Nordic dairy cow model Karoline in predicting methane emissions: 1. Model description and sensitivity analysis

Sensitivity analysis of the INRA 2018 feeding system for ruminants by hybrid local and global approaches: Comparing the contribution of dietary input variables to multiple response prediction in dairy cattle

We conducted sensitivity analysis (SA) of the INRA 2018 feeding system for ruminants applied to dairy cows. We evaluated which dietary input variables contribute most to changes in each output variable, considering the potential interactions presence among input variables. A quantitative analysis (one-at-a-time analysis, OAT; i.e., local SA) and a relative comparative analysis (global SA, GSA) through variance-based SA considering potential interactions and non-monotonicity were applied. The 5 likely influential dietary input variables were selected: CP, Gross energy (GE), OM apparent digestibility (OMd), effective degradability of nitrogen assuming a passage rate of 6%/h (ED6_N) and true intestinal digestibility of nitrogen. The sensitivity of 5 selected animal responses (output variables) to input variables was analyzed: DMI, milk protein yield (MPY), energy in methane (ECH4), nitrogen utilization efficiency (NUE), and ratio between urine and total N excretion (UN/TN). Six diets for dairy cattle, reflecting the diversity of diets commonly used in practice, were formulated to meet 95% of the potential milk production (37.5 kg/d) of a multiparous dairy cow at wk 14 of lactation. For each diet, the 5 input variables were randomly sampled around the INRA 2018 feed table values (reference point), and the animal responses around this reference situation were calculated using the rationing software INRAtion®V5. In OAT, the sensitivity of animal responses was quantified by calculating the normalized tangent value at the reference point, and in GSA, the Sobol indices were calculated for relative influence of each input and their interaction. The influence of the 5 key input variables on the 5 main animal responses predicted from the INRA feeding system was consistent across both SA approaches. With the 6 diets, GE and OMd appeared as the main contributors to changes in DMI, MPY, ECH4, and NUE. Crude protein was the main contributor to changes in UN/TN and another major contributor to changes in NUE. When considering OAT, the sensitivity of outputs showed differences depending on diet, more particularly for DMI and MPY. With grass hay-based diets (GH diets), DMI was less sensitive and MPY was more sensitive to variations in input variables than other diets. When considering GSA, interactions between input variables were also noticeable for DMI and MPY; the interactions were high with the GH diets for DMI, and with fresh ryegrass and grass silage diets for MPY. On the other hand, for MPY, the non-GH diets were less sensitive to variations in input variables and the interaction between inputs was higher than with GH diets. In both cases, the interactions were mainly related to energy-related inputs (i.e., OMd and GE). Our results support the hypothesis that MPY, unlike DMI, is more responsive to energy-related factors at high PDI/UFL ratio (i.e., between metabolizable protein and NEL, e.g., GH diets >117 g PDI/UFL), than at lower PDI/UFL ratio. Hence, hybridizing the SA methods can help to interpret the system and facilitate a more precise evaluation thereof, especially GSA, which is amenable to non-monotonic models such as those characterizing complex feeding systems integrating multiple nutritional and animal factors.

Integrating microbial abundance time series with fermentation dynamics of the rumen microbiome via mathematical modelling

The rumen represents a dynamic microbial ecosystem where fermentation metabolites and microbial concentrations change over time in response to dietary changes. The integration of microbial genomic knowledge and dynamic modelling can enhance our system-level understanding of rumen ecosystem's function. However, such an integration between dynamic models and rumen microbiota data is lacking. The objective of this work was to integrate rumen microbiota time series determined by 16S rRNA gene amplicon sequencing into a dynamic modelling framework to link microbial data to the dynamics of the volatile fatty acids (VFA) production during fermentation. For that, we used the theory of state observers to develop a model that estimates the dynamics of VFA from the data of microbial functional proxies associated with the specific production of each VFA. We determined the microbial proxies using CowPi to infer the functional potential of the rumen microbiota and extrapolate their functional modules from KEGG (Kyoto Encyclopedia of Genes and Genomes). The approach was challenged using data from an in vitro RUSITEC experiment and from an in vivo experiment with four cows. The model performance was evaluated by the coefficient of variation of the root mean square error (CRMSE). For the in vitro case study, the mean CVRMSE were 9.8% for acetate, 14% for butyrate and 14.5% for propionate. For the in vivo case study, the mean CVRMSE were 16.4% for acetate, 15.8% for butyrate and 19.8% for propionate. The mean CVRMSE for the VFA molar fractions were 3.1% for acetate, 3.8% for butyrate and 8.9% for propionate. Ours results show the promising application of state observers integrated with microbiota time series data for predicting rumen microbial metabolism.

