Livestock numbers are increasing to supply the growing demand for meat-rich
diets. The sustainability of this trend has been questioned, and future
environmental changes, such as climate change, may cause some regions to
become less suitable for livestock. Livestock and wild herbivores are
strongly dependent on the nutritional chemistry of forage plants. Nutrition
is positively linked to weight gains, milk production and reproductive
success, and nutrition is also a key determinant of enteric methane
production. In this meta-analysis, we assessed the effects of growing
conditions on forage quality by compiling published measurements of grass
nutritive value and combining these data with climatic, edaphic and
management information. We found that forage nutritive value was reduced at
higher temperatures and increased by nitrogen fertiliser addition, likely
driven by a combination of changes to species identity and changes to
physiology and phenology. These relationships were combined with multiple
published empirical models to estimate forage- and temperature-driven changes
to cattle enteric methane production. This suggested a previously undescribed
positive climate change feedback, where elevated temperatures reduce grass
nutritive value and correspondingly may increase methane production by
0.9 % with a 1
Global meat production has increased rapidly in recent years, from 71 million
tonnes in 1961 to 318 million tonnes in 2014 (FAOSTAT, 2016). This is due to
population growth and a transition to meat-rich diets across many countries
(Tilman and Clark, 2014). Grazing lands have expanded to support this
production, particularly across Asia and South America, and now cover
35 million km
Ruminants (cattle and small ruminants such as sheep and goats) consume 80 % (3.7 GT) of the plant material grown to feed livestock (Herrero et al., 2013), and grasses continue to comprise the largest proportion of livestock diets. For example, in the year 2000, 48 % (2.3 billion tonnes) of the biomass consumed by livestock was grass, followed by grains (1.3 billion tonnes). The remainder of livestock feed (0.1 billion tonnes) was the leaves and stalks of field crops, such as corn (maize), sorghum and soybeans (Herrero et al., 2013). The chemical composition and morphology of forage grasses determines their palatability and nutritive value to livestock, thus influencing the amount of feed consumed, the efficiency of rumination, the rates of weight gain, the quality and volume of milk produced and reproductive success (Herrero et al., 2015). Forage grasses generally enhance nutritive value for livestock if they contain a greater proportion of readily fermentable components, such as sugars, organic acids and proteins, and a lower proportion of fibre (Waghorn and Clark, 2004). Furthermore, highly nutritious forage can reduce ruminant methane production, since feed moves through the digestive system more rapidly (Knapp et al., 2014). Accordingly, regional and inter-annual variability in forage nutritive value generates corresponding variability in the production of meat and dairy products and variability in the magnitude of ruminant methane emissions (Thornton and Herrero, 2010).
Meat and dairy production in arid, equatorial and tropical regions is often lower than production in temperate regions due to the lower nutritional quality of forage grasses, lack of access to inorganic nitrogen (N) fertilisers, infertile soils and adverse climatic conditions (Thornton et al., 2011). Warmer regions are associated with taller, less nutritious and slow-growing grasses with low concentrations of protein, high concentrations of fibre and high plant dry matter content (DM; the proportion of plant dry mass to fresh mass) (Jégo et al., 2013; Waghorn and Clark, 2004). While extremely cold regions are also associated with grasses of low nutritive quality, cold regions are rarely suitable for ruminant livestock (Nielsen et al., 2013). The timing of grazing and forage harvesting are also important determinants of forage quality. For example, summer harvests frequently produce grasses of lower nutritive quality than spring harvests (Kering et al., 2011). Consequently, grasses of lower forage quality have low dry matter digestibility (DMD; the proportion of plant dry mass which is digestible; high DMD is positively associated with livestock productivity) (Lavorel and Grigulis, 2012; Pontes et al., 2007a). Greater grass nutritive value has been linked to cooler temperatures and N fertiliser addition due to phenological and physiological changes towards delayed flowering, modified stem : leaf ratios, thinner cell walls, reduced lignification and species turnover (Gardarin et al., 2014; Hirata, 1999; Kering et al., 2011).
