1 Introduction

To keep the global average surface temperatures from increasing by more than 2 °C above the pre-industrial level (UNFCCC 2010), global greenhouse gas (GHG) emissions will have to decrease greatly by the end of this century (Rogelj et al. 2011). However, emissions have yet to peak, and the expected growth in global population and per-capita income will continue to exert a strong upward pressure on emission levels for many decades to come.

In agriculture, several “dedicated,” technical options with significant mitigation potential exist, e.g. soil carbon sequestration, increased nitrogen (N)-use efficiency and nitrification inhibitors for reduction of nitrous oxide (N2O) emissions from soils (Snyder et al. 2009; Akiyama et al. 2010; Luo et al. 2010), drainage of paddy rice fields (Fumoto et al. 2010), and fat additives in ruminant feed rations to reduce methane (CH4) (Grainger and Beauchemin 2011). However, it is highly uncertain whether technical measures alone can offer sufficiently deep emission reductions in the long term. In the midterm, the global mitigation potential for CH4 and N2O in agriculture through such measures is estimated to be rather small (DeAngelo et al. 2006; Beach et al. 2008; Smith et al. 2008).

Aggregate productivity in the global agricultural sector has increased over time, in particular the past 50 years. This has entailed increased output of crops per unit of land, and meat per unit of feed consumed by livestock. Such productivity increases have also mitigated GHG emissions per unit of food, particularly if land-use-change effects on soil and vegetation carbon stocks are included (Burney et al. 2010). Increased productivity is likely to be a major GHG mitigation option in many low and middle-income countries, where crop yields and feed conversion efficiencies are well below biophysical limits (Lobell et al. 2009; Wirsenius et al. 2010; Tilman et al. 2011; Valin et al. 2013). However, it should also be recognized that yields of wheat and other major crops in some key regions have leveled off and been stagnant in the past 10–20 years (Grassini et al. 2013), which suggests that the potentials for further yield increases are limited. Furthermore, realizing the productivity potentials that do exist will be associated with higher N turnover per area unit and intensified livestock production, which both may counteract GHG emission reductions.

Dietary changes that substitute vegetable products for animal products may hold a large mitigation potential. However, the actual mitigation level is highly sensitive to which products that are substituted since there is very large variation in the GHG intensity per unit of food within both the animal and vegetable groups (see e.g. Berners-Lee et al. 2012). In addition, a substitution from e.g. meat to vegetables is not consistent with historical and current trends: Per capita consumption of meat continues to increase, also in many affluent countries (FAOSTAT 2013). In most analyses of diets as a mitigation option, assumed dietary changes are not based on specific changes in policy that influence demand (c.f., consumption taxes, as in Wirsenius et al. 2011 and Edjabou and Smed 2013), but instead rely on hypothetical changes in consumer preferences. There is, therefore, poor knowledge of the global mitigation potential through dietary changes under the constraints of consumer preferences.

To our knowledge, no previous study has assessed the combined effect on agricultural GHG emissions of the afore-mentioned mitigation categories increased productivity, technical measures, and dietary changes. Popp et al. (2010) estimated the impact of reducing meat and dairy demand, and Stehfest et al. (2009) found that the cost of reaching stringent climate targets would be significantly lower if consumption of livestock products was abandoned. However, none of these studies included analysis of the mitigation effect of above-baseline increases in agricultural productivity. Valin et al. (2013) carried out analyses of different long-term scenarios of increased productivity, but did not include assessments of the mitigation effect of dedicated mitigation technology or changes (from baseline) in human diets.

This paper aims to contribute to the understanding of global agricultural GHG mitigation by estimating long-term emissions under different mitigation scenarios and relate these emission levels to those compatible with global temperature targets. More specifically, this paper aims to:

  1. 1.

    Assess the global GHG mitigation potential in the agricultural sector from i) livestock productivity increases above baseline; ii) technical measures; and iii) dietary changes from baseline.

  2. 2.

    Assess the compatibility of emissions from the agricultural sector with the 2 °C target.

