Global mitigation potential and costs of reducing agricultural non-CO2 greenhouse gas emissions through 2030

Abstract Agricultural emissions account for 53% of 2010 global non-CO2 emissions and are projected to increase substantially over the next 20 years, especially in Asia, Latin America and Africa. While agriculture is a substantial source of emissions, it is also generally considered to be a potential source of cost-effective non-CO2 GHG abatement. Previous “bottom-up” analyses provided marginal abatement cost (MAC) curves for use in modeling these options within economy-wide and global mitigation analyses. In this paper, we utilize updated economic and biophysical data and models developed by the US Environmental Protection Agency (EPA) to investigate regional mitigation potential for major sources of agricultural GHG emissions. In addition, we explore mitigation potential available at costs at or below the estimated benefits of mitigation, as represented by the social cost of carbon. Key enhancements over previous regional assessments include incorporation of additional mitigation options, updated baseline emissions projections, greater spatial disaggregation, and development of MAC curves through 2030. For croplands and rice cultivation, biophysical, process-based models (DAYCENT and DNDC) are used to simulate yields and net GHG emissions under baseline and mitigation scenarios while the livestock sector is modeled by applying key mitigation options to baselines compiled by EPA. MAC curves are generated accounting for net GHG reductions, yield effects, livestock productivity effects, commodity prices, labor requirements, and capital costs where appropriate. MAC curves are developed at the regional level and reveal large potential for non-CO2 GHG mitigation at low carbon prices, especially in Asia.


Introduction
The agricultural sector is a substantial source of global greenhouse gas (GHG) emissions and the largest source of non-carbon dioxide (non-CO 2 ) GHG emissions, accounting for ABSTRACT Agricultural emissions account for 53% of 2010 global non-CO 2 emissions and are projected to increase substantially over the next 20 years, especially in Asia, Latin America and Africa. While agriculture is a substantial source of emissions, it is also generally considered to be a potential source of cost-effective non-CO 2 GHG abatement. Previous "bottom-up" analyses provided marginal abatement cost (MAC) curves for use in modeling these options within economy-wide and global mitigation analyses. In this paper, we utilize updated economic and biophysical data and models developed by the US Environmental Protection Agency (EPA) to investigate regional mitigation potential for major sources of agricultural GHG emissions. In addition, we explore mitigation potential available at costs at or below the estimated benefits of mitigation, as represented by the social cost of carbon. Key enhancements over previous regional assessments include incorporation of additional mitigation options, updated baseline emissions projections, greater spatial disaggregation, and development of MAC curves through 2030. For croplands and rice cultivation, biophysical, process-based models (DAYCENT and DNDC) are used to simulate yields and net GHG emissions under baseline and mitigation scenarios while the livestock sector is modeled by applying key mitigation options to baselines compiled by EPA. MAC curves are generated accounting for net GHG reductions, yield effects, livestock productivity effects, commodity prices, labor requirements, and capital costs where appropriate. MAC curves are developed at the regional level and reveal large potential for non-CO 2 GHG mitigation at low carbon prices, especially in Asia.
53% of global non-CO 2 emissions in 2010 (US Environmental Protection Agency [USEPA] 2012). Emissions in the agricultural sector result primarily from four sources: (1) cropland soil management (primarily nitrous oxide [N 2 O]), (2) rice cultivation (primarily methane [CH 4 ] from flooded rice paddies, although N 2 O can also be important under certain growing conditions), (3) ruminant livestock enteric fermentation (primarily CH 4 ), and (4) livestock manure management (both CH 4 and N 2 O, with CH 4 from anaerobic manure management systems dominating). Changes in soil carbon are also important determinants of net GHG emissions for soil management and rice cultivation.
However, this sector also offers the potential to provide relatively low-cost opportunities for GHG mitigation. In particular, agriculture may play an important role in GHG abatement portfolios as a cost-effective alternative to reductions in emissions from fossil fuel combustion, industrial activity, and other sources. The agricultural sector has been projected to potentially contribute 7-22% of cumulative abatement in the first few decades of long-run climate stabilization scenarios ). In addition, the sector could potentially receive billions of dollars in farm revenue annually from payments for undertaking mitigation activities (USEPA 2008;Baker et al. 2010). Given the important impact that the level of agricultural mitigation has on total cost estimates of reaching mitigation targets, as well as the implications for land use, agricultural commodity markets, and sectoral income, there is considerable interest in assessment of sectoral mitigation potential and associated costs.
However, there is wide variation across previous studies in regional, sectoral, and GHG coverage and disaggregation, as well as in the potential mitigation and cost reported. In order to build upon the existing literature and provide a detailed set of marginal abatement cost (MAC) curves using consistent methods across all countries for all significant non-CO 2 GHG emitting sectors, USEPA conducted a major study to update previous USEPA estimates of mitigation cost and potential (USEPA 2013(USEPA , 2014. In this paper, we use the data and methods developed in that report to generate MAC curves for the agricultural sector for seven regions covering the globe. We also calculate potential mitigation available at private costs at or below the social benefits of GHG mitigation, as represented by the social cost of carbon (Interagency Working Group on Social Cost of Carbon 2015).

