Comparison of models for describing the lactation curves of Chios sheep using daily records obtained from an automatic milking system

The objectives of this study were: (i) to compare five models (Wood, Cobby & Le Du, Wilmink, Cappio Borlino, Djikstra) for describing the lactation curve of Chios sheep, (ii) to identify variation in lactation parameters related to environmental factors (season) and animal factors (parity). A data base on 61,705 recordings of daily milk production obtained from an automatic milking system was used. The lactation models were individually adjusted for each lactation. Analysis of variance was performed for the comparison of the parameter estimates. The goodness of fit measures used for comparisons of the models was the coefficient of determination (R 2), mean of mean square error (MMSE), Akaike information criterion (AIC), corrected Akaike information criterion (AICc) and Bayesian information criterion (BIC). Wood model had the lowest values for information criteria (MMSE = 347.4681, AIC = 1,056.436, AICc = 1,056.733, BIC = 1,063.856) and the highest value for the coefficient of determination (R 2 =0.79). The highest values for information criteria were found for Djikstra's model (MMSE = 636.6438, AIC = 1,076.621, AICc = 1,077.117, BIC = 1,086.582) having the same time the lowest value for the coefficient of determination (R 2 =0.59). Overall, Wood (1967) model showed the best adjustment. Despite of being more recent, the model by Djikstra (1997) mechanist based and with a higher number of parameters showed a low convergence for the data used. Wood model (1967) has a greater advantage of producing a good fit measurement with only three parameters.