3.1. Ammonia Source Apportionment
 The ammonia emissions at the individual province level in 2006 from different sources are presented in Table 5. As shown, the total ammonia emission was 9.8 Tg, contributing approximately 15% and 35% to global and Asian emissions, respectively, based on the previous assessment at the global and Asian scale [Bouwman et al., 1997; Streets et al., 2003]. The ammonia emission from China is much higher than elsewhere, exceeding Europe (5 Tg) and the USA (4 Tg) by approximately 96% and 145%, respectively [Bouwman et al., 1997; Pain et al., 1998; Anderson et al., 2003].
Table 5. Regional-Specific Ammonia Emissions (Gg) From Different Sources in 2006
| ||Fertilizer||Agricultural Soil||N-Fixing Crop||Compost||Livestock||Biomass Burning||Human Excrement||Chemical Industry||Waste Disposal||Traffic|
 The most important contributor is livestock manure management (5.3 Tg), accounting for approximately 54% of the total budget. Next is the fertilizer application (3.2 Tg), which is responsible for 33% of emissions. The less important sources at the national scale are the following: soil 0.2 Tg (2%), N-fixing crops 0.05 Tg (1%), compost of crop residues 0.3 Tg (3%), biomass burning 0.1 Tg (1%), human excrement 0.2 Tg (2%), the chemical industry 0.2 Tg (2%), waste disposal 0.1 Tg (1%) and traffic 0.1 Tg (1%).
 The emission characteristics in China reflect its unique agricultural structure and farming practice. For Europe and the USA, livestock overwhelmingly dominate the inventory. In China, however, both fertilizer and livestock play significant roles. The types of fertilizer used in China are quite different from those used in developed countries. Urea and ABC, which are highly volatile compounds, are the two most widely used N fertilizers in China. In contrast, less volatile fertilizers such as anhydrous ammonia and ammonium nitrate are much more popular in North America and Europe (http://www.fertilizer.org/ifa/ifadata). The incredibly large consumption and high volatility of urea and ABC make the ammonia emission from fertilizer application as important as that from the livestock in China. At the same time, the feature of husbandry in China is also distinct, being characterized by multiple rearing systems. The most important is the free-range system, accounting for 65% of the total. In most rural areas, livestock are raised in a traditional way; that is, a small number of animals are fed by individual families and are allowed to roam freely instead of being contained in any manner. Next is intensive rearing (approximately 30%), which refers to the process of raising livestock in confinement at high stocking density; in this process, the farm operates as a factory. This type of farming is extensive in broiler (73%), dairy cow (57%) and pig (41%) production [EOCAIY, 2007]. Grazing, a relatively less important system, only dominates in the northern and western parts of China. Consequently, the free-range system in rural areas is the greatest contributor, accounting for 67% of the emissions of livestock. The largest NH3 emission livestock category is cattle, which emitted 1.9 Tg NH3 per year, followed by laying hens and pigs, which both emitted 0.7 Tg NH3 per year.
 Other sources were not as notable for the whole country, as shown in Figure 2. However, some of these sources could become significant at the provincial scale. For instance, the contribution of biomass burning is a bit higher in Heilongjiang and Jilin provinces (3%), where forest fires frequently occur. Human waste is responsible for greater portions in Guizhou and Shaanxi provinces because of the larger rural populations and the ubiquitous tatty latrines (sanitary latrines only account for only 30% and 35%, respectively). In Shanxi and Hainan provinces, 8% and 7% of ammonia emissions are caused by chemical industry production, respectively, which is higher than the average of 2% for the whole country. Both waste disposal and traffic source are far more prevalent in urban areas such as Beijing and Shanghai, with, on average, 12 and 7 times greater contributions than elsewhere, respectively.
