Mitigation potential of global ammonia emissions and related health impacts in the trade network – Nature.com

NH3 emissions

Agricultural sources of NH3 emissions refer to manure management, direct and indirect soil emissions, manure in pasture/range/paddock, and agricultural waste burning. The estimation of agricultural NH3 emissions at the country level is extremely challenging, due to the fact that a large amount of activity-level data and emission factors are hard to obtain. The EDGAR v4.3.2 emission database27 from Joint Research Centre, European Commission, has updated the bottom-up inventories of NH3 emissions of nations to the year of 2012, which make it possible for a more systematic study on consumption-based accountings of global NH3 emissions.

The global MRIO tables covering multiple regions of the world have been prepared by multiple organizations. Among these tables, the MRIO tables from the Eora database cover the most regions41,42, which have been widely used to analyze the embodied resource and environmental elements in international trade43,44,45. In this study, the Eora database is adopted to build the global MRIO table for 2012.

The global MRIO model, incorporating direct emission inventories, reveals the NH3 emissions induced by final demand and international trade. MRIO can trace the emissions back to the original source that produced the emissions even if products were intermediate constituents in a multiregional supply chain. To perform the MRIO modeling, we should extract the direct emission data that are related to economic activities and reallocate these data to each industrial sector of different economies. The resulting emissions at the sectorial level are used in the MRIO model to link the emissions to consumption and trade. Assume that the number of sectors of country s is ({k}_{s}), the number of countries is n, and denote (N=mathop{sum }nolimits_{s=1}^{n}{k}_{s}). According to the balance of the global MRIO model, the basic linear equation can be expressed as

$${{{{{bf{X}}}}}}={{{{{rm{A}}}}}}{{{{{bf{X}}}}}}+{{{{{bf{F}}}}}}$$

(1)

where ({{{{{bf{X}}}}}}) is the (Ntimes 1) gross-output vector, ({{{{{rm{A}}}}}}) is the (Ntimes N) technical coefficient matrix, and ({{{{{bf{F}}}}}}) stands for the (Ntimes 1) final-consumption vector.

After that, the Leontief inverse matrix can be obtained from Eq. (1):

Here, ({(I-A)}^{-1}) is the Leontief inverse matrix, which shows the total production of each sector required to satisfy the final demand in the region; (I) is the identity matrix. (D) refers to the (Ntimes N)matrix diagonalized from the (Ntimes 1) vector of sectorial NH3 emission intensities (NH3 emissions per output). Since we are interested in the embodied agricultural NH3 emissions in global trade, the emission intensities of nonagricultural sectors are assigned to be zero. The global NH3 emission flow matrix can be acquired by

where (hat{F}) is a (Ntimes N) matrix diagonalized from the vector ({{{{{bf{F}}}}}}). The NH3 emission flow matrix (C) can be written as the following block matrix

where ({C}_{st}) is a ({k}_{s}times {k}_{t}) matrix. When st, ({C}_{st}) denotes the emissions produced in region s that are related to the final consumption of region t. When s=t, ({C}_{st}) represents emissions related to final consumption produced locally. Let ({T}_{st}) refer to the sum of each element ({C}_{st}^{(ij)}) in ({C}_{st})

$${T}_{st}=mathop{sum }limits_{i,=,1}^{{k}_{s}}mathop{sum }limits_{j,=,1}^{{k}_{t}}{C}_{st}^{(ij)}$$

(5)

({T}_{st}) is a scalar that represents the total embodied agricultural emissions that are produced in region s and related to the final consumption of region t.

Based on the NH3 flows matrix, two key indicators that reflect the impacts of international trade on NH3 emissions can be deduced. The NH3 emissions embodied in international import and export are expressed as

$${{{{{mathrm{EE}}}}}}{{{{{mathrm{I}}}}}}^{s}=mathop{sum }limits_{t=1,tne s}^{n}{T}_{ts}$$

(6)

$${{{{{mathrm{EE}}}}}}{{{{{mathrm{E}}}}}}^{s}=mathop{sum }limits_{t=1,tne s}^{n}{T}_{st}$$

(7)

$${{{{{mathrm{EE}}}}}}{{{{{mathrm{B}}}}}}^{s}={{{{{mathrm{EE}}}}}}{{{{{mathrm{I}}}}}}^{s}-{{{{{mathrm{EE}}}}}}{{{{{mathrm{E}}}}}}^{s}$$

