Interactive comment on “ Agricultural ammonia emissions in China : reconciling bottom-up and top-down estimates ” by Lin Zhang et al

Comment: Zhang et al have done a terrific job using models and observations to improve understanding of ammonia emissions in China. Not only do they do a top-down analysis, using the GEOS-CHEM adjoint constrained by TES column NH3 measurements, to improve the seasonal and spatial variability in NH3 emissions, they then do a very thorough job improving past bottom-up inventories through careful analysis of fertilization practices and animal emissions. Combined, these make for a very strong paper – one of the best I have reviewed in some time. I recommend the authors attend to a few comments in revising the manuscript: Response: We thank Prof. Collett for the valuable comments. We have addressed all of them in the revised manuscript, and please see the itemized responses below.

1 Introduction Ammonia (NH 3 ) and its aerosol-phase product ammonium (NH + 4 ) exert important influences on atmospheric chemistry and biodiversity.They contribute to formation of fine particulate matter (PM) that has adverse effects on air quality and visibility (Park et al., 2004;Lelieveld et al., 2015) and causes a cooling climatic forcing (Martin et al., 2004;Henze et al., 2012).Their deposition to nonagricultural ecosystems can further lead to soil acidification and eutrophication (Stevens et al., 2004;Bowman et al., 2008).Quantifying these environmental consequences requires accurate knowledge of NH 3 sources, which are mainly associated with agricultural farming and livestock production (Bouwman et al., 1997).China, due to its intensive agricultural activities, is one of the largest NH 3 emitting countries in the world.However, current estimates of Chinese agricultural NH 3 emissions differ by more than a factor of 2 (see Sect. 2).Here we aim to better constrain agricultural NH 3 emissions in China using available NH 3 concentration and wet deposition flux measurements interpreted by the GEOS-Chem chemical transport model (CTM) and its adjoint.
As the main alkaline gas in the atmosphere, NH 3 reacts with sulfuric acid (H 2 SO 4 ) and nitric acid (HNO 3 ), which are produced by the oxidation of sulfur dioxide (SO 2 ) and nitrogen oxides (NO x ), to form ammonium sulfate and ammonium nitrate aerosols, respectively.These secondary inorganic aerosols account for 40-57 % of the fine PM concentrations in eastern China (Yang et al., 2011;Huang et al., 2014).Recent studies also highlighted the possible role of NH 3 in neutralizing aerosol pH that can strongly enhance formation of sulfate through heterogeneous oxidation of SO 2 (Wang et al., 2016;Cheng et al., 2016;Paulot et al., 2016).All this evidence leads to increasing concerns that the effectiveness of SO 2 and NO x emission controls on fine PM pollution over China may be undermined by unregulated NH 3 emissions (Wang et al., 2013;Fu et al., 2017).
Emissions of NH 3 are generally estimated from bottomup statistical methods or process-based models by considering activity data and emission factors (emissions per unit activity) of all possible sources.The two most important NH 3 sources are the application of synthetic fertilizers generated by the Haber-Bosch process (Erisman et al., 2008) and livestock production (volatilization of NH 3 from animal excreta).They together are estimated to contribute 57 % of global NH 3 emissions (Bouwman et al., 1997) and 80 % in Asia (Streets et al., 2003;Kurokawa et al., 2013).Bottomup NH 3 emission estimates highly depend on the accuracy of activity data and emission factors that require detailed spatial and temporal information on local agricultural practices and environmental conditions as will be discussed in Sect. 2.
Inverse modeling methods provide top-down emission estimates through optimizing comparisons of model simulations with measurements (Gilliland et al., 2003(Gilliland et al., , 2006;;Pinder et al., 2006;Zhu et al., 2013;Paulot et al., 2014).Top-down estimates of NH 3 emissions over China have been rare due to limited concentration or flux measurements of reduced nitrogen (NH x = gaseous NH 3 + aerosol NH + 4 ).A previous inversion study by Paulot et al. (2014) used NH + 4 wet deposition flux measurements from the Acid Deposition Monitoring Network in East Asia (EANET) that only included two sites in China over the study period.Satellite observations of atmospheric NH 3 concentration have been emerging in recent years.These satellite instruments include the Tropospheric Emission Spectrometer (TES) (Beer et al., 2008;Shephard et al., 2011), the Infrared Atmospheric Sounding Interferometer (IASI) (Clarisse et al., 2009;Van Damme et al., 2015), the Atmospheric Infrared Sounder (AIRS) (Warner et al., 2016(Warner et al., , 2017)), and the Cross-track Infrared Sounder (CrIS) (Shephard and Cady-Pereira, 2015), providing increasingly rich datasets to understand the spatial and temporal variability of NH 3 in the atmosphere.