Approaches for predicting dairy cattle methane emissions: from traditional methods to machine learning

Measuring dairy cattle methane (CH4) emissions using traditional recording technologies is complicated and expensive. Prediction models, which estimate CH4 emissions based on proxy information, provide an accessible alternative. This review covers the different modeling approaches taken in the prediction of dairy cattle CH4 emissions and highlights their individual strengths and limitations. Following the guidelines set out by the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA); Scopus, EBSCO, Web of Science, PubMed and PubAg were each queried for papers with titles that contained search terms related to a population of "Bovine," exposure of "Statistical Analysis or Machine Learning," and outcome of "Methane Emissions". The search was executed in December 2022 with no publication date range set. Eligible papers were those that investigated the prediction of CH4 emissions in dairy cattle via statistical or machine learning (ML) methods and were available in English. 299 papers were returned from the initial search, 55 of which, were eligible for inclusion in the discussion. Data from the 55 papers was synthesized by the CH4 emission prediction approach explored, including mechanistic modeling, empirical modeling, and machine learning. Mechanistic models were found to be highly accurate, yet they require difficult-to-obtain input data, which, if imprecise, can produce misleading results. Empirical models remain more versatile by comparison, yet suffer greatly when applied outside of their original developmental range. The prediction of CH4 emissions on commercial dairy farms can utilize any approach, however, the traits they use must be procurable in a commercial farm setting. Milk fatty acids (MFA) appear to be the most popular commercially accessible trait under investigation, however, MFA-based models have produced ambivalent results and should be consolidated before robust accuracies can be achieved. ML models provide a novel methodology for the prediction of dairy cattle CH4 emissions through a diverse range of advanced algorithms, and can facilitate the combination of heterogenous data types via hybridization or stacking techniques. In addition to this, they also offer the ability to improve dataset complexity through imputation strategies. These opportunities allow ML models to address the limitations faced by traditional prediction approaches, as well as enhance prediction on commercial farms.

Review: Towards the next-generation models of the rumen microbiome for enhancing predictive power and guiding sustainable production strategies

The rumen ecosystem harbours a galaxy of microbes working in syntrophy to carry out a metabolic cascade of hydrolytic and fermentative reactions. This fermentation process allows ruminants to harvest nutrients from a wide range of feedstuff otherwise inaccessible to the host. The interconnection between the ruminant and its rumen microbiota shapes key animal phenotypes such as feed efficiency and methane emissions and suggests the potential of reducing methane emissions and enhancing feed conversion into animal products by manipulating the rumen microbiota. Whilst significant technological progress in omics techniques has increased our knowledge of the rumen microbiota and its genome (microbiome), translating omics knowledge into effective microbial manipulation strategies remains a great challenge. This challenge can be addressed by modelling approaches integrating causality principles and thus going beyond current correlation-based approaches applied to analyse rumen microbial genomic data. However, existing rumen models are not yet adapted to capitalise on microbial genomic information. This gap between the rumen microbiota available omics data and the way microbial metabolism is represented in the existing rumen models needs to be filled to enhance rumen understanding and produce better predictive models with capabilities for guiding nutritional strategies. To fill this gap, the integration of computational biology tools and mathematical modelling frameworks is needed to translate the information of the metabolic potential of the rumen microbes (inferred from their genomes) into a mathematical object. In this paper, we aim to discuss the potential use of two modelling approaches for the integration of microbial genomic information into dynamic models. The first modelling approach explores the theory of state observers to integrate microbial time series data into rumen fermentation models. The second approach is based on the genome-scale network reconstructions of rumen microbes. For a given microorganism, the network reconstruction produces a stoichiometry matrix of the metabolism. This matrix is the core of the so-called genome-scale metabolic models which can be exploited by a plethora of methods comprised within the constraint-based reconstruction and analysis approaches. We will discuss how these methods can be used to produce the next-generation models of the rumen microbiome.

