Effects of fine grinding on mid-infrared spectroscopic analysis of plant leaf nutrient content

Abstract

Fourier transform mid infrared (FT-MIR) spectroscopy combined with modeling techniques has been studied as a useful tool for multivariate chemical analysis in agricultural research. A drawback of this method is the sample preparation requirement, in which samples must be dried and fine ground for accurate model calibrations. For research involving large sample sets, this may dramatically increase the time and cost of analysis. This study investigates the effect of fine grinding on model performance using leaf tissue from a variety of crop species. Dried leaf samples (N = 300) from various environmental conditions were obtained with data on 11 nutrients measured using chemical methods. The samples were scanned with attenuated total reflectance (ATR) and diffuse reflectance (DRIFT) FT-MIR techniques. Scanning was repeated after fine grinding for 2, 5, and 10 min. The spectra were analyzed for the 11 nutrients using partial least squares regression with a 75%/25% split for calibration and validation and repeated for 50 iterations. All analytes except for boron, iron, and zinc were well-modeled (average R² > 0.7), with higher R² values on ATR spectra. The 5 min level of fine grinding was found to be most optimal considering overall model performance and sample preparation time.

Figure 1

Top—Microscope images of dried rice leaf after 0, 2, 5, and 10 min of fine grinding. Middle—Average particle diameter (µm) of all plant species for each level of grinding (minutes). Bottom—Distribution of particle size for each level of fine grinding. Outliers are indicated by the black points.

Figure 2

Minimum, mean, and maximum spectra of all samples for ATR (top) and DRIFT (bottom). Wavenumber is measured in cm⁻¹.

Figure 3

95% confidence ellipses of sample spectral data for ATR (top) and DRIFT (bottom) spectroscopy in PC space for different levels of grinding (minutes). Measures of the center for each confidence ellipse are indicated by the ‘x’ corresponding to the category color.

Figure 4

Boxplots of R², root mean square error (RMSE), the ratio of performance to deviation (RPD), and bias for the prediction of 4 properties for each level of fine grinding for ATR (top) and DRIFT (bottom) spectroscopy techniques. Each boxplot is labeled with Tukey’s significance letters (p < 0.05) for comparison purposes. Fine grinding levels are in minutes.