In this study, we apply TES satellite observations of NH 3 column concentration to provide top-down constraints on NH 3 emissions in China for the year 2008 using the GEOS-Chem adjoint model at the 1/2 • × 2/3 • horizontal resolution.In order to reconcile with the bottom-up estimates and to better understand inversion results, we construct a new bottomup inventory of Chinese agricultural NH 3 emissions by using more practical fertilizer application rates and timing over different crop categories and by better considering the seasonal variability of emission factors.We further evaluate the topdown and the improved bottom-up Chinese NH 3 emissions using an ensemble of surface measurements of NH 3 gas concentration and NH + 4 wet deposition flux.  .The factor of 2 difference is not likely due to the different base years.Analyses of historical NH 3 emissions in China show relatively stable or weak increasing trends (less than 3 % yr −1 ) since 2000 (Xu et al., 2016;Kang et al., 2016), consistent with trends in atmospheric NH 3 concentration observed from satellites (Warner et al., 2017;Fu et al., 2017).Fertilizer application and livestock waste are the two largest NH 3 sources, together accounting for more than 82 % of the total anthropogenic emissions over China (Table 1).However, considerable differences exist in their emission totals and relative importance.Estimates of NH 3 emissions from fertilizer application in China range from 1.82 Tg a −1 in 2004 (Li and Li, 2012) to 9.82 Tg a −1 in 2010 (Zhao et al.,  2013).All these emission estimates are calculated by multiplying fertilizer use amounts with corresponding volatilization rates (emission factors), except for Fu et al. (2015) that considered bi-directional NH 3 fluxes over an agricultural model.Large differences are mainly due to uncertainties in NH 3 emission factors that are highly sensitive to fertilizer types, local soil, and meteorological properties (Bouwman et al., 2002;Søgaard et al., 2002).NH 3 emissions from livestock waste also range from 2.88 to 8.82 Tg a −1 .An important uncertainty is also attributed to emission factors from livestock waste that heavily relied on European-based measurements in earlier estimates (Streets et al., 2003).Moreover, some estimates, e.g., Yan et al. (2003) (2.48 Tg a −1 for 1995) and Xu et al. (2016) (3.8 Tg a −1 for 2008), only accounted for livestock manure spreading to cropland and omitted contributions from animal housing and manure storage.NH 3 from humans, including latrines and human perspiration and respiration, is another source with considerable differences (0.12-1.81 Tg a −1 ).The major component of this source is from rural excrement stored in roughly constructed latrines without sewage service, with uncertainties in estimates of rural population and the associated NH 3 emission factor.Other sources, such as agricultural burning, transportation, waste disposal, and industry also contribute 0.14-2.8Tg NH 3 a −1 , depending on inclusions of different source sectors in the emission inventories.For example, Dong et al. (2010) (0.14 Tg a −1 ) only estimated NH 3 emitted from the chemical industry.These sources are relatively small compared to agricultural sources of fertilizer ap-plication and livestock waste at the national scale; however, recent studies show that fuel combustion, (Pan et al., 2016), transportation (Chang et al., 2016;Sun et al., 2017), or local green space (Teng et al., 2017) can be dominant sources of NH 3 in the urban atmosphere.
We find substantial differences in spatial and seasonal variations of NH 3 emissions among the inventories.Figure 1 compares spatial distributions of anthropogenic NH 3 emissions in China from four commonly used inventories: the Regional Emission in Asia (REAS v2) inventory (Kurokawa et al., 2013), the Streets et al. (2003) inventory, the Huang et al. (2012) inventory, and the Emission Database for Global Atmospheric Research (EDGAR) (Olivier and Berdowski, 2001).Although they all show higher NH 3 emission rates in the east than the west, with the highest emissions occurring over North China, there are distinct regional differences of 50-200 %.In particular, EDGAR shows much more evenly distributed NH 3 emission rates spreading over China than the other three inventories.
Figure 1 also shows seasonal variations in these Chinese NH 3 emissions.Some inventories such as EDGAR and REAS v2 do not consider the seasonality of NH 3 emissions due to a lack of reliable information.More recent estimates account for information on the timing of fertilizer application and influences of meteorology on NH 3 emission factors.We can see that Huang et al. (2012) suggest a weak summer peak in Chinese NH 3 emissions, while the MASAGE inventory (Paulot et al., 2014) indicates largest emissions in April and July.NH 3 emission estimates of Streets et al. (2003) have a strong peak in June and are much higher than Huang 3 Model description

