Nordic dairy cow model Karoline in predicting methane emissions: 2. Model evaluation

• Performance of Karoline model predicting methane emissions was evaluated.
• Evaluation was based on a large dataset from respiration chamber studies.
• Karoline model predicted methane emissions accurately and precisely.
• Residuals were not related to most important factors influencing methane emissions.

Models are widely used to predict methane (CH4) emissions, and can be used to develop mitigation options and policies. The objective of the current study was to evaluate the performance of the Nordic dairy cow model Karoline in predicting CH4 emissions. Karoline is a dynamic, deterministic and mechanistic simulation model that describes the digestion and metabolism of nutrients, and production in a dairy cow. The model was evaluated against observed data from studies reporting CH4 emissions from respiration chamber studies. The dataset included a total of 184 treatment means from 31 published papers. The dietary and animal characteristics used for the model evaluation represent the typical range of diets fed to dairy cattle. When analyzed with a fixed model regression, there was a good relationship between predicted and observed CH4 emissions measured from respiration chamber studies with a small root mean square error of prediction (R2=0.93, RMSPE=10.1% of the observed mean). The mean bias was small (1.9%) but statistically significant, and there was no slope bias. Most of the error was due to random variation (96.4%), whereas the contributions of mean and slope bias were small. By considering study as the random effect in the model (mixed model regression analysis), the fit improved to R2= 0.98 and RMSPE decreased to 6.1% of the observed mean. The influence of some input variables such as total dry matter intake, proportion of concentrate, dietary concentrations of crude protein, neutral detergent fiber and ether extract, and organic digestibility (OMD) on the residuals (observed–predicted) of CH4 emissions were not significant. The residuals of both CH4 emissions and OMD were significantly related to each other, indicating the Karoline model requires accurate estimates of digestion kinetic parameters as input variables. When the laboratory was used as a class variable in the model, the residuals of CH4 emissions were significantly different both between the laboratories and also between experiments within individual laboratories. It is concluded that the Nordic dairy cow model Karoline is a useful tool in predicting CH4 emissions and understanding the system behavior. The model can also be used in developing mitigation strategies for the national inventories of CH4 emissions.