The First Global Map of Atmospheric Ammonia (NH₃) as Observed by the HIRAS/FY-3D Satellite

Atmospheric ammonia (NH₃) is an important trace gas that causes environmental problems and harms human health. As the most abundant alkaline gas, NH₃ reacts with atmospheric sulfuric acid and nitric acid, and it is a major source of aerosols (Na et al., 2007; Hao et al., 2020; Bao et al., 2021). The ammonium aerosols change the reflectivity of the Earth’s surface and play a significant role in the radiance balance of the Earth’s system (Xu and Penner, 2012). When aerosols are deposited on water surfaces, eutrophication can occur, damaging the aquatic ecosystem (Mahowald et al., 2017). Guo et al. (2020) estimated that ammonia-related aerosols could lead to 36 billion dollars of public health loss per year in the United States.

Ammonia (NH₃) in the atmosphere is mainly emitted from livestock animal wastes and fertilizer productions/applications (~85%), and partly from industrial manufacture, biomass burning, transports, plant decomposition, and volatilization from oceans and soils (Behera et al., 2013; Zhu et al., 2015; Van Damme et al., 2018). Due to its importance, NH₃ emission inventories have been developed (Bouwman et al., 1997; Crippa et al., 2020). However, the current NH₃ emissions remain highly uncertain at a global scale due to a lack of NH₃ observations (Luo et al., 2022). It is quite difficult to measure the NH₃ concentration by in situ techniques, as the NH₃ molecules will adsorb to the inlet and other surfaces of the in situ instrument (Twigg et al., 2022).

Satellite remote sensing techniques using the specific absorption lines of NH₃ in the thermal infrared spectral region can provide global NH₃ column measurements. The first global NH₃ map was successfully derived from the Infrared Atmospheric Sounding Interferometer (IASI) carried on the MetOp-A satellite (Clarisse et al., 2009). Later, NH₃ columns were also derived from several other hyperspectral infrared satellites, e.g., the Cross-track Infrared Sounder (CrIS) and the Thermal And Near infrared Sensor for carbon Observations-Fourier Transform Spectrometer (TANSO-FTS) (Shephard and Cady-Pereira, 2015; Someya et al., 2020). The NH₃ columns derived from satellite IASI and CrIS measurements have been compared and validated with ground-based Fourier Transform Spectrometer (FTS) retrievals, and the results show that there is a good correlation between ground-based and satellite NH₃ measurements and the bias between them is typically within their estimated retrieval uncertainties (Dammers et al., 2016, 2017).

In November 2017, the Hyperspectral Infrared Atmospheric Sounder (HIRAS) was launched into a sun-synchronous polar orbit at 863 km above the ground onboard the meteorological FengYun (FY)-3D satellite. It was the first Chinese satellite sensor to provide nadir hyperspectral infrared spectra of the upwelling radiance between 650 and 2550 cm^–1 (Yang et al., 2019). The performance of HIRAS has been well evaluated and calibrated in the optical laboratory before launch, and in space after launch (Qi et al., 2020; Wu et al., 2020). The brightness temperature noise level of the HIRAS spectra is about 0.15–0.4 K at 280 K (Qi et al., 2020), which is slightly larger than the IASI noise level of about 0.1–0.2 K at 280 K (Hilton et al., 2012). Atmospheric temperature and water vapor have been successfully derived from the HIRAS spectra (Zhang et al., 2021; Li et al., 2022). However, until now, the HIRAS spectra have never been used to retrieve atmospheric NH₃ concentrations.

In this study, we establish a retrieval algorithm to derive the NH₃ column from the HIRAS observed spectra and present the first HIRAS NH₃ global map. Section 2 describes the satellite data and the retrieval method. In section 3, we show the retrieval results and the comparisons between HIRAS and IASI NH₃ column retrievals. After that, the current issues affecting HIRAS NH₃ retrieval are discussed in section 4. Finally, conclusions are drawn in section 5.

