fusion of laser-induced breakdown spectroscopy (LIBS) and mid-infrared spectroscopy (MIR) coupled with random forest (RF) for the quantitative analysis of soil pH

Abstract

Soil pH is one of the important properties of soil. The quick and accurate determination of soil pH is key to realizing precision agriculture and understanding soil characteristics and fertility. Previous research established a soil pH measurement method based on laser-induced breakdown spectroscopy (LIBS) technology combined with random forest (RF) (the determination coefficient of cross validation (R_c²) was 0.9995, the root mean square error of cross validation (RMSEC) was 0.0201; the determination coefficient of prediction (R_p²) was 0.9687, and the root mean square error of prediction (RMSEP) was 0.1285). This study explored the impact of three different spectral preprocessing methods (first derivative (D1^st), multivariate scattering correction (MSC), and standard normal variation (SNV)) on the prediction performance of the RF correction model using 21 soil samples, as in the previous study. The input variables were optimized through variable importance thresholds. Then, a method was established based on mid-infrared (MIR) technology combined with RF for the qualitative analysis of soil pH (R_c² = 0.9887, RMSEC = 0.0875, R_p² = 0.9208, RMSEP = 0.1476, and the mean relative error (MRE) was 0.0168). Meanwhile, a soil pH measurement method based on a LIBS-MIR spectral data fusion strategy combined with RF was further established. The results showed that the RF calibration model based on intermediate spectral data fusion showed better prediction abilities (R_c² = 0.9997, RMSEC = 0.0163, R_p² = 0.9809, RMSEP = 0.0645, and MRE = 0.0065). Compared with a spectral analysis method based on LIBS or MIR alone, this study provides new ideas and new methods for the rapid, accurate, and quantitative analysis of soil pH.