Livestock increasingly drove global agricultural emissions growth from 1910–2015

Emissions from agricultural activities constitute 11% of global greenhouse gas emissions and are hard to abate. Here, we present and analyze a consistent empirical assessment of global emissions from agricultural activities from 1910–2015. Agricultural emissions increased 3.5-fold from 1910–2015, from 1.9 to 6.7 GtCO2eq yr−1. CH4 emissions, emissions from enteric fermentation and from livestock products contributed the highest fractions of emissions by gases, processes, and products, respectively. A decomposition analysis quantifies the contribution of major drivers of agricultural emissions dynamics. It reveals that globally and across the entire period, changes in population, agricultural production per capita (‘output’), regional distribution of production (‘regional mix’), and composition of final products (‘product mix’, i.e. a shift towards livestock production) all contributed to increasing agricultural emissions. Conversely, declining emissions per unit of production (‘emissions intensity’), particularly for livestock, partly counterbalanced the emissions increase. Significant variations prevail across regions and time periods. Most notably, the composition of final products counteracted agricultural emissions increase from 1910–1950, but growing livestock production has become an increasingly important driver of emissions growth in more recent periods. This finding unravels that increases in livestock production offset the improvements in emissions intensity of industrial agricultural intensification. Our findings underscore the large potential of reducing livestock production and consumption for mitigating the climate impacts of agriculture.


Introduction
Agriculture provides essential food, material and energy resources to humanity, but also generates major environmental impacts [1], including socioeconomic greenhouse gas (GHG) emissions: agricultural activities currently contribute 11% to GHG emissions [2], similar to the net emissions from deforestation [3].The emissions commonly attributed to agriculture include CH 4 (methane) mostly from livestock management and rice cultivation, and N 2 O (nitrous oxide) from fertilization, manure and soil management [4,5].In addition, the production and use of industrial inputs to agriculture, such as mineral fertilizers and tractors, rely on fossil energy inputs [6] that cause further CO 2 emissions.Because the dominant agricultural GHG CH 4 is so short-lived, a reduction of agricultural emissions can have immediate beneficial climate impacts [7,8].To develop effective climate-change mitigation strategies, it is therefore crucial to understand the dynamics and drivers of agricultural emissions.
While fossil energy combustion started causing CO 2 emissions of global relevance only with the Industrial Revolution in the 19th century [9], agriculture has been practiced for millennia since the Neolithic Revolution [10].Research studying longterm climate impacts of land use has focused more on quantifying emissions from deforestation, wood harvest and land degradation [11][12][13][14] than from agricultural activities.However, several recent studies quantify agricultural emissions in the long run [15][16][17][18]: Jones et al [16] find that between 1830 and 2021, the sum of CH 4 and N 2 O emissions from agriculture increased by a factor 7.8, while CO 2 emissions from fossil energy combustion grew by a factor of 416.Quite consistently, Dangal et al [17] and Zhang et al [18] find that since 1890, global CH 4 emissions from enteric fermentation [17] and the sum of enteric fermentation and manure management [18] increased by factors of 3.3 and 4, respectively.
Research investigating the drivers of agricultural emissions, however, exists only for the period since 1960, when FAOstat [19] provides online data on agricultural activities and emissions [20]: This research has shown that since 1960 [21], 1970 [22], and 1990 [2] respectively, increasing emissions from global agriculture, forestry and other land use have been driven by deforestation emissions in the tropics.At the same time, emissions intensity (measured in emissions per unit of both livestock [23] and crop products [22] and land demand of agricultural production [2,21]) declined across world regions.The time period since 1960 was also characterized by the large-scale spread of industrialized agriculture [24] and a modest reduction of global deforestation rates [25].Knowledge gaps therefore prevail on how the introduction of industrial inputs affected global agricultural GHG emissions in the long run, and on the ensuing implications for climate-change mitigation today.
At the level of individual countries, regions, and agricultural processes, a few studies point to the different drivers of emissions dynamics before 1960: agricultural emissions in France [26] and Austria [27] grew more rapidly in the mid-20th century than ever before or after, the period when also agricultural production increased most.Studies on Spain demonstrate that emissions from agricultural practices such as draught power [28] and fertilization [29] shifted in qualitative terms during this period, when traditional practices (livestock used for draught power and manure provision) were replaced by the use of industrial inputs (tractors and mineral fertilizer).An analysis of the U.S. Great Plains traces the GHG impacts of North American frontier agriculture, where soil and livestock emissions emerged decades before the introduction of industrial inputs [30].National case studies also highlight long-term efficiency increases: In Germany, emissions intensity per unit of livestock product roughly halved between the late 19th century and the present [31], and in Austria, agricultural emission intensity per unit of agricultural output almost halved in the period from 1830 to 1910 [32].
We here present an assessment of global GHG emissions from agricultural activities from 1910-2015, comprising emissions from livestock and soil management, as well as emissions from major industrial inputs to agriculture, but excluding emissions from land conversion (i.e.deforestation).We conduct decomposition analyses to quantify the effects of selected drivers on changes in agricultural emissions at the global and world regional levels.In addition, we scrutinize the roles of livestock and industrial agriculture emissions in shaping agricultural emissions and emissions intensity over time.We discuss our findings in view of opportunities for agricultural climatechange mitigation today.

