Effect of animal and diet parameters on methane emissions for pasture-fed cattle

Measurements from young livestock

The objective was to measure methane emissions of young cattle fed on fresh-cut pasture. Three groups of growing young (Hereford × Friesian) cattle aged 8, 12, and 16 months (the group numbers, sex and liveweight are described in Table 1) were transported from Aorangi farm (AgResearch, Manawatu, New Zealand) to the New Zealand Ruminant Methane Measurement Centre. The animals were adapted over 2–3 weeks to cut pasture, handling, and the indoor environment in covered yards or calf pens. The three groups of animals were adapted to individual confinement in metabolic crates for 2 days before placement in respiration chambers. Methane emissions and DMI were determined in the respiration chamber facility consisting of four chambers (Pinares-Patiño et al. 2018). The pasture was cut daily approximately 11:00 hours, divided into two meals, and stored at 4°C until offered to the animals in the afternoon and the next morning. A daily subsample of the cut grass was collected and analysed as described below. The feed was offered ad libitum and replaced with a fresh allocation twice a day at approximately 08:30 hours and 15:30 hours. The refusals were removed before fresh feed was offered and a subsample was taken for DM content determination. Fresh water was always available.

In a separate experiment, 24 female Hereford × Friesian calves were collected from a commercial farm at 4 days of age, transported to the New Zealand Ruminant Methane Measurement Centre, and randomly allocated into two groups of 12. All animals were housed in a calf shed on a bedding of bark chips. Transition milk was fed to ease the introduction of reconstituted calf milk replacer (CMR). One group of calves received a limited amount of CMR (4 L day⁻¹) and had unlimited access to concentrate starter pellets (SealesWinslow Calf 20% Pellets). The other group had a higher allowance of CMR (up to 8 L day⁻¹) and access to ryegrass-dominant pasture in a paddock adjacent to the calf shed and no concentrate supplementation. Pasture hay and fresh water were available ad libitum to both groups. The CMR and starter pellets were fed using a CalfSMART automated milk and pellet feeding system (Zeddy, Palmerston North, New Zealand). At 2.5 month of age, methane measurements were conducted in respiration chambers. These calves were transferred to the respiration chambers directly from the calf shed. While in chambers, the calves received milk together with either starter pellets or fresh pasture as described previously. Methane emissions of these animals were re-measured post-weaning at 4 months of age while being fed cut pasture as described above.

Effect of concentrate supplementation and three levels of intake

For this trial, a high-starch concentrate feed (multi-feed nuts; Sharpes Stock Feeds, Carterton, New Zealand) was used and fed at approximately 30% of the DM. The pasture was offered ad libiturm. The supplement was offered to three groups of cattle that had a different level of DMI. The first group consisted of 18-month-old Hereford × Friesian cattle representing a low-intake group with an average DMI of 6 kg day⁻¹. The second group (medium-intake group) consisted of late-lactating Jersey × Holstein cows with an average DMI of 12 kg day⁻¹, and the high-intake group was represented by early lactating Holstein dairy cows that consumed approximately 20 kg day⁻¹. Relative to the bodyweight, these intakes represented 1.7%, 2.9% and 3.3% of bodyweight for the low-, medium- and high-intake groups. The concentrate was offered as two equal meals in the morning and afternoon before new pasture was allocated. Lactating dairy cattle were milked twice a day by using mobile milking units (DeLaval, MacDougalls Agri Services, Palmerston North, New Zealand). The pasture was cut at Aorangi farm daily and delivered around midday to the trial site, where it was offered ad libitum and replaced with a fresh allocation in the afternoon and the next morning. Refusals were removed for DM determination before fresh feed was offered and fresh water was always available. The animals were weighted on arrival and after methane measurements. The weight at the end of the measurements was recorded (Table 1).

Methane and DMI measurements

The gases methane, hydrogen and carbon dioxide were quantified in four open-circuit respiration chambers. Each chamber is 15.4 m³ (3.5 m long × 2 m wide × 2.2 m high) with an airflow rate set at ~0.5–2.0 m³ min⁻¹ (depending on animal size), which was continuously monitored by measuring differential pressure with a Venturi flowmeter. The temperature inside the respiration chambers was, on average, approximately 20°C and the relative humidity was, on average, approximately 79%. All gases were measured at ~2.8 min intervals by using a 4900C Continuous Emission analyser (Servomex Group, East Sussex, UK), and the daily production of each gas was calculated from the difference between concentration flowing in and that flowing out of the chamber, corrected for temperature, pressure, and humidity (Pinares-Patiño et al. 2018). Respiration chambers were opened twice daily (~20 min each time for growing cattle and ~ 45 min for lactating dairy cattle) for cleaning, feed refusal collection, feeding, and milking for lactating cows. No measurements were performed during the period when chambers were opened, and missing data were interpolated by taking the average of the last 12 values (~45 min) before the doors were opened.