(8)

where ({{{{{mathrm{EE}}}}}}{{{{{mathrm{I}}}}}}^{s}) is the total emissions in other regions related to consumption in region s, while ({{{{{mathrm{EE}}}}}}{{{{{mathrm{E}}}}}}^{s}) is the total emissions in region s related to final consumption in other regions. The embodied NH3 emissions in international trade balance (EEB) can be obtained as the difference of import (EEI) and export (EEE). The EEB is also equal to the difference between the CBE emissions and its PBE emissions of a region. A positive value of EEB means that a regions CBE emissions are larger than its PBE emissions.

The 189 countries/regions in the original MRIO table are merged with the EDGAR database. Serbia and Montenegro are merged into one country. Seven regions cannot be merged with the EDGAR database and thereby are dropped (Andorra, Former USSR, Gaza Strip, Liechtenstein, Monaco, San Marino, South Sudan). Finally, we have 181 economies (Supplementary Data6) and 14,839 economysector pairs.

In the Full Eora database, most countries have more than two agricultural sectors. Crop and livestock sectors differ noticeably in their NH3 emission intensities (emissions per output). Therefore, we allocate NH3 emissions to crop and livestock sectors, respectively. More specifically, for each region, we assume the same emission intensity (emission per output) for different crop sectors, and allocate emissions to the detailed sectors based on their outputs. Similarly, we assume the same emission intensity for different livestock sectors and allocate emissions according to their outputs.

Furthermore, inputoutput analysis is susceptible to aggregation errors due to coarse sector classifications. For example, the ratio of export-related emissions to PBEs in Ethiopia is more than 90%, because it exports the agricultural products with smaller embodied emissions (such as coffee)14. Even though we use the Full Eora MRIO database that covers a comprehensive set of sectors for developed countries, some developing countries are only recorded in 26 sectors. To alleviate potential aggregation biases, we use product-level trade information from the United Nations Comtrade Database and follow Oita et al.14 to manually correct misallocation for countries susceptible to aggregation errors. Supplementary Table5 shows the countries and exported commodities to be adjusted for aggregation errors. Due to missing data in the Comtrade database, we do not adjust exports of five small countries (Bermuda, Brunei Darussalam, Cape Verde, Cayman Islands, Netherlands Antilles) that are considered by Oita et al.14 We adjust exports of another country (Papua New Guinea) that is also susceptible to aggregation errors due to large export of palm oil, coffee, and cocoa beans. The embodied emissions calculated with the Eora-26 MRIO database, which has coarser sector classifications, are quite close to the results by the Full Eora database with adjustment to aggregation errors (Supplementary Fig.4). This suggests that potential aggregation errors have a limited effect on the calculations and are unlikely to bias the analysis.

We use complex network indicators and the community detection method to analyze the global NH3 trade-related health-effect network characteristics.

Degree and degree distribution: in the health-effect network, definitions of out-degree and in-degree are analogous to those in trade network46,47,48. Out-degree is the number of economies to which a given economy bears health loss for exporting goods, and in-degree is the number of economies that a given economy is transferring health loss to by importing goods from them. The two indicators measure the extensive margin of the economy involved in international trade and are defined as

$${k}_{i}^{{{{{mathrm{out}}}}}}=mathop{sum }limits_{j,=,1(i,ne, j)}^{n}{a}_{ij},,{k}_{i}^{{{{{mathrm{in}}}}}}=mathop{sum }limits_{j,=,1(i,ne, j)}^{n}{a}_{ji}$$

(9)

where ({a}_{ij}) is a dummy variable indicating whether health effects are flowing from economy (i) to economy (j), (n) is the total number of economies (181 in the health-effect network), ({k}_{i}^{{{{{mathrm{out}}}}}}) and ({k}_{i}^{{{{{mathrm{in}}}}}}) represent the out-degree and in-degree, respectively.

To further analyze the heterogeneity among the 181 economies, we calculate the probability distribution of degree (k) as (p(k)={n}_{k}/n), where ({n}_{k}) is the number of economies that have the same degree (k). The network can be characterized as a scale-free network if its degree distribution is well fitted by a power-law distribution, i.e., (p(k)propto {k}^{-lambda }). A scale-free network implies the coexistence of a large number of nodes in the periphery that are loosely connected with others and a very few hub nodes that play central roles in connecting other nodes.

