In the empirical bottom-up method, emissions of NH₃ are calculated using statistical compilations of the products of the activity data (the sources) and their corresponding condition-specific emission factors (Zheng et al., 2021). This statistical bottom-up method is also sometimes referred to as an emission factor method, and uses the following general equation: E N H 3 = ∑ i ∑ p ∑ m A i , p , m × E F i , p , m where E(NH ₃ ) is the estimated annual total NH₃ emissions over some area (e.g., using provincial boundaries summed at the national scale of the country. i , p , and m represent the source type, the study area in China, and the month, respectively. A _i,p,m is the activity data for each specific category, and EF _i,p,m is the corresponding emission factors (Huang et al., 2012; Kang et al., 2016).

Agricultural activities represent the largest contributor of NH₃ emissions in China, and recent reports confirm that agricultural activities account for more than 80% of total NH₃ emissions nationally (Zhang et al., 2011; Huang et al., 2012; Xu et al., 2016). Cropland ecosystems are an important source of atmospheric NH₃ because of extensive nitrogen fertilizer applications, which lead to substantial NH₃ volatilization. Earlier research has pointed out that NH₃ volatilization rates from fertilizer application strongly depend on fertilizer types, rates, and application method (Huang et al., 2012). In addition, emissions depend on prevailing environmental conditions, such as soil properties and meteorological conditions (temperature, wind speed and precipitation) (Li et al., 2021). The adjusted EFs are then a function of the above parameters for specific conditions (Huang et al., 2012), as shown in the following equation: E F i = E F 0 i × C F p H × C F r a t e × C F T × C F m e t h o d where EF _i is the emission factor for a specific condition. EF _0i is the reference emission factor associated with fertilizer type i , and CF are the correction factors for each variable, such as pH, application rate, temperature and fertilization methods, including basal dressing and top dressing (Huang et al., 2012). These parameters used to adjust EFs are derived from peer-reviewed literatures, which is uniform European EFs or referring to existing native measurements.

Ammoniacal nitrogen (TAN) in livestock waste can be hydrolyzed to ammonium and subsequent NH₃ volatilization to the surrounding air, or lost through other pathways during various stages of manure management such as storage and spreading, represents another important source of emission that must be included in ammonia inventories. Most of the earlier studies calculated NH₃ emission from livestock waste as a product of domestic animal quantities multiply by an approximate annual EF per animal (Buijsman et al., 1987). More recent researches improved estimates using emission factors based on distinct phases of manure management for different livestock categories, including manure housing, storage, spreading and grazing stage (Dong et al., 2010; Zhang et al., 2010). Other studies compiled recent inventories of optimized NH₃ emission using a mass-flow approach, which regards domestic animals in several typical categories, and distinguishes emissions from animals raised by completely different agricultural practices, including free-range, intensive and grazing systems (Li et al., 2021). For each of livestock system, TAN inputted into manure management is the product of the daily amount of urine and excrement produced (kg (day capita)⁻¹, N content (%), and TAN content (%).The livestock NH₃ emissions are estimated by multiplying TAN at four different stages of manure management: outdoor, manure housing, storage, and spreading onto farmland with the corresponding EFs (Huang et al., 2012; Kang et al., 2016). The EFs from each livestock waste management are affected by many additional factors. For example, NH₃ emissions from the housing stage depends on factors such as housing conditions, humidity, temperature, from spreading stage depends on practices such as basal or deep application.

Although improved estimation methods with correction emission factor can reduce uncertainties, there remain several uncertainties. For example, due to a lack of well-defined EFs, the parameters applied in one study conducted under one set of environmental conditions are likely not fit to be directly deployed in a subsequent inventory or study under different conditions. For domestic animals, the potential EFs of each source cannot be exactly determined as conditions vary widely over time and distance (Wang et al., 2021). To narrow the gap between different methods of NH₃ emission estimates and the real emission, process-based models (e.g., DNDC, CHANS) have been introduced. These process-based models capture spatio-temporal variations of NH₃ emissions by incorporating input-output processes and the interactions between subsystems (Fu et al., 2015; Xu et al., 2018).