2 Methods and data

2.1 Overview of scenarios and methodology

In the analysis, we include CH4 from enteric fermentation, paddy rice, and manure management and N2O from agricultural soils and manure management. We consider CO2 from fossil fuels to be part of the energy system and not the food system, and it is therefore omitted. We also do not include CO2 from agricultural expansion and other land-use changes. Our methodological approach is purely biophysical and cannot capture the political and socioeconomic mechanisms behind deforestation and land use change.

To analyze future GHG emissions, we construct regionalized food consumption scenarios and calculate the corresponding crop and livestock production in each region. Application rates of N in fertilizer and manure in crop production are estimated from the N content in harvested crops and assumptions of N use efficiencies for different crops and regions. N2O from agricultural soils is estimated as an emission factor per mass unit N input in fertilizer and manure as well as crop residues left on the fields. Feed intake in livestock production is estimated from assumptions on feed energy conversion efficiencies and feed rations for different livestock systems and regions. Based on the feed intake we estimate N2O and CH4 from manure excretion and storage, as well as CH4 from enteric fermentation. For details on the scenario assumptions, see the online supplementary material.

We present GHG emission estimates for the years 2000, 2030, 2050 and 2070. The estimate for year 2000 is produced for validation purposes. Our main focus is 2050 and 2070 since the global aggregated emission trajectories to reach the 2 °C target at a chance more likely than not exhibit substantial emissions reductions by mid-century and beyond. Emissions of CH4 and N2O are converted to the CO2-equivalents using the 100-year global warming potential (GWP) that includes indirect climate-carbon cycle feedbacks as reported in Myhre et al. (2013).

We examine five different scenarios (see Table 1) that are partly interlinked. The scenarios are described in more detail in sections 2.22.5.

Table 1 Main characteristics of scenarios to 2070 of global agricultural greenhouse gas emissions

2.2 Methane and nitrous oxide emissions in the reference scenario

N2O emissions associated with fertilizer and manure application is estimated by calculating the required N input in crop and pasture production given the N content in the required biomass output and the N utilization efficiency of the applied N. The N efficiency is here defined as the proportion of applied N that ends up in the above-ground biomass. Regional estimates of N efficiencies range from 28 % for rice in China (Fan et al. 2009) to about 75 % in forages (Smil 2001; Lenssen et al. 2010). We use data from Cassman et al. (2002), Ladha et al. (2005), and Smil (2001) to estimate current N efficiencies for different crops in Europe. We then rescale the values for other regions based on Bouwman et al. (2009). The N efficiency is assumed to improve over time in most regions. In the reference (baseline) scenario, we assume that N efficiency remains constant in Europe and gradually converges to the European level in North America and Pacific OECD. In other regions, N efficiency is assumed to be 20 % lower than in Europe by 2050. Direct N2O emissions related to fertilizer and manure application were calculated using an emission factor of 1 % N2O-N per N applied (IPCC 2006).

CH4 emissions from enteric fermentation is estimated as a fraction of feed intake in energy terms (IPCC 2006), assumed to be 7 % for permanent pasture, 6.5 % for forages (silage/hay), 5 % for protein concentrates and 3.5 % for cereal grains in gross energy terms.

For N2O and CH4 from manure management, we estimate the fractions of different manure handling systems in different regions (based on IPCC 2006) and the respective emission factors (based on temperature-averages, according to IPCC 2006). Further, from the assumed feed–to-product ratios (Table 2) we estimate the amount of manure produced, which in combination with the assumptions on manure systems are used to estimate N2O and CH4 emissions.

Table 2 Feed-to-product ratios in year 2000 and 2050 for the reference (REF) and increased productivity (IP) scenarios (region acronyms are explained in the supplementary material)

2.3 Livestock productivity scenarios

Feed requirements are calculated using feed-to-product ratios, here defined as the amount of feed (in gross energy) required to produce one unit of product (in human-metabolizable energy). In addition to the feed-to-product ratio, the feed ration is of great importance for the overall land requirements for animal food production and the associated GHG emissions.

