Societal benefits of halving agricultural ammonia emissions in China far exceed the abatement costs

Mitigating agricultural ammonia (NH3) emissions in China is urgently needed to avoid further damage to human and ecosystem health. Effective and feasible mitigation strategies hinge on integrated knowledge of the mitigation potential of NH3 emissions and the associated economic costs and societal benefits. Here we present a comprehensive analysis of marginal abatement costs and societal benefits for NH3 mitigation in China. The technical mitigation potential of agricultural NH3 emissions is 38–67% (4.0–7.1 Tg N) with implementation costs estimated at US$ 6–11 billion. These costs are much lower than estimates of the overall societal benefits at US$ 18–42 billion. Avoiding unnecessary fertilizer use and protein-rich animal feed could provide 30% of this mitigation potential without additional abatement costs or decreases in agricultural productivity. Optimizing human diets with less animal-derived products offers further potential for NH3 reduction of 12% by 2050.


Supplementary Figure 3 The framework of input drivers in scenario analysis.
White rectangles with blue edges represent the main input parameters, black arrows stand for data flow, green rounded rectangles represent different scenarios, while orange rounded rectangles represent the corresponding activity data output.

C9
Soil amendment (AM) rice, wheat, maize, vegetable, fruits 31-54% 11,13 Note: Chinese specific data are preferred to improve the accuracy of assessment when many studies are available. When several studies are referenced, the range of abatement efficacy refers to mean values provided in the studies (meta-analysis results). Note: Chinese specific data are preferred to improve the accuracy of assessment when many studies are available. When several studies are referenced, the range of abatement efficacy refers to mean values provided in the studies (meta-analysis results).

Supplementary
CRF: 2500 CNY/t, NBPT:150CNY/kg, DCD:8CNY/kg China's total production of EENF was 21 million tons by 2015, with a total promotion area of 33 million hectares. 34,35 C3 OF increase the recycling of organic wastes to substitute synthetic fertilizer. organic fertilizers are incorporated into soil as basal fertilizers, the proportion of organic substitution for chemical fertilizer varies in different crops and regions, typically organic manure as basal fertilizer providing 30% of N nutrients to crops and 50% to fruits and vegetables. N content in typical organic manure fertilizers stands at 1.2%. average market price for organic fertilizer :1100 CNY/t, 33,36,37 C4 OPR this measure calls for a direct reduction in N fertilizer use for certain crops. The optimal N rate was 28% lower than the traditional N rate of the studies included in meta-analysis.
Reductions in crop-specific N fertilizer rates are based on recommended rates for main crops (see supplementary  Table 14) we assume no additional cost for OPR. the fertilizer utilization efficiency is assumed to increase to more than 40%. 4

C5
RBF the minimum percentage of BF reduction is 10% to avoid over basal N fertilization and increased N uptake along crops growth, thus reducing NH3 emission, N leaching, and runoff.
Assume 10% more labour input is required. Minimum labour cost :17CNY/hour 22,38,39 C6 SF compared to single application, split the total amount of N fertilizer into 3-4 applications for basal fertilization and top dressing, and shift from mid-season drainage to intermittent irrigation more labour input is required for corn, wheat, rice, fruits split fertilization. Minimum labour cost :17CNY/hour 4,37 C7 DP deep placement significantly decreased floodwater NH4 + -N concentration and NH3 volatilization compared to surface application, the minimum depth of the deep placement of fertilizer N was 5 cm below the soil surface, usually supplied at 20cm depth from soil surface below plants.
Increased machine investment and operation costs for maize/cotton/fruits deep fertilization. Average fertilizer applicator price: 22000 CNY per machine. 36,40 C8 IR Irrigation with at least 5 mm water immediately following fertilizer application has been shown to reduce NH3 emission by up to 70%. The irrigation amount and time should be reasonably determined to achieve the integration of water and fertilizer management.
High-efficient irrigation systems allow for labour and water savings in vegetable, fruits and cotton fields. Subsurface drip irrigation system costs include 15000 CNY/ha initial investment and installation cost (lifespan=10 years) and annual maintenance and renewal cost of smaller diameter polytube at 1500 CNY/ha (10%), 16 AGT adjust the grazing time for grazing animals according season: shorter grazing period in summer, prolong the grazing time in winter. During grazing, less NH3 was emitted because manure, especially the urine from grazing animals often bind relatively quickly with the soil and do not volatilize as much as in confined operations.
Assume only little additional labour cost generated since free grazing on pasture is most common in Chinese grassland systems. Extend the grazing time for grazing animals. Cost savings based on the reduced need for building floor scraping and slurry handling, together with reduced silage production costs 31 Supplementary Grazing L18 0 ---33% Note: '+', '-' and '0' indicate an increase, decrease and no change in emissions after application of control option, "NA" means not applicable or not available; a CH 4 emissions from rice cultivation, enteric fermentation and manure management; b N 2 O emissions include both direct and indirect N 2 O emissions from the application of synthetic fertilizers, organic manure and crop residues; c The overall GHG emissions are presented as kg CO2-eq, using the default values of 298 kg CO 2 -eq for N 2 O emissions and 34 kg CO 2 -eq for CH 4 emissions (IPCC, 2013) 69 . Note: the prediction of food consumption under BAU scenario is based on current high-income countries diet structure (animal food N ratio=60%); the prediction of food consumption under DIET scenario is based on the Dietary guidelines for Chinese residents 99 (animal food N ration=40%).