HIRAS is a Fourier transform spectrometer that records infrared spectra emitted by the Earth, its atmosphere, and the solar component. One HIRAS cross-track scan sequence has 33 measurements, including 29 Earth Scenes, 2 Deep Space, and 2 Internal Calibration Target observations (Wu et al., 2020). With a wide swath width of about 2250 km, it offers near-global coverage twice a day, with overpass times at 0130 and 1330 (local solar time). The field of view (FOV) of each HIRAS footprint is 1.1°, corresponding to a 16-km diameter projection on the ground at nadir (Qi et al., 2020). There are three bands observed by the HIRAS sensor: the long wave IR (LWIR) band from 650 to 1135 cm^–1, the middle wave IR (MWIR) band from 1210 to 1750 cm^–1, and the short wave IR (SWIR) band from 2155 to 2550 cm^–1. All bands are recorded at the same full spectral resolution of 0.625 cm^–1, corresponding to a maximum optical path difference (MOPD) of 0.8 cm. For the MWIR and SWIR bands, HIRAS also provides data with spectral resolutions of 1.25 and 2.5 cm^–1, respectively. As the NH₃ absorption lines lie mainly in the LWIR (Gordon et al., 2022), only the HIRAS LWIR spectra with a spectral resolution of 0.625 cm^–1 are used hereafter.

Figure 1a shows a typical HIRAS LWIR spectrum, along with the NH₃ retrieval window. The HIRAS spectra used in this study cover January and July 2020 (2 months; 62 days). According to the HITRAN 2020 spectroscopy, relatively strong NH₃ absorption lines are found near 930 and 967 cm^–1. We tested both the 925–935 cm^–1 and 960–970 cm^–1 spectral windows to check their sensitivity to the NH₃ columns. In the former window, we found that there are many strong CO₂ and H₂O lines that contaminate the NH₃ signal. Therefore, the latter (960–970 cm^–1) window was selected for retrieval. Figure 2 shows a typical transmittance spectrum from the surface to the top of the atmosphere between 960 and 970 cm^–1 using the US standard atmosphere (NOAA, 1976). Although there is still some interference from CO₂, H₂O, and O₃; NH₃ absorption lines are less affected by them, especially near 965.5 and 967.3 cm^–1.

Figure 1.  (a) A typical HIRAS observed spectra in the long wavelength range (648–1136 cm^–1; spectral resolution of 0.625 cm^–1) at 18.72°N, 70.50°E, for a satellite zenith angle of 3.27° and a solar zenith angle of 45.25°. The NH₃ retrieval window (960–970 cm^–1) is marked by a red shadow. (b) The observed (Obs) and simulated (Sim) spectra, and (c) the residual of the fitted spectrum (Obs – Sim).

Figure 2.  (a) The transmittances of the main species (CO₂, H₂O, O₃, and NH₃) in the NH₃ retrieval window, and (b) a zoom window in the transmittance range between 0.965 and 1.00 to establish a better view of NH₃ absorption lines.

To reduce the impact from clouds, we only perform the NH₃ retrieval under clear-sky conditions. The medium-resolution spectral imager-2 (MERSI-2) sensor onboard the FY-3D satellite provides cloud mask products with a 250-m spatial resolution (Xian et al., 2021). We calculate the cloud fraction based on the MERSI-2 cloud measurements for each HIRAS observation and select the clear-sky HIRAS measurements (cloud fraction equal to 0). The cloud masking procedure used here is the same as in Li et al. (2022).