Ruminant methane production is calculated using IPCC (2006) methodologies in GHG accounting (tiers 1, 2 and 3), and the more complex methods (tiers 2 and 3) incorporate the effects of nutritive value (Schils et al., 2007). However, few models have been developed to predict the effects of climate change on forage nutritive value (Kipling et al., 2016), and those which include climate or management have focussed on single plant species (Jégo et al., 2013) or regions (Graux et al., 2011). Quantifying the relationships between forage grass nutritive value, growing conditions and management more broadly and across many plant species provides an opportunity to make general projections of future changes to livestock and associated methane production. To our knowledge, such relationships have not been systematically assessed at the global scale.
We tested the hypothesis that increasing temperatures are associated with grasses of lower nutritive value, delivering higher concentrations of fibre, lower protein and lower DMD with N fertiliser addition having opposite effects. To quantify the variation in the nutritive value of forage species growing across a range of bioclimatic zones and to understand the influence of climate and fertiliser application, data were gathered from published literature sources in which field-derived nutritive data were reported. Neutral detergent fibre (NDF; structural plant components; cellulose, lignin and hemicellulose) and crude protein (CP; approximate protein content) are presented as the most commonly reported measurements of forage nutritive value. NDF and CP are generally negatively and positively correlated with livestock productivity, respectively. These data were combined with a range of potentially modifying variables, including temperature, rainfall, rates of N fertiliser addition and the photosynthetic pathway. Statistical models were then used to generate projections of future climate-induced changes to forage grass nutritive value and cattle methane production.
Data were obtained from peer-reviewed journal articles. These articles were
identified by systematically searching the ISI Web of Knowledge (WoK;
To ensure that the methods for measuring forage nutritive value were
consistent across the articles, data were included if NDF and CP analyses were
carried out on dried samples and presented in units of g kg
Descriptive data were included in the database for each data point. These
potential explanatory data described the site (latitude, longitude and
elevation), the experiment (degree of replication, experimental treatments and
whether the grassland was a monoculture or polyculture), the management (fertiliser
addition rate and grazing density), the soil (type and pH), the climate (mean annual
temperature (MAT) and mean annual rainfall (MAR)), the weather during the month of
sample collection (mean monthly temperature and total monthly rainfall) and
data describing the plant photosynthetic pathway system (C3 and C4). Data
were recorded from each article from text or tables. When this was not
possible, data were obtained from graphs using the digitizing software
DataThief (
Sites were allocated to a bioclimatic zone as defined by the
Köppen–Geiger climate classification system (Kottek et al., 2006) and
recorded in the database as arid (
In many cases, data were obtained from the articles analysed, but in some
cases there were gaps in the information available. Data most commonly
gathered from external sources were related to weather (sampling temperature and
rainfall) and climate (MAT and MAR), which were obtained from the closest
weather station to each site according to the National Centers for
Environmental Prediction database (
Statistical analyses were carried out using weighted, restricted
maximum-likelihood linear mixed-effects (LME) models (Pinheiro and Bates,
2000). Model selection was carried out by including NDF or CP as a response
variable with multiple potential explanatory variables added as fixed
effects to generate full (maximal) models. Fixed effects were mean
temperature during the sampling month or MAT, total rainfall during the
sampling month or MAR, elevation, rates of N addition and the photosynthetic
pathway. Grazing density, soil pH and whether the plants were grown in
monoculture
or polyculture were shown not to significantly relate to CP or NDF in the LME
models in preliminary analyses. To avoid over-fitting, these
variables were not included in the initial full models (all
For the random effects structure, the identity of grass species was nested within the experimental treatment and treatments were nested within sites and represented within LME models, thus accounting for cases in which several
measurements were taken for the same site, treatment or
species. This accounted for the differences between the species and between the sites
without making them the focus of our analysis. Any relationships identified
therefore included the effects of changes to species identity and
changes to physiology and phenology. However, a separate model was also
fitted for the best represented plant species in the database
(
The non-significant explanatory variables were removed from the full models as all
terms were found to reduce Akaike's information criterion (AIC). The
relative influence of each term on the model likelihood was assessed by comparing
the AIC of the current model with that of a simplified model with the terms
deleted until the AIC ceased to decline (Crawley, 2013; Richards, 2005).
