Fitting lactation curves in a Colombian Holstein herd using nonlinear models

Statistical characterization of dairy cows lactation curves allows prediction of total milk yield from one or a few test-day records. The objectives were the characterization of milk production in a Colombian Holstein herd under tropical dairy production and to evaluate the statistical performance of six different lactation curve models on test-day records. The data set included 425 full lactations and 244,876 test day records of 14 consecutive years. Records of adjusted milk yield at 305 days, lactation length, production peak, time to the peak, and lactation persistence were analyzed under an univariate mixed linear model, and six non-linear lactation curve models were evaluated on the basis of their goodness of fit on test day records. Cows averaged 5830 ± 59 kg of milk at 305 days, 330 ± 3 days of lactation, 27.7 ± 0.3 kg of milk at peak, 4.7 ± 0.1 wk to the peak and 63.1 ± 0.6 % of persistence calculated for a 6 month period. Adjusted milk yield, lactation length and production peak increased with the number of calving. The best goodness of fit for each lactation curve or minimum means square error for records of average daily milk production of each week was reached by the two-phase lactation curve of Grossman and Koops. Under the conditions of this tropical dairy herd, results provide substantial evidence favoring the two-phase lactation curve of Grossman and Koops for research and technical purposes, while the model of Wood would be the best choice for livestock use