Similar to the adaptation in crates, refusals were also collected and analysed during the time in the respiration chambers. Subsamples of the refusals and the feed offered were dried at 105°C for 24 h to determine the dry-matter (DM) content. Daily DMI of the animals were determined from the difference between the DM offered and refused. Feeding level as a multiple of maintenance metabolisable energy requirements was calculated using the Australian Feeding Standard (Freer et al. 2007).

Additional samples from each pasture delivery were collected, dried at 65°C in a ventilated oven for 48 h and ground to pass a 1 mm sieve. Diet samples were analysed for crude nutrient composition by using near-infrared spectroscopy by Massey Nutrition Laboratories (Palmerston North, New Zealand), as described by Corson et al. (1999) (Table 2).

Statistical analyses

The statistical analysis was performed using R (ver. 4.13, R Core Team 2021). In the exploratory data analysis step we used correlations to determine which numerical variables showed signs of a relationship with the response variable of interest, methane (CH₄) output.

In the modelling step, we fitted linear mixed models with the function lmer from the R package ‘lme4’, ver. 1.1-30 (Bates et al. 2015; Kuznetsova et al. 2017). The numerical variables evaluated as predictors for the linear mixed-effects model, based on the correlation analysis, were individual diet chemical components, feeding level, bodyweight, DMI and age. In addition, the factors breed, physiological state, sex and supplementation level were also considered as fixed effects. The animal ID was used as a random effect to account for variation among the animals.

Note that to make the variable bodyweight more comparable to the other variables in the model, we converted it to a more common scale. This was achieved by redefining bodyweight such that a unit change in weight is 10 kg instead of just one. The rescaled bodyweight variable is named BW.

Initially, we fitted a full model with all these variables as well as respiration-chamber ID and measurement day as fixed effects. We found that there were no differences in measurements among respiration chambers, on average, but the measurement day did have an effect. For the development of the final predictive model for CH₄ production of pasture-fed cattle, chamber ID was thus removed from the dataset and measurement day included in the model as a random effect.

The model selection was a two-step process. In the first step, we compared model of all possible subsets of the fixed-effect components of interest using the function dredge from the R package ‘MuMIn’, ver. 1.46 (Barton 2022). The models were compared using the corrected Akaike’s information criterion (AICc), where a lower value indicates a better model. The 10 models with the lowest AICc values are shown in Table 3. The model with the lowest AICc included the diet/supplement parameter, an indicator variable representing the animals that were supplemented with 30% concentrate. This model was excluded as the number of supplemented animals were too low to make it useful for prediction.

In the second step, we calculated the variance inflation factor (VIF) scores, using the VIF function from the ‘car’ R package, ver. 3.1 (Fox and Weisberg 2019), of the predictors in each of the listed models. If a component has a VIF value greater than 10, then this indicates that the model has a problem with collinearity. We then selected the highest-ranking model (with the lowest AICc value) that had no collinear predictors.

Inspection of the residual plot for the final model of the CH₄ production predictions highlighted a degree of heteroscedasticity, but the violation was not severe enough to warrant a log-transformation of the data. The emmeans function from the R package ‘emmeans’ 1.7.5 (Barton 2022; Russell 2022) was used to compute predicted values on the basis of the final model.

Note that a linear mixed model with CH₄ yield as the response was ruled out, given that in the exploratory data-analysis step, we found that none of the possible numerical predictors was significantly correlated with CH₄ yield.

Exploratory analysis of the combined dataset

The results of the animal trials described above were analysed with data generated previously (Jonker et al. 2016, 2017). These were two previous datasets of CH₄ emissions from growing beef cattle and non-lactating and lactating dairy cattle fed on 100% fresh-cut pasture, totalling 191 measurements over 16 different pasture-feeding periods. The new data presented here added another 81 measurements to the combined dataset. The distribution of the CH₄ production (g day⁻¹) data of the previous and the new dataset is highly correlated to DMI (Fig. 1). The new dataset extended the range mostly to the lower DMI ranges, but also added some high-DMI data.

Fig. 1.

Relationship between methane production (pM) and dry-matter intake (DMI) for the existing dataset reported previously (Jonker et al. 2016, 2017) and the new dataset reported here.