We use two scenarios that reflect different agricultural productivity developments. We construct one Reference (baseline) scenario (REF) and one Increased Productivity scenario (IP). In the REF scenario we assume moderate increases in livestock productivity, largely in line with FAO projections. In the IP scenario we assume faster livestock productivity growth, based on extrapolations from the ‘Increased Livestock Productivity’ scenario in Wirsenius et al. (2010). Table 2 presents the corresponding estimates of feed-to-product efficiencies.

2.4 Technical mitigation scenario

We construct a Technical Mitigation (TM) scenario, based on the IP scenario, where we assess mitigation potentials of different technologies and management practices, and how fast these technologies may be diffused in different regions.

We assume that the N efficiency in crop production gradually increases so that all regions reach the efficiency level of Europe by 2050. At the global aggregate level, this gives a reduction in N2O emissions of 12 % in 2070 compared to the IP scenario. No additional N2O mitigation is assumed. There is a rather large potential for mitigation options for CH4 from paddy rice production. We assume that emissions per kg rice are gradually abated over time; the abatement starts in 2030 and grows linearly to 80 % reduction by 2070 (Lucas et al. 2007).

We assume that CH4 from enteric fermentation may be reduced by 20 %, either by fat additives or other additives in non-pasture feed. We assume that this mitigation option is applied to all ruminant systems in Europe, Pacific Oceania and North America by 2070, and to 50 % of ruminant systems in the rest of the world. We assume a linear interpolation from 2030. This assumption is in line with previous estimates of the mitigation potential for ruminants (DeAngelo et al. 2006; Beach et al. 2008).

Mitigation of manure emissions is arguably most effectively done by using either anaerobic digester or coverage and flaring of methane in slurry systems. Both options reduce CH4 and N2O by around 70 % (Montes et al. 2013). There is an even greater N2O mitigation potential if solid manure systems are converted to slurry systems. We assume a gradual transition first from solid systems towards slurry systems, and thereafter to slurry systems with flaring. This means that regional aggregate emission factors of CH4 and N2O from manure management drop by 30–70 % to 2070 compared to year 2000.

2.5 Dietary change scenarios

We design two scenarios to explore the potential GHG mitigation from dietary changes. In both scenarios we use the TM scenario as a basis. The rationale for this is that substantial deviations from current dietary preferences are unlikely and would probably occur only as a result of policy interventions. However, policy-driven dietary changes are contentious and would almost certainly emerge only after productivity improvements and technical measures largely have been exhausted.

We construct a Climate Carnivore (CC) scenario where 75 % of the ruminant meat and dairy consumption is replaced by other meat (in kcal terms). This scenario represents an increase in total meat by consumption per capita by 45 % compared to the baseline, but the average GHG intensity of this meat consumption is lower. We also make a Flexitarian (FL) scenario, where 75 % of all meat and dairy products are replaced by cereals and pulses (in kcal terms). “Flexitarian” is a term assigned to people that often, but not always, choose to eat vegetarian food (Forestell et al. 2012). It should be noted that the degree of substitution of 75 % is not based on any systematic analysis of limiting factors, and should therefore be considered as a tentative basis for an estimate of the upper-end mitigation potential from dietary changes. In addition to the generally conservative nature of food preferences, factors that might limit a far-reaching shift away from meat and dairy consumption include, for example, preservation of landscapes and cultures related to grazing and pastoralism.

2.6 Emission pathways consistent with 2 °C stabilization

To estimate multi-gas emission pathways compatible with the 2 °C target, we use an integrated climate-economy model based on Johansson (2011). The emission pathways generated in the model represent the least-cost solutions for meeting the temperature target. The emissions of the three most important greenhouse gases CO2, CH4 and N2O are determined endogenously in the model. Abatement of these emissions is modeled by marginal abatement cost functions and with constraints on how rapidly emissions may fall over time. We generate two emission pathways, using two different assumptions on climate sensitivity (being either 3 or 4 °C for a doubling of the CO2 concentration). Climate sensitivity is defined as the equilibrium increase in the global annual average surface temperature from a doubling of the atmospheric CO2 concentration with respect to pre-industrial levels and its value is likely between 1.5 and 4.5 °C (IPCC 2013). Climate sensitivity is critical for the emission space compatible with a given climate target. The emission pathway generated with a climate sensitivity of 4 °C will have a larger chance of keeping the global temperature increase below 2 °C as compared to the emission pathways generated using a climate sensitivity of 3 °C, see supplementary information for further details.