The basic principle of process-based models is the use of a mass balance of model inputs and outputs of the system, quantifying all relevant Nr flows and their interactions with the linkages among air, soil and water subsystems (Gu et al., 2012; Luo et al., 2018). For calculation of atmospheric NH₃, studies mainly focus on the NH₃ subsystem, since the process-based modes consider all significant N outputs, including gas emissions (N₂O, NO, NH₃), dissolved soil N and crop uptake in response all relevant cropland N inputs (organic manure, N deposition, nitrogen fixation, crop residues and synthetic N fertilizer). The resulting total NH₃ emissions are then calculated as the product of NH₃ concentration in the soil liquid phase and the parameterized emission coefficients (Yang et al., 2022).

In summary, NH₃ emissions calculated by a process-based model approach is a product of corresponding parameterization factors. The model needs a corrected the EFs by considering relevant environmental conditions to accurately assess the emission of the main contributors to NH₃ emissions.

An inversion approach also provides insights into the spatial and temporal patterns of NH₃ emissions. Inverse modeling starts from an a priori bottom-up emission estimates, and refines it by assimilating observation datasets to obtain a posteriori emissions. This method is also referred to as a top-down method.

The current inverse estimates use either simple mass-balance methods or more complicated data assimilation schemes, such as a Kalman filter and variational data-assimilation. Such methods start with a model simulation that uses a priori emissions and translates these emissions into “simulated observations”. Thus, atmospheric chemistry transport models are an indispensable tool. The Model-3 atmospheric model represented by the Community Multiscale Air Quality Modeling System (CMAQ) and GEOS-Chem model have been widely used to simulate air pollutant concentrations (Pleim and Ran, 2011; Chen et al., 2021; Marais et al., 2021). Next, the mismatch between simulated and true observations is used to define a cost-function that is subsequently minimized. In the case of NH₃, the observation datasets used in the inversion can be either from surface monitoring of NH_X or satellite observations of NH₃ column concentrations.

There are some existing studies that optimized NH₃ emissions in China using datasets from surface monitoring networks. NH₃ emissions in China were constrained by assimilating NH₃ surface observations using an ensemble Kalman filter (Kong et al., 2019). This method is formulated as follows: x = c T , E T T x a = x b + P b H T H P b H T + R − 1 y 0 − H x b where x represents the augmented state. In this state, c and E represent the vectors of monthly NH₃ concentrations and emissions, respectively. b is the background state (a priori), and a is the update state (a posteriori). P ^b background error covariance matrix that is flow dependent and represented by an ensemble [50 in Kong et al. (2019)]. y ⁰ represents the vector of the observations with an error covariance matrix of R . H is the linear observational operator that maps the m-dimensional state vector x to a p- (number of observations) dimensional observational vector Hx ^b .

Paulot et al. (2014) improved the NH₃ emissions in China for 2005–2008 by minimizing a cost function using variational data-assimilation. NH₄ ⁺ wet deposition flux measurements from a monitoring network were used to constrain NH₃ emissions. The cost function (J) defined as J = 1 2 y s i m − y o b s T S o b s − 1 y s i m − y o b s + 1 2 η T S a − 1 η where y _obs and y _sim represent the vector of observed monthly wet deposition fluxes of NH₄ ⁺ and the collocated model values. S _obs and S _a represent the error covariance matrix of the observation system and that of the emissions. The second (background) term in the cost function represents the costs that are related with deviations from the a priori emissions. In their formulation, η is a vector of log-normal scaling factors with elements η i = ln ⁡ ⁡ E / E a , where E and E _a are the corresponding optimized and a priori NH₃ emissions for grid square i (Paulot et al., 2014).

Recent advancement in satellite remote sensing techniques enables the observation of atmospheric NH₃ vertical columns accurately across time and space. Several studies reported observations of the spatio-temporal distribution of NH₃ by utilizing the NH₃ column density retrieved from the Tropospheric Emission Spectrometer (TES) (Beer et al., 2008; Shephard et al., 2015), the Cross-track Infrared Sounder (CrIS) (Shephard and Cady-Pereira, 2015), the Atmospheric Infrared Sounder (AIRS) (Warner et al., 2016; Warner et al., 2017) and the Infrared Atmospheric Sounding Interferometer (IASI) (Van Damme et al., 2020; Chen et al., 2021; Luo et al., 2022).