If we consider the weighted health-effect network, links connecting any two economies are not regarded as binary indicators but weighted in proportion to the health-effect flows between them. Just analogous to out-degree and in-degree, out-strength and in-strength can be obtained by replacing ({a}_{ij}) with ({q}_{ij}), which indicates the volume of health effects transferred from (i) to (j):

$${s}_{i}^{{{{{mathrm{out}}}}}}= mathop{sum }limits_{j,=,1(i,ne, j)}^{n}{q}_{ij},,{s}_{i}^{{{{{mathrm{in}}}}}}=mathop{sum }limits_{j,=,1(i,ne, j)}^{n}{q}_{ji}$$

(10)

Betweenness centrality: betweenness centrality characterizes the connectivity and intermediality of a network and reflects the importance of a given node as the role of bridging other nodes by calculating the number of shortest paths that go through it:

$${b}_{k}=mathop{sum }limits_{i,=,1}^{n}mathop{sum }limits_{j,=,1}^{n}{sigma }_{ij}(k)/{sigma }_{ij}$$

(11)

where ({sigma }_{ij}) is the number of shortest paths between economy (i) and economy (j), ({sigma }_{ij}(k)) is the number of shortest paths between (i) and (j) that pass through economy (k), ({b}_{k}) is the betweenness centrality of economy (k). This measure indicates if economy (k) is on the shortest path between (i) and (j), then it counts towards the betweenness centrality of economy (k). In the health-effect network, an economy with high betweenness centrality implies its crucial bridging roles in transferring or receiving health effects.

For the weighted health-effect network, the path length from economy (i) to economy (j) is defined by the number of bilateral flows of health effects ({q}_{ij}), with which we likewise obtain the weighted betweenness centrality.

Eigenvector centrality: another prevalent centrality measure is the eigenvector centrality, which evaluates the importance of a node based on its neighboring nodes. The intuition behind this is that a node should have high centrality if it is connected with many other nodes that also have high eigenvector centrality. Eigenvector centrality is defined as:

$${{{{{{bf{v}}}}}}}_{{{{{{bf{i}}}}}}}={lambda }^{-1}mathop{sum }limits_{j,=,1}^{n}{a}_{ij}{{{{{{bf{v}}}}}}}_{{{{{{bf{j}}}}}}}$$

(12)

where (lambda) and ({{{{{{bf{v}}}}}}}_{{{{{{bf{j}}}}}}}) are the largest eigenvalue and the corresponding eigenvector.

The weighted average of nearest-neighbor degree: a similar hybrid network is a network in which nodes tend to be connected with other nodes of similar degree. To assess this tendency, we calculate the weighted neighboring degree of node (i):

$${omega }_{i}=mathop{sum }limits_{j,=,1}^{v(i)}({q}_{ij}+{q}_{ji})({k}_{j}^{{{{{mathrm{out}}}}}}+{k}_{j}^{{{{{mathrm{in}}}}}})/({s}_{i}^{{{{{mathrm{out}}}}}}+{s}_{i}^{{{{{mathrm{in}}}}}})$$

(13)

where (v(i)) is the number of the neighboring nodes of economy (i).

Given that the degree of economy (i) is (k), the average neighboring degree of all the nodes with degree (k) is defined as:

$$omega (k)=mathop{sum }limits_{iin {j|{k}_{j}^{{{{{mathrm{out}}}}}}+{k}_{j}^{{{{{mathrm{in}}}}}}=k}}^{{n}_{k}}{omega }_{i}/{n}_{k}$$

(14)

The network is a similar hybrid network if and only if (omega (k)) is monotonically increasing in (k).