The GEOS-Chem model
Here we will use the GEOS-Chem CTM and its adjoint to simulate the sources and sinks of NH 3 over China.GEOS-Chem is a global 3-D tropospheric chemistry model (http: //geos-chem.org)driven by assimilated meteorological data from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO).The GEOS-5 meteorological data have a horizontal resolution of 1/2 • latitude × 2/3 • longitude and a temporal resolution of 3 h (1 h for surface variables).We apply here a one-way nested-grid version of GEOS-Chem with the native 1/2 • × 2/3 • horizontal resolution over East Asia (70-140 • E, 15-55 • N) and 2 • × 2.5 • over the rest of the world (Wang et al., 2004;Chen et al., 2009).
The model simulates a detailed tropospheric ozone-NO x -hydrocarbon-aerosol chemistry as described by Park et al. (2004) and Mao et al. (2010).NH 3 in the atmosphere is partitioned to gas and aerosol phases based on the Regional Particulate Model Aerosol Reacting System (RPMARES) thermodynamic equilibrium model (Binkowski and Roselle, 2003).NH 3 is thermodynamically favored to combine with H 2 SO 4 to form ammonium bisulfate and ammonium sulfate, and excessive NH 3 can react with HNO 3 to form ammonium nitrate.GEOS-Chem simulations of secondary inorganic aerosols (ammonium, sulfate, and nitrate) over China have been validated by Wang et al. (2013) and Li et al. (2016) recently; both show high sensitivity of simulated nitrate concentrations to NH 3 emissions.GEOS-Chem also includes the ISORROPIA II thermodynamic equilibrium model (Fountoukis and Nenes, 2007).We find that differences in simulated monthly NH 3 concentrations over China by the two equilibrium models are less than 5 %, and RPMARES runs about 30 % faster in the GEOS-Chem adjoint.
The model wet deposition scheme is described by Liu et al. (2001) with updates from Amos et al. (2012) for soluble gases and Wang et al. (2011) for aerosols.It includes convective updraft scavenging as well as largescale precipitation rainout and washout.Uptake of gaseous NH 3 is estimated following Henry's law in warm clouds (T > 268 K), using a retention efficiency of 0.05 in mixedphase clouds (248 < T < 268 K) and zero efficiency in cold clouds (T < 248 K) (Wang et al., 2008), while aerosol NH + 4 is fully incorporated in all clouds.Dry deposition calculation follows a standard resistance-in-series model as described by (Wesely, 1989) for gases and Zhang et al. (2001) for aerosols.
We use the EDGAR global anthropogenic emissions overwritten by regional emission inventories including the US EPA 2005 National Emissions Inventory (NEI-2005), the European Monitoring and Evaluation Programme (EMEP) emissions, and the Canada Criteria Air Contaminants (CAC) inventory.Asian anthropogenic emissions are from Zhang et al. (2009) except for NH 3 as described below.These global and regional inventories are scaled to the simulation year of 2008 using the energy statistics as implemented by van Donkelaar et al. (2008).For the prior NH 3 emissions, we use the REAS v2 emission inventory that does not consider any seasonal variation (Kurokawa et al., 2013) so that the inverted emission seasonality is solely from satellite observations.We also follow Zhu et al. (2013) and increase NH 3 emissions from fertilizer use and livestock by 90 % in the daytime and reduce them by 90 % at night to account for their diurnal variability.Natural sources of NH 3 and NO x from biomass burning, soil, and lightning follow the settings of Zhao et al. (2015) and are relatively small over China (0.56 Tg NH 3 a −1 ; Zhao et al., 2017).

The GEOS-Chem adjoint
We use the adjoint of GEOS-Chem to optimize Chinese NH 3 emissions through assimilation of TES NH 3 column measurements as will be discussed in the next section.The model adjoint is first described by Henze et al. (2007) for aerosols and Kopacz et al. (2009) for carbon monoxide.It has been highly validated and applied in studies to analyze aerosol sensitivities and constrain aerosol sources in the United States (Henze et al., 2009;Zhu et al., 2013;Mao et al., 2015) and China (Kharol et al., 2013;Zhang et al., 2015Zhang et al., , 2016;;Qu et al., 2017).
The emission optimization is conducted by minimizing the cost function (J ), defined as where y obs is the vector of satellite observations, x is the vector of NH 3 emissions in the model, Y (x) represents simulated NH 3 concentration for comparison with y obs , x a is the vector of a priori emissions, σ is the vector of scaling factors (x/x a ) for optimizing, and S a and S e are the a priori and observational error covariance matrices, respectively.The adjoint model computes the gradient of the cost function (∇ x J ) numerically, and applies the quasi-Newton L-BFGS-B algorithm (Byrd et al., 1995) to minimize the cost function iteratively.It usually takes about 15 iterations to reach the convergence, identified as the iteration when the cost function decreases by less than 2 % relative to the prior one.To lower the computational expenses, we follow the approach of Paulot et al. (2014) and use an offline NH x (NH 3 + NH + 4 ) simulation for the adjoint inversion that only calculates the physical and chemical transformation of NH x driven by hourly simulated sulfate and total nitrate (HNO 3 + NO − 3 ) concentrations archived from the standard simulation (Sect.3.1).This approach would induce errors by not accounting for changes in total nitrate concentrations when NH 3 emissions change (Paulot et al., 2014).We find that a 30 % increase of Asian NH 3 emissions would increase the total nitrate concentration by about 10 %, but deviations of NH 3 concentrations in the offline simulation from the standard simulation due to the NH 3 emission change are less than 3 % over China.