3 Results

3.1 Emissions per unit of product

Figure 1a and b present GHG emissions per kg of product across scenarios and selected regions for ruminant meat and dairy products. The assumed increases in feed efficiency in the baseline scenario explain most of the difference between 2000 and the REF scenario. As can be seen, a substantial reduction in emissions per unit of output can be expected by 2050 in today’s low and mid-income regions. However, by assuming faster livestock productivity improvements than in REF, emissions can be reduced even more in IP. In the TM scenario, emissions are reduced even further, but the mitigation effect is less than that achieved by increased livestock productivity.

Fig. 1
figure 1

a Estimated GHG emissions per average unit of output in ruminant meat production in 2000 and scenarios for 2050. (REF Reference, IP Increased productivity, TM Technical mitigation). b Estimated GHG emissions per unit of output in dairy production in 2000 and scenarios for 2050. (REF Reference, IP Increased productivity, TM Technical mitigation)

There are a number of counteracting factors that exert upward pressure on emission levels as feed-to-product efficiencies increase. First, to achieve higher livestock productivity, larger shares of the rations have to consist of energy and protein-rich feed, such as cereals or soybeans. For ruminant systems, the fractions of permanent pasture and crop residues in the ration therefore typically decline as productivity increases. This means that a larger share of the land used for feed production has higher N fertilizer application rates, which contributes to higher overall N2O emissions. The smaller share of pasture-based feed in the rations also means that more manure is produced in stables, which implies higher CH4 emissions from manure storage facilities. Second, as milk yield in the dairy sector improves, less cattle meat (from surplus dairy calves and culled cows) per unit of milk is produced as a by-product. Most of the GHG emissions from dairy systems may be allocated to the milk output, since it represents roughly 90 % of the milk and meat output in sales value. Therefore, cattle meat from the dairy system may be considered to have lower GHG emissions per unit of meat compared to cattle meat from suckler cow beef systems. Given constant beef demand, this means that the higher the dairy cow milk yield, the higher the average GHG emissions per unit of total beef supply. Similarly, in regions where dairy consumption increases at a slower rate than beef consumption, the share in total beef supply of co-product meat from the dairy sector decreases, and therefore the average emission intensity of ruminant meat increases. These factors are behind the increase in Europe in emissions per unit of ruminant meat in 2050 compared to 2000, see Fig. 2.

Fig. 2
figure 2

Global food-related GHG emissions from agriculture in scenarios for 2050. (REF Reference, IP Increased productivity, TM Technical mitigation, CC Climate Carnivore, FL Flexiterian)

3.2 Emissions from the global food system

Emissions in the REF scenario amount to 12 Gton CO2eq in 2050; ruminant meat is responsible for around two thirds and animal products in total for about 80 %, see Fig. 2. Increased livestock productivity cuts the global emission by 2 Gton CO2eq in 2050 as seen in the IP scenario. In the TM scenario, were technical mitigation options are added to the IP scenario the emissions are reduced to about 8.3 Gton CO2eq/year in 2050. The potential for productivity improvements and mitigation measures is more restricted in the ruminant sector than for dairy, other meat and vegetables; notice that ruminant meat constitutes a larger fraction of the total emissions in TM than in REF. Changing diets cuts the emissions further, in the CC scenario the annual emissions are 4.9 Gton CO2eq, while in the FL scenario the total emissions are about 3.1 Gton CO2eq/year. Hence, a radical shift in diets with substantial productivity improvements and technical mitigation measures may cut baseline emissions by about 75 %. More detailed data and an uncertainty analysis for each scenario are presented in the supplementary material.