Community detection: in order to better visualize the health-effect network, it is useful to partition the complex network, which consists of 181 nodes and more than 30,000 edges, into several submodules or communities, within which the nodes are densely linked but sparsely connected with the nodes in other communities. We apply the modularity maximization method introduced by Girvan and Newman29 to find the community partition of the health-effect network. The modularity of partition compares the compactness of the links inside communities with the links between communities. A higher value of modularity suggests better quality of community partition. The modularity (Q) in our network is defined by:

$$Q=frac{1}{2m}mathop{sum }limits_{i,=,1}^{n}mathop{sum }limits_{j,=,1}^{n}left[{w}_{ij}-frac{{p}_{i}{p}_{j}}{2m}right]delta ({c}_{i},{c}_{j})$$

(15)

where ({w}_{ij}={q}_{ij}+{q}_{ji}) is the amount of health effects connection between economy (i) and economy (j), ({p}_{i}=mathop{sum }nolimits_{j,=,1}^{n}{w}_{ij}) is the sum of health effects attached to economy (i), ({c}_{i}) is the community to which economy (i) is assigned, (delta ({c}_{i},{c}_{j})) is an indicator function which equals to 1 if ({c}_{i}={c}_{j}) and 0 otherwise, and (m=mathop{sum }nolimits_{i,=,1}^{n}mathop{sum }nolimits_{j,=,1}^{n}{w}_{ij}/2). We use this algorithm to implement the method and extract community structures of the network.

The GEOS-Chem is a global three-dimensional CTM of the atmospheric compositions (version 12.0.0, http://www.geos-chem.org), and includes detailed ozoneNOxVOCaerosol chemistry49. The model was run at a horizontal resolution of 2 latitude by 2.5 longitude driven by the NASA Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2) meteorological fields. In GEOS-Chem simulations, NH3 emissions from anthropogenic sources were from EDGAR v4.3.2 for 2012, and emissions from soil, vegetation, and the oceans were from the Global Emissions Inventory Activity inventory50. Other global anthropogenic emissions of NOx, SO2, CO, black carbon (BC), and organic carbon (OC) from EDGAR v4.3.2 and speciated volatile organic compounds emissions from the RETRO, overwritten by the default regional emissions, were adopted. Other natural emissions follow the configuration of Li et al.51.

A baseline simulation was conducted driven by global anthropogenic and natural emissions described above. To quantify the impacts of export-related agricultural NH3 emissions on particulate air pollution, sensitivity simulations with deducted NH3 emissions embodied in international trade for 181 countries were also performed. For each country, the trade-related fraction of NH3 emissions is assumed uniform, and this method has been applied widely in the previous studies21,22,52. The fractional contributions of export-related NH3 emissions to PM2.5 were determined on a 2 latitude by 2.5 longitude grid, due to model resolution. Then, these spatially varying fractions were multiplied by the 0.10.1 global PM2.5 concentrations from GBD 201328 to get estimated PM2.5 concentrations induced by export-related NH3 emissions. All the simulations were conducted from January to December 2012 after a 6-month model initialization (JulyDecember 2011).

GEOS-Chem aerosol simulations have been extensively evaluated using ground-based measurements worldwide2,11,22,53,54,55,56,57, including the USA, Europe, China, and India. These previous studies have shown that the GEOS-Chem model can reasonably the response of PM2.5 formation to emission changes as well as the observed concentrations of PM2.5 components. For example, Zhang et al.22 reported that major PM2.5 components simulated by GEOS-Chem have an R2 of 0.52~0.78 when compared with those observed values over the US, Europe, and East Asia, while it tends to underestimate (overestimate) BC (nitrate). The high bias of nitrate and low bias of BC are the common issues in the GEOS-Chem model11,58. It means that further improvement of the models capability in capturing the dynamics of the sulfatenitrateammonium aerosol systems is needed. Here we used GBD-based PM2.5 concentrations in 2012 to validate the model-simulated PM2.5 concentrations (Supplementary Fig.5). The simulated and GBD-based PM2.5 concentrations (with a resolution of 0.10.1) have a correlation coefficient of 0.6 and normalized mean bias of 0.5%. These two datasets compare reasonably well for most regions, including high values of over 80gm3 over eastern China and northern India, higher values in the eastern US than the western US, and high values of about 35gm3 over some European areas.

Methods of the GBD study28 are followed to estimate the premature deaths from ambient PM2.5 exposure. Here the impacts due to the four leading causes of death: ischemic heart disease, chronic obstructive pulmonary disease, cerebrovascular disease, and lung cancer are considered. In 2012, these four major diseases together accounted for 18.4 million deaths (35% of all-cause mortality). We estimate 3.54 million premature deaths attributable to PM2.5 in 2012, which agrees well with some previous studies such as the GBD59 (3.44 million deaths) and Zhang et al.22 (3.45 million deaths).