Background
There has been increasing interest in mitigation of non-CO 2 GHG emissions from the agricultural sector both in the context of comprehensive climate policy discussions as well as in targeted initiatives, such as those described in the President's Climate Action Plan (Executive Office of the President 2013). For instance, the USEPA and the US Departments of Agriculture (USDA), Energy (USDOE), Interior, Labor, and Transportation are developing a comprehensive, interagency methane strategy, which includes reductions from agriculture. The strategy focuses on assessing current emissions data, addressing data gaps, identifying technologies and best practices for reducing emissions, and identifying existing authorities and incentive-based opportunities to reduce methane emissions. As one component, the USDA, USEPA, and the USDOE developed a Biogas Roadmap that identifies voluntary actions that can be taken to reduce livestock emissions through the increased use of biogas systems. The Biogas Roadmap outlines strategies to accelerate the adoption of methane digesters and other cost-effective technologies in order to achieve a 25% reduction in US dairy sector emissions by 2020 (USDA 2014;USEPA 2014;USDOE 2014).
In addition to strategies focused on non-CO 2 GHG mitigation nationally in the US, there is ongoing interest at the international as well as subnational levels. The recent Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5) identifies substantial opportunities for agricultural mitigation to play a part in the global effort to reduce emissions. The IPCC AR5 chapter on mitigation in the agriculture, forestry, and land use sector cites recent multi-model comparisons of idealized comprehensive climate policies that conclude agriculture can contribute substantially to the mitigation of non-CO 2 GHGs while reducing overall policy costs (IPCC 2014). California is also currently conducting research to improve the state's GHG inventory for agricultural emissions, particularly N 2 O emissions from agricultural ecosystems under California specific conditions (e.g. see http:// www.arb.ca.gov/ag/fertilizer/fertilizer.htm for a discussion of ongoing efforts).
Despite considerable interest in a better characterization of agricultural sector emissions and GHG mitigation potential, available data on global agricultural sector emissions lags behind data development on fossil fuel emissions (Tubiello et al. 2014). Mitigation potentials and MAC studies across the agricultural sector at the global level are even more difficult to find, particularly at a disaggregated level across the major sources of agricultural GHG emissions.
One reason for this is that there are unique challenges to developing agricultural data over large spatial scales (Beach et al. 2008). Among other challenges, agriculture is quite heterogeneous both spatially and temporally, necessitating consideration of biophysical and management conditions that will influence the effectiveness and cost of alternative mitigation options at a disaggregated level. However, obtaining data at this level of detail can be problematic. The agricultural sector also tends to have many activities that emit multiple types of GHGs, with potentially complex interactions between them. Finally, there are typically many implementation barriers, especially for smallholders in developing countries. Nonetheless, engineering or "bottom-up" abatement cost analyses have been developed for the agricultural sector for individual countries, regions, and the world (Hyman et al. 2002;DeAngelo et al. 2006;USEPA 2006;Beach et al. 2008;Moran et al. 2008;Smith et al. 2008;McKinsey & Company 2009;Schulte & Donnellan 2012;Pellerin et al. 2013). 1 These cost and potential estimates are crucial inputs into top-down modeling of multi-gas mitigation options, where they are incorporated via abatement supply curves or calibration (e.g. Rose et al. 2008;Hertel et al. 2009).
USEPA has recently completed an updated version of a global non-CO 2 GHG mitigation assessment and made the reports and underlying data available to the scientific community. This paper uses the updated MAC models to construct estimates of mitigation potential at the global regional level pertinent to the agriculture sector.

