Modelos não-lineares para ajuste de curvas de lactação de vacas gir leiteiro

Abstract

The information obtained from the variation in milk production during lactation are used to make the lactation curve. One aspect of this curve is how the milk yield falls after reaching the peak, which in dairy cattle happens around 45 days after calving, and that determines if an animal is persistent. The objective was to adjust the lactation curve of first order lactations of dairy Gir cows. The file used in the evaluation was provided by ABCZ and contained 125,992 records of the test-day yield of 16,646 cows from 1971 to 2016. Five nonlinear models were evaluated, Wood, Monophasic, Biphasic, Lactation Persistence Model (LPM) and Reduced Lactation Persistence Model (RLPM). Models were adjusted to the data using the iterative method of Gauss-Newton, by the NLIN procedure of the SAS program. The criteria used to determine the best model were the Akaike Information Criterion (AIC); Bayesian Information Criterion (BIC); Mean Square Error (MSE); Adjusted coefficient of determination (R²Aj); Durbin-Watson criterion (DW) and percentage of convergence (% C). The persistence, peak production and peak time were calculated according to the functions described for each model. Among the five adjusted models, the Wood model presented the lowest AIC (29.55391), BIC (29.39164) and MSE (1.3895), as well as the highest adjusted R² (0.66) and percentage of convergence (0.99). The Wood model presented the best goodness of fit for the curve of the first lactation of dairy Gir cows.