Evaluation of the precision and accuracy of models of enteric methane emissions from ruminants using Bayesian inference
Authors:
Preprint
Abstract
Reducing enteric methane (CH4) emissions from dairy cattle is essential to limit 18 global warming to the target of 1.5° C increase. Empirical models taken from the 19 literature are used to assess the impact of CH4 mitigation strategies. Numerous models developed are usually compared by quantifying the prediction error. However, the uncertainty evaluation of CH4 predictions reliability is rarely addressed. This work fills this gap by comparing the predictive accuracy and precision of ten total enteric CH4 production models, incorporating uncertainty using Bayesian inference. We assumed that integrating the uncertainty in model comparison can be useful to select models suited for specific conditions. To assess the accuracy and precision of models, three analyses were conducted: A1) Assessment of the probability of models to predict within target areas representative of several accuracy levels around the CH4 observed, A2) Assessment of the relationship between bias and standard deviation and A3) Assessment of the probability of models to predict lipid (EE)-level based dietary scenarios. These analyses were conducted using two restricted datasets RD1 (107 records) and RD2 (215 records), in particular RD2 explored a larger diversity of dietary treatments than RD1. Results indicated that a model associating dry matter intake (DMI), EE and neutral detergent fiber (NDF) digestibility was the most recommendable, as it was among the most accurate models in A1 and showed the best trade-off between accuracy and precision in A2, for both RD1 and RD2. Moreover, using gross energy intake alone highlighted a higher accuracy than using DMI alone in A1 and A2, with RD2. Including dietary composition variables (NDF and/or EE) in addition to DMI did not always improve accuracy, which was the case with RD1 but not with RD2. Among models using DMI and dietary composition variables, using NDF with DMI is recommended for a high-targeted accuracy, otherwise using DMI alone is recommended. Milk yield (MY) alone could be an alternative to DMI, as A1 highlighted a higher accuracy at 8% around the CH4 observed. Also, the most complex model had the highest accuracy in A1 and A2, raising question about the trade-off between accuracy and complexity. A model using EE and NDF performed poorly, confirming the importance of DMI for obtaining accurate predictions. Finally, A3 indicated that model predictive ability was different depending on the EE-level of the diet. Using DMI alone better predicted the low EE-level scenario, while MY alone better predicted medium and high EE scenarios, despite large over- and underestimation of observed CH4, respectively. The probabilistic framework presented was highlighted as relevant to provide recommendations on which models to use. Also, extending the A3 approach to reference diets (INRAE, USDA, FAO) could benefit future work in the field.Keywords
All Related Versions
This publication has 37 references indexed in Scilit:
- A global dataset of enteric methane mitigation experiments with lactating and non-lactating dairy cows conducted from 1963 to 2022Data in Brief, 2023
- Review: Fifty years of research on rumen methanogenesis: lessons learned and future challenges for mitigationAnimal, 2020
- Bayesian mechanistic modeling of thermodynamically controlled volatile fatty acid, hydrogen and methane production in the bovine rumenJournal of Theoretical Biology, 2019
- Evaluation of the performance of existing mathematical models predicting enteric methane emissions from ruminants: Animal categories and dietary mitigation strategiesAnimal Feed Science and Technology, 2019
- Prediction and evaluation of enteric methane emissions from lactating dairy cows using different levels of covariate informationAnimal Production Science, 2016
- Effects of nitrogen fertilisation rate and maturity of grass silage on methane emission by lactating dairy cowsAnimal, 2016
- A universal equation to predict methane production of forage-fed cattle in AustraliaAnimal Production Science, 2015
- Anthropogenic and Natural Radiative Forcing2014
- Enteric methane production, digestibility and rumen fermentation in dairy cows fed different forages with and without rapeseed fat supplementationAnimal Feed Science and Technology, 2013
- Improving the efficiency of energy utilisation in cattleAnimal Production Science, 2010