We applied IER functions developed by Burnett et al.59, which incorporated data from cohort studies of ambient PM2.5 pollution, household air pollution, and active and passive tobacco smoke, to fit the concentrationresponse relationship throughout the full distribution of ambient PM2.5 concentrations. Thus, high PM2.5 concentrations similar to those observed in China and India can be also accounted for. For each disease, the relative risk for mortality estimations for the all-age group was calculated as the following equation:

$${{{{{mathrm{RR}}}}}}(C) =1+alpha {1-exp [-gamma {(C-{C}_{0})}^{delta }]},{{{{{mathrm{for}}}}}},C , > , {C}_{0}\ ,{{{{{mathrm{RR}}}}}} =1,{{{{{mathrm{for}}}}}},Cle {C}_{0}$$

(16)

where C represents the annual mean PM2.5 concentration (on a 0.10.1 grid) in 2012, which was exponentially extrapolated from data for 2010 based on the GBD study by Brauer et al.28. The calibrated GBD PM2.5 data were estimated by a combination of satellite-based estimates, chemical transport model simulations, and ground measurements; C0 is the counterfactual concentration, representing a theoretical minimum-risk concentration above which there is evidence indicating health benefits of PM2.5 exposure reductions (range: 5.88.8gm3); and (alpha),(,gamma), and (delta) are parameters used to determine the overall shape of the concentrationresponse relationship, which are obtained from Burnett et al.59. We reported averaged mortality results using 1000 sets of coefficients and exposureresponse functions based on Monte Carlo simulations.

The grid-based (0.10.1) premature deaths attributed to ambient PM2.5 exposures were then estimated:

$${{{{{mathrm{Mort}}}}}}={y}_{0},times ,{{{{{mathrm{pop}}}}}},times ,(1-1/{{{{{mathrm{RR}}}}}})$$

(17)

where ({y}_{0}) is the Country-level baseline mortality for each disease for the all-age group from the Institute for Health Metrics and Evaluation (http://ghdx.healthdata.org/ihme_data) and pop is the population obtained from the Gridded Population of the World, version 3 at a resolution of 2.5min 2.5min, which was further aggregated to the same resolution of 0.10.1. The estimated 2012 population is linearly extrapolated from the 2010 and 2011 values.

Here, we estimated the mortality contribution from export-related agricultural NH3 emissions based on an assumption that the contribution of one source to the disease burden of PM2.5 is directly proportional to its share of PM2.5 concentration. The more recent GBD research, GBD MAPS, has demonstrated the scientific basis of such an assumption, which was adopted by early studies, such as another GBD study60 and Zhang et al.22. For a given country, the premature deaths from its export-related NH3 emissions can be calculated by multiplying its fractional contribution of export-related NH3 emissions to PM2.5 concentration by the total PM2.5 concentration-related mortalities for each 0.10.1 grid cell. The fractional contribution of export-related NH3 emissions to PM2.5 was estimated by the GEOS-Chem simulations.

Scenario analysis is conducted through import substitution and export transfer within the countries in each community.

Suppose that country i imports ({{{{{mathrm{I}}}}}}{{{{{mathrm{M}}}}}}_{ij}) agricultural products from country j, and the NH3 emission intensities of the two countries are ({{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{i}) and ({{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{j}). If ({{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{i} , < , {{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{j}), emission reduction can be achieved by country i to substitute the import ({{{{{mathrm{I}}}}}}{{{{{mathrm{M}}}}}}_{ij}) with its own production. Denote the amount substituted as ({{{{{mathrm{S}}}}}}{{{{{mathrm{T}}}}}}_{ij}). For any country i, the objective function of import substitution is to maximize the emission reduction, i.e., (mathop{sum }nolimits_{j=1,jne i}^{n}({{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{j}-{{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{i})times {{{{{mathrm{S}}}}}}{{{{{mathrm{T}}}}}}_{ij}). Nevertheless, complete substitution is unrealistic due to constraints in natural resources. Countries are unable to expand their agricultural production beyond their total capacity. We impose this constraint with their potential capacity in agricultural production (denoted as ({{{{{mathrm{A}}}}}}{{{{{mathrm{C}}}}}}_{i})), measured by the area of arable land (in hectare) multiplied by the value of agricultural products per hectare. Furthermore, farm goods produced in different countries are not perfectly substitutable either. To increase the reliability of the counterfactual analysis, we limit trade substitution within countries with annual average temperature (denoted as AVT) and precipitation (denoted as PCP) at similar conditions ((pm)5C for temperature and (pm)500mm for precipitation). Therefore, for country i, the objective function of import substitution is:

$$mathop{max }limits_{{{{{{{mathrm{S}}}}}}{{{{{mathrm{T}}}}}}_{ij}}}mathop{sum }limits_{j=1,jne i}^{n}({{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{j}-{{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{i})times {{{{{mathrm{S}}}}}}{{{{{mathrm{T}}}}}}_{ij}$$

(18)

$${{{{{mathrm{s.t.}}}}}}mathop{sum }limits_{j=1,jne i}^{n}{{{{{mathrm{S}}}}}}{{{{{mathrm{T}}}}}}_{ij}le {{{{{mathrm{A}}}}}}{{{{{mathrm{C}}}}}}_{i}$$

$$0le {{{{{mathrm{S}}}}}}{{{{{mathrm{T}}}}}}_{ij}le {{{{{mathrm{I}}}}}}{{{{{mathrm{M}}}}}}_{ij}$$

$$-5le {{{{{mathrm{AV}}}}}}{{{{{mathrm{T}}}}}}_{i}-{{{{{mathrm{AV}}}}}}{{{{{mathrm{T}}}}}}_{j}le 5$$

$$-500le {{{{{mathrm{PC}}}}}}{{{{{mathrm{P}}}}}}_{i}-{{{{{mathrm{PC}}}}}}{{{{{mathrm{P}}}}}}_{j}le 500$$

The import substitution involves two parties, i.e. the exporter and importer. More substantial emission reduction will be achieved if allowing the production of those exported products to be transferred to a third country with lower emission intensity than the original exporter and importer. The scenario of export transfer we consider is to minimize the total NH3 emissions in agricultural trade for each community. This is equivalent to taking total exports as given, reorganizing the production structure of exported goods within countries of the same community. Therefore, the objective function for export transfer is:

$$mathop{max }limits_{{{{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{{{{{mathrm{AF}}}}}}}}mathop{sum }limits_{i,=,1}^{n}{{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{{{{{mathrm{AF}}}}}}times {{{{{mathrm{E}}}}}}{{{{{mathrm{I}}}}}}_{i}$$

(19)

$${{{{{mathrm{s.t.}}}}}}mathop{sum }limits_{i,=,1}^{n}{{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{0}=mathop{sum }limits_{i,=,1}^{n}{{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{{{{{mathrm{AF}}}}}}$$

$${{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{{{{{mathrm{AF}}}}}}le {{{{{mathrm{A}}}}}}{{{{{mathrm{C}}}}}}_{i}+{{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{0}$$

$${{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{0}le mathop{sum }limits_{j,=,1}^{n}{I}_{ij}times {{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{j}^{{{{{mathrm{AF}}}}}}$$

where ({{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{0}) is the original export of country i before the adjustment of export transfer, ({{{{{mathrm{E}}}}}}{{{{{mathrm{X}}}}}}_{i}^{{{{{mathrm{AF}}}}}}) is the new export of country i after the adjustment, ({I}_{ij}) is a dummy variable which equals 1 if and only if (-5le {{{{{mathrm{AV}}}}}}{{{{{mathrm{T}}}}}}_{i}-{{{{{mathrm{AV}}}}}}{{{{{mathrm{T}}}}}}_{j}le 5) and (-500le {{{{{mathrm{PC}}}}}}{{{{{mathrm{P}}}}}}_{i}-{{{{{mathrm{PC}}}}}}{{{{{mathrm{P}}}}}}_{j}le 500). This first constraint keeps the total exports unchanged after the adjustment. The second constraint ensures the adjusted export of country i does not exceed its total production capacity, measured by the sum of potential capacity and original exports. The last constraint guarantees that the original export of country i is transferred to countries with similar annual temperature and precipitation.