Supplementary Note 1. Selection of NH 3 mitigation measures
Agricultural NH3 emissions in this study refer to the NH3 emissions from crop and livestock production systems. The mitigation potentials of livestock production and crop farming are described separately after taking into account their interaction through manure recycling to fields. Given the inherent differences among the various crops and livestock types, here we assess NH3 mitigation potential by crop and animal types to explore the feasible national emission reduction target 2 . For example, options for reducing NH3 losses from poultry housing and manure storage focus on rapidly drying and transferring the manure to the storage area. Measures for slurry storage from pigs and cattle generally aim at minimizing contact between manure surface and air. According to the agricultural characteristics and farming practice in China, 10 main crops and 10 animal types in China were selected for sector specific NH3 mitigation assessment.
In fact, not all measures can be applied to 100%, some techniques are restricted in applicability by their effectiveness or by practical limitation 14 . These limitations may be of very different natures, including local climate, soil conditions (pH, slope), farm size, financial and technical limitations 14 . Therefore, implementation of NH3 abatement measures should follow their applicability and be adjusted to local conditions 2 . Supplementary Tables 3 and 4 summarize the practical consideration of NH3 mitigation options applied to cropland and animal production in China. National policy and plans have been taken into consideration to determine the parameters or implementation level of different options in the future 33,42,47,55,[102][103][104] . Supplementary Table 5 lists the current farming practice in China and the applicability of the selected measures.
Note that some individual options (e.g. manure covered storage) do not really remove NH3 but merely preserve N in the manure, which may be emitted at later stages, (e.g. the stage of application 105 ). Therefore, these measures should be used in conjunction with other options to enhance the mitigation efficiency by positive measure interactions, e.g. manure treatment should generally be coupled with improved manure application methods to avoid N loss during application. Supplementary Table 6 summarizes the possible effects of NH3 mitigation measures on GHG (N2O and CH4) emissions. Some mitigation options for NH3 reduction may induce GHG emissions at certain stage. For example, the application of manure and straw cover, or injection of liquid fertilizers to reduce NH3 volatilization may lead to increased emissions of methane (CH4) by the anaerobic decomposition of manure, and of nitrous oxide (N2O) through nitrification and denitrification in livestock manure and urine 30 . However, optimal combinations of different measures could offset the stagespecific side-effects and achieve both NH3 mitigation and total GHG mitigation 28,66,67 . The selected packages of measures in Supplementary Table 7 could improve, or at least maintain, crop yields or animal productivity according to the results of existing studies and meta-analyses. However, due to limited Chinese-specific experimental data of yield change, the economic benefits of crop and animal productivity improvement are not quantified in the cost-benefit analysis.

