Rumen stoichiometric models and their contribution and challenges in predicting enteric methane production

Abstract

Fermentation in the rumen is a complex process involving microbial activities and degradable dietary components. Therefore, representation of this process using mathematical models is also complex. Besides degradation of dietary components and microbial growth, fermentation stoichiometry needs to be known in order to evaluate specific dietary components for type of volatile fatty acid (VFA), H2 and CH4 produced during rumen fermentation. The objectives were to evaluate extant VFA stoichiometric models for their capacity to predict VFA molar proportion and CH4 using independent data sources. Two data sets were organized from the published literature. The first contained 141 treatments of rumen digestion studies with lactating dairy cows collected from 43 published experiments. The second data set contained 18 treatments from 8 studies. The experiments reported information on diet composition, true rumen substrate digestibility, molar proportion of VFA and enteric CH4 production (the latter only for data set 2). Model comparison was based on mean square prediction error (MSPE), concordance correlation coefficient and regression analysis. In general, models had different prediction performances with respect to the type of VFA in rumen fluid with root MSPE (RMSPE, % observed mean) values from 5.2 to 43.2. Among the 4 models evaluated, that of Murphy et al. (1982, MUR) had the highest RMSPE value for propionate (25.7%) with 19.6% MSPE being random error. The model of Bannink et al. (2006, BAN) had the lowest RMSPE (10.7%) for butyrate with 97.8% MSPE being random error. Similarly, the model of Nozière et al. (2010, NOZ) had the lowest RMSPE (5.2%) for acetate with 83.0% MSPE being random error. Variations among stoichiometric models in predicting VFA molar proportions affected estimated CH4 production. Comparison of predicted versus measured CH4 production showed that BAN had the lowest RMSPE (9.8%) with only 18.1% of MSPE error due to deviation of the regression slope from unity (ER). The RMSPE was 11.2 and 12.2% for NOZ and MUR, respectively, with ER being 44.3 and 21.4%, respectively. Prediction of CH4 production using Sveinbjörnsson et al. (2006) model had the highest RMSPE (16.7%) with 41.2% MSPE being ER. Results indicate that there were unexplained variations in model predicted VFA molar proportions versus observed values. The variation among stoichiometric models in predicting VFA production has a major influence on the accuracy of estimated enteric CH4 production. Currently, CH4 inventory is usually based on IPCC Tier 2 approach which, compared to BAN, NOZ and MUR had a higher prediction error in estimating CH4 emissions. The IPCC Tier 2 approach had an RMSPE of 16.4% of observed mean with 56.9% of the error due to ER indicating proportional bias due to inadequate representation of relationships. There may be a need for more mechanistic approaches that consider nutritional and microbial factors rather than empirical models that relate VFA molar proportions to nutritional factors. Based on our analysis, mechanistic models, particularly BAN, may be preferred for CH4 inventory or mitigation purposes. Although current mechanistic models have a higher prediction accuracy and a demonstrably more adequate representation of relationships compared with the widely used IPCC Tier 2 approach, the prediction accuracy of current models requires further improvement and they should be used with care for regulatory purposes either to create enteric CH4 mitigation strategies or document impacts of mitigation strategies.This paper is part of the special issue entitled: Greenhouse Gases in Animal Agriculture- Finding a Balance between Food and Emissions, Guest Edited by T.A. McAllister, Section Guest Editors; K.A. Beauchemin, X. Hao, S. McGinn and Editor for Animal Feed Science and Technology, P.H. Robinson.