Comparison of Molly and Karoline models to predict methane production in growing and dairy cattle

Numerous empirical and mechanistic models predicting methane (CH₄) production are available. The aim of this work was to evaluate the Molly cow model and the Nordic cow model Karoline in predicting CH₄ production in cattle using a data set consisting of 267 treatment means from 55 respiration chamber studies. The dietary and animal characteristics used for the model evaluation represent the range of diets fed to dairy and growing cattle. Feedlot diets and diets containing additives mitigating CH₄ production were not included in the data set. The relationships between observed and predicted CH₄ (pCH₄) were assessed by regression analysis using fixed and mixed model analysis. Residual analysis was conducted to evaluate which dietary factors were related to prediction errors. The fixed model analysis showed that the Molly predictions were related to the observed data (± standard error) as CH₄ (g/d) = 0.94 (±0.022) × pCH₄ (g/d) + 31 (±6.9) [root mean squared prediction error (RMSPE) = 45.0 g/d (14.9% of observed mean), concordance correlation coefficient (CCC) = 0.925]. The corresponding equation for the Karoline model was CH₄ (g/d) = CH₄ (g/d) = 0.98 (±0.019) × pCH₄ (g/d) + 7.0 (±6.0) [RMSPE = 35.0 g/d (11.6%), CCC = 0.953]. Proportions of mean squared prediction error attributable to mean and linear bias and random error were 10.6, 2.2, and 87.2% for the Molly model, and 1.3, 0.3, and 98.6% for the Karoline model, respectively. Mean and linear bias were significant for the Molly model but not for the Karoline model. With the mixed model regression analysis RMSPE adjusted for random study effects were 10.9 and 7.9% for the Molly model and the Karoline model, respectively. The residuals of CH₄ predictions were more strongly related to factors associated with CH₄ production (feeding level, digestibility, fat concentrations) with the Molly model compared with the Karoline model. Especially large mean (underprediction) and linear bias (overprediction of low digestibility diets relative to high digestibility diets) contributed to the prediction error of CH₄ yield with the Molly model. It was concluded that both models could be used for prediction of CH₄ production in cattle, but Karoline was more accurate and precise based on smaller RMSPE, mean bias, and slope bias, and greater CCC. The importance of accurate input data of key variables affecting diet digestibility is emphasized.

Challenges and opportunities to capture dietary effects in on-farm greenhouse gas emissions models of ruminant systems

This paper reviews existing on-farm GHG accounting models for dairy cattle systems and their ability to capture the effect of dietary strategies in GHG abatement. The focus is on methane (CH₄) emissions from enteric and manure (animal excreta) sources and nitrous oxide (N₂O) emissions from animal excreta. We identified three generic modelling approaches, based on the degree to which models capture diet-related characteristics: from 'none' (Type 1) to 'some' by combining key diet parameters with emission factors (EF) (Type 2) to 'many' by using process-based modelling (Type 3). Most of the selected on-farm GHG models have adopted a Type 2 approach, but a few hybrid Type 2 / Type 3 approaches have been developed recently that combine empirical modelling (through the use of CH₄ and/or N₂O emission factors; EF) and process-based modelling (mostly through rumen and whole tract fermentation and digestion). Empirical models comprising key dietary inputs (i.e., dry matter intake and organic matter digestibility) can predict CH₄ and N₂O emissions with reasonable accuracy. However, the impact of GHG mitigation strategies often needs to be assessed in a more integrated way, and Type 1 and Type 2 models frequently lack the biological foundation to do this. Only Type 3 models represent underlying mechanisms such as ruminal and total-tract digestive processes and excreta composition that can capture dietary effects on GHG emissions in a more biological manner. Overall, the better a model can simulate rumen function, the greater the opportunity to include diet characteristics in addition to commonly used variables, and thus the greater the opportunity to capture dietary mitigation strategies. The value of capturing the effect of additional animal feed characteristics on the prediction of on-farm GHG emissions needs to be carefully balanced against gains in accuracy, the need for additional input and activity data, and the variability encountered on-farm.

Development of an equation to estimate the enteric methane emissions from Holstein dairy cows in Canada

The aim of this study was to use dietary factors, including the type of fats, and animal characteristics, to predict enteric methane (CH4) emissions from dairy cows under Canadian conditions. For this purpose, 193 individual observations from six different trials assessing the impact of dietary modification on enteric CH4 production were analyzed. Animal [milk yield (MY), milk fat content, milk protein content, days in milk, body weight (BW), and dry matter intake (DMI)] and dietary variables [organic matter, crude protein, neutral detergent fiber (NDF), acid detergent fiber (ADF), starch, ether extract (EE), rumen-inert fat, and unprotected fat (EE – rumen-inert fat)] were tested. A 5-fold cross validation was used to obtain the following equation: CH4 (g d−1) = −1260.4 + 1.9 × MY (kg d−1) + 62.8 × milk fat (%) –18.4 × milk protein (%) + 11.0 × DMI (kg d−1) + 0.3 × BW (kg) + 58.3 × NDF (% of DM) − 0.8 × NDF2 (% of DM) + 1.9 × starch (% of DM) − 2.5 × EE – rumen-inert fat (% of DM). The mean estimate from the proposed equation (474 g CH4 cow−1 d−1; r = 0.83, RMSE = 40.0) was close to the observed mean emission (476 g CH4 cow−1 d−1). The proposed model has a higher precision to predict CH4 emission from cows fed typical Canadian diets than other models, and it can be used to evaluate CH4 mitigation strategies.

