Spatial distribution and seasonal variability in atmospheric ammonia measured from ground-based FTIR observations at Hefei, China

Hefei, China Wei Wang1, Cheng Liu2,3,1,4,5*, Lieven Clarisse6, Martin Van Damme6, Pierre-François Coheur6, Yu Xie7, Changgong Shan1, Qihou Hu1, Huifang Zhang1, Youwen Sun1, Hao 5 Yin1, Nicholas Jones8,9* 1Key Laboratory of Environmental Optics and Technology, Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Hefei, 230031, China 2Department of Precision Machinery and Precision Instrumentation, University of Science and Technology of China, 230026 Hefei, China 10 3Center for Excellence in Regional Atmospheric Environment, Institute of Urban Environment, Chinese Academy of Sciences, Xiamen, 361021, China 4Key Laboratory of Precision Scientific Instrumentation of Anhui Higher Education Istitutes, University of Science and Technology of China, Hefei, 230026, China 5Anhui Province Key Laboratory of Polar Environment and Global Change, University of Science and 15 Technology of China, Hefei, 230026, China 6Université libre de Bruxelles (ULB), Atmospheric Spectroscopy, Service de Chimie Quantique et Photophysique, 1050 Brussels, Belgium 7Department of Automation, Hefei University, Hefei 230601, Anhui, China 8School of Earth, Atmospheric and Life Sciences, University of Wollongong, Northfields Ave, 20 Wollongong, NSW, 2522, Australia 9School of Physics, University of Wollongong, Northfields Ave, Wollongong, NSW, 2522, Australia Correspondence to: Cheng Liu (chliu81@ustc.edu.cn) Nicholas Jones (njones@uow.edu.au) 25


Introduction 50
Atmospheric ammonia (NH3) plays a significant role in the formation of fine particulate matter (PM2.5), as ammonia reacts rapidly with nitric acid and sulfuric acid to form ammonium salts (Behera et al., 2013;Meng et al., 2018). Ammonium salts constitute a large proportion of fine particulate matter, which has an adverse impact on air quality, climate change and human health (WHO, 2013). Moreover, ammonium salts have a longer atmospheric lifetime (several days) than that of gaseous NH3 (hours to days), so that 55 they can be subject to long-range transport away from NH3 sources. Globally, the main sources of atmospheric ammonia are related to agricultural activities, including farming and animal husbandry (Sutton et al., 2013). Atmospheric ammonia also originates from other sources, such as biomass burning, vehicle exhaust, natural vegetation and wild animals. Recently, industrial emissions have also been identified as important point source emitters . Major sinks of atmospheric 60 ammonia are dry deposition, wet removal by precipitation, and conversion to particulate ammonium salts via reaction with acids (Baek et al., 2005;Liu et al., 2011).
Although ammonia is a major player in various environmental and health issues, the ammonia budget and its contribution of specific sources to emissions still remain uncertain on regional scales. This is https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. mainly caused by lack of representative measurements of atmospheric NH3. In the last two decades, there 65 have been significant efforts to measure atmospheric NH3 around the world. In situ measurements based on passive samplers or dedicated denuders are usually performed with a time resolution of days to weeks (Erisman et al., 2001;Perrino et al., 2002;Adon et al., 2010;Cisneros et al., 2010;Pinder et al.,2011;Day et al., 2012;Heald et al., 2012;Zbieranowski et al., 2012;Makkonen et al., 2012;Benedict et al., 2013;Li et al., 2014;Li et al., 2017). However, atmospheric ammonia at ambient levels is difficult to 70 measure in situ, due to its reactive and sticky nature, the rapid gas-to-particle conversion in the atmosphere, and the strong spatial and temporal variations of concentration. Only a few sites use spectroscopic measurement techniques, such as quantum cascade laser absorption spectroscopy (QCLS), differential optical absorption spectroscopy (DOAS), cavity-ring down spectroscopy (CRDS), open path FTIR and photoacoustic spectroscopy to provide high temporal resolution data (Pogány et al., 2009;Von 75 Bobrutzki et al., 2010;Volten et al., 2012;Miller et al., 2014;Sun et al., 2014;Dammers et al., 2015;Sintermann et al., 2016;Berkhout et al., 2017;Benedict et al., 2017;Phillips et al., 2019). Compared to ground-based in situ observations, measurements of the NH3 vertical profiles from in situ observations are sparser. Recently, airborne and tower measurements to obtain vertical profiles of NH3 in the free troposphere have appeared (Yokelson et al., 2003;Nowak et al., 2012;Leen et al., 2013;Schiferl et al., 80 2014;Battye et al., 2016 ;Dammers et al., 2017a;Li et al., 2017).