Adjoint inversion of Chinese NH 3 emissions with satellite observations
We use satellite observations of NH 3 column concentration over China retrieved from TES, a high spectral resolution Fourier transform infrared spectrometer aboard the NASA Aura satellite launched in July 2004 (Beer et al., 2006).TES observations have a spatial resolution of 5 km × 8 km at nadir with a local crossing time of 01:30 and 13:30 LT and global coverage achieved every 16 days (Beer et al., 2008;Shephard et al., 2011).TES retrievals of atmospheric NH 3 concentration are estimated by the optimal estimation method as described in Shephard et al. (2011).We filter the TES NH 3 retrievals by only using daytime observations with degree of freedom for signal (DOFS) greater than 0.1.We also correct the positive biases (0.04-1.0 ppbv in the lower troposphere depending on the a priori profile type used in the retrieval) in TES NH 3 retrievals following Zhu et al. (2013)  October.We do not analyze the late fall and winter months (November-February) as the valid TES observations become very limited, which hinders a reliable emission inversion in those months.As can be seen in Fig. 2, the largest NH 3 column concentrations are observed over North China, reflecting intensive agricultural activities over this area.High NH 3 concentrations are also observed over the province of Xinjiang in Northwest China, which are likely emitted from animal grazing and remain mainly in gas phase due to lower NO x and SO 2 emissions to generate acids there.TES observations show a strong seasonality, with the national averaged NH 3 column concentration a factor of 2 higher in summer than spring, similar to other satellite NH 3 observations retrieved from AIRS (Warner et al., 2017) and IASI (Van Damme et al., 2015).We now assimilate TES-observed NH 3 columns into the model through minimizing the cost function defined by Eq. (1).The emission optimization is conducted for each month of March-October 2008.Model results are sampled along the satellite orbits and then processed with TES a priori profiles and averaging kernel matrices as a necessary process for comparing with satellite retrievals based on the optimal estimation method (L.Zhang et al., 2010;Zhu et al., 2013).For the error covariance matrices, we assume the a priori error covariance (S a ) to be diagonal and the uncertainties to be 100 %.The observational error covariance (S obs ) is assumed to be 40 % of averaged values of observations and model results, accounting for uncertainties in both observations and the model.We have also conducted sensitivity inversions by using different S a (50 and 200 %) or S e (20 and 60 %) for the month of July.
Figure 3 shows differences between TES-observed and model-simulated NH 3 column concentrations with both the prior and optimized NH 3 emissions over China for April, July, and October.It also shows the correction ratios of optimized emissions over the prior emissions (REAS v2 in Fig. 1).We can see that with the prior Chinese NH 3 emissions, model results largely underestimate TES observations in July with a mean bias of −47 %, while they overestimate observations in April and October by 10-35 %.Model results with optimized NH 3 emissions improve comparison correlation coefficients (from 0.25-0.55 to 0.49-0.63)and significantly reduce the mean biases (−8-0 %).For these inversions, the cost functions are generally reduced by 35-45 %.
The emission correction ratios reflect seasonally and spatially heterogeneous adjustments.Ratios in July show overall increases over China by factors of 1.5-3 except for some locations over Northeast China.In April and October, there are large decreases (up to 70 %) over North China and Central China, while increases are noted in Southeast China.Large emission increases are also shown over Northwest China in April and July.The optimized Chinese anthropogenic NH 3 emissions in July (1.90 Tg month −1 ) are 47-57 % higher than April (1.21 Tg month −1 ) and October (1.29 Tg month −1 ).Inversion results for March-October indicate that Chinese NH 3 emissions peak in summer (see Fig. 7).Similar to previous work of Mao et al. (2015), we also find that the top-down results can be moderately affected by the selection of a priori and observational error covariance matrices.Sensitivity inversions with different S a (50 and 200 %) or S e (20 and 60 %) for July show a range of 1.60-1.93Tg month −1 , with higher S a and lower S e values giving higher July emission estimates.

Improving bottom-up estimates of agricultural NH 3 emissions
The top-down estimates presented above show a stronger summer peak in Chinese NH 3 emissions than those represented in current bottom-up emission inventories (Fig. 1).
Reconciling the discrepancy then requires us to better un-derstand the bottom-up emissions from the underlying processes.Previous studies have shown that NH 3 emissions are highly sensitive to the magnitude and timing of fertilizer application as well as variations of meteorology (Søgaard et al., 2002;Gyldenkaerne et al., 2005;Paulot et al., 2014), but these factors are neither sufficiently represented nor well evaluated in the Chinese NH 3 emission estimates.Here we construct an improved bottom-up Chinese NH 3 emission inventory from fertilizer application and livestock waste, with the objective to better estimate fertilizer application practices and emission factors.Figure 4 shows the schematic diagram of the bottom-up methodology as will be described in detail below.

