A static model for estimating energy content of compound feeds in a dynamic feed evaluation system

The objective of the study was to develop a static empirical model for the estimation of net energy content of compound feeds in a dynamic feeding system using net energy for lactation at 20 kg of dry matter intake/d (NEL20) values calculated by the Nordic Feed Evaluation System (NorFor) model. In the NorFor system, NEL20 is a standardized value used to describe net energy content of feeds. The static model would allow prediction of the net energy value of compound feeds without access to the input data needed for the dynamic models. Our hypothesis was that NEL20 values of compound feeds can be predicted using organic matter digestibility (in vitro) and chemical components of the compound feeds as input variables. For this, 75 compound feeds and their 108 associated ingredients were collected across Scandinavia for model development. The proposed best model for prediction of compound feed NEL20 included crude fat, neutral detergent fiber, digestible organic matter measured in vitro, and crude protein (urea corrected) as independent variables. Lack of additivity of chemical components between values analyzed directly in the compound feed and values calculated by the weighted sum of ingredients was detected as the main source of error in the model, emphasizing the importance of accurate chemical analysis and sampling practices. Results from practical use of the model show that it may be a valuable tool that could be used by several actors in the feeding sector using the NorFor system. Feed manufacturers could use it to monitor the net energy content in their final product, and farmers could use it to check the net energy content of the purchased compound feed. However, validation of this model against an independent set of samples is lacking in this study and its prediction performance should be further evaluated. The model will need recalibration if the feed parameters used in the dynamic model for the estimation of reference values change, as this would not be reflected in the predicted values of the created model.