Baseline data
Although USEPA (2012) contains estimates of baseline emissions for agricultural sources, alternative baselines were developed for the purposes of the mitigation report. The primary rationale was to ensure consistency in the area, number of livestock head, production, and price projections used across the entire agricultural sector. Projections provided by the International Food Policy Research Institute (IFPRI) from their IMPACT model of global agricultural markets were used to adjust values for agricultural activities and associated emissions over time. In addition, detailed process-based models-Daily Century (DAYCENT) for croplands and DeNitrification-DeComposition (DNDC) for rice cultivation-were used for both the baseline emissions estimates and the GHG implications of mitigation options, thus allowing for a clear identification of baseline management conditions and consistent estimates of changes to those conditions through mitigation activities. 2 Emissions obtained using these detailed simulation models differ from those obtained in USEPA (2012), which relied upon IPCC default emissions factors (IPCC 2006). For emissions associated with livestock, this analysis relies on projections similar to those used in USEPA (2012), but with some small differences due to the adjustments made for consistency with IFPRI IMPACT projections across all agricultural sectors. Projected acreage changes from the IMPACT model (Nelson et al. 2010) reflect socio-economic drivers such as population growth and technological changes to meet global food demand that differ from those used in USEPA (2012).The baseline emissions were also disaggregated by livestock production system and intensity using data provided by the United Nations Food and Agriculture Organization (FAO) (Benjamin Henderson, personal communication, December 20, 2011).
The baseline agricultural emissions data utilized for the mitigation analysis in this paper, based on USEPA (2013), are considerably lower than the totals reported in USEPA (2012) or the IPCC (2014) and Tubiello et al. (2013) reports, as shown in Table 1. However, much of the differences are due to differences in the sources included in the total estimates. The baseline emissions for most of the agricultural sources in common across the studies (e.g. rice cultivation, enteric fermentation, manure management) are fairly consistent with the exception of emissions from croplands. There are several reasons for these differences in cropland emissions. For one thing, this study incorporated only the baseline emissions simulated by DAYCENT for the crops available within the model, which captures only about 61% of global cropland areas. Emissions from pasture area Table 1. comparison of agriculture sector non-co 2 emissions estimates, 2010 (million metric tons, co 2 e). notes: na = not applicable. a uSePa (2012) presents estimates of emissions from agricultural soils, which includes both croplands and pasture. b DaYcent baseline used for this study includes only maize, wheat, barley, sorghum, oats and related crops, and covers 61% of the global non-rice cropland areas reported by faoStat. c Includes emissions from dairy cattle, non-dairy cattle, buffalo, sheep, goats, camels, mules/asses, horses, market swine, breeding swine, and poultry. d other agricultural non-co 2 emissions in ePa (2012) include cH 4 from agricultural soils and forest clearing as well as cH 4 and n 2 o emissions from field burning of agricultural residues and prescribed burning of savannas.  were not included in this study due to a lack of DAYCENT results on productivity and emissions impacts associated with mitigation options for pasture. In addition, while the other inventory estimates relied primarily upon Tier 1 emissions estimates, our estimates were based on more complex Tier 3 calculations modeled using DAYCENT. Although there was variation across regions, emissions factors from DAYCENT tended to be lower than IPCC default emissions factors (IPCC 2006). Because we are not applying our mitigation options to the full quantity of cropland emissions from agriculture, it is important to keep in mind that the cropland mitigation estimates presented in this paper reflect reductions only in the portion of baseline emissions captured within the study. However, the emissions captured do reflect the majority of the emissions sources seen as having a high potential for emissions reductions.

Methods
Agricultural cropping systems are very complex, with soil conditions, microbial activity, and crop growth interacting through a number of processes. Therefore, the relationship between changes in practices and crop yields and GHG emissions are generally non-linear. The majority of assessments of the efficacy of GHG emissions mitigation from cropland management and rice cultivation have focused on site specific field studies. However, extrapolation of results from field studies to watershed, province, or national scales is enhanced by using spatially explicit process models. In this study, we rely on the DAYCENT and DNDC models in combination with literature review to characterize options for cropland management and rice cultivation, respectively. The mitigation measures selected for cropland management and rice cultivation were based on assessment of options that could be modelled at the global level using the currently available versions of the DAYCENT and DNDC models. There are a number of additional measures that have been identified in the literature, but that were excluded from this study because they lacked information on the necessary changes in practices and costs, were not globally applicable, or had large uncertainties regarding their potential to mitigate emissions.
For the livestock sector, we rely on data from USEPA, FAO, IFPRI, the International Institute for Applied Systems Analysis (IIASA), USDA, and others as well as information from the professional literature to develop assessments of mitigation costs and emissions reductions associated with alternative mitigation options. The livestock measures incorporated in this study were selected based on measures where mitigation potential and costs have most commonly been quantified in the literature. Similar to options for mitigating emissions from crop production, there are many additional mitigation measures that could potentially be used in selected regions, but they were excluded from this analysis due to considerable uncertainty regarding the mitigation potential, productivity, and animal health impacts across different regions of the world.
This information from the cropland management, rice, and livestock management sectors is used as an input into the International Marginal Abatement Cost (IMAC) model to estimate the associated costs of changing production practices, changes in net GHG emissions, and MACs for each of the four major sources of agricultural GHG emissions. The DAYCENT, DNDC, and IMAC models are described in more detail below (see USEPA 2013 for additional information).

