High-throughput prediction of eucalypt lignin syringyl/guaiacyl content using multivariate analysis: a comparison between mid-infrared, near-infrared, and Raman spectroscopies for model development

Table 3 Comparison of PLS predictive models using vibrational spectroscopy and pyrolysis molecular beam mass spectrometry

Method	SEL validation^1,a	SEP^2,a	RMSEP^3,a	r-Val⁴	R²Val⁵	Outliers⁶
Raman 2 ^nd deriv (19 pt) + SNV 32 scans	0.05	0.14	0.13	0.89 ± 0.04	0.79 ± 0.08	1
Raman 1 ^st deriv (7 pt) + EMSC 32 scans	0.05	0.13	0.13	0.91 ± 0.02	0.83 ± 0.04	1
Raman EMSC + 2 ^nd deriv (15 pt) 96 scans	0.05	0.14	0.15	0.90 ± 0.02	0.81 ± 0.04	0
Raman 2 ^nd deriv (15 pt) + SNV 96 scans	0.06	0.17	0.16	0.86 ± 0.02	0.74 ± 0.04	0
MIR EMSC + 2 ^nd deriv (15 pt)	0.05	0.14	0.13	0.87 ± 0.06	0.8 ± 0.1	1
MIR 2 ^nd deriv (17 pt) + MSC	0.05	0.14	0.14	0.91 ± 0.01	0.83 ± 0.01	1
MIR 2 ^nd deriv (17 pt) + SNV	0.05	0.15	0.15	0.87 ± 0.02	0.76 ± 0.03	1
NIR EMSC + 2 ^nd deriv (25 pt)	0.06	0.19	0.20	0.79 ± 0.01	0.62 ± 0.01	0
NIR 2 ^nd deriv (25 pt) + MSC	0.06	0.18	0.18	0.82 ± 0.04	0.67 ± 0.07	1
NIR 2 ^nd deriv (25 pt) + SNV	0.06	0.22	0.21	0.80 ± 0.04	0.65 ± 0.07	1

¹Standard Error of the Laboratory for the validation data.
²Standard Error of Prediction.
³Root Mean Standard Error Prediction.
⁴Correlation coefficient for the validation set.
⁵Pearson coefficient of determination for validation.
⁶Number of outliers removed from validation models.
^aAverage errors of three randomly generated models using data provided. Models were not statistically different.
The numbers listed parenthetically reflect the degree of Savitzky-Golay spectral smoothing.
Statistical values are the average of 3 independent models.
MIR = mid-infrared spectroscopy, NIR = near-infrared spectroscopy, EMSC = extended multiplicative scatter correction, MSC = multiplicative scatter correction, SNV = standard normal variate.

ISSN: 1754-6834