We solve the above linear optimization problems to find the maximized emission reduction under export transfer and import substitution. The area of agricultural land and arable land is obtained from FAO (http://www.fao.org/faostat/en/#data/RL). Average annual temperature and rainfall are measured by the country-level averages from 1990 to 2012, which are available from the Climatic Research Unit of University of East Anglia (http://www.cru.uea.ac.uk/).

We develop several NH3 emission control scenarios from the production side (reducing N fertilizer overuse, deep placement of fertilizers, enhanced-efficiency fertilizers, and improved animal manure storage and disposal) and consumption side (reducing food loss and waste, and replacing beef consumption with soy consumption).

Estimating the potential of reducing N fertilizer overuse: N fertilizers are applied in unnecessarily high amounts in much of Asia (especially China), India, USA, and Europe. Overuse of N is much more severe in vegetables and fruits production, e.g., in China. Management of N during fruits and vegetable production has a much larger potential to improve compared to management during grain crop production. Mueller et al.61 estimated that nitrogen-fertilizer application on maize, wheat, and rice could decrease globally by 28% without impacting current yields. Considering the share of crop grain crop N in total N fertilizer use, we conservatively estimate that globally total N fertilizer use can be reduced by 14% to remove oversupply of nutrients and associated Nr losses. It is also noted that N fertilizer might not be reduced if there is not a regime to replace existing inorganic fertilizer applications.

Estimating the potential of fertilizer deep placement and enhanced-efficiency fertilizers in grain crops: deep placement of N fertilizers can make N less susceptible to NH3 volatilization and more available to grain crops. A global meta-analysis found 55% NH3 emission reduction achieved through deep placement62. A meta-analysis for China found 35% NH3 emission reduction achieved for wheat and rice systems and 70% for maize. We conservatively estimate that 55% of grain production-related NH3 emissions can be reduced through deep placement. We obtain the contribution of grain production to total crop NH3 from Paulot et al.63. Urease inhibitors reduce the hydrolysis rate of urea fertilizers thus reducing NH3 emission rates by 4070% depending on crop types and N application rates64. A global meta-analysis of field experiments reported 54% NH3 emission reduction according to 198 observations62. A meta-analysis for China found 35% NH3 emission reduction achieved for wheat and rice systems and 70% for maize65. A field research in Germany estimated 70% NH3 emission reduction66. We conservatively estimate that urease inhibitors will reduce NH3 emissions from grain crops by 54%.

Estimating the mitigation potential from livestock production: A global meta-analysis found the highest mitigation potential in dietary additive (3554%), urease inhibitor (2469%), manure acidification (8995%), and deep manure placement (9499%). Manure storage management could also significantly reduce NH3 emission by 7082%67. These mitigation measures should be taken simultaneously to effectively reduce NH3 emissions from livestock sectors. Despite great technological potential to reduce NH3 emissions, currently, manure around the world has been poorly managed68. For example, two-thirds of manure N produced in China are released as air pollutants69. We thus conservatively estimate that moderate improvements in manure management can reduce NH3 emissions by 35% and drastic improvements can reduce 70%.

Estimating the potential of eliminating food loss and waste: globally around 1/3 of food produced are discarded during the food supply chain, food retail and consumption processes70. Reducing food waste and loss thus provides the opportunity for reducing agricultural emissions, especially in developed regions such as Europe and North America which already have relatively effective production management. We estimate the NH3 mitigation potential of eliminating food waste and loss for each country using the following equation:

$${{Reductio}}{{n}}_{{{NH}}_{3}} ={{Reductio}}{{n}}_{{{cro}}{{p}}_{{{NH}}_{3}}}+{{Reductio}}{{n}}_{{{livestoc}}{{k}}_{{{NH}}_{3}}}\ ={{Baselin}}{{e}}_{{{cro}}{{p}}_{{{NH}}_{3}}}times {{Wast}}{{e}}_{{{LossRati}}{{o}}_{{crop}}}\ kern1pc+, {{Baselin}}{{e}}_{{{livestoc}}{{k}}_{{{NH}}_{3}}}times {{Waste}}_{{Loss}}_{{Rati}}{{o}}_{{meat}}$$