DAYCENT model
DAYCENT is a process-based model that simulates biogeochemical carbon (C) and nitrogen (N) fluxes between the atmosphere, vegetation, and soil by representing the influence of environmental conditions on these fluxes including soil characteristics and weather patterns, crop and forage qualities, and management practices at a daily time step (Parton et al. 1998;Del Grosso et al. 2001). For example, plant growth is controlled by nutrient availability, water, and temperature stress. Nutrient supply is a function of soil organic matter decomposition rates and external nutrient additions. Daily maximum/minimum temperature and precipitation, timing, management events (e.g. fertilization, tillage, harvest), and soil texture data are model inputs. Key submodels include plant production, organic matter decomposition, soil water, soil temperature by layer, nitrification and denitrification, and CH 4 oxidation.
DAYCENT's simulation of indirect N 2 O emissions accounts for volatilization and leaching/runoff from all N in the soil system, regardless of the N source, according to specific environmental and management conditions. N 2 O is emitted indirectly from N applied as commercial fertilizer, sewage sludge, and livestock manure, and other management practices (e.g. plowing, irrigating, harvesting). Nitrogen from managed manure not applied to crops (or pastures) was assumed to volatilize before application to soils. The global spatial data (vegetation, soil, cropland management) was updated for this study to reflect the most recent data available based on a major FAO data collection and synthesis effort.
Global DAYCENT modeling was carried out for irrigated and non-irrigated production systems for maize, wheat, barley, soybean, and sorghum. Crop yields and GHG fluxes were simulated at the 0.5°grid resolution for periods 2000-2010 and 2011-2030 at five-year increments for areas where major crop types (e.g. wheat, maize, soybean, barley, sorghum, millet, rapeseed, dry beans, sunflower seed, and oats, which account for about 61% of global cropland) are grown. A baseline scenario is established for each crop production system assuming business-as-usual (BAU) management practices. Seven mitigation scenarios were then analyzed.
The mitigation options represent alternative management practices that would alter crop yields and the associated GHG emissions, including adoption of no-till management, split N fertilization applications, application of nitrification inhibitors, increased N fertilization (20% increase over BAU), 3 decreased N fertilization (20% reduction from BAU), optimal N fertilization, and 100% crop residue incorporation (see Table 2). The N management practices (split N fertilization, nitrification inhibitors, increased and decreased N fertilization, optimal Table 2. croplands management and rice cultivation mitigation options.

Croplands mitigation options
Rice production mitigation options (combinations of the following) Irrigation na midseason drainage, continuous flooding, alternative wetting/drying, dry seeding, dryland rice cropping 100% residue incorporation, no till 100% residue incorporation, 50% residue incorporation, no till fertilization reduced fertilization 20%, increased fertilization 20%, optimal fertilization, nitrification inhibitors, split nitrogen fertilization ammonium sulphate fertilizer, reduced fertilization 10%, reduced fertilization 20%, reduced fertilization 30%, optimal fertilization, nitrification inhibitors, slow release fertilizer N fertilization) influence N 2 O emissions in addition to soil organic C stocks due to reduced or enhanced C inputs associated with the level of crop production. Although soil organic C stock fluxes are negligible in the DAYCENT baseline because they are already in equilibrium, there is considerable opportunity to modify stocks by changing practices. Levels of soil organic matter and soil C both influence and are influenced by cropland productivity. Other things being equal, higher crop yields tend to increase soil C because there is more crop residue available to be incorporated into the soil.

