-
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).Retrieved parameters ($ \mathbf{x} $) Prior 1σ NH3 column CAMS model SD of CAMS monthly means O3 column CAMS model 50% CO2 column Carbon Tracker 10% H2O column ERA5 reanalysis data 100% Surface temperature ERA5 reanalysis data 3% Spectral shift 0 10% Table 1. The state vector (
$ \mathbf{x} $ ) together with their prior and variance (1σ) in the HIRAS NH3 retrieval algorithm. -
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.
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.
Region (latitude/longitude) North America Europe East China South America Africa India [15°N, 55°N]/
[120°W, 80°W][30°N, 60°N]/
[20°W, 30°E][15°N, 45°N]/
[100°E, 120°E][60°S, 0°]/
[80°W, 30°W][0°, 18°N]/
[20°W, 60°E][10°N, 35°N]/
[60°E, 90°E]Jan-2020 Mean (%) –2.0 –3.1 –17.2 35.6 –5.5 –5.2 Std (%) 52.9 65.1 65.1 82.4 51.7 46.6 R 0.28 0.31 0.69 0.55 0.68 0.73 P-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 DOF 0.47 0.48 0.61 0.53 0.49 0.57 Jul-2020 Mean (%) 42.5 11.4 –3.0 –9.9 –11.0 –3.6 Std (%) 78.7 53.4 82.3 71.5 51.8 60.8 R 0.50 0.32 0.31 0.32 0.47 0.58 P-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 DOF 0.57 0.66 0.44 0.45 0.42 0.54 Table 2. The relative mean and std of the difference between HIRAS and IASI measurements [(HIRAS−IASI)/IASI × 100%] over six regions with high NH3 columns in January and July 2020, together with their Pearson correlation coefficients (R), p-values, and mean DOFs.
Retrieved parameters ($ \mathbf{x} $) | Prior | 1σ |
NH3 column | CAMS model | SD of CAMS monthly means |
O3 column | CAMS model | 50% |
CO2 column | Carbon Tracker | 10% |
H2O column | ERA5 reanalysis data | 100% |
Surface temperature | ERA5 reanalysis data | 3% |
Spectral shift | 0 | 10% |