Jump to Main Content
Comparison of Near Infrared Reflectance Spectroscopy and Raman Spectroscopy for Predicting Botanical Composition of Cattle Diets☆
- Altangerel, Narangerel, Walker, John W., González, Piedad Mayagoitia, Bailey, Derek W., Estell , Rick E., Scully, Marlan O.
- Rangeland ecology & management 2017 v.70 no.6 pp. 781-786
- Cynodon dactylon, Phleum pratense, Prosopis glandulosa, Raman spectroscopy, Sorghum bicolor subsp. drummondii, Triticum aestivum, botanical composition, cattle, discriminant analysis, ecosystems, feces, hay, least squares, lignin, lipids, near-infrared spectroscopy, prediction, quantitative analysis, starch, wheat, Sudan
- Diet selection is an important driver of ecosystem structure and function that is difficult to measure. New spectroscopic instruments are available for evaluating their applicability to ecological field studies. The objective of this study was to compare near-infrared reflectance spectroscopy (NIRS) to Raman spectroscopy of fecal samples for predicting the percentage of honey mesquite (Prosopis glandulosa) in the diet of ruminally fistulated cattle fed three different hay diets and compare them for their ability to discriminate among the three base diets. Spectra were collected from feces from a feeding trial with mesquite fed at 0%, 1%, 3%, and 5% of the diet and base hay diets of timothy hay (Phleum pratense), Sudan hay (Sorghum sudanense), or a 50:50 combination of Bermudagrass hay (Cynodon dactylon) and beardless wheat hay (Triticum aestivum). NIRS and Raman spectra were used for partial least squares regression calibrations with the timothy and Sudan hays and validated with the Bermudagrass/ beardless wheat hay diets. NIRS spectra provided useful calibrations (r² = 0.88, slope = 1.03, intercept = 1.88, root mean square error = 2.09, bias = 1.95, ratio of performance to deviation = 2.6), but Raman spectra did not. Stepwise discriminant analysis was used to select wavenumbers for discriminating among the hays. Fifteen of 350 possible wavenumbers for NIRS spectra and 29 of 300 possible wavenumbers for Raman spectra met the P ≤ 0.05 entry and staying criteria. Canonical discriminant analysis using these wavenumbers resulted in 100% correct classification for all three base diets, and the Raman spectra provided greater separation than NIRS spectra. Discrimination using Raman spectra was primarily associated with wavenumbers associated with undigestible constituents of the diet (lignin). In contrast, discrimination using fecal NIRS (f.NIRS) spectra was primarily associated with wavenumbers associated with digestible constituents in the diet (protein, starch, and lipid). We believe that Raman spectroscopy deserves further investigation as a quantitative technique in ecological field studies.