Optimization of measuring procedure of farmland soils using laser‐induced breakdown spectroscopy

Abstract

Laser‐induced breakdown spectroscopy (LIBS) is an emerging multi‐elemental analytical technique offering fast and simultaneous quantification of soil properties with minimal sample preparation and effective cost. Due to soil heterogeneity, spectral variation however limits the quantitative robustness. In this study, 348 soil samples were collected and prepared for acquisition of LIBS spectra. Influences of shot layer and number on LIBS quality were evaluated by spectral intensity and relative standard deviation (RSD). Effects of shot layer and number and five normalization procedures on LIBS ability to measure soil organic matter (SOM), total nitrogen (TN), and total soluble salt content (TSC), were evaluated using partial least squares regression (PLSR). Increasing shot number reduced LIBS spectral variance, thereby improving the quantitative accuracy of selected soil properties. Deep shot layers (4th or 5th shot layers) reduced the intensities of soil spectra and thereby decreased the quantitative accuracy for TSC. However, deep shot layers improved the SOM and TN prediction performances. Among the normalization approaches, the method based on the correction of Si line (DS) showed superior performance for improving quantitation of SOM and TN. The arithmetic average method (AA) was best for TSC prediction. Optimization of shot layer, number and normalization procedures of LIBS spectra resulted in fair prediction of SOM (residual prediction deviation of validation set, RPDV = 1.608), good prediction of TN (RPDV = 1.836), and very good quantitative analysis of TSC (RPDV = 2.456). Therefore, our findings illustrate very good potential for improving the quantitative accuracy of the LIBS soil spectra.

Modified self‐adaptive model for improving the prediction accuracy of soil organic matter by laser‐induced breakdown spectroscopy

“…Consequently, variations in the spectra at each site are significantly different, which is mainly responsible for the low repeatability of LIBS analysis (Wainner, Harmon, Miziolek, McNesby, & French, 2001). To optimize the LIBS measuring procedure, spectral normalization and calibration or prediction modeling are effective ways to improve accuracy, as found in our previous study (Xu et al., 2020). We concluded that in‐depth effort is needed to overcome the poor accuracy caused by spatial variations in the soil.…”

“…The soil tablets were placed over an X–Y–Z manual–automatic micrometric platform with a 1.0‐μm stage of travel at every coordinate. According to the optimized soil measuring procedure (Xu et al., 2020), the LIBS spectra of soils for predicting SOM were collected at five depths with a 5‐ by 5‐mm window cut. The soil spectra of the deepest shot layer are the best for predicting SOM, according to Xu et al.…”

“…The soil spectra of the deepest shot layer are the best for predicting SOM, according to Xu et al. (2020). Thus 25 single‐shot spectra in total were obtained for each soil sample.…”

“…The spectra with the first principal component scores of below –10,000 were identified as outliers and were removed. The multiple‐measured spectra were then normalized via the method based on correction of the Si line (Xu et al., 2020) to obtain a characteristic spectrum for each soil sample. Fifty soil samples (10 fluvo‐aquic soils, 10 paddy soils, 10 red soils, 10 black soils, and 10 chestnut soils) were randomly selected as the validation set for verifying the prediction performance.…”