Supplementary Note 2. Reduction efficiency of mitigation measures for crop farming
The abatement effect of single option is easy to access due to a lot of previous field experiments, measurements and meta-analyses 2 . However, in practice, implementation of a single option alone has limitations to its abatement effectiveness and cost. For example, the use of NBPT (a kind of Urease inhibitor) alone may not be sufficiently effective in inhibiting NH3 emissions, while NBPT with recommended N fertilizer type and optimal irrigation management may achieve the desired results of decreasing N losses and increasing N use efficiency 106 . Note that the abatement rate and cost-effectiveness of individual measures may change when applied with other measures jointly, so the actual abatement efficiency and cost-effectiveness of the combined packages of mitigation options need to be identified or recalibrated to explore the maximum mitigation potential, and to inform the best cost-effective abatement strategies 2 . Due to limited information on the effectiveness of combinations of measures for NH3 mitigation in cropping systems, here the interactions are addressed by assigning implementation priorities to selected mitigation options. For example, if measure C1 and C4 allow N application rates to decrease from 200 kg ha -1 to 100 kg ha -1 , the mitigation effect of deep placement (measure C7) will be based on the N rate of 200 kg ha -1 . The potential of adding organic manure (measure C3) to rice paddies is quantified under the improved irrigation by measure C8. If there are no supporting data available from previous field experiments or meta-analysis results for quantitative analysis of the NH3 mitigation potential in cropping system, the combined efficiency for a package of two (A+B) and three (A+B+C) mitigation options was assessed following the below Equation (1-2): where A, B, C are the control technologies included in the combination, , , is the reduction efficiency of a given mitigation option.

Supplementary Note 3. Reduction efficiency of mitigation measures for livestock production
NH3 emissions from livestock production are based on the concept of the N mass flow balance in livestock system (Supplementary Figure 2). NH3 emission at each stage of the management systems, including housing, storage, grazing and application to the land was calculated following Equation (3)(4)(5)(6). It is therefore clear that the volume of NH3 emissions from the later stages are affected by emissions from previous stages. For example, reducing NH3 volatilization at previous stage, such as covered storage stage, will likely create a more N-rich waste and therefore greater potential of NH3 loss at a later stage, such as land application 107 . It makes sense to adopt low emission techniques during manure application. Thus, optimal package of mitigation options that covers all stages N flow need to be designed to systematically reduce NH3 emissions.
In this study, results of meta-analyses or currently available research about combinations of multiple options in China are mainly referred to explore the maximum mitigation potential. For instance, combined NH3 mitigation options on feed, housing, manure storage, and land application could reduce the farm-scale NH3 emission by up to 89.3% in broiler production systems 89 . Combined effect of the selected measure at a given stage on NH3 emission factor was estimated following the Equation (7-10) when data is not available.
where 1,2,3,4 ′ is the new NH3-N emission factors at specific emission stages, 1,2,3,4 is the NH3 reduction efficiency at specific emission stages. For livestock production combinations of measures within a stage may interact, including overlapping application of measures with similar effects or subordinating relationships. For example, during animal housing, optimal bedding, floor adaptation and air purification could be applied separately or jointly. Information about the combined abatement efficiency within same stage is mainly collected from current research (i.e. field experiment, model simulation and meta-analysis) 20,30 .