During the last decade, satellite have shown improving abilities to monitor global and regional distributions of NH3. NH3 column densities have been obtained by the Tropospheric Emission Spectrometer (TES) instrument on the NASA EOS Aura satellite (Beer et al., 2008;Shephard et al., 2011;Pinder et al., 2011), the Infrared Atmospheric Sounding Interferometer (IASI) instrument on the Metop 85 satellite Clarisse et al., 2010;Van Damme et al., 2014a), the Cross-track Infrared Sounder (CrIS) instrument on the Suomi National Polar-orbiting Partnership (NPP) satellite (Shephard and Cady-Pereira, 2015), and the Atmospheric InfraRed Sounder (AIRS) instrument on the NASA EOS Aqua satellite (Warner et al., 2016). NH3 concentrations in the upper troposphere were detected by Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) on board the Envisat satellite by 90 analyzing the infrared limb-emission spectra (Höpfner, et al., 2016). Satellite observations have been applied in air quality monitoring, quantification of source emissions, trend analysis, and model evaluation (e.g., Van Damme et al., 2018;Clarisse et al., 2019;Dammers, et al., 2019;Lachatre, et al., 2019).
Nonetheless, satellite data are limited by large uncertainties due to atmospheric conditions (mainly https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. thermal contrast, and cloud coverage), and usually less accurate than ground-based measurements. 95 Satellite data need to be validated by high-precision and high-accuracy data obtained independently by ground-based instruments. Ground-based FTIR observations have been commonly used to validate satellite data products, such as carbon dioxide (CO2), methane (CH4), carbon monoxide (CO), and nitrous oxide (N2O) (Dils, et al., 2006;Morino et al., 2011;Reuter et al., 2011;Schneising et al., 2012). More recently, FTIR measurements have been shown to also provide total column and vertical profiles of 100 ammonia at a high temporal resolution, and are now also used for validation of satellite NH3 observations (Dammers, et al., 2015;Dammers, et al., 2016;Dammers, et al., 2017b). Moreover, FTIR NH3 data have been used to measure ammonia emissions from biomass burning and demonstrate long-range transport of NH3 (Paton-Walsh et al., 2005;Lutsch et al., 2016;Lutsch et al., 2019).
High levels of ambient ammonia has become one of the most prominent air pollution problems in recent 105 years and given rise to growing concerns in China. The ammonia emission inventory developed by the Peking University reveals that the national NH3 emissions increased by 64.6%, from 5.9 to 9.7 Tg from 1980 to 2012 (Kang et al., 2016). According to the EDGAR emission inventory, the total NH3 emissions grew by 357.8%, from 3.06 to 14.01 Tg from 1970in China (EDGAR, 2016. The 14-year AIRS satellite data record indicates the significant increasing trends of NH3 concentration for this country, with 110 the rate of 2.27 % yr -1 from 2002 to 2016 (Warner et al., 2017). The two major contributors of ammonia emission in China, livestock manure and synthetic fertilizer application contributed to 80-90% of the total emissions from 1980 to 2012 (Kang et al., 2016). NH3 emissions are predicted to continue to increase in the next few years, owing to ongoing increases in fertilizer application and intensive livestock farms. From the spatial distribution of NH3 emissions derived from the inventory, it is seen that virtually 115 all high levels of NH3 concentrations are above the agricultural regions of China, such as the North China Plain (Hebei, Shandong, Henan, Jiangsu, Anhui) and Sichuan provinces Kang et al., 2016). This is in agreement with the spatial pattern of NH3 distributions observed by satellite (Van Damme et al., 2014a;Warner et al., 2016;Warner et al., 2017). Emission inventories and satellite observations of NH3 need validation from ground-based remote sensing in China. Furthermore, ground-120 based observations of NH3 have been sparse throughout China (Liu et al., 2011;Xu et al., 2015).
