Table 1 Accuracies for different pre-processing methods achieved in model validation (mean and standard deviation from 100 runs).

From: Fine root lignin content is well predictable with near-infrared spectroscopy

Method

nLV

RPD

RP2

RMSEP [%]

SEP [%]

BIAS [%]

Raw

8.17 ± 0.75

1.83 ± 0.32

0.71 ± 0.08

4.11 ± 0.62

4.05 ± 0.59

−0.43 ± 0.95

B.als

6.62 ± 1.48

1.73 ± 0.27

0.69 ± 0.07

4.34 ± 0.64

4.27 ± 0.59

−0.44 ± 1.05

B.irls

7.96 ± 1.40

1.69 ± 0.30

0.68 ± 0.09

4.47 ± 0.69

4.40 ± 0.65

−0.49 ± 1.04

MSC

7.34 ± 1.61

1.95 ± 0.37

0.74 ± 0.09

11.64 ± 8.74

3.84 ± 0.64

−1.83 ± 13.97

SNV

7.37 ± 1.28

1.96 ± 0.35

0.74 ± 0.08

3.83 ± 0.57

3.79 ± 0.56

−0.22 ± 0.86

D1

7.95 ± 2.08

1.84 ± 0.39

0.71 ± 0.10

4.14 ± 0.73

4.09 ± 0.70

−0.35 ± 0.98

D2

5.04 ± 2.31

1.56 ± 0.17

0.58 ± 0.09

4.77 ± 0.64

4.71 ± 0.59

−0.02 ± 1.18

  1. Given are measures of accuracies as number of latent variables used in PLSR (nLV), residual predictive deviation (RPD), measure of determination between predicted and observed lignin contents (Rp²), root mean squared error of prediction (RMSEP), standard error of prediction (SEP) and deviation from the line of equality of linear regression between predicted and observed values (BIAS). We used raw data (Raw) and six different pre-processing methods: asymmetric least squares baseline offset correction (B.als), iterative restricted least squares baseline offset correction (B.irls), multiplicative scatter correction (MSC), standard normal variate (SNV), first derivative (D1) and second derivative (D2). The best model is highlighted in bold.
Back to article page