NH 3 emission from fertilizer application
The MASAGE inventory recently developed by Paulot et al. (2014) provides spatial-resolved and crop-specific estimates of fertilizer application practices over the globe.We follow the methodology of MASAGE for NH 3 from fertilizer application but include detailed refinements for China.NH 3 emissions from fertilizer application (E NH 3 −F ) are calculated as the product of synthetic fertilizer application mag-  nitude (F ) and corresponding emission factors (α F ): (2)

Fertilizer application magnitude
We estimate fertilizer application amounts for 18 crop categories (including early/late rice, spring/winter wheat, spring/summer maize, cotton, potato, and others as shown in Fig. 5).The fertilizer application magnitude (F ) is calculated as where A c is the planting area of crop c, and ψ represents fertilizer application rate at time t (day of the year).We use the EarthStat dataset of crop harvest area (Monfreda et al., 2008;EarthStat, 2015), which provides global crop harvest areas and yields at 5 min × 5 min resolution for the year 2000.
Here we regrid them to the model 1/2 • × 2/3 • resolution, and scale the harvest area for each of the 18 crop categories to the year 2008 using the province-level data from the National Bureau of Statistics of China (NBSC, 2015).We use the harvest areas as the crop-planting areas except for rice and tobacco.Their planting areas are about 5 % of the harvest areas during the seeding period, and then they move to the transplanting fields till harvest.
Estimating the fertilizer application rate c (t) requires consideration of the planting schedule and fertilizer application practice for each crop.Paulot et al. (2014) distribute the annual fertilizer application amount over three stages (at planting, at growth, and after harvest) by assuming cropspecific application ratios.Here we consider much more detailed fertilizer application practices over China.Each crop requires the basal fertilizer applied at planting and several top dressing fertilizers during its growth (up to five application times).We construct tables of the fertilizer dates and rates (Tables S1 and S2 in the Supplement) for the main crop categories in China based on Liao (1993) and Zhang and Zhang (2012).The timing of fertilizer application to each crop is based on its planting date or the calendar day (Table S1).We use the crop-planting dates of Sacks et al. (2010).Following Paulot et al. (2014), a Gaussian distribution function (Gyldenkaerne et al., 2005) is applied to account for uncertainties and interannual variations of application dates: where µ c is the crop-specific fertilizer application date, and σ c is the deviation from the mean planting date estimated from the dataset of Sacks et al. (2010) as summarized in Table S1.
The procedures above allow us to estimate fertilizer application rates for each calendar day for the main crops listed in Tables S1 and S2, and we sum them up at the monthly scale.Fertilizers applied through the injection and broadcast modes are estimated separately.Based on fertilizer application practices, the first fertilizer application at plant is typically through injection (for rice and tobacco the first applications at both seeding and transplanting fields) and the rest by broadcast.For other crops, fruits, and vegetables, we use the annual fertilizer use amounts from the International Fertilizer Industry Association (IFA, 2013) and NBSC ( 2015), and then we distribute them spatially using the crop yield data of EarthStat ( 2015) and apply monthly variations proportional to the number of daylight hours following Park et al. (2004).
Figure 5 shows seasonal variations of fertilizer application to each crop category in China through injection and broadcast, separately.We estimate that 9.2 Tg N a −1 fertilizers are used through injection and 15.8 Tg N a −1 through broadcast.They show strong but different seasonal variations as resulted from variations of application timings for different crop categories.The use of injected fertilizers peaks in spring (April and May) and has a second peak in fall mainly due to winter wheat, while the use of broadcast fertilizers maximizes in late spring and summer.The annual total fertilizer applica-  S4 in the Supplement.
tion is estimated to be 25.0 Tg N, compared with 22.4 Tg N in Huang et al. (2012) for 2006 and the FAO (Food and Agriculture Organization of the United Nations) estimate of 28.5 Tg N for 2008 (FAOSTAT, 2015).