Despite the importance of NH3 in the formation of particulate ammonium, NH3 emission in China has not been routinely monitored in contrast to sulfur dioxide (SO2), nitrogen oxides (NOx), CO, and fine particles, due to the lack of specific regulatory requirements for its measurement. The National program that the concentration of the inhalable particles reduces more than 10 % between 2012 and 2017 in cities at the prefectural level and above, and the fine particle concentration reduces 25 %, 20 % and 15 % in the Beijing-Tianjin-Hebei region, Yangtze River Delta, and Pearl River Delta, respectively (http://www.gov.cn/zhengce/content/2013-09/13/content_4561.htm). Reduction of NH3 emissions is considered an effective way to lower PM2.5 pollution (Erisman et al., 2004;Wang et al., 2015;Wu et al., 135 2016).
Hefei, located in eastern China, is a highly populated and polluted region, with intensive agricultural production, heavy traffic and transportation. Hefei has suffered severe haze and poor visibility in recent years (Hong, et al., 2018;Tan et al., 2019). Concentrations of PM2.5 often exceed the Ambient Air Quality Standard in autumn and winter. Although many studies are aiming to understand particulate matter 140 pollution in China, little is known about the role and contribution of NH3 in fine particulate formation on a regional or local scale. In this study, we present and analyze temporal and spatial distribution, seasonal trends, emission sources and potential source areas retrieved from two years of ground-based FTIR measurements of NH3. This paper is organized as follows. Materials and data are described in Section 2, in particular, the 145 measurement site and instrumentation, the retrieval methods of NH3, and IASI satellite data are introduced. Results and discussion are presented in Section 3. The vertical distribution of NH3 and characteristics are shown in Section 3.1. Time series, seasonal trends and annual variability are analyzed in Section 3.2. Comparisons of ground-based measurements with satellite data for NH3 are made in Section 3.3. The relation of NH3 with surface CO concentrations is discussed in Section 3.4. Then, the 150 potential sources that contribute to NH3 columns over Hefei are identified based on analysis of the relationship of NH3 with meteorological parameters (Section 3.5). Conclusions are presented in Section 4. https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.

Site description and instrumentation 155
The Hefei site (31°54´ N, 117°10´ E, 29 m above sea level), part of the Anhui Institute of Optics and Fine Mechanics, is operated by Key Laboratory of Environmental Optics and Technology, Chinese Academy of Sciences. It is located in the north-western rural area of Hefei city in eastern China (Fig. 1), adjacent to the Dongpu lake in a flat terrain. The Hefei urban area, about 10 km south-east of the site, is densely populated with about 7.7 million people. The site is surrounded by wetlands or cultivated lands in other 160

directions.
A Bruker IFS 125HR FTIR spectrometer and a solar tracker are combined to routinely measure trace gases since January 2014. The FTIR spectrometer and the solar tracker are detailed in Wang et al. (2017).
The spectrometer uses a liquid-nitrogen-cooled MCT/InSb detector in combination with a KBr beamsplitter and a suit of optical filters to record mid-infrared (MIR) solar absorption spectra (700-4000 165 cm -1 ) since July 2015. The solar spectra in the 700-1350 cm -1 filter region, obtained with the MCT detector at a spectral resolution of 0.005 cm -1 are used to retrieve NH3.
Additionally, a weather station (ZENO, Coastal Environmental Systems, USA) mounted near the solar tracker on the roof recorded meteorological parameters, such as surface pressure, air temperature, relative humidity, wind speed, wind direction, solar radiation, rain, snow and leaf wetness since September 2015. 170 At the same time, the indoor pressure, temperature and relative humidity are logged continuously.