The optimal estimation method (OEM; Rodgers, 2000) is applied to retrieve the NH₃ column from the observed HIRAS spectra. A cost function [$ \mathrm{J}\left(\mathbf{x}\right) $] is defined by Eq. (1):

where $ \mathbf{y} $ is the observed spectra; $ \mathbf{F}\left(\mathbf{x},\mathbf{b}\right) $ is the forward model to simulate spectra; $ \mathbf{x} $ is the state vector (retrieved parameters); $ \mathbf{b} $ represents the model parameters that are not retrieved, such as the surface emissivity, temperature profile, and satellite-Earth geometry; subscripts $ \mathrm{a} $ and $ \epsilon $ represent the prior and the measurement noise, respectively; $ {\mathbf{x}}_{\mathrm{a}} $ is the prior, presenting the best estimation of the state vector based on the a priori knowledge; $ {\mathbf{S}}_{\epsilon } $ is the measurement covariance matrix, determined by the signal-to-noise-ratio (SNR) of the HIRAS observed spectrum (the diagonal values of the $ {\mathbf{S}}_{\epsilon } $ are calculated as $ 1/{\mathrm{S}\mathrm{N}\mathrm{R}}^{2} $, and the non-diagonal values are set to 0); $ {\mathbf{S}}_{\mathrm{a}} $ is the a priori covariance matrix of the state vector, derived from an atmospheric chemistry transport model. The Newton iteration is applied to find the approximation of the true state which agrees best with both the measurement and the a priori information. It follows:

where G is the contribution matrix, K is the Jacobian matrix, representing the sensitivity of the observed spectra to the parameters, and subscript i is the iteration index. Finally, the optimal state vector ($ \widehat{\mathbf{x}} $) is given by Eq. (4):

where $ \mathbf{A} $ is the averaging kernel matrix, representing the sensitivity of the retrieved parameters to the true state; $ \epsilon $ is the retrieval uncertainty.

Table 1 lists the parameters in the state vector ($ \mathbf{x} $), together with both their a priori and variance settings. Apart from the NH₃ column, the columns of the interfering species (CO₂, H₂O, and O₃) are retrieved as well. In addition, the spectral shift and the surface temperature are included. Since the spatiotemporal variations of the H₂O column and the surface temperature are very large, their a priori values are derived from the ERA5 hourly reanalysis data (Hersbach et al., 2020).

We use the ASIMUT model to simulate the infrared radiance transmitted from the Earth’s surface to the HIRAS satellite sensor. The ASIMUT is a radiative transfer model (Vandaele et al., 2006) developed by the Royal Belgian Institute for Space Aeronomy (BIRA-IASB) which calculates the spectrum and analytical derivation of the Jacobians. The ASIMUT code has been applied for the dust and methane retrievals from the IASI satellite (Vandenbussche et al., 2013; De Wachter et al., 2017). The ASIMUT software has been coupled to the SPHER/TMATRIX (Mishchenko and Travis, 1998) and (V)LIDORT (Spurr, 2008) to compute the atmospheric scatterings.

In the NH₃ retrieval window (960–970 cm^–1), we neglect the scattering and consider the thermal emissions under local thermodynamical equilibrium. The line-by-line (LBL) method to calculate the cross-sections was implemented in the ASIMUT model, but it consumes large computing resources. To speed up the retrieval, we create look-up-tables (LUTs) for CO₂, H₂O, O₃, and NH₃ cross-sections in the spectral range from 900 to 1000 cm^–1 based on the HITRAN 2020 database (Gordon et al., 2022), for various pressures (between 7 × 10^–4 hPa and 1081 hPa) and temperatures (between 148 and 328 K). Numerous simulations with different pressure and temperature conditions have been carried out with both the LBL method and the LUT method, and the relative radiance differences between the LBL and LUT simulations are all within 0.005%, which is much smaller than the noise level of the HIRAS spectra.

The ASIMUT model includes 39 vertical layers between the surface and the top of the atmosphere (40 vertical levels: 0.0, 0.1, 0.3, 0.5, 0.7, 0.9, 1.0, 1.3, 1.5, 1.75, 2.0, 2.5, 3.2, 3.75, 4.5, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 17.0, 19.0, 21.0, 23.0, 25.0, 27.0, 30.0, 35.0, 40.0, 45.0, 50.0, 60.0, 70.0, and 100.0 km). The global spectrally dependent surface emissivity datasets are provided by Zhou et al. (2011). The HIRAS instrument line shape (ILS) has been taken into account in the forward model, which is characterized by a sinc function with a 0.625 cm^–1 spectral resolution.

