MODELING LACTATION CURVES OF DAIRY-COWS WITH EMPHASIS ON INDIVIDUAL VARIABILITY

Five published equations were compared on their ability to adjust diff erent patterns of lactation curves. the equation y(t) = a exp (b(1)t'( 2)/2+b(2)/t-c(1+t'/2)t') with t' = (t-21.4)/100 (i) was retained becau se of the quality of the adjustment and the absence of convergence pro blems when applied on individual curves. Including an effect of season (SE) and an effect of pregnancy (PE) improved the quality of individu al adjustments (no. = 339). The final equation was equal to (i) + SE PE with SE = a + b cos (2 pi(w(c) + w - 1)/52) + c sin(2 pi(w(c) + w- 1)/52), with w(c) = week of the year of calving, w = week after calvi ng and a = 0.0065, b = -1.26 and c = 0.374, and PE = a(w(p) - 18) e(bw p), with w(p) = week of pregnancy and a, b fixed parameters. Individua l cow characteristics were used to analyse equation (i) parameters. Th e predictions obtained with this equation and several predictive funct ions of the equation (i) parameters, which differed in the way they in cluded or not the potential of production, were compared. With no indi cation of this potential, the prediction was very poor. With the initi al production (mean production of the 4th, 5th and 6th days of lactati on) as an estimate of this potential, 75% of the lactations had the me dian of absolute values of errors less 2.95 kg/day. The results were h ighly improved by using the yield during the 5th week of lactation. th e threshold of 2.95 was reduced to about 2 kg/day. The quality of the individual prediction was better for primiparous than for multiparous cows, and for French Friesian and Montbeliarde cows than for pure or c rossbred Holstein cows. although individual predictions were not alway s satisfactory, they provided excellent agreement when averaged per gr oup (20 cows).