John S. Shenk

Investigation of a LOCAL calibration procedure for near infrared instruments

A new procedure (LOCAL) for local calibrations is presented. LOCAL selects spectra from a library of samples and computes a PLS calibration equation for each constituent of the sample. This study evaluated the performances of LOCAL on the prediction of ground corn grain and haylage using several different combinations of data transformations, wavelength segment reduction, number of PLS factors and samples used in calibration. LOCAL resulted in lower SEP values for all the constituents of corn and dry matter of haylage with improvements ranging between 6 to 13%. Global calibrations had only a small advantage over LOCAL (1–2%) in the prediction of acid detergent fibre and crude protein in haylage. The two most important variables controlling the accuracy of predictions were number of samples in the calibration and number of PLS factors in the solution. Best results were obtained using 150 to 225 samples and more than 20 PLS factors per calibration equation. The speed of the LOCAL procedure is 0.5–2 s per sample on a 90 MHz computer. With this speed and accuracy, LOCAL is now available for real-time routine operation on a Windows platform.

LOCAL prediction with near infrared multi-product databases

This study evaluated the use of an algorithm (LOCAL) for local calibration using multi-product databases. Four different databases were used: forages (hay, corn silage, haylage, small grain silage and total mixed ration; n=2924), grain (barley, corn, oats and wheat; n=1464), meat (meat and bone meal, fish meal and poultry meal; n=693) and feed (bakery products, mixed feed, poultry feed and soya products; n=1518). One-tenth of the samples were selected for validation from each database. Predictions of validation samples using generic and specific global calibrations were compared to the predictions generated by LOCAL. Standard errors of prediction for LOCAL calibrations were always lower than those of generic global calibrations and similar to those of specific global calibrations. However, LOCAL predictions were further improved by using different settings for each constituent. The analysis of the samples selected by LOCAL showed that for heterogeneous products such as total mixed rations and corn silage, LOCAL optimised predictions by choosing samples from different products. LOCAL calibration was then used with one database (n=6599) comprising all the samples. Standard errors of prediction were similar to those obtained with the four different databases. LOCAL can accurately predict the composition of different products using multi-product databases. Routine analysis can be simplified by using LOCAL calibration combined with large databases. In addition, LOCAL can provide accurate predictions of spectra from remote standardised instrument without the operator identifying the sample.

The development of near infrared wheat quality models by locally weighted regressions

A large data base of near infrared and protein data was used to examine the utility of a data base regression technique know as LOCAL in the ISI International software package. A total of 2203 samples of wheat from five classes with protein values by combustion analysis comprised the data base. One half of the samples came from hard red spring and hard red winter wheats. The number of samples in the individual classes ranged from 235 for Durum to 694 for hard red spring. These samples were collected over a five year period and represented wheat in commercial trade. Calibrations were determined for each class, the entire data base as a “global” calibration, a data base regression which selected appropriate samples for a unique calibration for each sample and the standard equations of NIR spectroscopy regulatory analysis. Results will be reported which show data base regression technique to be as good as specific calibrations by partial least squares regression for sub-classes of products and precise enough for use for regulatory purposes. The use of this data base regression technique over normal learning sets has other advantages, such as easier calibration update, easier transferability and the possibility to include authentication and classification as part of the model.

Comments on Standardisation: Part 2

Many standardisation methods were evaluated before the patented method was selected. We evaluated a method similar to piecewise direct calibration. The method worked very well on mathematically created wavelength shifts, baseline offsets and band pass changes. When actual instrument differences were evaluated, however, the coefficients in the standardisation models were unrealistically large. We concluded that higher order models were dependent on small spectral changes (on the order of sample rescan repeatability). We then concentrated on simple linear models and developed the method that was later patented. This method of standardisation is represented in Figure 2. The patented method was developed and optimised for Pacific Scientific 6250 monochromators. Although satisfactory for those instruments, the

Determination of Soil Separates with near Infrared Reflectance Spectroscopy

The use of near infrared (NIR) reflectance spectroscopy to evaluate soil properties has started to receive more attention in recent years. The technology is evolving and research on NIR spectroscopic analysis using natural state samples is increasing. There is no method available today, besides NIR spectroscopy, that could simultaneously evaluate physical and chemical properties of a soil sample without processing the sample and affecting the visual quality of the site. More samples can be scanned in their natural undisturbed form resulting in a variety of particle sizes. Research on the effect of scanning products with different particle sizes is essential. The differences in the particle size of the soil separates may lower the prediction accuracy of NIR spectroscopy. In this study, we evaluated the ability of NIR spectroscopy to predict soil separates from artificial soil samples. Feldspar and silica sands and silts, kaolinite and montmorillonite clays, and reed sedge and Canadian sphagnum peat moss organic matters were used as separates. They were scanned alone, and in different mixture percentages, from 400 to 2500 nm with a total of 116 samples. The absence of linearity in the binary mixtures, preventing accurate calibration, was noticed and required the development of a transformation model to generate new laboratory values from a laboratory weight scaling factor generated for each soil separate. The adjustment of the laboratory values improved the prediction accuracy of the mixtures. The coefficient of determination ranged from 0.95 to 0.99. The standard error of cross-validation ranged from 2.09 to 5.82%.