Temperature and rainfall could not be included together in the LME models because
these variables were shown to covary strongly (
Methane production projections were based on published, experimentally
derived relationships between forage NDF content or daily NDF intake (NDFi)
and enteric methane production, as measured in cattle. A suite of equations
was acquired from published articles with all but one being the product of
a meta-analysis (Table 1). These equations summarise many measurements of
cattle enteric methane production across Africa, Asia, Australasia, Europe,
North America and South America, and relate the magnitude of methane
production to the nutritive quality of forage and, in some cases, total feed
intake. In total, 303 studies were included across these meta-analyses with
methane production measured by hood, mask and whole animal calorimetry,
respiration chamber and sulfur hexafluoride (SF
A summary of the published equations used to model changes driven by grass nutritive quality in methane production, giving the details for cattle type (D is dairy and B is beef), regions covered (AF is Africa, AS is Asia, AUNZ is Austalia and New Zealand, EU is Europe, NA is North America and SA is South America) and the number of studies included in each analysis. The values for root mean square prediction error (RMSPE) are also presented.
NDF and NDFi were calculated using the parameters identified by our LME models,
which describe the relationship between NDF and MAT (see “Results”)
multiplied by the estimated daily feed intake or DMI (dry matter intake). The initial
modelling based on equations A–E assumed that cattle DMI was 18.8 kg
DMI d
Projections of temperature-driven changes to cattle methane production used
the HadGEM2 (Hadley Centre Global Environment Model version 2) family of
climate models (IPCC, 2014) by applying low and high representative
concentration pathways (low
HadGEM2 has been identified as a robust model which is valuable for
predictions across climate change scenarios including biogeochemical
feedback (Collins et al., 2011). The estimated increases in cattle methane
production were calculated as ratios of methane production based on
projected 2050 mean temperatures compared with production based on current
temperatures (Hijmans et al., 2005). HadGEM2 models based on RCP 2.6 assumed
that GHG mitigation policies are widely adopted and livestock numbers will
decline, resulting in a reduction in GHG emissions after 2020. The models based
on RCP 8.5 assume that GHG mitigation policies are not adopted, that
livestock numbers will increase and that GHG emissions will continue to increase
unabated. RCP 2.6 and RCP 8.5 therefore represented the lower and upper
projections of future climate and forage-driven increases in cattle methane
production. The regions which are unsuitable for ruminant livestock were excluded
(Robinson et al., 2014), as were regions which are predicted to exceed
30
There was a large range in mean neutral detergent fibre (NDF) across the
forage grass species (for a full list of species and a summary of the nutritive values for each
species, see the Supplement, Table S2) from the lowest,
NDF varied between bioclimatic zones, and grasses growing in cooler temperate or tundra zones had a mean NDF 21 % lower than in warmer arid and equatorial zones (Fig. 1a), but there was no difference between NDF values recorded from arid and equatorial zones. CP also varied between bioclimatic zones, and grasses growing in cooler temperate or tundra zones had a mean CP 8 % greater than grasses growing in equatorial zones (Fig. 1b). However, there were no differences between the CP contents of grasses growing in arid zones when compared with the other bioclimatic zones.
Box plots of
Higher temperatures during the sampling month were associated with increasing
NDF across the grasses (Fig. 2), and NDF increased by 0.4
The linear relationship between forage-neutral detergent fibre (NDF)
content and temperature (
The minimum adequate linear mixed-effects models for forage-neutral detergent fibre (NDF) and crude protein (CP). The values represent slopes, except the C4 pathway values, which represent the absolute differences between the C3 pathway (intercept) and the C4 pathway. The site numbers differ between response types since temperature at the time of sampling and NDF and CP were not always available from all articles.