Methods
We develop a consistent dataset of global emissions from agricultural activities for ten time cuts between 1910-2015, discerning ten activities (table 1), three GHGs (CH 4 , N 2 O and CO 2 ), and two product types (crop and livestock products).Our assessment combines agricultural census data with accounting procedures as defined in IPCC guidelines [4], the authoritative scientific basis for the preparation of national GHG inventory reports.Our input data are agricultural census data derived from FAOstat (1961-2015) [19] and statistical yearbooks from FAO and its precursory organization   [33][34][35][36], reporting country-level data on livestock numbers and production, harvested area, primary crop production, machinery and fertilizer use.Many of these data were compiled and consistently scaled up to world regions in previous work [37].Updates and revisions are described below and in the supplement.Our dataset is established at the level of countries (1962,1970,1980,1990,2000,2010,2015) and eleven world regions (1910, 1930 and 1950), which are then consistently aggregated to eight world regions, see supplementary excel 1 table 13.

Quantification of global agricultural emissions, 1910-2015
To quantify emissions from agricultural activities (table 1), we adhere to IPCC guidelines [4] that provide formulae and default emissions factors.In addition to the eight agricultural activities causing CH 4 and N 2 O emissions defined in the IPCC guidelines [4], we also quantify CO 2 emissions from the most important fossil energy inputs to agriculture [6], i.e. emissions from on-farm energy use for machinery and upstream emissions from fertilizer production.This enables us to summarize emissions from industrial agriculture activities defined conservatively as the sum of emissions from on-farm energy use, synthetic fertilizer production and application.We convert all emissions into Gt CO 2 eq yr −1 , using  [19] (1962-2015, countries) and precursory publications [33][34][35][36] as harmonized in Krausmann et al. [37] (1910-1950, regions) for input data, INRAE et al [39] and Sauvant et al [40] for data on feed digestibility and composition and the most recent revision of the IPCC guidelines [4] for equations, emission factors and coefficients.See supplement for detailed information.'Tiers' represent the respective levels of methodological complexity applied according to IPCC guidelines [4] global warming potentials (GWP-100) of 27 for CH 4 and 273 for N 2 O [38].Feed intake is a major data input for the quantification of agricultural emissions from enteric fermentation, manure deposited on pastures, manure management and manure applied to soils.Data on livestock numbers and livestock productivity serve as input to assess feed intake for all years in t DM/yr, based on the 'feed balance' approach developed by Krausmann et al [37], which we updated and advanced (for a detailed description, see supplement 'feed intake quantification').For the major livestock categories pigs, cattle and poultry, we estimate feed demand as a function of livestock production (i.e.milk, beef, pork and egg production) in the period 1962-2015.For all other livestock species (e.g.sheep, goats, horses) and earlier time points, we use coefficients for feed intake per head and year differentiating between traditional and industrialized production systems.We assess feed supply based on data and estimates for market feed supply, fodder crop production, and crop residue availability.The difference between feed demand and supply is considered to be grazed biomass.
Other input data for the quantification of agricultural emissions include area under rice cultivation, residues burnt on fields, application of synthetic fertilizers, tractor numbers, and land under crop cultivation (table 1).For detailed information on quantification and estimation procedures and formulae applied, see the respective activity descriptions in the supplement, for factors used see supplementary excel 1.

