Populations Structuring of Near Infrared Spectra and Modified Partial Least Squares Regression
- John S. Shenk and
- Mark O. Westerhaus
The computer programs CENTER and SELECT have been presented as a way to establish population boundaries and choose samples for near infrared calibrations. This study was conducted to evaluate calibrations derived on samples chosen by CENTER and SELECT from broad groups of hay, haylage, corn (Zea mays L.), wheat (Triticnm aestivum L.), and barley (Hordeum vulgare L.) samples. Population boundaries were established with a maximum standardized H distance from the average spectrum of 3.0. Every fifth sample was reserved for equation validation. Calibration samples were selected with a minimum standardized H distance between samples of 0.6. Forage samples were found to have more diverse spectra and chemistry than grain samples. Average r2 values were smaller, numbers of eigenvectors were larger, and standard deviations of laboratory reference values were larger for forages than for grains. The standard error of performance (SEP) for all samples and SEP for samples chosen by SELECT with a limit of 0.6 were similar for four of five products. Calibrations were developed using five different math treatments with and without multiplicafive scatter correction (De-trend). First derivative was the best math treatment for protein in all products. Second derivative was best for acid-detergent fiber (ADF) in forage products, but no single math treatment was superior for ADF in grain products. De-trend improved SEP in 28 of 50 calibrations.Please view the pdf by using the Full Text (PDF) link under 'View' to the left.
Copyright © . .