Emission factor from fertilizer application
We estimate emission factors of NH 3 from fertilizer application as a function of soil properties and agricultural activity information and further modulated by meteorological conditions (Paulot et al., 2014).The emission factor is first calculated as where the factors (f ) represent effects of soil pH, cation exchange capacity (CEC), fertilizer type (e.g., urea, ammonium bicarbonate (ABC), ammonium sulfate (AS)), and application mode (broadcast and injection) on NH 3 volatilization mainly based on Bouwman et al. (2002) (Table S3 in the Supplement).Monthly scalars are then applied to account for the seasonality driven by meteorology: where T i and W i are 2 m (meter) air temperature in • C and 10 m wind speed in m s −1 for month i, respectively (Søgaard et al., 2002;Gyldenkaerne et al., 2005).We use the gridded (0.5 • × 0.5 • ) soil pH data from the University of Wisconsin Nelson Institute Center for Sustainability and the Global Environment (SAGE, 2015), and the soil CEC data (0.5 • × 0.5 • ) from the ISRIC-World Soil Information (ISRIC WSI, 2015) of the World Data Center for Soils.Crop categories and application modes follow the calculation of fertilizer application rates as described www.atmos-chem-phys.net/18/339/2018/Atmos.Chem.Phys., 18, 339-355, 2018 above.Percentages of different fertilizer types applied to cropland are estimated based on the statistics from Zhang and Zhang (2012).
We have collected an ensemble of published measurements of NH 3 emission factors (volatilization rates) from fertilizer application that cover different regions of China and consider different fertilizer types and application modes, as summarized in Table S4 in the Supplement.We compare these measurements with corresponding emission factors derived by Eq. ( 5).This comparison does not include meteorological effects (Eq.6) due to a lack of relevant measurements for the published emission factors.As shown in Fig. 6, calculated and measured values are in good agreement with a correlation coefficient of 0.80 and a mean bias of 10 %, supporting calculations of NH 3 emission factors using Eq. ( 5) for fertilizer application practices in China.

Livestock waste
The Chinese NH 3 emissions from livestock waste are commonly derived as the product of livestock population and emission factors (Streets et al., 2003;Paulot et al., 2014).Here we follow Huang et al. (2012) and Kang et al. (2016) that adopted a more process-based mass-flow approach by considering the transformation of nitrogen in animal husbandry.As shown in Fig. 4, a pool of total ammoniacal nitrogen (TAN) as input to manure management is estimated by animal excreta from three main raising systems (free-range, intensive, and grazing).We use the gridded livestock (e.g., pork, beef, dairy, sheep, poultry) population from the Gridded Livestock of the World (GLW, 2015), and then we adjust them to match the province-level annual records of NBSC (2015) for 2008.The parameters of annual TAN excretion per animal considering both urine and feces and their nitrogen contents for each livestock category are given in Huang et al. (2012).
We estimate the content of TAN produced outdoors and indoors separately by assuming percentages of time spent outside and inside buildings for each livestock category (Huang et al., 2012).The outdoor TAN is directly deposited in the open air, while the indoor TAN can flow through the stages of housing, storage, and spreading to cropland as basal fertilizer with depletion of TAN from processes such as immobilization and leaching at each stage.NH 3 emissions from livestock are calculated as the product of TAN at the four stages and corresponding emission factors.We use the emission factors of Huang et al. (2012) and further account for the meteorological influences as represented by Eq. ( 6).We consider both air temperature and wind speed for outdoor NH 3 emissions, while we only account for air temperature for indoor emissions.Monthly emission factors are calculated using the GEOS-5 assimilated meteorological data at the model 1/2 • × 2/3 • resolution.

