Accuracy comparison of dry matter intake prediction models evaluated by a feeding trial of lactating dairy cows fed two total mixed rations with different forage source

Dry matter intake (DMI) prediction models of NRC (2001), Fox et al. (2004) and Fuentes-Pila et al. (2003) were targeted in the present study, and the objective was to evaluate their prediction accuracy with feeding trial data of 32 lactating Holstein cows fed two total mixed rations with different forage source. Thirty-two cows were randomly assigned to one of two total mixed ration groups: a ration containing a mixed forage (MF) of 3.7% Chinese wildrye, 28.4% alfalfa hay and 26.5% corn silage diet and another ration containing 33.8% corn stover (CS) as unique forage source. The actual DMI was greater in MF group than in CS group (P=0.064). The NRC model to predict DMI resulted in the lowest root mean square prediction error for both MF and CS groups (1.09 kg d−1 vs. 1.28 kg d−1) and the highest accuracy and precision based on concordance correlation coefficient for both MF and CS diet (0.89 vs. 0.87). Except the NRC model, the other two models presented mean and linear biases in both MF and CS diets when prediction residuals were plotted against predicted DMI values (P<0.001). The DMI variation in MF was caused by week of lactation (55.6%), milk yield (13.9%), milk fat percentage (7.1%) and dietary neutral detergent fiber (13.3%), while the variation in CS was caused by week of lactation (50.9%), live body weight (28.2%), milk yield (8.4%), milk fat percentage (5.2%) and dietary neutral detergent fibre (3.8%). In a brief, the NRC model to predict DMI is comparatively acceptable for lactating dairy cows fed two total mixed rations with different forage source.