Inverse modeling techniques assimilating these satellite observations have provided unique opportunities to improve NH₃ emission estimates with high spatial and temporal resolution using continuous, near real-time, large-scale measurement (Chen et al., 2021). This top-down method has been applied to derive NH₃ emissions in other countries and at the global scale (Cao et al., 2020; Chen et al., 2021; Luo et al., 2022; van der Graaf et al., 2022), but such inversion studies are still very limited in China. Zhang et al. (2018) optimized Chinese NH₃ emissions by assimilating TES satellite observations of NH₃ column concentration for March–October 2008. A state-of-the-art study reported long-term IASI-derived NH₃ emission in China by the mass-balance method (Liu et al., 2022b). This mass-balance method was used to exploit the ratio of NH₃ emissions to NH₃ concentrations. The satellite-derived monthly top-down NH₃ emissions ( E _sat ) are derived as follows: E s a t = Ω s a t × E m o d e l Ω m o d e l where Ωsat represents either satellite observations of NH₃ monthly columns or surface NH₃ concentrations derived from satellite columns; E _model are the monthly NH₃ emissions used in model, and Ω _model represents either the monthly NH₃ columns or surface concentration during the satellite overpass simulated by ATM.

Besides, some research validated the inverse model results by combining data from monitoring networks with aircraft remote sensing NH₃ observations (Shephard et al., 2015; Sun et al., 2015).

Results of interannual variation of NH₃ emission for the past few decades in China derived in the ten investigated different studies are given in Figure 1 and Table 1. According to these studies, the national annual emission has roughly doubled since the 1980 from 6.4 (5.2–8.5) Tg yr⁻¹ to 14.0 (9.8–20.8) Tg yr⁻¹ (median, minimum-maximum value) in 2015 (Figure 1, Supplementary Table S1). From an overall perspective, the emission trends can roughly be divided in two stages. From 1980 to 1996, the emissions increased steadily by approximately 5.2 (1.9–6.2) Tg yr⁻¹, with a rapid annual growth rate of about 2.1% yr⁻¹. This is due to an increased synthetic fertilizer application and the fast increase in livestock production, which increased in that time period by a factor of 2 (Kang et al., 2016; Fu et al., 2020; Yu et al., 2020).

NH₃ emissions experienced a temporary slow-down in 1997 and then began to fluctuate with slower growth rates in later years. The emission decline in 1997 can be attributed to the Asia financial crisis which caused a significant reduction in both the livestock and fertilizer industries. From 1998 to 2015, emissions increased by about 1.9 (0.7–2.3) Tg yr⁻¹. A gradual decline in the growth rate of emissions starting in 1998 is likely due to a variety of national strategies and government policies aimed at mitigating NH₃ emissions (Ma, 2020). In that period, various emission reduction policies were introduced, such as the corrected air quality standard, national action plan on air pollution control, fertilization recommendation and shut down of small thermal power plants. Encouraged by the Chinese government, a plan on zero increase in fertilizer use was raised in 2015 (Liu et al., 2020). Under the government’s initiative and the effects of a growing market economy, traditional free-range systems for the livestock industry were gradually replaced by large-scale intensive methods, and significant changes in farming practices were implemented. Another influence is in the replacement of ammonium bicarbonate (ABC) with urea, and since NH₃ volatilization from ABC is more than two-fold that from urea, a measurable decline in emissions followed (Kang et al., 2016). It should also be noted that changes in the proportions of livestock categories occurred. For example, the class proportion of intensively reared animals for beef cattle, pigs and laying hens significantly increased (Kang et al., 2016). These changes explain why fertilizer production and annual population of poultry have both doubled after 1998 without a corresponding increase in national NH₃ emissions.

Contrary to slower growth rates in NH₃ emission based on bottom-up methods, Liu et al. (2022a) suggested a rapid growth rate in NH₃ emission from 2009 to 2015, with a peak in 2015, using a top-down approach. This increase is driven mainly by observed increases of IASI NH₃ column concentrations. This observed increase in NH₃ columns over China is likely largely due to a decrease in SO₂ emissions in China since 2013 because of air pollution control measures, reducing the transformation of alkaline NH₃ to (NH₄)₂SO₄ (Luo et al., 2022).