3.3 Agricultural emissions compared to emission pathways compatible with the 2 °C target

The two emission pathways compatible with the 2 °C target, but based on different assumptions on climate sensitivity, are shown in Fig. 3. The emission pathway generated using the lower climate sensitivity (3 °C for a CO2 doubling) has about 50 % chance of staying below the temperature target, while the emission pathway generated with the higher sensitivity, gives a higher chance of staying within the limits of the 2 °C target. ,

Fig. 3
figure 3

Emission pathways consistent with the 2 °C target under different climate sensitivities (lines), and agricultural emissions in the different scenarios for 2030, 2050 and 2070 (bars)

As can be seen in Fig. 3, food-related emissions, if they remain unabated, are on track to take up most, if not all, of the long-term annual emission space allowable under the 2 °C target irrespective of the two emission pathways included here. By 2070 in the REF scenario, food-related GHG emissions alone are as high as, or higher than, the annual emissions in either one of the two emission pathways. However, that does not necessarily mean that the 2 °C target is out of reach; if overall emissions can be cut faster than in our illustrative pathways, this could compensate for higher sustained agricultural GHG emissions in the long-run, see section 4.4 for further discussion.

4 Discussion

4.1 Limitations of data and method

The approach in this study is bio-physical and does not explicitly include any economic factors. The scenario results presented here should be seen as descriptive rather than predictive.

This study excluded some GHG emission sources (see 2.1), most notably CO2 from changes in land use. Plausible levels of long-term CO2 emissions from land-use change span over a wide range, since they depend on several highly uncertain factors. Valin et al. (2013) estimated these at 1.9 billion ton CO2/year in 2050 in their baseline scenario—this corresponds to about 16 % of total emissions in the REF scenario in this study.

4.2 Comparison with previous studies

To make valid comparisons with earlier studies, in this section we use the GWP values reported in Forster et al. (2007) (which were used in earlier studies) instead of Myhre et al. (2013), which were used in this study. Popp et al. (2010) estimated agricultural CH4 and N2O emissions in 2055 at about 15 Gton CO2eq/year in an increased-meat-consumption scenario, and at 10 Gton CO2eq/year in a scenario with technical mitigation. Corresponding numbers in this study are 10 Gton CO2eq/year (REF scenario) and 8.3Gton CO2eq/year (TM scenario), respectively. Lack of transparency in Popp et al. (2010) prevents a closer investigation of these differences. The generally lower numbers in our study may be due to lower increases in per-capita meat consumption and/or higher productivity growth in the livestock sector.

In contrast, Valin et al. (2013) produced estimates far lower than those in this study. In their baseline scenario, agricultural CH4 and N2O emissions reach about 4.5 Gton CO2eq/year in 2050, which is less than half of our baseline (REF) scenario, and less than a third of the corresponding scenario in Popp et al. (2010). Different approaches in the modeling of the livestock sector’s feed intake and GHG emissions are likely to be the main explanation to these diverging results. This is illustrated by the lower base-year estimate in Valin et al., which at 3.5 Gton CO2eq/year (in year 2000) is far lower not only than in this study (5.9 Gton CO2eq/year), but also in comparison with other studies, including the recent comprehensive FAO study (Gerber et al. 2013), which estimated livestock-related emissions at 5.2 Gton CO2eq/year.

4.3 Constraints in long-term emission reductions in the energy and food systems

To meet the 2 °C target with a probability larger than 50 %, global GHG emissions have to drop to about 10 Gton CO2eq/year or less by the second half of this century. The prospects for achieving such very deep emission cuts vary across sectors. As indicated in this study, deep cuts in emissions from food and agriculture do not seem plausible without large changes in consumption towards less GHG intensive food, in particular less ruminant meat and dairy.

Stronger advancement of mitigation technology than that assumed in this study could avoid the assertion that dietary changes may be needed for meeting the 2 °C target. Our estimates of the potential for technical mitigation of CH4 from ruminants and N2O from soils are conservative by design; currently known mitigation technology does not seem to promise large reductions. Moreover, these sources constitute microbe-controlled pathways in the carbon and nitrogen cycles in ruminants and soils, and any use of technology to block such pathways will face the likelihood that microorganisms over time evolve to circumvent blocked pathways. For instance, nitrification (which produces N2O as a by-flow) is an exothermic reaction, and therefore represents an energetic potential for soil microorganisms to exploit. It therefore cannot be ruled out that microorganisms would evolve to bypass any pathway blocked through the use of technology such as nitrification inhibitors. On balance, the prospects for deep emission cuts in agriculture through technology seem unfavorable.