Attributing emissions to final products
After quantifying emissions from agricultural activities, we attribute them to agricultural final products, i.e.(a) crop final products for food and fibres, or (b) livestock final products.Because this is not straightforward (e.g.manure applied to cropland could be attributed to both crop and livestock production), we develop a 'best guess' estimate, a 'high livestock' and a 'low livestock' estimate, following different attribution logics (table 2).
For the 'best guess' estimate, we attribute emissions emerging from activities linked solely to livestock production (i.e.enteric fermentation, manure deposited on pastures) to livestock products entirely.All other emissions are attributed partially to livestock and crop products respectively, because part of crop production ultimately serves as feed for livestock.We use the region-and year-specific share of market feed in total primary crop production to quantify the emissions from cropping that ultimately serve livestock production.The remainder of emissions is attributed to crop products.This attribution relies on data for agricultural production at the level of world regions.We define crop products as total crop production minus market feed supply (both derived from Krausmann et al [37], with a correction of rice production in Eastern Asia, see supplement 'Rice cultivation').For livestock products, we develop a consistent dataset of meat, milk and egg production in tDM/yr, based on data derived from FAOstat and precursory sources [33,34], reduced by the respective water contents [43].
This procedure allows to establish a consistent time series of emissions by final products, considering that some crop products are fed to livestock.However, it has the following limitations: (1) it neglects cropland related emissions from livestock feeding on other cropland products (e.g.residues); (2) it is sensitive to market feed input data, which are available only from 1962 onwards, while data for earlier years rely on estimates (see supplement 'feed intake quantification'); (3) it does not explicitly account for whether market feed is imported or produced domestically at the world regional level.This leads to underestimation of crop products emissions in regions importing market feed, such as Western Europe and, in recent decades, Eastern Asia, and to an overestimation in exporting regions, such as North America and Latin America [44,45].
The 'low livestock' and 'high livestock' estimates enable to quantify the uncertainty range inherent to the best guess attribution: In the 'low livestock' estimate, we attribute to livestock only those emissions that directly emerge from livestock metabolism (i.e.enteric fermentation) and emissions from activities not related to crop production (i.e.manure deposited on pasture).In the 'high livestock' estimate, livestock emissions include emissions from all livestock activities (i.e.enteric fermentation, manure deposited on pasture, manure applied to soils, manure management), as well as 50% of on-farm energy use, but exclude emissions from feed production.