(20)

where ({{{{{mathrm{Baselin}}}}}}{{{{{mathrm{e}}}}}}_{{{{{{mathrm{cro}}}}}}{{{{{mathrm{p}}}}}}_{{{{{{mathrm{NH}}}}}}_{3}}}) and ({{{{{mathrm{Baselin}}}}}}{{{{{mathrm{e}}}}}}_{{{{{{mathrm{livestoc}}}}}}{{{{{mathrm{k}}}}}}_{{{{{{mathrm{NH}}}}}}_3}}) are national total NH3 emissions from N fertilizer application and livestock manure handling, respectively, for the year 2012 from the EDGAR inventory, ({{{{{mathrm{Waste}}}}}}_{{{{{mathrm{Loss}}}}}}_{{{{{mathrm{Rati}}}}}}{{{{{mathrm{o}}}}}}_{{{{{mathrm{crop}}}}}}) and ({{{{{mathrm{Waste}}}}}}_{{{{{mathrm{Loss}}}}}}_{{{{{mathrm{Rati}}}}}}{{{{{mathrm{o}}}}}}_{{{{{mathrm{meat}}}}}}) are the ratios of food loss and waste of cereal crops and animal meat products in this country. ({{{{{mathrm{Waste}}}}}}_{{{{{mathrm{Loss}}}}}}_{{{{{mathrm{Ratio}}}}}}) are estimated to include food loss and waste during agricultural production, postharvest handling and storage, processing and packaging, distribution and supermarket retail, and food consumption for each major region provided by Gustavsson et al.71.

Estimating the potential of dietary shifts: beef, compared to other animal meat products, have much heavier nitrogen and water use footprints and greenhouse gas emissions72. NH3 emissions from beef manure alone contributes to 30% of agricultural NH3 emissions globally63. Replacing beef protein with other animal meat products or plant-based soybean protein can help reduce NH3 emissions. Here we consider the dietary change strategy of reducing beef consumption by 20% and 50%, replacing that beef protein with soybean protein. The additional NH3 emissions brought by increased soybean cultivation are negligible. This is because the reduced animal feed production (mostly soy) should more than enough cover the increased soybean consumption, given that protein in animal feed does not end 100% in beef products. In addition, soybean is a N-fixing crop and requires moderate N fertilizer application and emits little NH3 emissions. We obtain the contribution of beef production to NH3 emissions from animal production in the US, Europe, China and the world from Paulot et al.63. Then on top of EDGAR NH3 emission inventory, we impose a 20% and a 50% decrease of beef NH3 emissions by country.

The MTFR scenario from GAINS model: we also applied the MTFR scenario by 2050 calculated from the GAINS (Greenhouse gas-Air pollution Interactions and Synergies) model (freely online from the website: https://gains.iiasa.ac.at/models/gains_models3.html), which is developed by the International Institute for Applied Systems Analysis (IIASA)36. The GAINS model has been widely applied for assessing strategies of ammonia emission abatement73,74. The MTFR scenario in GAINS model assumes implementation of best available measures ignoring political or economic constraints but considering technical applicability that might vary regionally.

Export-related NH3 emissions link local producers to global consumers along the entire supply chain. Based on multimodels, the uncertainty in this study mainly lies in the emissions inventory, economic data that includes the national accounts and interregional trade, and atmospheric transport model, and atmospheric model. Previous studies quantifying the uncertainty of national consumption-based carbon emissions (including imports and excluding the exports) are in the range 515%75 and 216%76. The comparable uncertainty range of production- and consumption-based accounts indicate that a major source of uncertainty of MRIO is mainly the emission inventory rather than the economic and trade data77. According to Crippa et al.27, the uncertainty of EDGAR NH3 emissions is within a factor of 23 for global major regions. This is mainly due to the uncertainty of adopted emission factors. Van Damme et al.3 has shown that the EDGAR inventory is able to capture NH3 emissions in the large source regions while it fails to capture strong point sources. Other observation-based constraints of NH3 emissions also have confirmed the validation of EDGAR NH3 inventory in China, USA, and Europe63,78. Future studies to further improve the accuracy of NH3 emissions in global inventories are urgently called for.

Further information on research design is available in theNature Research Reporting Summary linked to this article.

Read more:

Mitigation potential of global ammonia emissions and related health impacts in the trade network - Nature.com