Retrieval methods
Two spectral micro-windows were chosen to retrieve atmospheric ammonia, similar to the NH3 retrieval strategies in Dammers et al. (2015). The first micro-window (MW1) covers the spectral range of 929.4-931.4 cm -1 . The interfering species in MW1 are H2O, O3, CO2, and two isotopologues of CO2 ( 13 CO2, 175 and C 16 O 18 O). The second micro-window (MW2) spans the spectral range of 962.1-970.0 cm -1 and is characterized by the same interfering species. The retrieval is performed using the SFIT4_0.9.4.4 algorithm (an updated version of SFIT2, Rinsland et al, 1998), which is based on the optimal estimation method to retrieve vertical profile of concentration and total columns of NH3. A priori information, including gas vertical profiles and covariance matrices are used to constrain the retrieval. A priori profiles 180 of NH3 and interfering gases are taken from the Whole Atmosphere Community Climate Model (WACCM, v.6_120_99) in combination with initial measurement values. The a priori covariance matrix https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.
for ammonia was constructed to be diagonal, with standard deviations of 100% for all layers. The temperature and pressure profiles for the meteorological parameters are taken from the National Centers for Environmental Prediction (NCEP) analysis for each day. The profiles were separated into 48 discrete 185 layers for the forward model calculations, from the surface up to 120 km. The HITRAN 2012 spectral database is used for the spectroscopic line parameters. Figure 2 shows a typical spectral fit of NH3 in the spectral windows centered at 930.4 and 966.05 cm -1 , respectively. The measured spectrum is shown in blue, the fitted spectrum in red and the residual in black. The RMS value of the residuals is used to judge the quality of the fits for each of the retrievals. The RMS of the residuals is about 0.498 % and 0.505% 190 in the two spectral windows, respectively. The IASI NH3 data used here are part of the ANNI-NH3-v3R retrieval product (Van Damme et al., 2014a;Whitburn et al., 2016;Van Damme et al., 2017;Franco et al., 2018). A few comparison studies have been performed to validate the IASI-NH3 data product using independent ground-based or airborne measurements (Van Damme et al., 2015a;Dammers et al., 2016). These validations indicate a general good agreement, but also the possible presence of small biases. The IASI data products have been used 205 to estimate NH3 emissions from agricultural sources or biomass burning, to evaluate model simulations, and to identify small emission sources (Van Damme et al., 2014b;Whitburn et al., 2015;Fortems-Cheiney, et al., 2016;Schiferl et al., 2016;Li et al., 2017;Van Damme et al., 2018). In our study, only the IASI data collected from the morning orbit are considered, as the sensitivity of thermal nadir measurements near the surface is higher at this time, owing to a larger thermal contrast in most places. 210 https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.

Error analysis of NH3 retrieval
An error analysis was performed on the basis of the error estimation method described in Rodgers (1990). 215 The error calculation is based on attributing uncertainties to all parameters used in the profile retrieval.
The influence of such parameters as the temperature profile, solar zenith angle (SZA), spectroscopic line parameters and interfering species has been studied. The error budget can be divided into three contributions: the model parameter error due to the inaccurately-described forward model parameters, the measurement error due to the measurement noise, and the smoothing error due to the low vertical 220 resolution of the retrieval. Table 1 lists the uncertainties of the parameters assumed in the retrievals.
The results of error analysis for a typical NH3 retrieval are summarized in Table 2. The total errors are about 11.42 % based on the combination of random and systematic errors. The random error is mainly due to temperature uncertainty and measurement noise, with an error of 2.56 %. As for the systematic error, it amounts to an error of 11.13 %, dominated by uncertainties in spectroscopic line parameters, 225 with small contributions from uncertainties in temperature, SZA, and phase. It is clear that uncertainties of the line intensity parameter for the ammonia absorption lines are the main error sources for the NH3 retrieval.