Analysing drivers of emissions
To quantify the effects of selected drivers on trends in agricultural emissions, we conduct additive log mean divisia decomposition analyses [46].In addition to emissions data at the level of world regions and final products, we use data on population [37] and agricultural final production (section 'attributing emissions to final products') as input for this analysis.
At the global level, we discern the following drivers of changes in emissions (E i,r ): population (P); output (O), i.e. agricultural final production (Prod) per capita; regional mix, i.e. the regional distribution of agricultural final production as fraction of regional production (Prod r ) in global production; product mix, i.e. the fraction of livestock and crop final products (Prod i,r ) in total production in each world region; and emissions intensity i.e. the regionand product-specific emissions per unit product in each world region: For crop and livestock products, and for the periods 1910-1950, 1950-1990, 1990-2015, and 1910-2015, we quantify the effect of changes in each of these drivers on emissions changes: For better comparability of the dynamics in different periods, we divide each period by the respective number of years and present data in Mt CO 2 eq yr −1 .
To quantify drivers at the level of individual world regions, we apply regional values of population (P r ) and output (Prod r /P r ), but exclude the regional mix as a driver: In contrast to other studies quantifying the drivers of emissions [2,21,22], our decomposition analyses purposefully refrain from using a term that refers to agricultural land.This is because our dataset includes only emissions from agricultural activities, excluding emissions from deforestation and other land conversions.Therefore, the effect of agricultural area change is not a major explanatory factor in our analysis.In addition, the area extent of extensive agricultural practices, particularly of extensive livestock rearing, is very hard to demarcate, and extensive land uses were much more common in the early 20th century than they are today [47,48], obstructing a clear distinction between agricultural and non-agricultural land.We therefore focus on those drivers which provide explanatory power on emissions from agricultural activities.Finally, to investigate the correlations between total agricultural emissions and emissions from livestock or emissions from industrial agriculture activities, respectively, we quantified the coefficient of determination (r 2 ) across world regions in the entire time period and for  and 1990-2015.

Global emissions trends
Global agricultural emissions from livestock and soil management, as well as major fossil energy inputs in agriculture increased continuously from 1.9 GtCO 2 eq yr −1 in 1910 to 6.7 GtCO 2 eq yr −1 in 2015, i.e. by a factor of 3.5.The lowest average annual growth rate occurred in the period 1950-1962 (0.6% yr −1 ) and the highest in the period 1970-1980 (2.5% yr −1 ).
CH 4 accounted for the absolute majority of emissions throughout the period and increased by a factor 2.7.Therefore, its relative contribution declined from 89% in 1910 (1.7 GtCO 2 eq yr −1 ) to 70% in 2015 (4.7 GtCO 2 eq yr −1 ), figure 1(a).Emissions of N 2 O and CO 2 , the latter comprising emissions from on-farm energy use and fertilizer production, grew much more rapidly.Enteric fermentation was the dominant process across time (increasing by a factor 2.5), its share in total emissions declining from 74% in 1910 to 60% 1970, and then to 54% in 2015.Emissions associated to manure management and manure applied to soils increased more than total agricultural emissions across the time period (factors 5.1 and 7.2 respectively), while emissions from manure deposited on pasture increased only moderately (factor 1.7), and thus decreased in relative importance.The contribution of rice cultivation to total emissions remained relatively stable between 10% and 15% across the time period.By contrast, emissions from industrial agriculture activities (figure 1(b)) accounted for less than 1% of emissions in 1910, reached 8% in 1962, and grew to 21% in 2015.
Emissions from livestock products dominated over emissions from crop products and increased by a factor 2.9 in our best guess estimate (2.5 and 2.8 in the 'low' and 'high' livestock estimates respectively).Their relative contribution declined from 85% in 1910 (82% and 87%) to 69% in 2015 (58% and 70%), figure 1(c), because emissions from crop production, particularly due to mineral fertilizer production and application, increased at a higher rate.Agricultural emissions were slightly more evenly distributed across world regions in 2015 than in 1910 (figure 1(d)).The most pronounced absolute emissions increase occurred in South Asia (SAsia, total of 1 GtCO 2 eq across the period), and the strongest relative increase in Eastern Asia (EAsia, factor 7.9).The region with the lowest absolute and relative increase (0.1 GtCO 2 eq and factor 1.4 between 1910 and 2015) was the Former Soviet Union and Eastern Europe (FSU-EE), because of the strong drop in emissions between 1990 and 2000, following the collapse of the Soviet Union [49].

