Evaluation of five models predicting feed intake by dairy cows fed total mixed rations

The objective of this study was to evaluate the accuracy of five models predicting dry matter intake (DMI) in dairy cows fed total mixed ration (TMR). The five models were the North American model from NRC, and the Northern European models: NorFor (Denmark, Iceland, Norway, and Sweden), TDMI (Finland), Zom (the Netherlands), and Gruber (Austria, Germany, Switzerland). The evaluated models represent different approaches to predict DMI. One approach uses only animal characteristics; a second uses the interaction between animal and dietary characteristics, and a third uses no production characteristics, such as body weight or milk yield. These different modelling approaches results in very different substitution rates, where only two of the models demonstrate direct or indirect relation to concentrate allocation. Accuracy of DMI prediction was evaluated by mean square prediction error (MSPE), root mean square prediction error (RMSPE), together with the decomposition of error into error of central tendency (ECT), error of regression (ER), and error due to disturbance (ED). The evaluation was performed on data from 12 Scandinavian production experiments with a total of 917 lactating dairy cows in 94 treatment means. The NorFor model was evaluated on only 9 of the experiments as 3 experiments had been used in the development of this model. The five models predicted DMI in groups of dairy cows fed TMR with RMSPE ranging between 1.2 kg dry matter (DM) per day for the Gruber model to 3.2 kg DM per day for the Zom model. Evaluated across the experiment the ECT and the ER ranged between 0.3% and 65% and between 3% and 38% of MSPE, respectively. Error associated to ED ranged between 31% and 93% of MSPE. When all five models were evaluated for prediction of DMI both across and within experiments, results revealed that all five models predicted differences between diets within experiments better than differences across experiments. The Gruber model, which predicted DMI most accurately did so due to its negligible systematic error (ECT, ER) resulting in 93% of the error located in ED. (C) 2015 Elsevier B.V. All rights reserved.