Time series and seasonal trend of NH3
The time series of the ammonia column observed by the FTIR from December 2016 to November 2018 at the Hefei site are plotted in Figure 5. The data are not continuous, with gaps due to adverse weather conditions. Many spectra ranging from 700 to 1350 cm -1 are saturated in summer (due to high humidity), 250 causing the retrieved NH3 data to be sparsely sampled relative to those in other seasons. The seasonal and inter-annual variations of ammonia are clearly identifiable. A combination of sine and cosine trigonometric functions was used to fit the seasonal variation of NH3 columns, expressed in Eq. (1) (Keeling et al., 1976;Thoning et al., 1989), where X represents the individual NH3 columns, t is the elapsed time in years, A0 denotes the initial state of NH3 column, A1 is the slope of the linear part, and 255 A2-A5 are the fitting coefficients describing the seasonal cycle. The parameters in Eq. (1) are detailed in the study of Bie et al. (2018).
The seasonal amplitude of NH3 column over the Hefei site is comparable for the two years, with value of 5.33×10 16 and 5.42×10 16 molec cm -2 , respectively. The maximum and minimum NH3 appear in 260 summer and winter, respectively. The annual mean NH3 column is 1.31×10 16 and 1.60×10 16 molec cm -2 , respectively, with an increase rate of about 22.14 %.
The summer maximum and winter minimum of NH3 over the two years indicate that agricultural practices maybe the main source of NH3 over the Hefei site. In the recent study by Shephard et al. (2011), the representative volume mixing ratio (RVMR) of NH3 observed by TES satellite over southeast China 265 (22°N to 42°N, 99°E to 121°E) exhibited a distinct seasonal cycle, with peak concentration in summer. Van Damme et al. (2015a) found that NH3 columns observed by IASI and concentrations from the surface NH3 agrees with 2.7 % yr -1 increase in the use of fertilizer in China. The annual increasing rate of ammonia columns in Hefei estimated by our two-year FTIR measurements (22.14 % yr -1 ) is much larger than the reported value by satellite observations over China. This is likely due to the different sampling years. The increasing trend of NH3 in Hefei is likely caused by either an increased fertilizer use, or increasing air temperature, or decreased sulfur emissions due to strict SO2 control measures. 285

Comparison with satellite data
Here we present a comparison with the IASI satellite measurements. The FTIR dataset is suitable for comparison with satellite data given high concentrations observed and the flat geography surrounding the Hefei station. For comparison with ground-based FTIR measurements, IASI Level 2 product data within 0.5° latitude/longitude radius of Hefei station were considered. We set the collocation time to 90 290 minutes. We remove the data with negative IASI-NH3 columns due to large retrieval error. Table 3 details the data filtering criteria, which follows the criteria adopted in Dammers et al. (2016).
In order to compare two measurements from different remote-sensing instruments directly, their different vertical sensitivity and a priori profiles should be accounted for (Rodgers and Connor, 2003). Since the IASI-NH3 retrieval does not provide averaging kernels and vertical profiles ( Van Damme, et al 2014a), 295 this method for comparison is not applied. Here we therefore compare the IASI satellite and FTIR data directly, without considering the effect of different a priori profiles and averaging kernels. https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. Figure 6 depicts the direct comparison of our data with respect to the co-located IASI V3R data. The IASI data are averaged if multiple satellite overpasses match one single FTIR observation. Although there are a few data matching these coincidence criteria, it is found that the IASI data are in broad 300 agreement with the FTIR data. The mean relative difference between IASI and ground-based FTIR columns are computed (satellite minus FTIR, divided by FTIR), and the standard deviation of the relative differences are also calculated. The Relative differences larger than 100% were considered as outliers from the data. There are 230 and 264 pairs of matched data for columns of NH3 for IASI-A and IASI-B satellite data, giving mean relative difference of 4.51% and 0.33%, with standard deviation of 44.44% 305 and 41.00%, respectively. The correlation coefficients R are 0.86 and 0.78, respectively. The scatter graphs of the retrieval results of FTIR and IASI in Figure 6  Additionally, the distributions of the relative difference of the two dataset show that, the relative bias mainly range from -60% to 80% for IASI A data, from -60% to 60% for IASI B data, and the bias from -20% to 0% as well as -40% to -20% has the highest frequency in both respective bins (Figure 6 (c) and 310 (d)). Dammers et al. (2016) first validated the IASI NH3 data product using ground-based FTIR observations from nine NDACC stations, and showed that the mean relative difference between the satellite NH3 total columns and FTIR data were -32.4 ± (56.3) %, with a correlation coefficient of 0.8. Dammers et al. (2017b) compared CrIS NH3 column and profile data with FTIR measurements from seven co-located 315 NDACC stations. The correlation coefficient (R) between the CrIS NH3 total columns and FTIR data was 0.77, and the relative difference is 0-5 % with a standard deviation of 25-50 % for comparison of high levels of NH3. The average relative difference between the CrIS and FTIR profile was in the range of 20 to 40 %. So the relative differences between IASI total columns and our FTIR data and standard deviations of the differences are within the range of comparison results from other NDACC site data, 320 and the correlation coefficients are comparable to that of other comparison results.