The Copernicus Atmosphere Monitoring Service (CAMS) global ECMWF Atmospheric Composition Reanalysis 4 (EAC4) model simulations are applied to generate the a priori information of NH₃. The CAMS EAC4 model has a horizontal resolution of 0.75° × 0.75°, with 60 model vertical levels between the ground and 0.2 hPa. The CAMS model uses anthropogenic emissions from the MACCity inventory (Stein et al., 2014), fire emissions from the Global Fire Assimilation System (GFAS) (Kaiser et al., 2012), and the Model of Emissions of Gases and Aerosols from Nature (MEGAN) driven by the MERRA reanalyzed meteorology to generate the monthly mean volatile organic compound emissions (Sindelarova et al., 2014). For more information about the CAMS model simulations refer to Inness et al. (2019) and the references therein. Previous studies demonstrate that there are large day-to-day and month-to-month variabilities of NH₃ globally (Van Damme et al., 2015; Wang et al., 2021). Therefore, for the HIRAS NH₃ retrieval, we use the CAMS model monthly means between 2015 and 2020 (6 years) to generate the a priori profile of NH₃. Moreover, the standard deviation (std) of the daily NH₃ concentration is calculated in each grid cell to set the variability of the NH₃ (Sa = 1$ {\sigma }^{2} $). Figure 3 shows the CAMS simulated NH₃ mole fraction near the surface (model bottom level). The NH₃ mole fractions are relatively high in Southeast Asia, South Asia, Europe, Australia, North America, and South America, and the NH₃ mole fractions are relatively low above the ocean and in the polar region, which is generally consistent with the IASI satellite observations (Van Damme et al., 2015). Figure 4a shows typical NH₃ vertical profiles over land and ocean derived from the CAMS model. The NH₃ mole fraction generally decreases with altitude and becomes less than 0.01 ppb above 10 km.

Figure 3.  The CAMS model global simulated NH₃ mole fraction near the surface in January, April, July, and October between 2015 and 2020.

Figure 4.  Typical NH₃ vertical profiles from the CAMS model over land and ocean. (a) Typical averaging kernel matrix (solid lines; lines are colored with different altitudes; in units of ppb (ppb)⁻¹ and column averaging kernel scaled with 0.1 (red dashed line; in units of (molecules cm^–2) (molecules cm^–2)⁻¹ of the HIRAS NH₃ retrieval over (b) land and (c) ocean.

As shown in Table 1, we perform the column retrieval (only a scaling factor of NH₃ column in the state vector) for the HIRAS NH₃ retrieval. To understand the vertical sensitivity of the NH₃ retrieved column, we carried out profile retrieval for NH₃ above several areas, such as in India and the tropical ocean, to derive the NH₃ profile averaging kernel (AVK) matrix in units of ppb (ppb)⁻¹. Note that we have tuned the Sa values to make the sum of the AVK matrix (DOF) from the profile retrieval close to what we get from the column retrieval. Then the column-averaging kernel (CAVK) is calculated based on the AVK matrix as follows:

where $ {\mathbf{x}}_{\mathrm{a},\mathrm{p}} $, $ {\mathbf{x}}_{\mathrm{t},\mathrm{p}} $, and $ {\mathbf{x}}_{\mathrm{r},\mathrm{p}} $ are the a priori, true, and retrieved NH₃ vertical mole fraction profiles, respectively; $ \mathrm{T}{\mathrm{C}}_{\mathrm{a}} $ and $ \mathrm{T}{\mathrm{C}}_{\mathrm{r}} $ are the a priori and retrieved NH₃ column, respectively, in units of molecules cm^–2; $ {\bf{PC}}_{\mathrm{a}} $ and $ \bf{P}{\bf{C}}_{\mathrm{t}} $ are the a priori and true partial column vertical profiles of NH₃, respectively, in a unit of molecules cm^–2; $ \bf{C}\bf{A}\bf{V}\bf{K} $ is the column averaging kernel in units of (molecules cm^–2) (molecules cm^–2)^–1; $ {\bf{P}\bf{C}}_{\mathrm{a}\mathrm{i}\mathrm{r}} $ is the partical column vertical profile of the dry air.

Figures 4b and 4c show a typical averaging kernel (AVK) of HIRAS retrieved NH₃ over land and ocean. Due to the weak absorption lines and low concentrations of NH₃, the retrieved HIRAS NH₃ column is mainly sensitive to the mid-troposphere (2–8 km), which is similar to GOSAT and TES satellite retrievals (Clarisse et al., 2010; Someya et al., 2020). The NH₃ mole fraction is high in the boundary layer (0–2 km), but it also shows relatively high values in the mid-troposphere (2–6 km; Fig. 4a), especially over land. Consequently, such an AVK still allows us to derive NH₃ column information. Note that the AVK value varies strongly with surface type and atmospheric conditions. For instance, the NH₃ signals captured by HIRAS over the ocean are very weak leading to a degree of freedom (DOFs) close to 0, and the DOFs can be up to 0.9 over several polluted land areas.

As an example, Fig. 5 shows the NH₃ global maps observed by the HIRAS satellite during daytime and nighttime on 30 January 2020. Note that we filter out the retrieved NH₃ columns with a DOF < 0.2, root mean square error of the fitting residual (RMSE) > 0.55%, and an NH₃ column < 0. More valid NH₃ column data are available over land than over the ocean. Moreover, the uncertainty of the NH₃ retrievals is relatively large during the nighttime as compared to the daytime, because the NH₃ retrieval quality depends on the thermal contrast between the surface and the lowest atmospheric layer; therefore, it is more challenging to capture the NH₃ signal near the surface when the thermal contrast is small during nighttime (Clarisse et al., 2010).

Figure 5.  The retrieved NH₃ columns on 30 January 2020 from the HIRAS observations under clear-sky conditions during daytime (a) and nighttime (c), together with the differences between the retrieved columns and a priori columns (re-ap) during daytime (b) and nighttime (d), respectively.

The HIRAS measurements show several NH₃ hotspots around the world, e.g., India, West Africa, and East China. These hotspots observed in the HIRAS data are generally consistent with the CAMS model simulations (Fig. 3) since there are large NH₃ emissions in these regions (Stein et al., 2014). Compared to the CAMS model monthly means between 2015 and 2020, the HIRAS retrieved NH₃ columns during the daytime on 30 January 2020 are larger than the model simulations in middle Africa and lower than the model simulations in East China.

To assess the uncertainty of the HIRAS NH₃ retrievals, we compare HIRAS with IASI/Metop-B NH₃ measurements. The IASI measurements have been validated with nine ground-based FTIR sites around the world, and the relative means and stds of the differences (IASI – FTIR) are –32.4% ±56.3% (Dammers et al., 2016). Figures 6 and 7 show the global monthly NH₃ columns during daytime observed by HIRAS and IASI satellites in January and July 2020, respectively. The HIRAS and IASI data pairs are selected based on their co-located gridded monthly values. Although HIRAS provides lower data density than IASI, especially over the ocean and in the high-latitude regions, the two satellites see a similar global map of NH₃ columns. High NH₃ columns are observed in regions with high human activities or strong biomass burnings, which agrees well with bottom-up emission inventories (Crippa et al., 2020).