Drivers of global emissions
At the global level, increases in population were concurrent with agricultural emissions growth, though population grew more strongly than emissions (figure 2(a)): in the absence of any counteracting drivers, global population growth would have led to emissions increases of 50 MtCO 2 eq yr −1 across the time period (instead of 45 MtCO 2 eq yr −1 ).The effects of output, regional mix and product mix added to global agricultural emissions increase, contributing respectively 19, 7 and 21 MtCO 2 eq yr −1 on average in 1910-2015.This indicates that percapita production increased, agricultural production shifted towards world regions with higher emissions per unit product, and that livestock production increased more strongly than crop production.Conversely, reductions in emissions intensity counteracted emissions growth by 51 MtCO 2 eq yr −1 on average (figure 2(b)), particularly since the mid-20th  century, signalling that emissions per unit of agricultural product declined.
From 1910-2015, the increase in livestock emissions contributed 63% to total agricultural emissions growth (48% and 63% in the high and low livestock emissions estimates).The underlying drivers, i.e. the enhancing effects of output, regional mix and product mix, as well as the counteracting effect of emissions intensity, were much more pronounced for livestock products than for crop products (figure 2(b)).
Across time periods, the effect of the product mix changed direction: from 1910-1950, a slight shift towards crop production resulted in a negative signal of the product mix, while changes in the regional mix and increasing population and output drove emissions growth.Emissions intensity changes were least pronounced in this period.From 1950-1990, further output increase and a change in the product mix (i.e. more livestock production) emerged as major drivers of growing emissions beyond population growth.Reductions in emissions intensity counteracted emissions growth to a much greater degree than before.From 1990-2015, a further shift towards livestock production turned the product mix to the second most important driver of emissions increase after population growth, consistent across attribution procedures.At the same time the effect of the regional mix increased and that of output declined.Reductions in emissions intensity further counteracted emissions growth to an even stronger extent than in the previous period.

Drivers of regional emissions
Global trends in agricultural emissions are the sum of diverse regional constellations of drivers (figure 3).In the period 1910-1950, the population effect was positive in all regions, and the dominant driver of increased emissions in Northern America (NAmerica), Latin America and the Carribean (LAmerica), Southeastern Asia and Oceania (SEA-Oc) and Southern Asia (SAsia).The effect of emissions intensity was negative in all regions in this period, except in SEA-Oc.The effects of output (positive) and product mix (negative) were strong enough in the Americas to counteract opposing drivers in other parts of the world, most notably Africa and Western Asia (Africa-WAsia, product mix), SEA-Oc (output) and Eastern Asia (EAsia, output).In the early 20th century, frontier agriculture of the Americas [30], in which extensive livestock rearing was gradually replaced by increasingly efficient crop production to supply global markets [50,51] thus importantly shaped the drivers of global agricultural emissions.
In 1950-1990, the drivers that dominated global trends (i.e.population, output and product mix driving emissions increase, and emissions intensity counteracting it) acted, with varying relative importance, in LAmerica, FSU-EE, SAsia and EAsia.This combination of drivers reflects the effects of the green revolution [24], with increasing output of both livestock and crop products, but a relative shift towards livestock production, an effect particularly pronounced in EAsia [52].In NAmerica, Western Europe (WEurope), and SEA-Oc, by contrast, the product mix had a negative effect in this period, indicating that in these regions, increases in crop production (heavily subsidized by industrial inputs [24]) outpaced growth in livestock production.
In the period 1990-2015, the global combination of positive and negative drivers was the same as in the previous period, and was observed in four regions: LAmerica, SAsia, EAsia, and SEA-Oc.In this period, the relative effect of output was lower than in the previous period, with even negative signals in WEurope, FSU-EE, and Africa-WAsia, due to very diverse reasons: in WEurope, policies led to a stabilization of agricultural production at a very high level [53].In FSU-EE, the collapse of the Soviet Union resulted in a decline of livestock numbers and agricultural production [49].In Africa-WAsia, declining agricultural production per capita was linked to the worrisome trend towards reduced food self-sufficiency, jeopardizing food security in some African countries [54].The effect of the product mix was higher than in earlier periods, driven by trends in four regions in the Global South, connected to the dietary transition particularly in Asia [55].Emissions intensity continued to act as a driver counteracting emissions increase to an even stronger extent than in the previous period, particularly due to dynamics in LAmerica, SAsia and EAsia.