Relationship of NH3 with surface CO
The number of cars in mid-2017 reached more than 1.5 million in Hefei according to the report of the Hefei Traffic Management Bureau, and vehicle exhaust has become an important source of urban air pollution. The tunnel studies carried out in China and other countries indicate that motor-vehicle exhausts 325 constitute an important source of NH3 in urban areas (Perrino et al., 2002;Ianniello et al., 2010; https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. al., 2014;Chang et al., 2016). The correlation between NH3 and CO concentrations reported in these studies suggests a common source for urban NH3 and CO, as CO is a traffic emitted primary non-reactive pollutant. To examine the contribution of traffic to NH3 columns, we analyze the relationship of NH3 However, NH3 columns show high correlation with CO concentrations in summer, as displayed in Figure   7(a). The correlation coefficient (R) is 0.77, although the summer data are sparse. Because NH3 is mainly distributed in the boundary layer, NH3 columns represent the surface concentrations of NH3 to some 340 extent. The close link between NH3 columns and CO concentrations indicates that NH3 has common sources with CO in summer over the Hefei site. Atmospheric CO is regarded as a primary pollutant mainly emitted from vehicles in urban areas, and there is no significant biomass burning source around the Hefei site, thus the elevated NH3 columns are likely partly caused by urban emissions from vehicles.

Identification of potential source of NH3
The variability of NH3 columns is strongly affected by emission strengths from agricultural sources, and meteorological and atmospheric conditions, such as air temperature, wind speed, wind direction and relative humidity. To assess the impact of meteorological parameters on the variation of NH3 columns, 350 we analyze the relationship between NH3 columns with these meteorological and atmospheric conditions. High correlation of NH3 columns with air temperature is obvious from their diurnal variation during the observation period, as seen in Figure 8. Our measurements are performed generally from 9:00 to 16:00 local time. The whole data are averaged per hour during the two years. The diurnal variation shows that averaged NH3 column increased with temperature in the morning until it peaked at the time interval from 355 https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.
14:00 to 15:00, when temperature reaches the maximum. Then NH3 column reduced with the decrease of air temperature. High temperature promotes the volatilization of N-based fertilizer applied to the local cropland, and at the same time, increased temperature favors the phase transition of particulate ammonium to gas ammonia. Further, the scatter plot of NH3 columns with air temperature in spring and autumn season displays relatively high correlation, with correlation coefficients of 0.53 and 0.48 (Fig.  360 9), respectively. However, there is no clear correlation between NH3 columns and air temperature in the summer and winter season over Hefei from the scatter plots (not shown). In Hefei, the cropland in spring is characterized by the growing season of wheat and early rice, and synthetic fertilizer is applied during this period. The fertilizer application also tends to occur in autumn, when winter wheat is sowed and a second rice crop is growing. The agricultural practice may explain the correlation between NH3 columns 365 and temperature over Hefei over these two seasons, which suggests that agriculture was the main source of ammonia in spring and autumn.