DNDC model
DNDC was originally developed to estimate C sequestration and trace gas emissions for non-flooded agricultural lands. The model has been applied in a number of applications to simulate the fundamental processes controlling the interactions among ecological drivers, soil environmental factors, and relevant biochemical or geochemical reactions, which collectively determine the rates of trace gas production and consumption in agricultural ecosystems (Li et al. 1992(Li et al. , 1994(Li et al. , 1996. Details of management (e.g. crop rotation, tillage, fertilization, manure amendment, irrigation, weeding, and grazing) have been parameterized and linked to the various biogeochemical processes (e.g. crop growth, litter production, soil water infiltration, decomposition, nitrification, denitrification, fermentation) embedded in DNDC. Water management has a major influence on soil conditions in rice paddies. The frequent changes between saturated and unsaturated conditions in paddy soils lead to substantial fluctuations in soil redox potential (Eh), which has a large influence on the activity of methanogenic bacteria. Because CH 4 and N 2 O are produced or consumed under certain Eh conditions (−300 to −150 mv for CH 4 and 200-500 mv for N 2 O), soil Eh dynamics play a key role in CH 4 and N 2 O production and consumption (Li et al. 2006). Given the differences in conditions that produce these GHGs, they are generated during different stages of soil Eh fluctuations and some management practices that mitigate one GHG may increase another. To enable DNDC to simulate C and N biogeochemical cycling in paddy rice ecosystems, the model was modified by adding a series of anaerobic processes. 4 By tracking Eh dynamics, DNDC is able to link the soil water regime to trace gas emissions for rice paddy ecosystems. The model simulates the dynamics of biomass growth, which is a major factor affecting CH 4 transport from the soil to the atmosphere. DNDC simulates daily CH 4 and N 2 O fluxes from rice paddies through the growing and fallow seasons as fields remain flooded or move between flooded and drained conditions during the season.
A modified version of the DNDC 9.5 Global database was used to simulate crop yields and GHG fluxes from global paddy rice cultivation systems. The DNDC 9.5 global database contains information on soil characteristics, crop planted area, and management conditions (fertilization, irrigation, season, and tillage) on a 0.5 by 0.5 degree grid cell of the world. The model considers all paddy rice production systems, including irrigated and rainfed rice, and single, double and mixed rice as well as deepwater and upland cropping systems. Whereas USEPA (2006) included only major rice-producing countries in Asia, model scenarios conducted for this study were run for all countries in the world that produce a substantial amount of rice.
Twenty-six scenarios were run using DNDC 9.5 (see Table 2). The scenarios addressed management techniques in various combinations hypothesized to reduce GHG emissions from rice systems: flood regime (continuous flooding [CF], mid-season drainage [MD], dry seeding [DS], alternate wetting and drying [AWD], and switching to dryland (upland) rice), residue management (partial removal or 100% incorporation), conventional tillage or no till, and various fertilizer alternatives (conventional/urea, ammonium sulfate in place of urea, urea with nitrification inhibitor, slow release urea, 10% reduced fertilizer, 20% reduced fertilizer, 30% reduced fertilizer, and DNDC optimization of fertilizer application to maximize yields). The water management system under which rice is produced is one of the most important factors influencing CH 4 emissions. Specifically, switching from continuous flooding of rice paddy fields to draining flooded fields periodically during the growing season-a water conservation practice that is increasingly adopted in the baseline to reduce water use-would significantly reduce CH 4 emissions. 5 Other practices (e.g. fertilizer applications, tillage practices and residue management) also alter soil conditions and hence affect crop yields and the soil C-and N-driving processes such as decomposition, nitrification and denitrification (Neue & Sass 1994;Li et al. 2006). Due to the complex interactions, changes in management practices would trigger changes in multiple GHG fluxes. For instance, while drainage of rice fields during the growing season would significantly reduce CH 4 emissions, emissions of N 2 O actually increase (Cai et al. 1997;Zheng et al. 1997Zheng et al. , 2000Zou et al. 2007).

Livestock mitigation
A significant number of livestock GHG mitigation measures can be identified in the literature (e.g. UNFCCC 2008; Archibeque et al. 2012;Hristov et al. 2013;Whittle et al. 2013). However, developing consistent and regional-specific cost estimates for emerging mitigation measures or options that are not widely adopted has proven a challenging task. Cost data for mitigation measures are scarce and often reflect anecdotal experience reported in a specific country, region, or livestock production system. Assumptions have to be made to extrapolate the estimates in other countries, regions and production systems. This review uncovered only a few studies where cost information was presented in addition to associated emission reductions for a number of mitigation measures. Moreover, for some mitigation measures, such as those that potentially reduce livestock enteric fermentation CH 4 emissions, the literature varies on the estimated magnitude of emissions reductions as well as the long-term mitigation effects and animal and human health impacts. Based on the availability and quality of mitigation measure cost and emission reduction efficiency information, this analysis evaluates six mitigation options for enteric fermentation CH 4 emissions and ten options for manure management CH 4 emissions as summarized in Table 3.