Figure 6.  The NH₃ column monthly means in January 2020 observed by (a) HIRAS/FY3D and (b) IASI/MetopB re-gridded onto 1° × 1° (latitude × longitude). The scatter plots between HIRAS and IASI NH₃ columns in (c) North America, (d) Europe, (e) East China, (f) South America, (g) Africa, and (h) India. In each scatter plot, the dots are colored according to the data density. The red dashed line is the linear regression. The solid black line is the one-to-one line. R is the Pearson correlation coefficient and N is the number of the data points.

Figure 7.  Same as Fig. 6, but for July 2020.

Six regions (North America, Europe, East China, South America, Africa, and India) are selected to investigate the correlations and relative differences between HIRAS and IASI NH₃ monthly mean columns (Table 2). Positive correlations between both satellite data sets are identified in all these regions, with the Pearson correlation coefficient (R) ranging from 0.28 to 0.73 in January, and from 0.31 to 0.58 in July. The R values are relatively lower in July as compared to those in January, especially in East China, Africa, and India, because the HIRAS has significantly fewer clean-sky pixels in these regions after the cloud filtering. Table 2 shows the means and standard deviations (stds) of the differences between HIRAS and IASI NH₃ columns. In January, the mean relative differences (HIRAS-IASI) are within 5.5% in North America, Europe, Africa, and India, –17.2% in East China, and 35.6% in South America. The stds of their relative differences are between 46.6% and 82.4%. In July, the mean relative differences (HIRAS – IASI) are within 12% in Europe, East China, Africa, India, and 42.5% in South America. The stds of their relative differences are between 51.8% and 82.3%. These values are comparable to the means and stds of the differences between IASI and ground-based FTIR measurements.

Currently, the HIRAS satellite after quality filtering provides a smaller data density than the IASI, and the HIRAS NH₃ measurements are mainly over land and in mid- and low-latitude regions. Figure 8 shows all the convergent retrievals derived from the HIRAS spectra in January 2020 before applying the RMSE and DOF filtering. We found a strong latitudinal dependence in terms of the RMSE, with relatively low values in low-latitude regions and high values in high-latitude regions. Figure 9 illustrates the relationship between the RMSE and DOF of HIRAS NH₃ retrievals in January 2020 and latitude. Similar to Fig. 8, we find that the RMSE is pretty large and the DOF is generally less than 0.05 in the high-latitude regions. The SNR of HIRAS spectra is low in high-latitude regions, which is probably caused by low surface temperatures and significant snow coverage. With our quality filtering criteria, the HIRAS NH₃ retrievals in the high-latitude regions are mainly filtered out. In Fig. 8b, several areas in the tropics and near the Inter-tropical Convergence Zone (ITCZ; Waliser and Gautier, 1993) still have a relatively high RMSE which indicates remaining cloud effects. Therefore, an improved cloud screening method is needed in the future to better select clear-sky HIRAS measurements. For example, we could apply the nitrous oxide (N₂O) infrared absorption lines to retrieve cloud parameters and to do a cloud mask (Siddans et al., 2017), or design a new cloud detection algorithm based on a neural network (Whitburn et al., 2022). Apart from the ITCZ regions, there is no obvious RMSE difference between land and ocean retrievals in tropical areas. However, we find that the DOF over the ocean is quite low even with a low RMSE. The reason is that the NH₃ concentration over the ocean is so low that the NH₃ signal is too weak to be captured from the HIRAS observed spectra. Regarding the HIRAS measurements in the high-latitude regions, the RMSE is generally larger than 0.5%, leading to a low DOF as well. Overall, for all the retrievals with a low DOF due to a bad fitting or a low NH₃ column, the retrieved NH₃ columns are pretty close to their a priori values as the retrieval is dominated by the a priori information (Rodgers, 2000).

Figure 8.  (a) The NH₃ columns, (b), the RMSE, and (c) the DOF for all convergent retrievals derived from the HIRAS spectra in January 2020.