The roles of livestock and industrial agriculture in explaining agricultural emissions
Emissions from agricultural activities correlate with livestock emissions consistently and significantly across world regions (figure 4(a)), with an average r 2 of 0.85 in the period 1910-2015, and increasing r 2 values over time (figure 4(f)).Emissions intensity correlates negatively with the share of livestock emissions (figure 4(b)), but with a very low r 2 of 0.14 across the time period, and low levels of significance (figure 4(e)).
Emissions from industrial agriculture activities also correlate positively with total agricultural emissions (figure 4(c)), but show large deviations across regions and a relatively low r 2 value across the time period of 0.51 (figure 4(e)).Yet, higher shares of emissions from industrial agriculture activities in agricultural emissions coincide with lower agricultural emissions intensity in most periods and regions (figure 4(d)), with a low, but increasing r 2 value (0.23 in 1910-2015, increasing from 0.05 in the first to 0.44 in the last period, figure 4(e)).These findings indicate that while industrial agriculture inputs contributed to reducing emissions intensity in the long run (e.g. by increasing crop productivity) with varying effectiveness across world regions, total livestock emissions increasingly determine absolute levels of agricultural emissions across world regions.

Discussion
Our study presents an analysis of global long-term trends in emissions from agricultural activities and their drivers.Compared to existing quantifications (Supplement, Extended Data figure 1), the sum of CH 4 and N 2 O emissions in our data lies above the results by Jones et al [16], and below those by Hong et al [21], after adjustment for differences in the global warming potentials applied.Our results are higher than other estimates, but consistent in temporal trends, for agricultural CH 4 emissions [16,21], for livestock emissions [18], and for emissions from enteric fermentation [17,56].The relatively high estimate of livestock emissions is due to our quantification of feed intake based on the livestock metabolism.By contrast, our assessment of N 2 O falls well within the range of Hong et al [21] and Jones et al [16].Thus, while our results align with estimates in the literature, this comparison underscores the persisting uncertainties related to estimates of global agricultural emissions, emphasizing the need for intensified scientific endeavours aimed at narrowing them down.
Our findings demonstrate that improvements in emissions intensity realized through the industrialization of global agriculture from 1910-2015 were literally 'eaten up' by increasing livestock production.They thus provide long-term evidence of 'rebound effects' regarding more recent efficiency increases in agricultural land productivity [57].They also underscore the observation that the industrialization of agriculture went along with concurring increases of fossil energy use and growing livestock production, described as the 'energy trap of industrial agriculture' [58].
Our study informs debates about options for agricultural climate-change mitigation today [59,60].Firstly, reducing livestock production emerges as a potentially powerful lever for mitigating the climate impacts of agricultural activities.According to our results, the share of livestock in agricultural production increasingly drove emissions growth in many regional and historical contexts from 1910-2015.While acknowledging the historical relevance and the multiple values of traditional livestock rearing in many parts of the world [61,62], we consider reducing livestock production and consumption as an effective way to reduce future agricultural emissions [63].This pertains particularly to industrialized contexts where the traditional agricultural functions of livestock have been replaced by industrial inputs [28,29], and where consumption levels of animal products are high.A decline of livestock production and consumption would additionally reduce the land demand of agriculture ensuing lower deforestation emissions [42,64].
Secondly, our findings confirm that throughout the 20th century, industrial intensification was effective in reducing agricultural emissions intensity with large regional variations [27,31], even without factoring in the effect of reduced land demand per unit of agricultural product [65].Particularly since 1950, reductions in emissions intensity counteracted agricultural emissions growth, and higher emissions from industrial agriculture increasingly coincided with lower emissions intensities.However, reducing emissions intensity never resulted in declining agricultural emissions in absolute terms: Instead, increases in agricultural production-particularly livestock production-consistently overcompensated efficiency gains.While some industrial intensification may help alleviate deprivation in specific regional contexts, we therefore consider a reliance on industrial intensification as an insufficient strategy to reduce future agricultural emissions globally.Instead, our findings confirm that reducing livestock production and consumption can enable to achieve a shift towards more agroecological practices to overcome pressing social and ecological sustainability challenges [66][67][68].