IMAC model
As described in Beach et al. (2008), the break-even price for each mitigation option is calculated by setting total benefits (e.g. higher yields, coproducts) equal to total costs of a given mitigation option and solving for the present-value break-even price within the IMAC model. We have updated the model and moved from an Excel spreadsheet format to a GAMS model to provide additional flexibility and more rapid assessment of alternative specifications. To develop MAC curves, we apply a set of mitigation options identified in the literature for each of the four emissions categories. Emissions, yields, productivity changes, labor requirement changes and other factors from the mitigation scenarios are being compared with baseline conditions for the years 2010, 2020, and 2030, and for all agricultural regions globally with available data. If a mitigation option is considered technically feasible for a given region, it is assumed to be adopted immediately, i.e. in data year 2010, and the change in management is continuous for the entire 2010-2030 period. Mitigation estimates therefore represent the technical potential for GHG reductions, with associated costs, without accounting for implementation barriers that would slow adoption of technically feasible options. 6 As described above and in USEPA (2013), DAYCENT and DNDC are used to estimate baseline and mitigation option emissions of CH 4 , N 2 O, and soil C, as well as yield and water resource changes for cropping systems. These factors are drawn from the literature for the livestock sectors. Revenue changes are estimated by using the percentage yield changes from biophysical estimates and baseline yield and commodity prices drawn from FAO, IFPRI, IIASA, and other data sources. Some rice mitigation options require soil amendments (e.g. phosphogypsum) or an alternative fertilizer (e.g. ammonium sulfate instead of urea). Baseline input shares are drawn from the Global Trade Analysis Project (GTAP) database (http://www. gtap.org) and labor requirements to implement the mitigation options are drawn from available sources. Regional prices for fertilizer were obtained from FAOSTAT. The cost implications of any labor requirement changes are calculated using agricultural wages for each region. For more detail on the assumptions incorporated within the IMAC model, refer to USEPA (2013).
Each set of mitigation options for each emission category in each region was assumed to be implemented simultaneously, but without any overlap among the options. This is a simplistic method that avoids double-counting among options but likely underestimates potential penetration of low-cost options and overestimates potential penetration of highcost options.

Reference case and mitigation scenarios
For our reference case, we define management practices consistent with our best estimate of typical management practices in 2010, along with baseline improvements in productivity.
One key change in management that has been taking place in the baseline is increased adoption of midseason drainage in China and other Asian countries as a water management strategy. This has substantially reduced baseline emissions, while reducing mitigation potential.
Under each mitigation scenario, we simulate cost and GHG emissions impacts associated with adoption of the specified mitigation options, with restrictions placed on adoption of selected options in some regions. For instance, options that involve intensive management of livestock, such as application of large-scale anaerobic digesters for manure management, are only assumed to be available within the portion of the livestock emissions generated by intensively managed livestock operations within each country (based on livestock production system and intensity data provided by Benjamin Henderson, personal communication, December 20, 2011). These results are compared with the baseline emissions to generate estimates of mitigation potential. We compute a break-even price for each abatement option for 195 countries to construct MAC curves illustrating the net GHG mitigation potential at specific break-even prices for 2010, 2020, and 2030.

