Assessing the differences of two vineyards soils' by NIR spectroscopy and chemometrics

Abstract

Soil properties influence greatly the status of vine plants which consequently influences the quality of wine. Therefore, in the context of viticulture management, it is extremely important to assess the physical and chemical parameters of vineyards soils. In this study, the soils of two vineyards were analysed by near-infrared (NIR) spectroscopy and established analytical reference procedures. The main objective of this study was to verify if NIR spectroscopy is a potential tool to discriminate the soils of both vineyards as well as to quantify differences of soil's parameters. For that, a total of eight sampling spots were selected at each vineyard taking into consideration the soil type and sampled at different depths. The data analysis was performed using analysis of variance (ANOVA), principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) and partial least squares (PLS) regression. The ANOVA results revealed that 12 out of the 18 parameters analysed through the reference procedures can be considered statistically different (p < 0.05). Regarding PCA, the obtained results revealed a clear separation between the scores of both vineyards either considering NIR spectra or the chemical parameters. The PLS-DA model was able to obtain 100 % of correct predictions for the discrimination of both vineyards. PLS regression analysis using NIR spectra revealed R²_P and RER values higher than 0.85 and 10, respectively, for 8 (pH (H₂O), N, Ca²⁺, Mg²⁺, SB, CEC, ECEC and GSB) of the 18 chemical parameters evaluated. Concluding, these results demonstrate that it is possible to discriminate the soils of the different vineyards through NIR spectroscopy as well as to quantify several chemical parameters through soils NIR spectra in a rapid, accurate, cost-effective, simple and environmentally friendly way when compared to the reference procedures.

Keywords: Chemometrics; NIR spectroscopy; PCA; PLS; PLS-DA; Soils.

Fig. 1

Diagram of Portuguese wine regions with the location and photographs of both vineyards and the respective sampling spots.

Fig. 2

Score plot of the first two principal components using soils parameters which captured 67.8 % of the total variance (data were auto scaled).

Fig. 3

Loadings of PC1 of the PCA model performed using soils parameters.

Fig. 4

Class predictions for the 2 LV PLS-DA model calibrated from the soil's chemical parameters for QC (a) and QL (b).

Fig. 5

Regression coefficient vector for the PLS-DA model considering soils parameters pre-processed with auto scaling and using 2 LV.

Fig. 6

Score plot of the first two principal components using NIR soils' spectra which captured 93.3 % of the total variance (spectra were pre-processed with Savitzky-Golay (using 15 points filter width, second polynomial order and first derivative) followed by SNV and then mean centered).

Fig. 7

Loadings of PC1 of the PCA model performed using soils NIR spectra.

Fig. 8

Class predictions for the 2 LV PLS-DA model calibrated from the soils NIR spectra for QC (a) and QL (b).

Fig. 9

Regression coefficient vector for the PLS-DA model considering the NIR spectra pre-processed with mean centring and using 2 LV.

Fig. 10

Experimental values versus the cross-validation (●) and prediction (■) model estimated obtained for Ca²⁺(a), pH (H₂O) (b), N (c), CEC (d), Mg²⁺(e), SB (f), ECEC (g) and GSB (h).

Fig. 11

Regression coefficient vectors for the PLS models for Ca²⁺(a), pH (H₂O) (b), N (c), CEC (d), Mg²⁺(e), SB (f), ECEC (g) and GSB (h).