MIR and Vis/NIR spectroscopy cannot authenticate organic bulk milk

Figures & data

Table 1. Total mixed ration composition (least square means ± standard error) administered to the cows during lactation.

Trait^a	Organic farms	Conventional farms	p Value
DM, %	49.55 ± 3.66	51.40 ± 3.66	.724
Ether extract, % of DM	3.03 ± 0.24	3.34 ± 0.24	.380
Crude protein, % of DM	13.67 ± 0.65	14.94 ± 0.65	.181
Starch, % of DM	24.79 ± 1.71	23.73 ± 1.71	.667
Ash, % of DM	7.38 ± 0.34	7.24 ± 0.34	.763
aNDFom, % of DM	38.81 ± 1.53	38.64 ± 1.53	.940
ADFom, % of DM	23.34 ± 1.10	23.04 ± 1.10	.848
Lignin (sa), % of DM	4.05 ± 0.35	4.04 ± 0.35	.990

^a DM: dry matter; aNDFom: neutral detergent fibre assayed with a heat stable amylase expressed exclusive of residual ash; ADFom: acid detergent fibre; lignin (sa): lignin determined by solubilisation of cellulose with sulphuric acid.

Figure 1. Standardised principal component 1 (PC1) and principal component 2 (PC2) plot with a normal confidence ellipse (95% confidence level) for each category built: (a) using mid-infra-red spectroscopy and (b) using Vis/NIR spectroscopy. In green, conventional farms; in orange, organic farms. Graph created with R-software.

Figure 2. Absorbance of organic (grey dots) and conventional (black dots) milk, and variable in projection (VIP) score (black) for two components after removing signal-to-noise ratio wavelength of the spectra collected using: (a) MIR spectroscopy and (b) Vis/NIR spectroscopy. Black dashed-line indicates the threshold value of 1 for VIP score.

Table 2. Confusion matrix and statistical measures of performance of the partial least squares-discriminant analysis model for two components for the classification of organic and conventional bulk milk in the MIR region.

	Predicted		Predicted
	Train set (n = 151)		Test set (n = 74)
Original class	Organic	Conventional	Organic	Conventional
Organic	42	31	11	26
Conventional	21	57	8	29
Model performance, %
Error rate	0.34		0.46
Accuracy	0.66		0.54
Sensitivity	0.58		0.30
Specificity	0.73		0.78
False positive rate	0.27		0.22
Precision	0.67		0.58

The model was built selecting wavelengths with variable in projection greater than one after removing low signal-to-noise ratio wavelengths.

Table 3. Confusion matrix and statistical measures of performance of the partial least squares-discriminant analysis model for two components for the classification of organic and conventional bulk milk in the Vis/NIR region.

	Predicted		Predicted
	Train set (n = 121)		Test set (n = 62)
Original class	Organic	Conventional	Organic	Conventional
Organic	32	26	14	17
Conventional	23	40	6	25
Model performance, %
Error rate	0.40		0.37
Accuracy	0.60		0.63
Sensitivity	0.55		0.45
Specificity	0.63		0.81
False positive rate	0.37		0.19
Precision	0.58		0.70

The model was built selecting wavelengths with variable in projection greater than one after removing low signal-to-noise ratio wavelengths.

Data availability statement

The data presented in this study are available free of charge for any user upon reasonable request from the corresponding author (CLM).