3.2. Spatial and Temporal Distribution
 The spatial pattern of ammonia emission is illustrated in Figure 3. It is easy to see the high emission rates in Henan, Shandong, Hebei, and Jiangsu provinces and in Eastern Sichuan. The emission contributions from Henan, Shandong, Hebei and Sichuan provinces, all of which have high agriculture production, accounted for 12.7%, 9.2%, 8.2% and 6.5%, respectively. The highest emission rates of NH3 were mainly caused by fertilizer application and livestock, which jointly contributed 93%, 88%, 91% and 87% for these four major emitters, respectively.
 Table 5 lists the emission amounts from fertilizer application in each province (the pattern is shown in Figure S1 in 1 km grid cells). The intensity of NH3 volatilization from synthetic N fertilizers was concentrated in Henan, Shandong, Hebei and Jiangsu province, where the crop cultivation is most developed over the whole country. The crop production in these four provinces is the greatest in China, with a total contribution of 33% of China's crop production, and they consumed approximately 30% of the N fertilizer for the whole country in 2006 [NBSC, 2007e]. For example, the ammonia emission rates in Henan province reached a maximum value of 7.2 kg per square kilometer of arable land. The farmers here usually adhere to fertilization during plant cultivation. The application rate in this region was incredibly high, and consequently, the nitrogen loss was tremendous [Ju et al., 2009]. According to an investigation carried out by the government [NBSC, 2007c], the application rate of N fertilizer in Henan province is 1.1–3.4 times higher than the national average for staple crops. For ABC, the gap was even greater (1.1–9.3 times). Among the various kinds of crops, sugarcane and vegetable have the highest fertilization rates (335 and 264 kg N per hectare, respectively), which are double the rates of rice and wheat and are approximately tenfold higher than those of beans and peanuts. In addition, vegetables are cultivated extensively in China, with a sowing area that is just behind cereal crops. Over 20% of ammonia loss originates from vegetable fertilization. The North China Plain, with a 25,624,000 and 5,754,000 ha area under cultivation of cereal crops and vegetables, respectively, is responsible for 43% of the NH3 emissions from fertilization in China. The cultivated land in Guangdong, Fujian, Hunan and Jiangxi provinces also shows large NH3 volatilization. Paddy fields are common in this region, with rice being the dominant crop and contributing the most emissions (approximately 40% for Jiangxi province). The smaller emitters are located mostly in western China, with a minimal amount of arable land or low use of synthetic nitrogenous fertilizers.
 The ammonia emissions from livestock during 2006 are also presented in Table 5 and Fig. S2. The high NH3 emission areas, with rates over 2000 kg NH3 per square kilometer, were concentrated in Henan, Shandong, Hebei and East Sichuan province. Approximately 5.3 Tg NH3was released in China due to livestock rearing, 0.6 Tg of which was emitted from Henan province, followed by Hebei (0.5 Tg), Shandong (0.5Tg) and Sichuan (0.4 Tg). These four provinces are well-known for their large animal population, providing approximately 36% of China's husbandry production [NBSC, 2007e]. Many kinds of animals are extensively bred in Henan, Hebei and Shandong provinces, including those raised for beef, dairy, pork and poultry. Among them, the greatest contributors are laying hens and beef, which contribute 26% and 24%, respectively, to the total emissions of these four provinces. Inner Mongolia, Tibet and Xinjiang province, where sheep are widely raised, also emit remarkable ammonia emissions related to sheep manure management, accounting for over 40% of emissions of these provinces. Cattle are widely raised in Northeast China and are responsible for 53% and 51% of the ammonia emissions in Liaoning and Heilongjiang provinces, respectively. Sichuan province is another large emitter, with an emission rate of 0.4 Tg NH3 per year. Cattle and pig production released 54% and 20% NH3 among all livestock emission in Sichuan province, respectively.
