Accuracy and precision in DM intake prediction models for lactating dairy cows

Predicting the daily DM intake (DMI) of lactating dairy cows is an essential criterion for formulating diets according to requirements, which limits the application of safety margins in economically and environmentally sensitive nutrients, such as energy, protein, and phosphorus. An accurate estimation of nutrient excretion, which is necessary for good practice in crop fertilisation, is also highly dependent on DMI predictions. The study aimed to assess the accuracy and precision of the Swiss model developed in 1994 by Agroscope (2021), the North American model by National Research Council (NRC, 2001) and its update from National Academies of Sciences, Engineering, and Medicine (NASEM, 2021), the French model by Institut national de recherche agronomiques (INRA, 2018), the German model by Gesellschaft für Ernährungsphysiologie (GfE, 2023), and the Australian model by Commonwealth Scientific and Industrial Research Organization (CSIRO, 2007). The evaluation was based on routine Agroscope dairy herd data recorded between November 2015 and March 2021. The sample consisted of 138 primiparous (12.4 ± 9.7 weeks of lactation (WOL), 28.4 ± 5.5 kg/d milk yield (MY), 614 ± 57 kg BW) and 135 multiparous (16.3 ± 11.2 WOL, 32.8 ± 7.6 kg/d MY, 701 ± 63 kg BW) lactating Holstein cows, resulting in 413 partial lactations. Milk and diet composition were available on a monthly basis, and DMI, MY, and BW were collected on a daily basis. The models were assessed for RMSE of prediction, including its decomposition into error of central tendency (ECT), error of regression, and error due to disturbance. Moreover, the models were evaluated using the concordance correlation coefficient (CCC) analysis. Globally, DMI was overestimated by NRC and NASEM and underestimated by INRA and GfE. The accuracy of DMI prediction using the RMSE of prediction metric ranged from 2.50 to 4.37 kg/d in primiparous and from 3.02 to 4.98 kg/d in multiparous cows. In both cow groups, the highest precision values were obtained, with the Agroscope (ECT = 0.001 and 0.01%, respectively) model. The highest CCC was exhibited by the Agroscope model in primiparous cows (0.53) and by the INRA model in multiparous cows (0.70). Finally, the 30-year old Agroscope model emerged as the most accurate and precise in predicting DMI in lactating dairy cows fed a diet consisting of 90-95% of a mixed basal diet (dry and ensiled herbage and corn silage) and of 5-10% concentrates (DM basis).