Effects of different barley and oat varieties on methane production, digestibility, and fermentation pattern in vitro

The objective of this in vitro study was to determine the effects of different barley and oat varieties on CH₄ production, digestibility, and rumen fermentation patterns in dairy cows. Our hypothesis was that oat-based diets would decrease CH₄ production compared with barley-based diets, and that CH₄ production would differ between varieties within grain species. To evaluate this hypothesis, we conducted an in vitro experiment using a fully automated gas production technique, in which the total gas volume was automatically recorded by the system. The experiment consisted of triplicate 48-h incubations with 16 treatments, including 8 different varieties of each grain. The grain varieties were investigated as a mix with an early-cut grass silage (1:1 ratio of grain to silage on a dry matter basis) and mixed with buffered rumen fluid. We estimated predicted in vivo total gas production and CH₄ production by applying a set of models to the gas production data obtained by the in vitro system. We also evaluated in vitro digestibility and fermentation characteristics. The variety of grain species did not affect total gas production, CH₄ production, or fermentation patterns in vitro. However, in vitro-determined digestibility and pH were affected by variety of grain species. Grain species affected total gas and CH₄ production: compared with barley-based diets, oat-based diets decreased total gas production and CH₄ production by 8.2 and 8.9%, respectively, relative to dry matter intake. Grain species did not affect CH₄ production relative to in vitro true dry matter digestibility. Oat-based diets decreased digestibility and total volatile fatty acid production, and maintained a higher pH at 48 h of incubation compared with barley-based diets. Grain species did not affect fermentation patterns, except for decreased molar proportions of valerate with oat-based diets. These results suggest that replacing barley with oats in dairy cow diets could decrease enteric CH₄ production.

A parsimonious software sensor for estimating the individual dynamic pattern of methane emissions from cattle

Large efforts have been deployed in developing methods to estimate methane emissions from cattle. For large scale applications, accurate and inexpensive methane predictors are required. Within a livestock precision farming context, the objective of this work was to integrate real-time data on animal feeding behaviour with an in silico model for predicting the individual dynamic pattern of methane emission in cattle. The integration of real-time data with a mathematical model to predict variables that are not directly measured constitutes a software sensor. We developed a dynamic parsimonious grey-box model that uses as predictor variables either dry matter intake (DMI) or the intake time (IT). The model is described by ordinary differential equations.Model building was supported by experimental data of methane emissions from respiration chambers. The data set comes from a study with finishing beef steers (cross-bred Charolais and purebred Luing finishing). Dry matter intake and IT were recorded using feed bins. For research purposes, in this work, our software sensor operated off-line. That is, the predictor variables (DMI, IT) were extracted from the recorded data (rather than from an on-line sensor). A total of 37 individual dynamic patterns of methane production were analyzed. Model performance was assessed by concordance analysis between the predicted methane output and the methane measured in respiration chambers. The model predictors DMI and IT performed similarly with a Lin's concordance correlation coefficient (CCC) of 0.78 on average. When predicting the daily methane production, the CCC was 0.99 for both DMI and IT predictors. Consequently, on the basis of concordance analysis, our model performs very well compared with reported literature results for methane proxies and predictive models. As IT measurements are easier to obtain than DMI measurements, this study suggests that a software sensor that integrates our in silico model with a real-time sensor providing accurate IT measurements is a viable solution for predicting methane output in a large scale context.

Addressing Global Ruminant Agricultural Challenges Through Understanding the Rumen Microbiome: Past, Present, and Future

The rumen is a complex ecosystem composed of anaerobic bacteria, protozoa, fungi, methanogenic archaea and phages. These microbes interact closely to breakdown plant material that cannot be digested by humans, whilst providing metabolic energy to the host and, in the case of archaea, producing methane. Consequently, ruminants produce meat and milk, which are rich in high-quality protein, vitamins and minerals, and therefore contribute to food security. As the world population is predicted to reach approximately 9.7 billion by 2050, an increase in ruminant production to satisfy global protein demand is necessary, despite limited land availability, and whilst ensuring environmental impact is minimized. Although challenging, these goals can be met, but depend on our understanding of the rumen microbiome. Attempts to manipulate the rumen microbiome to benefit global agricultural challenges have been ongoing for decades with limited success, mostly due to the lack of a detailed understanding of this microbiome and our limited ability to culture most of these microbes outside the rumen. The potential to manipulate the rumen microbiome and meet global livestock challenges through animal breeding and introduction of dietary interventions during early life have recently emerged as promising new technologies. Our inability to phenotype ruminants in a high-throughput manner has also hampered progress, although the recent increase in “omic” data may allow further development of mathematical models and rumen microbial gene biomarkers as proxies. Advances in computational tools, high-throughput sequencing technologies and cultivation-independent “omics” approaches continue to revolutionize our understanding of the rumen microbiome. This will ultimately provide the knowledge framework needed to solve current and future ruminant livestock challenges.