The polar plots of NH3 columns with wind in Figure 10 show that wind direction mainly ranged from 0° to 270° during the measurement period, indicating that wind was mainly from the north, east, south and south-west. High NH3 columns are associated with wind directions from 45° to 180°, corresponding to 370 wind originating from the east and south-east. The Hefei urban area is located to the east and south of the Hefei site, while the observation site is surrounded by wetlands or cultivated lands to the north and west directions. So the high levels of NH3 partly resulted from transport of air from the urban area. NH3 columns greater than 3×10 16 molec cm -2 correspond to wind speeds less than 1.65 ms -1 , while wind speeds beyond 1.65 ms -1 correspond to NH3 columns below 3×10 16 molec cm -2 . This result reflects the well-375 known phenomenon that large wind speeds increase the mixing and dispersion of the air mass, diluting the concentration of pollution gases. A correlation between NH3 column and relative humidity in air is not observed. Overall, the results indicate that air temperature, wind direction and wind speed are the main factors that influence gaseous NH3 concentrations in Hefei.
In order to get an insight into the potential source regions influencing NH3 concentrations over the Hefei

Conclusions
Atmospheric ammonia plays an important role in formation of fine particulate matter, affecting air quality and climate. Ground-based FTIR observations have great potential to improve our understanding of the spatial distribution and seasonal variations in atmospheric NH3 on regional scales. In this study, the spatial distribution and temporal variation, seasonal trends, emission sources and potential sources of 395 In order to assess the impact of meteorological parameters on the variation of NH3 columns, we analyzed 415 the relationship between them. High correlation of NH3 columns with air temperature is obvious from their diurnal variation during the observation period. Furthermore, there was a clear correlation between NH3 columns and air temperature in spring and autumn over Hefei, with correlation coefficients of 0.53 and 0.48, respectively. The agricultural practice may explain the correlation between NH3 columns and temperature over Hefei during these two seasons, which suggests that agriculture was indeed the main 420 source of ammonia in spring and autumn. In addition, wind direction and wind speed clearly influenced the gaseous NH3 concentrations over Hefei.
Further, the back trajectories of air masses calculated by the HYSPLIT model confirmed that agriculture was the dominant source of ammonia in spring, autumn and winter, while urban anthropogenic emissions contributed to the high level of NH3 in summer over the Hefei site. The potential source areas influencing 425 the NH3 columns were distributed in the local area of Hefei, the northern part of Anhui province, as well as Shangdong, Jiangsu and Henan provinces.
Although NH3 is currently not included in China's strict emission control inventory, we need to investigate the spatial distribution and temporal variation of NH3 together with the driving mechanism behind them to improve air quality. This study helps to identify the emission sources and potential sources 430 that contribute to NH3 columns over Hefei, a highly populated and polluted area. Our findings have potential implications for reduction of PM2.5 pollution in the urban atmosphere over Hefei. This is the first time that ground-based FTIR remote sensing of NH3 columns and comparison with satellite data are reported in China. Future work include the comparison of ground-based FTIR data with in-situ measurement and model simulations, and to estimate regional emissions of NH3 based on the 435 combination of many measurement techniques.
Data availability. The data used in this study are available from the author upon request (wwang@aiofm.ac.cn).

440
Supplement. The supplement related to this article is available online at: xxxxx.
Author contributions. WW and NJ worked on the NH3 retrieval methods. LC, MVD, and PFC provided the IASI-NH3 data and contributed to the discussion of the paper. CL, YX, and QH helped explain the results. CS, HZ, YS and HY took part in the FTIR measurements. 445 https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. Table 1. Random and systematic uncertainties used in the error estimation. Table 2. Typical random and systematic errors for each parameter in the retrieval of NH3.    the blue dots represent the daily averaged NH3; the red line is the fitting curve.  (a) (b) Figure 9. Scatter plot of NH3 column (molec cm -2 ) with air temperature (°C) in spring (a) and autumn (b).  Figure 10. Polar plots of NH3 columns with wind. Radial axes represent the individual NH3 columns (molecule cm -2 ) in relation to wind directions (theta, degrees). The colors denote wind speed (m s -1 ). https://doi.org/10.5194/amt-2020-39 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.