Conclusions
Our study sheds new lights on the global, centennial trajectory of emissions from agricultural activities and their drivers.Our findings confirm that emissions increase in agriculture was much less pronounced than that of other production processes because significant efficiency gains were realized particularly after 1950, when industrial agriculture inputs were increasingly introduced.However, a shift towards more and more livestock production increasingly drove emissions growth from 1910-2015, despite the fact that industrial inputs gradually replaced essential traditional livestock functions (e.g. for draught power and fertilization).In this sense, the green revolution represents an opportunity foregone for reducing agricultural emissions in absolute terms.Our findings thus emphasize the significance of reducing livestock production and consumption for reducing agricultural emissions.

Figure 1 .
Figure 1.Trends in global emissions from agricultural activities, 1910-2015 (a) by gasses, (b) by processes, (c) by products (red and yellow areas display emissions attributed in our 'best guess' estimate, lines present livestock emissions in the high and low livestock emissions estimates, respectively), (d) by regions.

Figure 2 .
Figure 2. Global trends in drivers of agricultural emissions: (a) temporal dynamics of major underlying factors, indexed to the year 1910; (b) drivers of global agricultural emissions change from 1910-2015 quantified in a decomposition analysis.Whiskers indicate the deviations in the product mix effect resulting from applying the low and high livestock emissions estimates respectively.

Figure 3 .
Figure 3. Regional trends and drivers of emissions as quantified in a decomposition analysis discerning the effects of population, output (agricultural production per capita), product mix (crop vs. livestock products), and emissions intensity of crop and livestock production (emissions per unit of crop and livestock product respectively).Whiskers indicate the deviations in the product mix effect resulting from applying the low and high livestock emissions estimates respectively.Trends in underlying drivers are displayed in Supplement, extended data figure 2.

Figure 4 .
Figure 4. Correlations between agricultural emissions (a) and (c) and agricultural emissions intensity (b) and (d) with potential levers of emissions reduction, and their respective coefficients of determination (e): (a) emissions from livestock production, (b) share of livestock emissions in agricultural emissions, (c) emissions from industrial agriculture activities, (d) share of industrial agriculture activities in agricultural emissions.The brightest coloured dots refer to data point 1910, the darkest to 2015.The grey lines in (a) and (b) display equal values of x and y axes.In the heat map (e), darker shades indicate higher coefficients of determination.Levels of significance are indicated by asterisks: * 0.01 < p < 0.05; * * p < 0.01.

Table 1 .
Procedures and input data applied in this study to quantify agricultural GHG emissions.If not stated otherwise in the table, sources are FAOstat , depending on data availability: Tier 1 is basic and Tier 2 is intermediate.
cropland, region-/country and species-specific leaching-and volatilization fractions (indirect emissions) and country-/region-specific emission factors for direct and indirect emissions.Rice cultivation CH4 1 Area under rice cultivation and country-/region-specific emission factors.Residues returned to or left on fields N2O 1 Country-/region-specific N in total residues, residues used as feed, fraction of burnt residues and country-/region specific emission factor.Burnt residues CH4, N2O 1 Burnt residues in each country and emissions factors.1962-2015: burnt residues derived from FAOstat [19], 1910-1950: derived from regional fraction of burnt residues in total residues of 1962.Application of synthetic N fertilizers N2O 1 Application of synthetic N fertilizers in each country, fraction of N volatilizing/leaching (indirect emissions) and country-/region-specific emission factors.Production of fertilizers (NPK) CO2 1 Application of synthetic N, P and K fertilizers in each country and world-region averages of emissions intensity of fertilizer production.[42] On-farm energy use CO2 1 Energy for field operations.2000-2015: agricultural land areas of major crop types and per-area emissions at world-region level.[42] 1910-1990: scaled back based on trend in tractor numbers at world-region level.

Table 2 .
Three variants for attributing emissions from agricultural activities conducted in this study.