Comparison and evaluation of mechanistic rumen models.

Mechanistic rumen models of Baldwin (1995), Danfaer (1990) and Dijkstra et al. (1992) were compared on identical inputs that were derived from trials with lactating dairy cows fed on grass herbage. Consistent differences were detected between models and between predicted and observed outputs. None of the models seemed to predict all nutrient flows best. The models particularly differed in the representation of microbial metabolism: degradation of insoluble substrate, fermentation of substrate into volatile fatty acids, and incorporation of substrate into microbial matter. Differences amongst models in the prediction of these processes compensated for each other and consequently all models predicted the duodenal flow of non-NH3 N, microbial N and organic matter reasonably well. Large differences remained in the prediction of individual nutrient flows, however, and it was stressed that in order to enhance prediction of the profile of nutrient flows, the mechanisms of microbial metabolism need to be tested on their ability to describe the intraruminal transactions. However, this requires more-detailed information on individual nutrient flows and on the microbial or non-microbial origin of duodenal contents. Parameter inputs for physical and chemical feed properties were identified that are improperly defined in extant models or susceptible to error. The description of these feed characteristics needs to be developed further and become identifiable for a wide range of dietary conditions.