Description
Near infrared reflectance spectroscopy (NIRS) for mineral analyses in feeds, forages and animal faeces has produced mixed results. Attempts to develop calibration equations have resulted in low prediction accuracy. Sample presentation have been shown to affect accuracy and precision of NIRS calibrations (Lovett, et al., 2005). Studies have also shown that vital information can be gleaned by extending the wavelength range for spectra data collection to the visible range (400 – 700 nm) (Cozollino and Moron, 2004). However, information on the impact of wavelength range of measurement and sample volume on accuracy of calibrations for the prediction of minerals is insufficient. A total of 111 faecal samples collected from horses with variation in type, gender, diet and age were used in this study to determine the impact of wavelength region of measurement (400-2500 nm and 1100-2500 nm) and sample volume (half-full cup and full cup) on accuracy of prediction of minerals. Scanning of samples to collate spectra data resulted in four spectra categories. Each category was subjected to four scatter corrections and five mathematical treatments to give 80 predictions per mineral. Improved predictions, shown by the ratio performance deviation (RPD) and coefficient of determination of validation (R2val) were 2.67 (0.90), 3.49 (0.92), 3.01 (0.89) and 2.33 (0.81) for Ca, Cu, Zn and Fe respectively, at 400-2500 nm with full cup. A similar trend was observed for P, Mg, S and Mo [RPD and R2val were 1.91(0.72), 2.45 (0.84), 2.16 (0.81) and 1.56 (0.6) respectively], although, with half-full cup. Wavelength region of measurement seemed to have more influence than the sample volume for prediction. Overall, accuracy of calibration models for the prediction of macro and micro minerals can be improved if spectra data is collected at the combination of visible and NIR region with either half-full or full cup of sample.
| Consider for full paper in JNIRS | Yes, please |
|---|