Supplementary Note 4. Data source and projection of future agricultural activity under different mitigation pathways
The years 2000-2015 are used as the reference years in this study and 2020-2050 in five-year intervals are set as the target years. According to the definition, classification and applicability of previously selected mitigation options, five emission scenarios were proposed in this study and combined with CHANS and GAINS models to explore the mitigation potential and the costs and benefits of each mitigation pathway in the next 30 years (2020-2050).
For data sources, historical data for 2000-2015 such as population, urbanization, gross domestic product (GDP), land use, fertilizer use, crop/livestock production, and resource consumption in the agricultural sector of China, were mainly collected from the National Bureau of Statistics of China (NBSC) 100 and Food and Agriculture Organization of the United Nations (FAO) statistics 108 . Prediction of future agricultural activity data for 2020-2050 is based on the N demand-supply balance framework (Supplementary Figure 3) i.e. N supply from crop and livestock production should meet the demands of human consumption. This framework combines various input drivers and parameters with previous scenario studies (e.g. Gu et al. (2015); Ma et al.(2019) 109 ). National plans and regulations (e.g. Zero Increase Action Plan on Fertilizer Use by 2020 110 ) are also considered in the prediction.
We first predicted the future human population, GDP, diet preference and urbanization; then, the demands of crop and animal feed were estimated; third, the required crop and livestock production, cropping area, fertilizer use and manure production were calculated for subsequent scenario analysis.
Most future prediction of basic activity data were directly derived from results of a series of models and scenarios derived from previous researches 109,111,112 where various sources were used, including national plans or targets (National population development plan (2016-2030) 113 , National plan for agricultural modernization (2016-2020) 114 ; National plan for sustainable agricultural development(2015-2030) 55 ), which are the most reliable data sources in China. Meanwhile, associated data from international organizations including the FAO 108, 115 , the World Bank 96 , the IIASA 116 were also collected to support the projection of future agricultural activity.
To make the scenarios clearer we extracted the important activity level indicators, coefficients and parameters shown in Supplementary Tables 8 -19. Under the BAU scenario we assume a population of 1.38 billion with an urbanization level of 75% and a PGDP of US$ 39,900 in 2050. The projection of the Chinese population in the future takes into account the universal two-child policy adopted in 2015 112,117 . The predictions of GDP and PGDP in China are based on the past stable economic growth in China 96, 109, 111 and socio-economic models and data in IPCC scenarios 118 . Human food consumption would increase due to the population and PGDP growth 51,119 . The projections of sowing area and crop productivity in China during 2020-2050 are based on requirements of the plant-based food and animal-based food consumption 120 . The forecast of fertilizer use is based on the fertilizer policy issued in 2015 "Zero-growth Action Plan" for chemical fertilizer application by 2020 55 . The NUEs and nutrient recycling rates of agricultural subsystems (e.g., cropland and livestock) are assumed to remain current level under BAU scenario. International trade (e.g. grain imports) are assumed to remain constant during 2020-2050.
It should be noted that there are numerous uncertainties in this scenario-based analysis of the future NH3 emissions and mitigation in China. Although our estimates in this study are based on current best available information, the accuracy and robustness of our estimates are still limited by the quality of the data, the applicability of the mitigation options and the validity of the assumptions made in terms of future activity levels. Estimates of the combined effects of a group of mitigation options could also be an important source of uncertainty because the practical NH3 removal efficiencies for NH3 and implementation levels in the future are unattainable and the lack of validation and optimization will also bring large uncertainty in the predictions. In addition, considerable uncertainties are also recognized regarding the estimation of abatement costs and societal benefits of NH3 mitigation. Such uncertainties are inevitable and need to be addressed and assessed with more bottom-up research in support of policies on agricultural NH3 emissions.

Supplementary Note 5. WRF/CMAQ model simulation of PM 2.5 concentration to NH 3 emission control
A detailed sensitivity simulation of PM2.5 reduction by NH3 mitigation using the WRF-CMAQ model in China was conducted by Xu et al (2017) 1 . The main description and validation of WRF-CMAQ simulation are listed as follows: (1) Simulation period: January, April, July and October of 2015, and the time interval of the output is 1h.
(2) Simulation area: the CMAQ model adopts the Lambert projection coordinate system, the central longitude is 103°E, the central latitude is 37°N, the two parallel latitudes are 25°N and 40°N, respectively. The horizontal simulation range for the X direction is -2690 to 2690 km, and for the Y direction it is -2150 to 2150 km with a grid spacing of 20 km. The whole of China was divided into 270×216 grids. A total of 14 pressure layers were set in the vertical direction, and the layer spacing gradually increased from bottom to top.
(3) Meteorological simulation: the meteorological field required by the CMAQ model is provided by the mesoscale meteorological model WRF. The WRF model and the CMAQ model adopt the same simulation period and spatial projection coordinate system, but the simulation range of WRF is larger than that of CMAQ. The horizontal simulation range for the X direction was -3600 km to 3600 km, and for the Y direction it was -2520 km to 2520 km, with a grid spacing of 20 km. (4) Parameterization scheme of WRF-CMAQ was summarized in Supplementary Table 20.
The performance of WRF-CMAQ model simulation in simulating PM2.5 chemical components at different geographic locations and times have been statistically evaluated in Xu et al. (2017), which generally shows good correlation of simulated ambient PM2.5 chemical composition relative to observed values. The correlation coefficient between observed and simulated annual average PM2.5 concentration was 0.82 (n=302, p<0.05), with normalized mean biases (NMB) of -21.67 and Normalized mean error (NME) of 29.49.
Responses of different compositions of PM2.5 to NH3 mitigation was investigated by WRF-CMAQ model simulation. Results showed that reducing NH3 emissions could significantly decrease annual average concentration of nitrate, ammonium and PM2.5 concentrations (Supplementary Figure 4). Sulfate concentration is not sensitive to NH3 mitigation because of its low vapor pressure and thermodynamic stable nature 121,122 . In contrast, nitrate concentration is very sensitive to changes in NH3 emissions because its high saturated vapor pressure and thermal stability 123