Calibrating fecal NIRS equations for predicting botanical composition of diets
Keywordshear infrared reflectance spectroscopy
MetadataShow full item record
CitationWalker, J. W., McCoy, S. D., Launchbaugh, K. L., Fraker, M. J., & Powell, J. (2002). Calibrating fecal NIRS equations for predicting botanical composition of diets. Journal of Range Management, 55(4), 374-382.
PublisherSociety for Range Management
JournalJournal of Range Management
AbstractThe objectives of this study were to investigate the use of near infrared spectroscopy (NIRS) of fecal samples for predicting the percentage of mountain big sagebrush (Artemisia tridentata Nutt. ssp. vaseyana (Rydb) Beetle) in sheep diets and to quantify the limitations of using NIRS of fecal samples to predict diet composition. Fecal material from a sheep feeding trial with known levels of sagebrush and several background forages was used to develop fecal NIRS calibration equations validated with fecal material from 2 other sheep feeding trials with known levels of sagebrush in the diets. The 1996 calibration trial varied the level of sagebrush, alfalfa, and grass hay in the diets. The 1998 trial compared frozen to air-dried sagebrush. The Wyoming trial was a metabolism study using frozen sagebrush. Trials used different levels of sagebrush varying from 0 to 30% of the diet in increments of 4 to 10 percentage points. Internal validation of the 1996 trial with a subset of the samples not used for calibration showed that when predicted samples are from the same population as the calibration samples, this procedure can accurately predict percent sagebrush (R2 = 0.96, SEP = 1.6). However, when predicted samples were from a different population than calibration samples, accuracy was much less, but precision was not affected greatly. Low accuracy was caused by a compression of the range of data in the predicted values compared to the reference values, and the predicted sagebrush levels in the diet should be considered to represent an interval scale of measurement. Modified partial least squares regression resulted in better calibration than stepwise regression, and calibration data sets with only high, low, and no sagebrush resulted in calibrations almost as good as data sets with several intermediate levels of sagebrush. High values of the H statistic were related to low precision but did not affect the accuracy of predictions. We believe the interval scale of measurement will contain sufficient information for the purpose of addressing many questions on rangelands.