Social cost of carbon
One key measure of the global social benefits of GHG mitigation is the social cost of carbon (represented in terms of the estimated marginal cost of climate change damages per metric ton of CO 2 ), as presented in the Technical Support Document: Technical Update of the Social Cost of Carbon for Regulatory Impact Analysis Under Executive Order 12866 prepared by the Interagency Working Group on Social Cost of Carbon (May 2013; Revised July 2015). 7 We refer to these estimates as "SC-CO 2 estimates." The SC-CO 2 is a metric that estimates the monetary value of social impacts associated with marginal changes in CO 2 emissions in a given year. It includes a wide range of anticipated climate impacts, such as net changes in agricultural productivity and human health, property damage from increased flood risk, and changes in energy system costs, such as reduced costs for heating and increased costs for air conditioning. It is typically used to assess the avoided damages as a result of regulatory actions (i.e. benefits of rulemakings that lead to an incremental reduction in cumulative global CO 2 emissions). The SC-CO 2 estimates represent global measures because of the distinctive nature of the climate change.
As an additional comparison between the estimated benefits of GHG mitigation and potential role of agriculture in providing cost-effective mitigation, we calculate potential mitigation that could be provided by the global agricultural sector with private costs at or below social cost of carbon. Interagency Working Group on Social Cost of Carbon (2015) provides four primary sets of SC-CO 2 values that are suggested for use in US regulatory analyses of impacts. values for SC-CO 2 across these four estimates are $12, $44, $64, and $128 per tonne of CO 2 emissions in 2020 and $17, $52, $76, and $158 per tonne of CO 2 emissions in 2030. The first three sets of values are based on the average SC-CO 2 at discount rates of 5, 3, and 2.5%, respectively. SC-CO 2 estimates for several discount rates are included because the literature shows that the SC-CO 2 is quite sensitive to assumptions about the discount rate, and because no consensus exists on the appropriate rate to use in an intergenerational context (where costs and benefits are incurred by different generations). The fourth value is the 95th percentile of the SC-CO 2 at a 3% discount rate. It is included to represent higher-than-expected impacts from temperature change further out in the tails of the SC-CO 2 distribution (representing less likely, but potentially catastrophic, outcomes).
We focus on values using a 2.5% discount rate for brevity. Results for alternative social cost of carbon calculations would differ, but the relative rankings of mitigation potential across regions would remain similar. values for the 2.5% discount rate are in the mid-range of the four values presented in Interagency Working Group on Social Cost of Carbon (2015).

Results and discussion
Based on our calculations using each of the models described above, we find substantial potential mitigation from the global agriculture sector, especially in Asia. As shown in Figure  1, global mitigation potential from agriculture at a small positive break-even price of  $5/tCO 2 e is over 210 MtCO 2 e in all years. Mitigation potential is almost doubled at the SC-CO 2 value of $64/tCO 2 e with mitigation potential just over 413 MtCO 2 e in 2020. In our analysis, mitigation potential declines slightly over time. The primary reason for that finding is that while we focused on options that reduce non-CO 2 emissions, there are sizable changes in soil C that are being captured within the MACs as well for cropland management and rice cultivation. Changes in practices cause an immediate change in soil C sequestration and then smaller fluxes over time as the soil moves to a new equilibrium. Figure 2 shows the mitigation potential identified at the global level by major subsector considered. Livestock management offers the greatest mitigation potential at negative prices as well as at higher prices, though rice cultivation has a similar potential between $0/ tCO 2 e to $20/tCO 2 e. Croplands management offers relatively large mitigation at negative costs, but the MAC curves are relatively vertical within the range displayed so mitigation from rice cultivation and livestock management become considerably larger than croplands management as the carbon price rises. Table 4 summarizes the abatement potential calculated at $5/tCO 2 e and the maximum technical potential calculated as a percentage of baseline emissions. These results suggest that there are significant opportunities for net GHG reductions in the agriculture sector.
Overall, the analysis suggests that there are opportunities to reduce agricultural emissions by 5-7% with incentives equivalent to a low carbon price of $5/tCO 2 e, which is equivalent to over 200 MtCO 2 e/year. The technical potential calculated achieved reductions of 13-16%, or over 520 MtCO 2 e/year. However, it is important to consider potential implications for food security with increasing adoption of mitigation options. Our analysis emphasized options that were considered feasible in terms of the tradeoffs required (e.g. water and other input requirements, yield impacts), but there may be impacts from adoption of options that lower yields on large areas. For instance, rice is a staple crop produced in areas with fast-growing populations that have been plagued by food shortages. Beach et al. (2014) examine the tradeoffs between GHG mitigation and food security for global rice cultivation.

Mitigation potentials at the global region level
Mitigation potential varies substantially by country/region. Table 5 summarizes mitigation potential by region. Asia has large opportunities for low-or-negative-cost GHG mitigation. At a low carbon price of $5/tCO 2 e, Asia has the highest potential of 129.2 MtCO 2 e, which amounts to 58% of the global total of 223.8 MtCO2e. By 2030 (still at $5/tCO 2 e), Asia's mitigation potential grows to 141.1 MtCO 2 e, which is 65.8% of the global total of 214.6 MtCO 2 e. At SC-CO 2 values, Asia's share of the total mitigation potential is also dominant. In 2020, Asia has 262.5 MtCO 2 e, which amounts to 64% of the global potential of 413.1 MtCO 2 e.
Regional disaggregation in Figures 3-6 shows very clearly that Asia has the highest mitigation potential in every agricultural sector, with large opportunities for low-or-negative cost mitigation options in major agricultural sectors included in all years. Table 5. agricultural non-co 2 gHg mitigation potential by region at $5/tco 2 e and Sc-co 2 e, 2010-2030 (mtco 2 e). a the primary value used for Sc-co 2 is based on average values presented in Interagency Working group on Social cost of carbon (2015) using a 2.5% discount rate and adjusted to 2010$. the Sc-co 2 value increases over time because the marginal impacts of gHg emissions are expected to become more negative as atmospheric gHg concentrations rise due to greater stress on impacted systems with more pronounced climate change.