Figure 9.  The RMSE (left) and DOF (right) as a function of the latitude bin (every 10 degrees) as derived from all the HIRAS NH₃ retrievals (grey) and the retrievals after quality filtering (orange) in January 2020. The error bar is the standard deviation.

To better understand why there are almost no HIRAS NH₃ retrievals left over the ocean after quality filtering, we have applied our retrieval algorithm developed for HIRAS (section 2) to IASI L1C data. Figure 10 shows typical spectra over the ocean derived from two adjacent HIRAS and IASI observations on 30 January 2020, respectively. Note that the same a priori profiles coming from the CAMS model are used for the HIRAS and IASI NH₃ retrievals. Both retrievals are convergent. However, the DOF of the retrieved NH₃ column from the HIRAS spectrum is very low (DOF = 0.03) causing the retrieved NH₃ column to be almost the same as the a priori column. The RMSE of the IASI residual is 0.03%, which is about eight times better than that of the HIRAS retrieval (0.25%). Thanks to the high SNR, we can successfully retrieve the NH₃ column (DOF = 0.78) from the IASI spectrum. This experiment demonstrates that the SNR of the HIRAS spectrum is currently too weak to derive the NH₃ column above regions with a low NH₃ column, such as the ocean.

Figure 10.  Typical NH₃ retrievals over the ocean using the (a) HIRAS observed spectrum and (b) the IASI observed spectrum. Note that the a priori NH3 in the title denotes the NH₃ mole fraction near the surface.

In this study, a retrieval (algorithm and strategy) based on the optimal estimation method is established to derive the NH₃ column from the LWIR nadir radiance spectra measured by HIRAS onboard the FY-3D satellite. We use the ASIMUT atmospheric radiative transfer model as the forward model to simulate the thermal radiation emitted from the surface and the atmosphere. The CAMS model is applied to create the a priori information of NH₃. The retrieval window of 960–970 cm^–1 is carefully selected to perform NH₃ column retrievals from the HIRAS spectra.

The first global map of atmospheric NH₃ columns observed by the HIRAS/FY-3D satellite shows several NH₃ hotspots around the world, for example, India, West Africa, and East China, where there are large NH₃ emissions (Stein et al., 2014). The HIRAS NH₃ columns are also compared to IASI observations. Both satellites observe similar NH₃ column spatial distributions. In January 2020, the mean relative differences are within 5.5% in North America, Europe, Africa, and India, –17.2% in East China, and 35.6% in South America. The stds of their relative differences are between 46.6% and 82.4%. In July 2020, the mean relative differences are within 12% in Europe, East China, Africa, India, and 42.5% in South America. The stds of their relative differences are between 51.8% and 82.3%. These values are comparable to the differences between IASI data and collocated reference ground-based FTIR data.

Finally, the remaining issues in the HIRAS NH₃ retrievals are discussed:

1) the SNR of the HIRAS spectrum in the high-latitude regions is too weak to perform NH₃ retrieval;

2) an improved cloud screening method is needed to better select clear-sky HIRAS pixels, especially near the ITCZ;

3) the SNR of the HIRAS spectrum is currently too low to derive reliable NH₃ columns for regions with a very low NH₃ column, even in low-latitude areas such as the tropical oceans.

Acknowledgements. The authors would like to thank Ann Carine VANDAELE (BIRA-IASB) for supporting the ASIMUT model. This study is supported by the FengYun Application Pioneering Project (FY-APP-2022.0502) and the National Natural Science Foundation of China (Grant No. 42205140).

Data availability. The HIRAS Level 1 data is publicly available via https://www.nsmc.org.cn/. The CAMS NH₃ model is publicly available via https://ads.atmosphere.copernicus.eu/cdsapp#!/dataset/cams-global-reanalysis-eac4?tab=overview. The IASI NH₃ data are publicly available at https://iasi.aeris-data.fr/. The IASI L1C data are provided by the EUMETSAT https://navigator.eumetsat.int/. The HIRAS NH₃ retrievals are available upon request to the authors.