Irrespective of the development in the food system, other sectors will need to achieve very large GHG reductions, of the order of 90 % or more, if the 2 °C target is to be met with a large chance. However, given the substantial constraints for emission reductions in the food system, it might be cost-effective to seek even deeper cuts in other sectors. This applies particularly to energy and transport, for which the prospects for deep reductions by technology are more favorable compared to the food system. However, also in these sectors, technology has limitations, and totally CO2-free systems in 40–50 years from now are imaginable only under optimistic assumptions of technological advancements. More importantly, even if CO2-neutral technologies exist, their diffusion may be a relatively slow process (Wilson 2010).

4.4 Trade-offs between short and long-lived gases in long-term emission reductions

The temperature impact of short-lived GHGs (as CH4) and long-lived GHGs (as CO2) differ, which has implications for the long-term emission spaces of the different gases. The temperature impact of short-lived gases correlates well with the annual emissions, whereas that of long-lived gases correlates well with the cumulative emissions (Smith et al. 2012). Hence, for a given temperature stabilization level, any change in the cumulative amount of long-lived GHGs emitted will affect the allowable sustained long-run level of emissions of short-lived GHGs. This means that, for a given temperature-increase target, any lowering of the cumulative CO2 emissions will give room for higher sustained emissions of CH4.

This study indicates that, for baseline diets, long-term GHG emissions from the global food system will be dominated by CH4, with a strong link to ruminant meat and dairy consumption levels. Given the trade-off between CH4 and CO2, in particular its asymmetrical nature with respect to time, society’s net influence on the cumulative CO2 emissions will be decisive for the long-term emission space of CH4, and hence also for the scope for continued dairy and ruminant meat consumption. Thus, the more cumulative CO2 emissions are depressed, the larger room there will be for perpetual CH4 emissions from ruminants.

In addition to deep cuts in CO2 emissions from the energy and transport systems as discussed above, atmospheric CO2 levels could also be mitigated by technology and management options that sequester atmospheric CO2. One of these is increased carbon sequestration in vegetation and/or soils. However, using land for carbon sequestration is likely to compete with food, fiber and bioenergy production. Another option is to use bioenergy together with carbon capture and storage (BECCS), which could in theory offer net negative emissions from the energy system. Such negative emissions could open up the possibility to have larger sustained emissions of CH4 (and N2O) for a given temperature target (Azar et al. 2013). However, due to lack hitherto of operational experience, and the inertia in the diffusion of these capital-intensive technologies, the prospect of a full-blown BECCS industry at the scale needed by the second half of the century is highly uncertain.

5 Conclusions

This paper indicates that, under current trends, food-related agricultural emissions of CH4 and N2O may increase to about 12.7 Gton CO2eq/year by the year 2070. This is likely to be larger than the total CO2-equivalent emission level compatible with meeting the 2 °C limit at chance larger than 50 % (on the order of 10–13 Gton CO2eq/year or less in 2070). Under policies that favor larger increases in livestock productivity as well as substantial implementation of technical mitigation measures, we estimate that emissions can be kept closer to what is required for the target (7.7 Gton CO2eq/year). However, only by also assuming reduced meat and dairy consumption do we find agricultural emission levels that do not take more than half of the total emissions space in 2070. We therefore conclude that dietary changes are crucial for meeting the 2 °C target with high probability. This conclusion carries even more weight when one considers that other GHG-emitting sectors, in particular energy, also face significant constraints in achieving very large reductions.

Stronger advancement of technology than assumed in this study could significantly relax the need for dietary changes. This applies to technology for mitigation of CH4 from ruminants and N2O from soils, for which assumed potentials in this study were conservative, on the basis that known mitigation technology does not seem to promise large reductions. It also applies to technologies such as BECCS which has the potential to sequester atmospheric CO2 and thereby make room for larger sustained agricultural GHG emissions. However, due to lack of operational experience of BECCS, and the inertia in its diffusion, a full-blown BECCS industry in 40–50 years at the scale needed is unlikely.