Data for individually recorded silage dry-matter intake (SDMI), concentrate dry-matter intake (CDMI), live weight, milk yield and milk composition of lactating dairy cows offered silage ad libitum and concentrates on a flat-rate basis, together with data for silage composition from experiments conducted at four sites, were used to obtain simple and multiple regressions of SDMI on other variables.
Simple regressions showed that the most important variables affecting SDMI were, in order of importance: silage ammonia nitrogen, fat yield, CDMI, silage digestible organic matter concentration (DOMD) and live weight. The best multiple regression for the mean SDMI over weeks 4 to 13 of lactation accounted for proportionately 0·649 of the variation. Examination of week by week data for weeks 3 to 20 of lactation showed that two models for early and mid lactation were required to give a reasonable pattern of residual variances. These models accounted for 0·627 and 0·581 of the variation respectively. It was necessary to fit time effects explicitly in early lactation. Live weight was best represented by fitting post-calving live weight and deviations from post-calving live weight separately. A number of models requiring fewer input variables were also obtained to allow for use in situations where the full range of measurements is not made.
The new models were tested using independent data from three sites. They performed better than a number of previously published models but the best model still gave a prediction error of proportionately 0·17 about the mean actual silage intake in early lactation and 015 in late lactation.
The results suggest that there is little to be gained from further refinement of the functional form of the models and that the construction of a number of models for specific food and management situations is preferable to the use of global models.