PLS modeling performances on the validation sets are summarized
in Table I. For each calibration condition, optimal model
performance, defined as the best combination of high r2 (validation)
and low SEP, generally occurred at the same number (±1)
of factors for each laboratory. As such, the statistics in Table
I for each condition are reported at the number of factors collectively
deemed as optimal among the three laboratories. These
full wavelength-region PLS models, regardless of laboratory or
of whether a second-difference pretreatment was performed on
the spectra, resulted in r2 (validation) values of 0.93 or higher.
The accompanying SEP values were -1% AAC, with negligible
bias, which are comparable with recently reported results on AAC
determination by NIR transmittance (800-1,050 nm wavelength
range) through bulk samples of milled rice (Villareal et al 1994).