Limitations
As with any MAC analysis, it is important to consider potential barriers that may need to be overcome in order to achieve adoption options that are identified with low or negative costs. In a MAC analysis, negative costs occur when the net revenues associated with an option are positive in particular instances, indicating that the practice would be profitable even in the absence of mitigation incentives. Thus, the fact that seemingly profitable practices are not being adopted implies that there are other costs that are not fully captured, market imperfections, institutional problems, risk aversion among agricultural producers, regulatory and/ or legal issues, or other barriers that are preventing adoption of these seemingly attractive  options. This suggests that GHG mitigation policy is necessary to induce adoption of these practices. In practice, many macroeconomic models incorporating MAC curves to characterize abatement potential assume there are missing costs for the negative cost options and employ ad hoc adjustments to add costs that shift the MAC curve up to eliminate the negative cost portion of the curve.

Conclusions
This study contributes updated regional agricultural non-CO 2 MAC curves for the period 2010-2030, reflecting additional mitigation options, more recent biophysical and management data, and a greater level of regional disaggregation than has been available to date. We find that regional mitigation potential varies considerably, with Asia dominating the potential agricultural mitigation estimated for each of the major agricultural sources considered. In addition, the combination of biophysical process-based models and models of production and abatement costs to estimate costs and mitigation potential for the agricultural sector at a regionally and sectorally disaggregated level for individual GHGs provides an important contribution to the literature. These data can be used in numerous multi-sector models, where the level of disaggregation will facilitate custom aggregations of individual countries, agricultural sources, and gases for consistency with individual models. In addition, our ongoing efforts to account for alternative rates of technological improvement and to reflect sequential and simultaneous adoption of multiple mitigation options for the same emissions stream in future research will lead to continued improvements in the characterization of agriculture mitigation estimates.  2. Note that our baseline projections assume continuation of historical practices in the absence of mitigation incentives. Thus, the level of adoption of mitigation measures is generally not increasing over time in the baseline projections. 3. Increased fertilization is expected to increase N 2 O emissions in many crop/region combinations, but was included as a potential mitigation option for two reasons. First, there are some countries in which baseline soil carbon levels are low enough that increased fertilization results in increases in soil carbon sequestration outweighing the higher N 2 O emissions, at least initially. In addition, it is possible that increased fertilization results in higher emission per unit area but lower emissions per unit of output, especially in regions where baseline fertilization levels are very low. Although we focus on changes in emissions per unit area in this analysis, changes in emissions intensity may be of interest in some applications. In generating the MAC curves, only mitigation options that result in a reduction in emissions within a given region are included as an option for that region. 4. The paddy-rice version of DNDC has been described and validated for a number of different world regions and is being used for national trace gas inventory studies in various countries in North America, Europe, and Asia (e.g. Li et al. 1992;Zhang et al. 2002;Cai et al. 2003;Li et al. 2004). 5. Water management options (e.g. shifting from continuous flooding to midseason drainage, etc.) are only applicable to irrigated systems. No water management options are available for rainfed, deepwater, or upland rice. 6. Note that this will tend to overstate the speed at which mitigation could be achieved in practice. Baseline emissions projections reflect an assumption that barriers prevent expansion of mitigation options in the absence of incentives, but adoption of these practices is assumed to take place under the mitigation scenario, reflecting an implicit assumption that greater focus on mitigation and provision of incentives would induce adoption of negative and low-cost options that are not being used in the baseline. 7. Interagency Working Group on Social Cost of Carbon (2015) presents SC-CO 2 in 2007$ per metric ton. The unrounded estimates from this report were adjusted to 2010$ using GDP Implicit Price Deflator.

Funding
This work was supported by the US Environmental Protection Agency, Climate Change Division [contract number EP-BPA-12-H-0023, Call Order EP-B14H-00217]. The views and opinions of the authors herein do not necessarily state or reflect those of the US government or the US Environmental Protection Agency.