Soil Particle Size Thresholds in Soil Spectroscopy and Its Effect on the Multivariate Models for the Analysis of Soil Properties

Abstract

This study focused on one of the few but critical sample preparations required in soil spectroscopy (i.e., grinding), as well as the effect of soil particle size on the FTIR spectral database and the partial least squares regression models for the prediction of eight soil properties (viz., TC, TN, OC, sand, silt, clay, Olsen P, and CEC). Fifty soil samples from three Moroccan region were used. The soil samples underwent three preparations (drying, grinding, sieving) to obtain, at the end of the sample preparation step, three ranges of particle size, samples with sizes < 500 µm, samples with sizes < 250 µm, and a third range with particles < 125 µm. The multivariate models (PLSR) were set up based on the FTIR spectra recorded on the different obtained samples. The correlation coefficient (R²) and the root mean squared error of cross validation (RMSECV) were chosen as figures of merit to assess the quality of the prediction models. The results showed a general trend in improving the R2 as the finer particles were used (from <500 µm to 125 µm), which was clearly observed for TC, TN, P₂O₅, and CEC, whereas the cross-validation errors (RMSECV) showed an opposite trend. This confirmed that fine soil grinding improved the accuracy of predictive models for soil properties diagnosis in soil spectroscopy.

Keywords: FTIR; partial least squares regression; particle size; soil properties.

Figure 1

Location map of the study sites, viz., Rhamna, Beni Mellal, and Larache regions of Morocco.

Figure 2

Workflow for building the database used in the study of the effect of particle size on soil spectroscopy.

Figure 3

FTIR spectra of the soil samples ground and sieved at 500 µm (A), 250 µm (B), and 125 µm (C) in the mid-infrared range (600 to 4000 cm⁻¹); (D) represents the total spectra after the preprocessing (first derivative). The different colors represent the different samples.

Figure 4

The average particle size curves of the samples belonging to the different particle size groups <500 µm, <250 µm, and <125 µm.

Figure 5

(A) Variation in the correlation coefficients according to the size of sample particles prepared before the acquisition of the FTIR spectra of the fifty soil samples. (B) Variation in the root mean squared error of cross validation (RMSECV) for eight selected soil properties according to the particle size range of the forty prepared soil samples.

Figure 6

PLSR models of the eight properties of interest, e.g., TC, TN, OC, sand, silt, clay, Olsen P, and CEC, depending on the size of sample particles.

Figure 7

Contact phenomenon between the infrared beam in the FTIR and the soil particles of the sample to be analyzed.