Comparison of the Optimum Lactation Curve Models in Holstein Population of Japan

Abstract

The objective of this study was to compare the fitness of lactation curve models for milk, fat, SNF and protein yields, and fat, SNF and protein contents, and somatic cell score (SCS). Data consisted of 5,645,976 test day records from 1st to 11th lactations obtained from 346,221 cows of Holstein and collected between 2000 and 2004 in Hokkaido Dairy Cattle Milk Recording and Testing Association. Lactation curves with the optimum fitness for each trait were selected from 23 models as follows ; Legendre polynomials of order 1 to 5 (5 models ; L1, L2, L3, L4 and L5), Legendre polynomials of order 1 to 5 combined with Wilmink's exponential function (5 models ; L1W, L2W, L3W, L4W and L5W) and polynomials of order 3 to 5 combined with logarithm and/or periodic functions (13 models). Lactation curve models were compared by using an Akaike information criterion (AIC), a Bayesian information criterion (BIC), a coefficient of determination (R2) and mean residual. In those four criterions, BIC was mainly used, because BIC accounted for not only number of parameters but also number of observations. The optimum lactation curve models for milk, fat and SNF yields, and fat, protein and SNF contents were selected out any one of L4, L3W, L4W and L5W. On the other hand, the optimum lactation curves for protein yield and SCS were polynomial models with logarithm and/or periodic functions. The use of lactation curve models selected in this study would be profitable as a standard lactation curve used for feeding management.