Prediction of enteric methane production, yield, and intensity in dairy cattle using an intercontinental database

Enteric methane (CH₄ ) production from cattle contributes to global greenhouse gas emissions. Measurement of enteric CH₄ is complex, expensive, and impractical at large scales; therefore, models are commonly used to predict CH₄ production. However, building robust prediction models requires extensive data from animals under different management systems worldwide. The objectives of this study were to (1) collate a global database of enteric CH₄ production from individual lactating dairy cattle; (2) determine the availability of key variables for predicting enteric CH₄ production (g/day per cow), yield [g/kg dry matter intake (DMI)], and intensity (g/kg energy corrected milk) and their respective relationships; (3) develop intercontinental and regional models and cross-validate their performance; and (4) assess the trade-off between availability of on-farm inputs and CH₄ prediction accuracy. The intercontinental database covered Europe (EU), the United States (US), and Australia (AU). A sequential approach was taken by incrementally adding key variables to develop models with increasing complexity. Methane emissions were predicted by fitting linear mixed models. Within model categories, an intercontinental model with the most available independent variables performed best with root mean square prediction error (RMSPE) as a percentage of mean observed value of 16.6%, 14.7%, and 19.8% for intercontinental, EU, and United States regions, respectively. Less complex models requiring only DMI had predictive ability comparable to complex models. Enteric CH₄ production, yield, and intensity prediction models developed on an intercontinental basis had similar performance across regions, however, intercepts and slopes were different with implications for prediction. Revised CH₄ emission conversion factors for specific regions are required to improve CH₄ production estimates in national inventories. In conclusion, information on DMI is required for good prediction, and other factors such as dietary neutral detergent fiber (NDF) concentration, improve the prediction. For enteric CH₄ yield and intensity prediction, information on milk yield and composition is required for better estimation.

Effects of ruminal digesta retention time on methane emissions: a modelling approach

The reasons for among-animal variations in methane (CH4) emissions are not fully understood. There is experimental evidence that ruminal digesta mean retention time (MRT) can affect CH4 emissions. The objective of the present study was to evaluate the contribution of among-animal variations in MRT on CH4 emissions and nutrient supply for dairy cow (default MRT = 34 h) and sheep (default MRT = 41 h), using the mechanistic Nordic dairy cow model Karoline. The simulations (n = 100) were made for a cow (bodyweight 600 kg) and for a sheep (bodyweight 60 kg) eating 20 kg and 1.0 kg DM/day, respectively. The diet for the dairy cow consisted of grass silage, barley and rapeseed meal (60 : 30 : 10 on a DM basis; crude protein 156 g/kg DM, neutral detergent fibre 450 g/kg DM) and the sheep diet was grass alone. Normal distribution of MRT values was assumed. Variability (coefficient of variation (CV) = 0.086) on default MRT was introduced by random-number generator of Excel. Intake, diet composition and digestion kinetic parameters were constant in all simulations, only ruminal MRT variables were changed in each simulation. Predicted CH4 emission increased with increased MRT for dairy cow (range from 407 to 488 g/day) and sheep (from 25.0 to 29.2 g/day). Increases in predicted CH4 emissions were partly associated with enhanced organic matter (OM) digestibility in dairy cow (from 0.715 to 0.758) and sheep (from 0.731 to 0.773). Greater CH4 emissions per kilogram digested OM with increased MRT were mainly related to reduced efficiency of microbial cell synthesis in the rumen both for dairy cows (22.8 ± 0.91 g N/kg OM truly digested; CV = 0.040) and for sheep (20.7 ± 0.92 g N/kg OM truly digested; CV = 0.044). Predicted CH4 yield was 20% and 17% greater in dairy cow and sheep, respectively, with the short (n = 10) compared with the long (n = 10) ruminal digesta MRT. Linear regression indicated that CH4 emissions increased by 0.37 (dairy cow) and 0.33 (sheep) g/kg DM intake per 1 h increase in ruminal digesta MRT. It is concluded that among-animal variation in MRT can markedly contribute to among-animal variation in CH4 emissions from ruminants.