NDF was also influenced by the photosynthetic pathway, with the NDF content of C4
species a mean of 9 % greater than C3 species. These C4 grasses were more
commonly recorded at warmer sites, and NDF content was recorded from C4
grasses growing in mean monthly temperatures greater than 15
CP was positively related to the rates of N addition, with a
100 kg ha
The estimated change in cattle methane production with temperature-derived declines in grass nutritive quality. The dotted lines represent the six model outputs as defined by equations A–F (defining the relationships between grass nutritive quality and methane production) when combined with the inverse relationship between temperature and grass nutritive quality presented in this article. The continuous line represents the mean weighted model, which is the mean methane production predicted by all six equations weighted by the number of contributing datasets.
Applying models A to F to the positive relationship between NDF and MAT
resulted in a range of projections for forage- and temperature-driven changes
to methane production (Fig. 3). Models A to E projected increased methane
production with rising temperatures, assuming a mean cattle size and DMI, with
model A projecting the largest increase in methane production (2.9 % for
a 1
The effect of simulating changes to cattle size by modifying DMI had
contrasting effects across the different models (Fig. 4). In the case of
model A, increasing cattle size consistent with the current global trend
towards larger cattle (Herrero et al., 2013) increased the rise in projected
methane production with temperature (0.8–3.7 % for a 1
The estimated change in enteric methane production with temperature
change for
When the statistical models were combined with future temperature scenarios, potential hotspots of forage-driven increases in methane production were identified. The low emissions scenario predicted increases in methane production for mean sized cattle by 1–2 % across most regions, whilst hotspots in North America, central and eastern Europe and Asia saw predicted increases of approximately 3–4 % (Fig. 5a). The high emissions scenario resulted in a larger area experiencing high increases in cattle methane production with many regions across North and South America, Europe, central and southern Africa, Asia and Australasia increasing by 6–8 % (Fig. 5b). These projections represent the estimated change in methane production for each animal. Simulated decreases and increases in the global cattle inventory are included in climate projections; RCP 2.6 and 8.5, respectively (IPCC, 2014).
Predictions of climate- and forage-driven increases in cattle methane
production (%) under temperatures predicted for 2050 using
Global food consumption patterns are shifting from traditional diets to
diets rich in refined sugars, fats, oils and meats (Tilman and Clark, 2014).
Assessments suggest that agricultural GHG emissions need to be reduced by
Forage grass nutritive values varied substantially between and within species and across bioclimatic zones, with our data indicating that 34–90 % of the dry weight of the grass that livestock consume is fibre and 5–36 % is protein. These ranges are greater than those presented elsewhere; for example, NDF has been shown to range from 35–67 % (O'Donovan et al., 2011) and CP from 14–24 % across several European grass species and cultivars (Pontes et al., 2007b), but these greater ranges are to be expected given the wider biogeographic coverage of our study.
NDF values were generally higher and CP values were generally lower in warmer bioclimatic zones than in cooler zones, and this is likely to be one reason why livestock productivity is lower across arid, equatorial and tropical regions. The reduced nutritive value in these zones may be driven by increased abundances of plants with adaptations to prevent heat stress and avoid water loss. These adaptations could include greater stem : leaf ratios, narrowly spaced veins, greater hair densities, thicker cell walls, a higher proportion of epidermis, bundle sheaths, sclerenchyma, vascular tissues and greater concentrations of lignin and silica (Kering et al., 2011). The C4 photosynthetic pathway is also an adaptation to heat and water stress, and C4 plants were more commonly recorded in warmer conditions than C3 plants. C4 plants were also associated with lower nutritive value. This is in line with studies that have measured elevated enteric methane production in cattle consuming high-fibre C4 grasses compared with those consuming C3 grasses (Ulyatt et al., 2002). Across warmer bioclimatic zones, reduced forage nutritive values may be driven by increased abundances of C4 species and of taller, slow-growing species with a conservative growth strategy (Martin and Isaac, 2015; Wood et al., 2015). The large variation within and between species highlights the potential for the cultivation and breeding of grasses to enhance livestock nutrition, which may promote resistance to future environmental changes.
