Author
 |
REEVES III, JAMES |
|
Submitted to: Applied Spectroscopy
Publication Type: Peer Reviewed Journal Publication Acceptance Date: 3/11/2009 Publication Date: 6/1/2009 Citation: Reeves III, J.B. 2009. Does Spectral Format Matter in Diffuse Reflection Spectoscopy. Applied Spectroscopy. 63:669-677. Interpretive Summary: Spectroscopy uses light to determine the composition of materials such as animal feedstuffs. Over the last several decades, near-infrared spectroscopy which uses light in the visible and beyond, has come to be a major tool for the analysis of materials ranging from forages to drugs. More recently, mid-infrared spectroscopy which uses light well beyond human sight, has also been shown to be capable of being similarly used. In both of these methods, spectra are generally collected as light radiation reflecting off a sample, compared to a background with no sample and converted to a specific format using a log transformation. This is done because according to known principles (Beer-Lambert Law) it is the log transform which is related to concentration of the parameter of interest. However, some researchers do not transform the data, but use the data directly as reflectance. While the Beer-Lambert Law holds that the results should not be as accurate without the required transformation, the question arises as to why then does not everyone transform the data? The objective of this work was to investigate this question using mid- and near-infrared spectra in various forms. Calibrations were developed using spectral data from forages in several formats: reflectance, log (1/reflectance), non-background corrected spectra, and others. Results indicate that, contrary to the Beer-Lambert Law, calibrations using partial least squares regression (statistical method for relating spectral information to parameter concentrations), do not require any specific data format. Accurate calibrations were developed for fiber, digestibility and protein measures in forages using any of the aforementioned spectral formats including non-background corrected interferograms (frequency domain format for Fourier Transform spectrometers). While calibrations could be developed using any of the formats, results indicate those using Kubelka-Munk (Format developed specifically for reflectance spectra in the mid-infrared spectra range) and especially interferograms, do not perform as well as the others, although they were still quite good. In conclusion, results using forage spectra indicate that accurate and equivalent calibrations can be developed using reflectance data, with or without background correction, or log (1/reflectance) at least when using partial least squares regression for calibration development. Technical Abstract: Near- and more recently, mid-infrared diffuse reflectance spectroscopy have come to be extensively used to determine the composition of products ranging from forages to drugs. In these methods, spectra are generally collected as (Reflectance or R) and transformed to log (1/R) according to the Beer-Lambert Law which states that concentration of the absorbing species is directly related to spectral absorbance, equivalent to log (1/R). However, some researchers do not transform the data, but use the data directly as reflectance. While the Beer-Lambert Law holds that the results should not be as accurate without the required transformation, the question arises as to why then does not everyone transform the data? The objective of this work was to investigate this question using mid- and near-infrared spectra in various forms. Calibrations were developed using spectral data from forages in several formats: R, log (1/R), non-background corrected single beam spectra, interferograms and Kubelka-Munk transformed data. Calibrations were developed using both non-pretreated spectral and using data pretreatments such as derivatives. Results indicate that, contrary to the Beer-Lambert Law, calibrations using partial least squares regression do not require any specific data format. Accurate calibrations were developed for fiber, digestibility and protein measures in forages using any of the aforementioned spectral formats including non-background corrected interferograms. While calibrations could be developed using any of the formats, results indicate those using Kubelka-Munk and especially interferograms do not perform as well as the others, although they were still quite good. In conclusion, results using forage spectra indicate that accurate and equivalent calibrations can be developed using reflectance data, with or without background correction, or log (1/R) at least when using partial least squares regression for calibration development. |