The exploratory data analysis showed that the highest correlation of CH₄ production was observed for DMI (r = 0.956), followed by feeding level, expressed as multiples of the maintenance energy requirements (r = 0.852), BW of the animal (r = 0.773) and age (r = 0.735) (Fig. 2). However, some of these variables were highly intercorrelated, with pair-wise r values ranging from 0.490 to 0.913.

Fig. 2.

Pearson correlation matrix of animal and pasture parameters. CH₄, methane production; yM, methane yield; dm, dry-matter content; ca, crude-ash content; cp, crude-protein content; cl, crude-lipid content; ss, soluble sugar content; ndf, neutral detergent fibre content; adf, acid detergent fibre content; me, metabolisable energy content; weight, bodyweight; fl, feeding level as multiples relative to maintenance energy requirements; dmi, dry-matter intake; age, animal age; bw, body weight.

The correlation of all diet composition parameters with CH₄ production were low. Methane production was only slightly positively correlated with crude protein (r = 0.19) and soluble sugar content (r = 0.20), while it was a slightly negatively correlated with the fibre fractions acid detergent fibre (r = −0.24) and neutral detergent fibre (r = −0.23) and there was a negligible relation with the lipid (r = 0.06) and the DM content of the diet (r = 0.11). Diet composition was expressed as a proportion of total DM and therefore all these parameters are intercorrelated, with r values ranging from −0.75 to 0.93. The highest correlation of CH₄ production with any diet parameter was observed for ME content of the diet (r = 0.43) (Figs 2, 3).

Fig. 3.

Relationship between methane production (CH₄) and metabolisable energy (ME) content of the pasture in male and female growing, lactating and non-lactating beef and dairy cattle.

When DMI was accounted for by expressing CH₄ production per kilogram of DMI, neither dietary nor animal factors were strongly correlated with CH₄ yield. The highest correlation (r = 0.242) was observed between CH₄ yield and BW of the animals.

Methane prediction model

The selected model for predicting CH₄ emissions contains DMI, physiological state, BW and age as main effects (Table 5) because it had a low AICc and no collinearity was identified, as indicated by the general VIFs (Table 6), which were all below the critical value of 10. The model equation is given as follows:

The units for dry-matter intake (DMI) and rescaled bodyweight (BW) are kg day⁻¹ and kg respectively. The basal value for the physiological stage is growing animals.

The estimated model coefficients and their uncertainty are shown in Fig. 4. On the basis of the model parameters, DMI and the lactating animals have the largest influence on CH₄ production (Fig. 5). Male animals are predicted to have a higher CH₄ production than females, all other things being equal. The predictions indicate that lactating animals had the highest CH₄ production at the same intake level (Fig. 6). The comparison of predicted with measured CH₄ production data showed that the predictions from the model follow the observations well (Fig. 7).

Fig. 4.

Model coefficients and their errors for dry-matter intake (DMI), bodyweight (BW), physiological stage (Phys.state) and sex. Note: coefficient for BW is based on 10 kg bodyweight increase.

Fig. 5.

Residuals (observed – predicted) of methane production (g day⁻¹).

Fig. 6.

Predicted means of methane production (g day⁻¹) for animals in different physiological states that consume 10 kg DM per day.

Fig. 7.

Predicted versus measured methane production (g day⁻¹) from the whole cattle dataset (n = 272). The line represents x = y.

Predicted CH₄ yield

When the predictions of CH₄ production (g day⁻¹) are converted into CH₄ yield (g kg⁻¹ DMI), it becomes evident that the model predicts a lower CH₄ yield because the DMI increases for each class of animal (Fig. 8). When the classes are compared at a fixed DMI, for instance 10 kg, the CH₄ yields of the growing females, growing males, lactating females, non-lactating females and mature males were 21.30, 22.50, 23.62, 20.91 and 22.11 g kg⁻¹ respectively. These data indicated that at a given DMI, lactating cattle have the highest CH₄ yield across the groups. The predicted CH₄ yield averaged over the observed DMI range was 24.08, 26.21, 22.12, 21.00 and 22.23 g kg⁻¹ DMI for growing females, growing males, lactating females, non-growing females and non-growing males respectively.

Fig. 8.

Raw predictions of methane yield (yM; g kg⁻¹ DMI; i.e. predicted methane production divided by dry-matter intake) in response to increasing DMI levels for growing females (Gf), growing males (Gm), lactating females (Lf), non-lactating females (Ngf), and non-growing males (Ngm) at intake levels measured.