Improved emissions and evaluation with surface measurements
Figure 7 presents our improved bottom-up estimates of NH 3 emissions from fertilizer application, livestock waste, anthropogenic totals, and their seasonal variations.We adopt estimates of other NH 3 sources (agricultural burning, chemical industry, transportation, and waste disposal) from Huang et al. (2012) for calculating the total anthropogenic NH 3 emissions.Our bottom-up estimates show Chinese NH 3 emissions of 5.05 Tg a −1 from fertilizer application and 5.31 Tg a −1 from livestock and reach 11.7 Tg a −1 with the addition of other anthropogenic sources.High NH 3 emission rates occur over the North China Plain (over 80 kg ha −1 a −1 in parts of Hebei and Henan provinces) and the Sichuan Basin.Y. Zhang et al. (2010) also reported similar high NH 3 emission (with highest value up to 198 kg ha −1 a −1 ) in the North China Plain in the year 2004.The spatial distribution and seasonal variations of our bottom-up NH 3 emission inventory are also presented in Fig. 1 for comparison with previous estimates.The regional distribution of our bottom-up estimates is broadly similar to Huang et al. (2012) and REAS v2 but exhibits some important regional differences.The total anthropogenic emission estimate of 11.7 Tg a −1 is in the middle of previous bottomup estimates as summarized in Table 1; however, our NH 3 emissions show much more distinct seasonal variations than previous estimates (e.g., Streets et al., 2003), with emissions a factor of 3 higher in summer than winter.As shown in Fig. 7, this strong seasonality is consistent with the adjoint optimized emission totals for March-October (50 % higher in summer than spring and fall), considering uncertainties in the inversion results and satellite retrievals.The improved bottom-up Chinese NH 3 emissions are ∼ 15 % higher than the top-down estimates in May and June and ∼ 20 % lower in other months.This can also be seen from the comparison of simulated NH 3 columns using the improved bottom-up inventory with the TES measurements (Fig. S1 in the Supplement).
For further and independent evaluation of these NH 3 emissions, we use an ensemble of surface measurements of NH 3 gas concentration and NH + 4 wet deposition flux compiled by Zhao et al. (2017).The dataset includes monthly averages from a nationwide measurement network over China for 2011-2012 (Xu et al., 2015) We show in Fig. 8 comparisons of NH + 4 wet deposition flux measurements with model results using improved bottom-up, prior (REAS v2), and optimized Chinese NH 3 emissions.Measured NH + 4 wet deposition fluxes indicate a strong peak in summer, with a national averaged monthly flux of 2.8 ± 1.6 kg ha −1 month −1 and a minimum in winter (0.4 ± 0.3 kg ha −1 month −1 ).Part of this seasonality is driven by heavier precipitation in summer as model results with the prior REAS v2 emissions also capture some of the summer vs. winter deposition flux differences.However, model results with the prior REAS v2 still show underestimates of the flux measurements in summer and overesti-mates in winter and spring, while the adjoint optimized NH 3 emissions reduce the model biases (e.g., mean biases reduced from 18 to 9 % in spring and from −18 to 11 % in summer).We can see that model results with our improved bottom-up NH 3 emissions reproduce the spatial and seasonal variations in measured NH + 4 wet deposition fluxes well (correlation coefficients r = 0.41-0.70).
Figure 9 shows comparisons for the surface NH 3 gas concentration.Comparing model results with these surface concentration measurements requires addressing the inconsistency in the altitude they represent.The lowest model layer is centered at 70 m above surface, while all these Chinese surface sites used here measure at 3 m above surface.Zhang et al. (2012) previously quantified the vertical gradient of HNO 3 concentrations in the lowest model layer based on the dry deposition resistance-in-series formulation and the Monin-Obukhov similarity theorem.Here we follow the same approach to estimate the 3 m/70 m gradient of NH 3 concentrations driven by net NH x flux in each grid.Implied NH 3 concentrations at 3 m are on average 20-30 % higher than those at 70 m.As a result, model results with our improved bottom-up emissions show better agreement (with seasonal mean biases within ±20 %) with surface NH 3 concentration measurements than REAS v2 and optimized emissions.The relatively low correlation coefficients (0.14-0.39) may reflect difficulties for the model to fully capture the heterogeneity in NH 3 concentration due to its short lifetime, uncertainties gaseous NH 3 and aerosol NH + 4 partitioning, and also interannual variations in NH 3 measurements.Furthermore, while land-atmosphere exchange of NH 3 is bidirectional, the model here treats it as one-way emission and dry deposition processes.Zhu et al. (2015) previously implemented a bi-directional NH 3 exchange algorithm in GEOS-Chem, and they found that it led to small changes in wet deposition fluxes but had large impacts on emission estimates and surface concentrations over eastern China.Future work is required to improve the bi-directional exchange processes in the model.

Conclusions
In summary, we have applied both bottom-up and top-down methods to better understand agricultural NH 3 emissions in China.A review of recent bottom-up estimates of Chinese NH 3 emissions shows substantial differences, not only in annual estimates of NH 3 emissions from fertilizer application and livestock waste, but also in their spatial and seasonal variations.The large differences mainly reflect limited information on fertilizer application practices and uncertainties in emission factors of NH 3 from agricultural activities.
We conduct top-down estimates of NH 3 emissions in China by assimilating TES satellite observations of NH 3 column concentration with the GEOS-Chem adjoint model for March-October 2008.The optimized Chinese NH 3 emis-sions show a strong summer peak that is generally underestimated in current bottom-up emission estimates.Optimized monthly emissions in July are 1.90 Tg (1.60-1.93Tg considering different configurations of error covariance), ∼ 50 % higher than these in April (1.21 Tg) and October (1.29 Tg).
To interpret the top-down emission estimates, we revisit the bottom-up estimate of agricultural NH 3 emissions, aiming to better estimate the fertilizer application practices and NH 3 emission factors in China.We improve the emission inventory of Paulot et al. (2014) for NH 3 from fertilizer application with more realistic estimates of fertilizer use magnitudes and growth schedules of main crop categories in China.Emission factors of NH 3 from fertilizer application are calculated as a function of fertilizer type, application mode, soil property, and meteorological condition.Our validation of calculated values with an ensemble of published emission factor measurements shows a good agreement.For NH 3 emissions from livestock waste, we follow the mass-flow approach of Huang et al. (2012) and further account for meteorological influences (air temperature and wind speed) on NH 3 emission factors.
We find in our improved bottom-up inventory for the year 2008 that annual Chinese NH 3 emissions are 5.05 Tg a −1 from fertilizer application and 5.31 Tg a −1 from livestock waste.Addition of other anthropogenic NH 3 sources from Huang et al. (2012) (1.3 Tg a −1 ) suggests annual anthropogenic emissions of 11.7 Tg a −1 .The improved bottom-up estimates of Chinese anthropogenic NH 3 emissions now display a strong seasonality, with emissions in summer ∼ 50 % higher than spring and a factor of 3 higher than winter, similar to the seasonality in the top-down emission estimates.
We further evaluate the improved bottom-up and top-down Chinese NH 3 emissions using available surface measurements of NH 3 gas concentration and NH + 4 wet deposition flux.We find that model results with improved bottom-up emissions reproduce the spatial and seasonal variations in these surface measurements well, demonstrating improvements in NH 3 emissions resulted from inclusion of detailed information on fertilizer application practices and seasonal variations of NH 3 emission factors.We acknowledge that measurements used in the study (both surface and TES satellite measurements) are still sparse in spatial coverage.Future studies using satellite NH 3 observations from AIRS, IASI, and CrIS that have better spatial data coverage will be valuable to constrain the spatial variability of NH 3 emissions.

