Production responses of lactating dairy cows fed silage-based diets to changes in nutrient supply

Abstract

The objective of this meta-analysis was to develop empirical prediction equations for milk and milk component yields using energy and protein intake and dietary composition as independent variables. A dataset of 1125 dietary treatment means from 245 production studies in dairy cows was collected. The diets were mainly based on grass silage or grass silage partly or completely replaced by leguminous or whole crop silages. Concentrates were fed on a flat rate basis. They were composed of cereal grains, fibrous by-products and different protein supplements. Dietary characteristics [crude protein (CP), ether extract (EE), neutral detergent fibre (NDF), starch, non-fibre carbohydrates (NFC) and silage fermentation] and feed values [metabolizable energy (ME) and protein (MP), ruminal CP balance and Feed MP] were recorded in the dataset. Production variables showed wide ranges: days in milk (38–232), dry matter (DM) intake (9.9–25.2kg/d), and yields of milk (MY; 13.0–45.8)kg/d, energy corrected milk (ECM; 12.8–43.5kg), fat (479–1830g/d) and protein (PY, 359–1449g/d). Mixed model regression analysis with a random study effect was used to develop prediction equations for MY, ECM and PY. The best-fit models for MY and ECM included linear and quadratic effects of ME intake available for production (MEI) and concentrations of NFC and concentrate fat and linear effect of Feed MP concentration as significant variables, the effect of MEI intake being most pronounced. Correcting ME intake for feeding level and associative effects of dietary composition further improved the fit of the models. The best-fit equation for PY included linear effect of MEI and linear and quadratic effects of Feed MP intake and dietary concentrations of NFC and concentrate fat. The adjusted root mean squared errors were 0.424kg/d, 0.497kg/d and 15.2g/d for MY and ECM and PY equations, respectively. Cross-validation errors were only marginally greater than calibration errors with almost all error variance due to random variation. The basic structure of the models confirmed the well established concept of diminishing milk and milk component returns with incremental increases in nutrient intake. The results demonstrated that the models predicted production responses of dairy cows to the changes in nutrient supply with a reasonable accuracy. It is concluded that the models can be used in the optimization of dairy cow rations on practical farms to maximize milk income over feed costs.