NDF was positively related to MAT and temperatures at the time of sampling.
Links between higher temperatures and declining nutritive values and between
declining nutritive values and increasing enteric methane production have
been established under controlled conditions (Knapp et al., 2014). Our results
indicate that the same mechanisms may operate at a global scale. MAT
represents prevailing climatic conditions, and elevated NDF is likely driven
by a shift towards grasses with heat- and drought-stress adaptations and
conservative functional traits associated with slow growth (Gardarin et al.,
2014). The positive relationship between the sampling temperature and NDF may
also be linked with changes to phenology, such as advanced flowering dates
and rapid tissue aging (Hirata, 1999). The timing of the measurements may have also
played a role in increasing NDF, since later harvests generally produce
grasses of lower nutritive quality (Kering et al., 2011). Temperature-driven
reductions in forage grass nutritive value are consistent with mechanistic and
empirical models (Barrett et al., 2005; Kipling et al., 2016). However, our
results contrast with a meta-analysis of temperature manipulation
experiments, which did not reveal any relationships between warming and
nutritive value, although this study was across a relatively small
temperature gradient (Dumont et al., 2015). The relationships between forage
nutritive value and both sampling temperatures and MAT imply that the
compositional (i.e. turnover in species identity), phenological and
physiological changes of species each play a role. The patterns generated by these different
processes were not directly disentangled in our study. However, there were
relationships between both MAT and sampling temperatures and NDF when
measured from one species,
N fertiliser addition generally increases the productivity of grasslands,
since the majority of these ecosystems are N limited (LeBauer and
Treseder, 2008; Lee et al., 2010). We present data which suggest that N
addition may also increase grass nutritive value, decreasing NDF by around
3–11 % (low to high fertiliser application rates) with an
associated increase in CP by 2–7 %. Increased rates of N addition have
been previously linked to increased abundances of grass species with `fast'
functional traits, reduced fibre and increased protein content (Pontes
et al., 2007a). N addition did not alter the nutritive quality of
Our estimates suggest that future cattle enteric methane production may
change by a mean weighted value of 0.9 % (
The trend towards larger cattle across many regions could also influence the magnitude of changes to enteric methane production because larger cattle have greater feed and fibre intakes (Knapp et al., 2014). Model predictions for larger animals were more variable, and therefore both the magnitude of emissions and the uncertainty surrounding these estimates increases with cattle size. The magnitude of projected change across the different models was also dependent on whether NDF or DMI was the dominant term. Furthermore, our projections are limited to cattle. However, there is emerging evidence that reductions in the nutritive value of forage also lead to increased enteric methane production from sheep (Ramin and Huhtanen, 2013) and buffalo (Patra, 2014). Together cattle, buffalo and sheep contribute > 95 % of global GHG emissions from enteric fermentation (FAO, 2013). If our projections hold across the global ruminant inventory, then overall enteric methane production will increase to a greater magnitude than we predict. Our calculations are also limited to cattle that consume grass. We therefore do not account for the trend towards permanently housed cattle, particularly across Europe and North America. This may further increase emissions because the mixed diets of housed cattle increase enteric methane production by around 58 % (March et al., 2014; O'Neill et al., 2011).
Hotspots of future increases in enteric methane production were identified across North America, central and eastern Europe and Asia using a low GHG emissions scenario combined with our weighted mean model. Hotspots became more widespread and of greater magnitude in a high GHG emissions scenario. At present, the greatest densities of cattle can be found in parts of Asia, North and South America, Europe and across Australasia (FAOSTAT, 2016). Many of these regions are projected to experience the greatest forage nutrition-driven increase in cattle methane production. Added to this, meat production has increased by 3.6 % across Africa and 3.4 % across Asia over the past decade compared with a 1 % increase across Europe (FAOSTAT, 2016), indicating greater future growth across these regions. Losses in forage quality could drive farmers into more extensive farming systems across many regions because larger land areas will be required for each animal. Therefore, it may be necessary to limit the growth of livestock production systems in warmer and drier regions, particularly those likely to experience future warming, if significant losses in livestock production efficiency and increases in methane emissions are to be avoided.
