Development of a fecal near-infrared spectroscopy model to estimate nutritional parameters in beef cattle and sheep

Studies highlight the potential of fecal near-infrared spectroscopy (fNIRS) to estimate nutritional parameters in grazing ruminants but bi-species calibrations remain scarce. Therefore, the present study aimed to evaluate the quality of models developed using samples from beef cattle and sheep fed on native grassland from the Pampa Biome in southern South America, Italian ryegrass ( Lolium multiflorum L.), and Pearl millet ( Pennisetum americanum (L.) Leeke). The database comprised 225 fecal samples (78 cattle, 147 sheep) obtained from 23 metabolism assays (13 cattle, 10 sheep) conducted across six independent experiments. The fNIRS models were developed for the parameters of organic matter intake (OMI), organic matter digestibility (OMD), and dietary crude protein concentration (dCPc). The models were constructed by Partial Least Squares regression, using 80% of the samples for calibration and 20% for prediction, in addition to interspecies validation. The preprocessing methods tested were Standard Normal Variate, Extended Multiplicative Scatter Correction, Detrending, Savitzky-Golay, second derivative, and Smoothing. For OMD and dCPc, the models showed satisfactory performance, based on relative prediction error of prediction (RPEp) of 5.02% and 8.91%, respectively, and calibration coefficient of determination (R 2 c) of 0.82 and 0.92; however, the model for dCPc presented a proportional error (slope ≠ 1, P < 0.05). In contrast, for OMI, the performance was unsatisfactory, with an RPEp of 22.15% and R 2 c of 0.68. In summary, bi-species models have shown promise when compared to mono-species models, exhibiting better accuracy. However, it is recommended that users evaluate the conditions under which the proposed models will be used.