 The seasonal distribution of ammonia emissions was in general agreement with that of temperature and agricultural timing (Figure 4). Generally, the maximum emission was in summer, and nearly 42% and 28% of the fertilization and livestock emissions, respectively, occurred during this season because of the high temperature and dense fertilization. Streets et al. also showed a roughly similar seasonal pattern of ammonia emission, except for fluctuations from June to August and September to November. The disparity was attributed to the timing of different plants. Various crops were planted on the same farmland in rotation over large areas in China, particularly south part, such as the winter wheat-summer maize rotation system practiced in the North China Plain. Summer plants such as maize are usually seeded in June with the application of base fertilizer, and the topdressing fertilizer is applied two months later, and the peak emissions correspondingly occur during June and August (http://www.zzys.gov.cn/). Moreover, fertilizer is densely applied for semi-late rice in June with emissions of 0.16 Tg NH3 (42% of total emissions in June), and cotton is intensively fertilized in August (0.14 Tg, 35% of total emissions in August), which also leads to peak volatilization in June and August. The high emission rates in September and November could be attributed to the basal dressing and top dressing of wheat, with 0.11 Tg and 0.15 Tg NH3 emissions respectively. The high ammonia volatilization rate from livestock also occurs in summer. There is little variation of animal population among the different seasons. The increasing emissions during the time period from June to August were caused by larger EFs associated with the substantial increase of temperature. This tendency could be mostly attributed to emissions from laying hens, which were responsible mostly for the monthly variation because their EFs change more rapidly with temperature than other species [Mannebeck and Oldenburg, 1991]. Although other emission sources such as biomass burning also have seasonal variations, temporal disparities were not considered in our estimate due to their relatively small contributions.
3.3. Comparison With Existing Result
 Our estimation was compared with the previous emission inventories listed in Table 6. As presented, the results are generally 40–50% higher than ours. These disparities are caused mainly by the emission estimations from the fertilizer volatilization. Our estimate has smaller emissions and larger spatial disparities. The fertilizer consumptions are comparable, and the discrepancy is mostly caused by the distinct EFs employed. Zhao and Wang , Streets et al. , Yan et al.  and S. Wang et al.  used uniform EFs for the whole country, and these values were based on European experiments or foreign expert judgments rather than native measurements in China. Zhang et al.  spatially and temporally characterized the EFs at the county level using the National Ammonia Reduction Strategy Evaluation System (NARSES) model developed for agriculture emissions in the UK. Similarly, many parameters have been introduced based on measurements in Europe. As mentioned above, the ammonia volatilization from chemical fertilizer strongly depends on numerous environmental conditions. The soil pH values, temperatures and agricultural practices are completely different between Europe and China. Our estimation made full use of local experiments, and it adjusted the EFs for different crops under local conditions. For instance, the EFs for Urea and ABC are identical, at 15% and 30%, respectively, in Streets et al.'s paper, while in our calculation, the EFs can reach up to 35–40% for non-glutinous rice, which is top-dressed during June to July, and then decrease to lower than 5% for early rice that is basal-dressed during February to March. Consequently, the disparity between their results and the estimates made in this study are to be expected. The comparisons of province-level emissions are demonstrated inFigure 5a. The largest emitters are concentrated in the Hebei, Henan, Shandong, Jiangsu and Sichuan provinces, in accordance with previous results [Streets et al., 2003; Zhang et al., 2011]. Henan and Shandong were given more weight in our budget, with 18% and 10% of the total fertilizer emissions, while the corresponding contributions in the inventory of Streets et al. are 10% and 7%, respectively. Simultaneously, the contribution of Jiangsu Province declines from 10% in Streets et al.'s result to 8% in our estimate. This difference could be attributed to the involvement of the soil acidity and the application rate. More emissions occurred in Henan and Shandong provinces because of alkaline soil and higher application rates, which were not considered in previous studies and might cause underestimation in this area. Jiangsu has fewer emissions in our results due to the lower soil pH and the smaller application rate.
