Evaluation of nonlinear models to predict milk yield and composition of beef cows: A meta-analysis

Abstract

The accurate prediction of milk yield is necessary to estimate the nutrient requirements of cows and calves. It still necessary the development of reliable equations that include data from beef cows experiencing different conditions in different years. The objective of this study was to evaluate the milk yield of grazing Nellore cows under tropical conditions using a meta-analytical approach. The dataset was collected from 16 studies, which were carried out between 2014 and 2022, under tropical conditions (Brazil), totaling 1503 cow-day observations. Milking procedures were based on a controlled suckling period before the calf separation, with daily production estimated by two methods: sum of morning and afternoon milking (AM_PM milk), and morning milking corrected to 24 h (AM_Corr milk). A sample of milk was collected from each cow to evaluate milk composition. The methods of milk collection AM_PM milk and AM_Corr milk were compared within the AM_PM dataset using F-test of the analysis of variance. A regression analysis was performed according to milk composition and week using the PROC MIXED procedure of SAS considering study and cow(study) as random effects. Four nonlinear alternative equation forms to predict milk yield were fitted to the data from 9 experiments of multiparous cow’s data (Mult; > 3 years) and to data from 4 experiments composed of primiparous cows (Prim; ≤ 3 years): Wood, (1967) [WD], Jenkins & Ferrell, (1984) [JF], Wilmink, (1987) [WK], and Cobby & Le Du, (1978) [CL]. Parameters from equation forms were fitted using the PROC NLMIXED procedure of SAS and considering study and cow(study) as random effects. For the independent validation, an external database was used for testing the models which was composed of 3 different studies previous separated randomly from the main dataset composed (n = 290 cow-day observations). The milk yield estimated by the extant equations were compared with the observed values on the independent dataset using the Model Evaluation System software. The equation of the last edition of the Nutrient Requirements of Zebu and Crossbred Cattle (BR-CORTE) was also included for comparison. The AM_PM and AM_Corr approaches were deemed equivalent, with an average of 6.51 and 6.70 kg ( ± 0.098), respectively (P = 0.261). The equation JF did not estimate milk yield of the independent data well (P > 0.05). The equations of WD, WK, and CL had the best estimations of milk yield and description of the lactation curve of grazing multiparous cows. All milk components changed during the lactation period, in which total solids, protein, and fat increased linearly throughout lactation, while lactose decreased linearly (P < 0.05).