L. Zhang et al.: Agricultural ammonia emissions in China
The Supplement related to this article is available online at https://doi.org/10.5194/acp-18-339-2018-supplement.
Competing interests.The authors declare that they have no conflict of interest.Special issue statement.This article is part of the special issue "Regional transport and transformation of air pollution in eastern China".It is not associated with a conference.

Figure 1 .
Figure 1.Spatial and seasonal variations of anthropogenic NH 3 emissions in China from different bottom-up inventories.Numbers inset are annual totals of Chinese anthropogenic NH 3 emissions.See Table1for references of the emission inventories.

Figure 2 .
Figure 2. TES observations of NH 3 column concentration over China from March to October in the years 2005-2010.Each point represents a TES observation with a footprint resolution of 5 km × 8 km.Values inset are number (n) of valid observations for assimilation in each month.

Figure 3 .
Figure 3. Differences between GEOS-Chem-simulated and TES-observed NH 3 column concentrations over China for April, July, and October.Panels (a, b) show results from model simulations with prior and optimized NH 3 emissions, respectively.Correlation coefficients (r) and normalized mean biases (NMB) are shown inset.Panel (c) shows monthly correction ratios relative to the prior emissions with optimized Chinese emission amounts shown inset.

Figure 4 .
Figure 4. Schematic diagram for estimating agricultural NH 3 emissions from fertilizer application and livestock waste.

Figure 5 .Figure 6 .
Figure 5. Fertilizer application through injection (a, b) and broadcast (c, d) techniques in China for 2008.Panels (a, c) show annual total fertilizer application at the 1/2 • × 2/3 • model resolution with the annual totals given inset.Panels (b, d) show monthly fertilizer application amounts over China for the 18 crop types.

FertilizerFigure 7 .
Figure 7. Improved bottom-up NH 3 emission estimates from fertilizer use (a), livestock (b), and total anthropogenic emissions (c) in China.Values inset are emission totals.Panel (d) shows seasonal variations in Chinese NH 3 emissions from different source categories.The bars represent our bottom-up estimates and the black line shows adjoint optimized anthropogenic totals for March-October.The vertical black line denotes the range of top-down estimates from inversions with different error configurations for the July month as described in the text.

Figure 8 .
Figure 8.Comparison of simulated vs. measured seasonal mean NH + 4 wet deposition fluxes over China.Spatial distributions (a) and scatterplots (b) are shown.Values inset are seasonal mean deposition fluxes and standard deviations for measurements (black), and model results with prior REAS v2 (orange), adjoint optimized (green; for spring and summer), and improved bottom-up (blue) NH 3 emissions.The correlation coefficients (values in parentheses) and the 1 : 1 line (dashed line) are also shown.

Figure 9 .
Figure 9.The same as Fig. 8, but for surface NH 3 concentration.
, 10 sites in the North China Plain monitored by the Chinese Academy of Sciences for 2008-2010 (Pan et al., 2012), and two sites in China from EANET (2015).It should be noted that these measurements are compared with simulated results for 2008, inducing uncertainties from interannual variations.To test the influences from varying meteorology, we have conducted a model simulation over 2008-2012 with our improved bottom-up NH 3 emissions and other SO 2 and NO x anthropogenic emissions fixed to the year 2008 conditions.Our results show small differences in simulated seasonal mean NH + 4 wet deposition fluxes and NH 3 gas concentrations between the 2008 and 5year averaged model results except for the wintertime surface NH 3 concentrations for which the 2008 model results are 14 % lower (Figs.8 and 9 as discussed below vs. Fig.S2 in the Supplement).

Table 1 .
Bottom-up estimates of ammonia anthropogenic emissions in China 1 .
Zhu et al. (2013)have previously applied the adjoint inverse model assimilating TES NH 3 data to constrain US NH 3 emissions, and Paulot et al. (2014) used wet NH x deposition data to constrain East Asian NH 3 emissions, both at a coarser 2 • × 2.5 • model resolution.
. Available TES observations for assimilation then become very spatially sparse for a single month.We assemble TES observations over the years of 2005-2010 for better spatial data coverage.