The nordic dairy cow model, karoline - evaluation

This study presents evaluation results of the model Karoline, which is described in another study (see Danfær et al., Chapter 32, this volume). The validations comprise behavioural analyses, testing against experimental data and comparison of Karoline with the Cornell Net Carbohydrate and Protein System (CNCPS) model. The model behaviour is evaluated from simulated responses to stage of lactation, feed intake, dietary protein level, dietary fat level and ratio of concentrate to forage in the diet.The response parameters are milk yield (MY), milk composition and live weight (LW) gain. In most cases, Karoline behaves similarly to what is generally observed in vivo. The digestion part of Karoline is tested against 61 treatment means from studies with cannulated dairy cows and growing cattle carried out in the Nordic countries. Simulated (X) and observed (Y) parameter values are compared in meta-analyses using both an unadjusted regression and a mixed model regression procedure with the individual experiment as a random factor. The parameters compared are duodenal flows of organic matter (OM), neutral detergent fibre (NDF), total crude protein (CP), dietary CP and microbial CP as well as faecal flows of OM, NDF and CP. In the mixed model analyses, the fixed regression coefficient is between 0.86 and 1.16, the fixed intercept is between -0.30 and 0.15 kg/day and the overall R2 is between 0.95 and 1.00. The full Karoline model is tested against 142 treatment means from production studies with lactating cows carried out in the Nordic countries. Unadjusted as well as mixed model regression analyses compare simulated (X) and observed (Y) parameter values:MY, energy-corrected milk (ECM) yield, milk fat yield and milk protein yield.The fixed regression coefficient is between 0.74 and 0.83, the fixed intercept is between 0.20 kg/day (milk protein) and 6.1 kg/day (milk), and the overall R2 is between 0.87 and 0.94. Karoline is further validated by a comparison with the CNCPS model. Both models are tested against a Nordic data-set of 75 treatment means from dairy cow production experiments. A deviation index based on mixed model regression analyses is calculated for both models. In this analysis, the deviation index is 0.19 and 0.07 for CNCPS and Karoline, respectively. For a perfect model, i.e. identical simulated and observed parameter values, the deviation index equals zero. It is concluded that the Karoline model is a useful tool to evaluate feed rations for dairy cows by predictions of nutrient digestibility and MY. However, predictions of the milk composition, i.e. fat and protein contents, need to be improved. Possible reasons for problems in the model are discussed