Table 2.
Modified partial least-squares statistics of calibration and cross-validation for two calibration models developed using near-infrared reflectance spectroscopy (NIRS) for crude protein content (CP, in%) based on different sample sets and combinations of cowpea leaves from Eastern Africa
Model | N | Mean | SD | Min | Max | SECV | R2 | G-H | RPD |
---|---|---|---|---|---|---|---|---|---|
Initial1 | 103 | 30.9 | 5.45 | 14.7 | 47.3 | 0.60 | 0.82 | 3.46 | 11.2 |
New | 261 | 32.1 | 4.52 | 18.6 | 45.7 | 0.74 | 0.93 | 0.71 | 6.84 |
N, number of spectra in the calibration set; Mean, estimated by NIRS (expressed as CP, in %); SD, standard deviation; Min, lowest value of reliably estimating samples; Max, maximum value of reliably estimating samples; RPD, ratio performance deviation; SECV, standard error of cross-validation; R2, determination coefficient of calibration; G-H, global-H value, where H is the Mahalanobis distance.
Initial calibration model (Tefera 2006) of 107 samples had four outliers removed.