NH₃ emissions in existing databases show visible differences both in terms of total annual emissions and long-term trends. The levels of NH₃ emission in the same years are widely different depending on the methods used to obtain the estimates (Figure 1 and Supplementary Table S1). Some results showed gradual stabilization of emissions, while others observed a consistent and stable growth pattern. Such large variations are primarily caused by the following reasons:

1) Agricultural activities represent the largest contributor of NH₃ emissions in China, and the range of variation in agriculture emission mainly determines the differences in the estimates of total NH₃ emission (Figure 2 and Supplementary Table S3). The activity level of specific sources and emission factors (EFs) are the two principal factors directly affecting NH₃ emission estimates, most of these results are based on bottom-up methods. The activity data of agriculture such as fertilizer consumption and livestock numbers are comparable, because data sources in most previous studies cited the same statistical yearbook of each province. The discrepancies are thus mainly a result of inconsistent EFs applied in different inventories, especially some studies use uniform European EFs without adjust by parameters for specific conditions.

In this review, we included studies that mostly employed constant European-based EFs without considering specific agricultural practices and local environmental factors. It should be noted that many recent studies in China to derived emission factors (functions) for fertilizer and manure application as a function of management (e.g. fertilizer type, fertilization method), crop type, livestock type, climate (rainfall, temperature) and/or soil properties (e.g. SOC, clay, pH, CEC, bulk density) such as (Huang et al., 2012; Xu et al., 2015a; Zhou et al., 2015; Xu et al., 2016; Xu et al., 2017a; Wang et al., 2018; Zhang et al., 2018). Considering high volatilization of ABC compared to urea, and interannual variation of synthetic fertilizer types, means some previous studies adopted much lower EFs than those who adjusted for the local environmental conditions, or those who estimated emission from livestock manure employing a mass-flow approach (Huang et al., 2012; Kang et al., 2016). The corrected EFs differed by more than a factor of two when the full range of influences is considered. For instance, the final corrected EFs used in Zhang et al. (2017) are approximately 16.6% higher than in other studies.

2) The source categories included in the compiled inventories are not identical (Supplementary Table S4). Notably, the non-agriculture sectors were usually not considered in some compiled ammonia inventories, mainly because of their relatively small contribution to total emissions. Consequently, several inventories reported that cropland and livestock emissions gradually stabilized since the beginning of the 21st century, but total emissions still show a steady growth because of miscellaneous non-agricultural sources that rose sharply during this period (Fu et al., 2020; Ma, 2020). The emissions from traffic, waste disposal and residential sectors are expected to be dominant in urban and industrial areas, where contributions from the agricultural sector are relatively small (Chang et al., 2019; Fu et al., 2020; Feng et al., 2022). Overlooking these non-agriculture sectors explains thus further the differences in trends and the large range in NH₃ emission estimates.

Figure 3 shows that the inventories report substantial seasonal variations of NH₃ emissions from six existing studies. The tendency of monthly emissions to rise from January to July and then decline agrees with the temperature patterns and the timing of known agricultural practices. In general, peak emissions occur from June to August, and the highest and lowest monthly emissions were recorded as 1.1 (0.7–3.2) Tg/month in July and 0.7 (0.5–1.4) Tg/month in January. Emissions in summer are approximately two times as high as in winter. These features coincide with seasonal distributions based on studies using an inversion model approach (Paulot et al., 2014; Kong et al., 2019; Evangeliou et al., 2021). The satellite-based monthly spatial distribution in China from (Liu et al., 2022b) show that high NH₃ emissions in North China Plain (NCP) occur in June and July.