 Comparisons of emissions related to animal manure management at the national scale and the province level are indicated in Table 6 and Figure 5b. Our estimation is comparable to the results of Zhao and Wang , Streets et al.  and Ohara et al. , at approximately 5.0 Tg/yr. The value given by S. Wang et al.  was 7.0 Tg/yr. The differences among these inventories mainly resulted from the selection of EFs. For previous livestock ammonia inventories, the EFs were either presented in loss rates per capita and then multiplied by population [Zhao and Wang, 1994; Streets et al., 2003; Ohara et al., 2007] or were recalculated for distinct stages as the portion of ammonia [S. Wang et al., 2009; Zhang et al., 2010]. However, most of the previous inventories used Europe-based EFs, introducing significant inaccuracies [Streets et al., 2003]. Our research considered three different rearing systems with Chinese characteristics. The EFs were also disaggregated both spatially and temporally according to the climate conditions and local manure treatment practices. Region-specific EFs made it possible to accurately represent the emission distribution. Spatially, Henan and Hebei provinces, two biggest emitters, released 0.6 and 0.5 Tg NH3 in our estimate, respectively, comparable with the results given by Zhang et al. ; the corresponding value in Streets et al.'s paper is 0.3 and 0.3 Tg, as shown in Figure 5b. Higher emissions in our study were caused by larger animal population in these two provinces in 2006 than 2000 (base year of Streets et al.'s inventory) [NBSC, 2007a]. In contrast, Sichuan province plays a much more important role in Streets et al.'s inventory due to the high assessment of N excrement from pig production. N excrement is age-dependent, especially for pigs [Liu et al., 2008]. Sichuan, with a large amount of pig production, might be overemphasized in Streets et al.'s budget due to the neglect of disparities among excrements in different growth phases.
 Compared with the previous ammonia inventories, many more sources were considered, including biomass burning and waste disposal. Although the total emissions of these miscellaneous sources were relatively small, some play an important role at the regional scale. For instance, waste disposal contributed almost 32% of the ammonia emission in Shanghai. Therefore, it is not acceptable to neglect these sources during calculation. Most existing inventories ignored these small sources, thus resulting in inaccurate estimates of some areas.
 Our results are also compared with the satellite data revealing global NH3 column distribution [Clarisse et al., 2009]. Elevated columns are observed in Northeast China and the North China Plain through satellite monitoring, similar to our inventory. However, little ammonia is found in South China and Sichuan province according to satellite observation. The difference between our estimation and satellite observed distributions is first attributed to the fact that satellite monitoring suffers signal impairments caused by meteorological effects such as water vapor, clouds [Garcia et al., 2008]. In South China and Sichuan province, higher amount of rain and cloud cover might introduce some uncertainties. Second, concentration distribution could not be always consistent with emission pattern. As mentioned in section 1, ammonia is very reactive and consequently short-lived in the atmosphere. It could react with many acid materials like SO2. Sichuan and Guizhou province are highly polluted areas with large amounts of SO2 emission [Zhang et al., 2009]. Simultaneously, both temperature and humidity are high there, which further accelerate NH3 depletion. The disparities between our estimation and satellite results are to be expected.
 Emission uncertainty is associated with activity data and EFs. Activity data were obtained through various first-hand investigations and from statistical information. Local conditions and situations were also considered in the process of EF correction. Nevertheless, there are still large uncertainties, particularly for fertilization and livestock emissions, due to the extremely large values and the numerous parameters involved in the EF adjustment. Generally, for economic activities such as fossil fuel use and industrial and agricultural production, which are often reasonably accurate, the coefficient variation (CV) is usually 5–10%. Activities such as biofuel use, waste burning and landfilling are less well-documented [Zhao et al., 2011]. The distribution and uncertainties for activity levels for various sources are listed in Table S3 based on existing publications or assumptions made for this study. The EFs show a substantial variation according to the type of process [Olivier et al., 1998]. The EFs for NH3 emissions from biofuel combustion and chemical industry production have larger fluctuations (CV ≥ 100%), and others have smaller uncertainties, such as livestock (CV: 50%) and waste burning (CV: 50%). Further details can be found in Table S3. We ran 20 000 Monte Carlo simulations to estimate the range of fire emissions with a 95% confidence interval. The estimated emission range was 6.5–12.5 Tg/yr.