Cattle methane production can be reduced by growing more nutritious forage plants, adding N fertiliser, adding feed supplements (e.g. macroalgae and fats), adjusting rumen pH, increasing concentrate feeding, practicing genetic selection and feeding methane inhibitors (Duin et al., 2016; Machado et al., 2014). However, implementing many of these measures is not feasible at a global scale and is unlikely to result in sufficient reductions in GHG emissions to meet ambitious GHG reduction targets. These measures may also promote other negative environmental effects such as biodiversity loss, nitrous oxide emissions and pollution in the air and water (Manning, 2012; Wollenberg et al., 2016). Ruminant meats (beef and lamb) produce around 250 times greater GHG emissions per gram of protein than legumes (crops from the family Leguminosae), and the production of eggs, seafood, poultry and pork as well as the practice of aquaculture all involve lower emissions than ruminant meats (Tilman and Clark, 2014). A global switch in human diets and a transition to more sustainable agricultural practices, as well as a greater prevalence of organic and silvopastoral farming, may reduce our reliance on intensively farmed cattle and other ruminants. In countries with high or increasing meat consumption, these measures could reduce the environmental impacts of agriculture and contribute to GHG emissions cuts with an associated improvement in human health (Springmann et al., 2016).
There are many uncertainties associated with modelling plant and livestock
systems, and all of the relevant factors could not be considered in our
analysis. Future attempts to refine our predictions therefore require
additional processes to be represented mechanistically and data to
parameterise these processes (Hill et al., 2016). Current livestock models
require many inputs which are not universally available and do not account
for variation across all individuals, breeds and regions. Furthermore,
current mechanistic vegetation models do not quantitatively consider
climate-driven changes to forage nutritive quality (Kipling et al., 2016).
Recent work has addressed the knowledge gaps in empirical models, such as by
quantifying the methane produced by cattle across Africa and other tropical
regions, thus improving the coverage of these models (Jaurena et al., 2015;
Patra, 2015). However, there continues to be low geographic coverage of
forage quality data in equatorial and tropical regions, where the nutritive
quality of forage is typically lower than in temperate regions (Nielsen et al.,
2013). Furthermore, the effects of heat stress on enteric methane production
have not been fully quantified (Kadzere et al., 2002), and the anticipated
near-doubling of the global livestock inventory was also not included in our
projections because future changes in the distribution of cattle and
technological advances are currently unknown (Herrero et al., 2015). If
livestock numbers increase in rapidly warming regions, then we predict that
there will be an associated rise in enteric methane production. Increased
grazing pressure may also alter plant species composition, thus changing the
nutritive value and extent of grazing lands (Gardarin et al., 2014). Other
global environmental changes, such as elevated CO
We present preliminary evidence of future temperature-driven declines in
forage nutritive quality and the corresponding increases in enteric methane
production. Upscaling the GHG footprint of the current livestock inventory to the
2050 projected inventory increases annual GHG emissions from enteric sources from 2.8 to 4.7 GT CO
Data can be obtained by contacting the lead author directly. Some of our data have been obtained from journals which are not open access and cannot be freely distributed.
Mark A. Lee and Pete Manning designed the approach, and Mark A. Lee carried out data collection and analyses. Mark A. Lee developed the predictive models and maps and prepared the paper with contributions from all co-authors.
The authors declare that they have no conflict of interest.
This paper was produced following consultation with the members of the RBG Kew Plant Nutrition and Traits Database steering committee. Thanks to Gerhard Boenisch, Jens Kattge, Charlie Marsh and Alex Papadopulos for editorial advice and discussions on the presentation and analyses. The authors would like to thank two anonymous reviewers for comments which improved the paper. Edited by: P. Stoy Reviewed by: two anonymous referees