The seasonal variation of NH₃ emission is mainly dominated by agriculture activities and temperature, especially synthetic fertilizer application (Figure 4). In China, the spring farming begins in April, and the NH₃ emissions continuously increase due to intensive fertilizer application coupled with higher temperatures in the following 1–2 months. Some summer plants such as maize are usually seeded in June with base and topdressing fertilization of other plants. Associated cropland emissions thus peak during June and August (Huang et al., 2012; Kang et al., 2016). From autumn onwards, most of the crops are harvested, which leads to an overall decline of emissions. Conversely, less NH₃ volatilization related to lower temperatures and rare cultivation occurs during the winter. The NH₃ emission from livestock manure shows a minor seasonal pattern, mainly dominated by seasonal temperature patterns (Figure 4) (Zhang et al., 2018; Li et al., 2021). As with NH₃ emissions from livestock and fertilizer applications, the NH₃ emissions from biomass burning or forest fires also show a distinct seasonal distribution. However, these effects do not significant influence the temporal disparities of total emissions in these studies due to their relatively small overall contributions (Kang et al., 2016).

Apart from the obvious seasonal variations, some modelling studies report significantly different monthly emissions. For example, the emission peaks often occur in different months. In particular, Paulot et al. (2014) used a new agricultural emissions inventory (MASAGE_NH₃) to suggest that the largest emissions occurred in April and July due to the timing of fertilizer practices. This phenomenon is consistent with some earlier studies that focused on cropland fertilization only (Zhang et al., 2011; Wang et al., 2021). Others studies reported peak NH₃ emissions that occurred in May and September due to inaccurate grasp of the timing of fertilizer application (Cao et al., 2011; Hoesly et al., 2018). The seasonal distribution from CEDS show a more uniform emission peak compared to other studies. In CEDS studies, regional procedures and practices of agriculture fertilization were ignored, leading to large variation in monthly emissions. Crop types are another confounding factor. There is considerable spatial heterogeneity in crop-categories in China, with maize mainly in the north and rice mainly in the south. Moreover, even within a single crop type category, variations exist. For example, maize crops include spring maize and summer maize. Spring maize is fertilized in March-April but the optimal date of summer maize fertilization is much later, from June to August depending on the region (Li et al., 2021). The actual farming conditions drive these differences in the fertilization dates of different provinces. In most previous studies, fertilization dates are fixed to a specific month without considering spatial heterogeneity, and they ignore the different crop types and agricultural practices on the ground. However, recent studies consider quite some variations (e.g. Li et al. (2021)), thus explaining differences in the temporal variation.

The spatial patterns of NH₃ emission data are compared for the year 2015, being available from four gridded datasets revealing a large spatial variability over China (Figure 5). The highest NH₃ emission intensities are concentrated in the NCP area, and in eastern and south-central China, due to intensive agricultural practices in the provinces of Shandong, Hebei and Henan. Although six provinces (Shandong, Henan, Hebei, Tianjin, Jiangsu and Anhui) account for only 8% of mainland China, they contribute almost 30% of the total NH₃ emission in China. The high emission intensity in the NCP is associated with a rapid growth rate in fertilization and livestock practices in that region. In addition, predominately alkaline soil properties in NCP further result in higher NH₃ volatilization (Zhang et al., 2010; Fu et al., 2020). The Beijing-Tianjin-Hebei region shows remarkable NH₃ emissions, dominated by non-agriculture sources consistent with the result of satellite remote sensing observations (Van Damme et al., 2015; Liu et al., 2017).

The intensive agriculture of the Sichuan Basin plays a significant role in high NH₃ emission in southwest China, with the highest observed growth rates of ∼2.3% yr⁻¹ (Table 2). The emission contribution from Sichuan accounts for 6.5% of the national total emission, mainly caused by fertilizer application and livestock, which jointly contribute 87% for this major regional emitter (Huang et al., 2012). In contrast, the regions with low NH₃ emission are primarily located across northwest and northeast China. These regions are characterized by dry climates with infrequent application of synthetic fertilizer, lower population densities and less overall industrial activity (Ma, 2020).

Some studies show a higher emission intensity in NCP, eastern and Sichuan basin compared to other regions (MEIC and REAS). This is similar to spatial distribution derived by IASI, and satellite-based studies also show a high spatial variation and hotspot areas (Chen et al., 2020; Liu et al., 2022b). Other datasets show a more uniform distribution in most areas in China (EDGAR and CEDS). The major reason of this discrepancy is again that inconsistent EFs of agricultural activities are applied in different inventories. The neglected non-agriculture sectors and subcategories are also explanations for discrepancies in the spatial distribution of NH₃ emissions in the different inventories.