The High Spectral Infrared Atmospheric Sounder (HIRAS) aboard the FY-3 satellite series is a Fourier Transform Michelson interferometer. It measures upwelling infrared radiative signals across the short-wave, mid-wave, and long-wave infrared bands. HIRAS-II is the second generation of the HIRAS sensor and is carried onboard both FY-3E and FY-3F, which were successfully launched in 2021 and 2023, respectively (Zhang et al., 2022a, b). FY-3E/HIRAS-II is deployed in a dawn-dusk orbit, with an equatorial overpass time of 05:30/17:30 LST, while FY-3F/HIRAS-II operates in a mid-morning orbit featuring an overpass time of 10:00/22:00 LST. Combined with CrIS aboard the Joint Polar Satellite System-1 (JPSS-1) platform, which has an overpass time of 01:30/13:30 LST, these three sensors form a global constellation observation system (Fig. 1a). Note that IASI's overpass times, approximately 09:30 and 21:30 LST, are close to those of FY-3F/HIRAS-II at 10:00 and 22:00 LST and thus can provide additional mid-morning orbit observations. For this concept study, however, only FY-3F/HIRAS-II is used. Their overpass times are shown in Fig. 1b, and the corresponding histograms are presented in Fig. 1c.

Both HIRAS-II instruments aboard the FY-3E and FY-3F satellites, together with CrIS, share key technical specifications with a nadir spatial resolution of 14 km and an unapodised spectral resolution of 0.625 cm-1 (Zhang et al., 2024). Specifically, HIRAS-II provides continuous spectral coverage across 650–2550 cm-1 via 3041 channels, while CrIS matches this spectral resolution, ensuring all three instruments fully cover the characteristic rotational-vibrational ν2 band of NH3 centered near 10.5 µm. For NH3 retrieval, we use the 955–975 cm-1 band extracted from observations of all three instruments. No empirical sensor-specific radiometric bias correction or additional spectral harmonization is applied before the retrievals. We randomly select 1 d of observations from the three satellites over the North China Plain and estimate their spectral noise. The noise equivalent differential temperature (NedT) of HIRAS-II aboard FY-3E and FY-3F is approximately 0.08 K, while that of CrIS is around 0.05 K. These noise levels are of a comparable magnitude and sufficiently sensitive to detect variations in NH3, ensuring uniform observational quality for cross-instrument analysis (Zavyalov et al., 2013; Shephard and Cady-Pereira, 2015; Zhang et al., 2024). While CrIS already has well-established and mature NH3 products from the CrIS Fast Physical Retrieval (CFPR) dataset developed by Environment and Climate Change Canada (Shephard et al., 2020), we adopt a unified retrieval algorithm for all three instruments. This choice is driven by the need to ensure consistency across the satellite constellation, enabling direct inter-comparability between FY-3E/HIRAS-II, FY-3F/HIRAS-II, and CrIS observations. For CrIS retrievals, we use JPSS-1 CrIS Level 1B Full Spectral Resolution V3 data as input, maintaining methodological coherence with our processing of HIRAS-II observations.

For the retrieval, only cloud-screened spectra from FY-3E/HIRAS-II, FY-3F/HIRAS-II, and CrIS are used. To remove cloud-contaminated pixels and ensure consistency across sensors, we apply a unified cloud-screening procedure following the brightness-temperature-difference method of Wells et al. (2020, 2022). Specifically, for each observation, the observed brightness temperature near 900 cm-1 is compared with the nearest-hourly surface skin temperature obtained from European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5). The cloud-screening threshold is adjusted according to the ERA5 water vapor column, and no fixed cloud-fraction threshold is used. Observations for which the 900 cm-1 brightness temperature is lower than this water-vapor-dependent threshold relative to the surface skin temperature are flagged as cloud-contaminated and excluded. The same cloud-screening procedure is applied to FY-3E/HIRAS-II, FY-3F/HIRAS-II, and CrIS to maintain consistency across different satellites.

The NH3 retrieval is based on the FengYun-Low Earth Orbit Atmospheric Infrared Retrieval algorithm (FY-LeoAIR), which adapts the optimal-estimation framework and forward-model structure of the FY-GeoAIR algorithm originally developed for FY-4B/GIIRS NH3 retrievals by Zeng et al. (2023). FY-LeoAIR has previously been applied to trace gas retrievals from FY-3E/HIRAS-II observations, including CO and VOCs (e.g., Zeng, 2025b; Hua et al., 2025). In this study, FY-LeoAIR is extended to global NH3 retrievals from FY-3E/HIRAS-II, FY-3F/HIRAS-II, and CrIS.

The retrieval is performed in the 955–975 cm-1 band, which contains strong NH3 absorption features and is fully covered by all three sensors. Sensor-specific spectral noise estimates are used to construct the diagonal measurement-error covariance matrix. The same forward model, retrieval method, and quality-control procedure are used for all three sensors to maintain internal consistency.

FY-LeoAIR combines a clear-sky infrared radiative transfer forward model with an optimal-estimation-based inverse model. The forward model uses atmospheric temperature, water vapor (H2O), ozone (O3), and surface pressure from ERA5, CO2 from ECMWF Copernicus Atmosphere Monitoring Service (CAMS) greenhouse-gas products, and other weakly interfering trace-gas profiles from standard atmospheric profiles, following the configuration of Zeng et al. (2023). Surface skin temperature is initialized from ERA5 and retrieved simultaneously with the trace gases. The a priori land surface emissivity used in the forward model is obtained from the University of Wisconsin-Madison global infrared land surface emissivity database (UOW-M), as in Zeng et al. (2023). It is adjusted within the retrieval window using a multiplicative Legendre polynomial expansion, in which the zeroth-order term is fixed to unity and four coefficients corresponding to the first- to fourth-order terms are retrieved simultaneously. The parameters in the retrieval state vector are summarized in Table S1 in the Supplement.

The main modification relative to Zeng et al. (2023) is that NH3 is retrieved using a single multiplicative profile-scaling factor applied to a fixed a priori NH3 profile, rather than as independent layer-by-layer NH3 elements. The same profile-scaling strategy is also applied to H2O in this implementation. This approach is adopted because the NH3 information content in individual thermal infrared observations is generally limited, with the degree of freedom for signal mostly below 1, making independent layer-resolved retrievals insufficiently constrained for global multi-sensor processing. The profile-scaling approach also improves computational efficiency for the global constellation retrieval. A similar retrieval strategy has also been adopted by FY-4B/GIIRS NH3 retrievals (Sheng et al., 2026).

The fixed NH3 a priori profile is constructed following Zeng et al. (2023), based on Goddard Earth Observing System composition forecast (GEOS-CF) NH3 simulations over representative polluted land regions in East Asia and South Asia. The same normalized profile shape is used for all retrievals, while the retrieved scaling factor determines the retrieved NH3 abundance. This choice helps avoid introducing artificial spatial or temporal variability from a time-varying prior into the retrieved columns. Retrieval uncertainty and the column averaging kernel are used to characterize the information content of each retrieval. We also conduct an a priori profile sensitivity test (see Sect. S1 and Figs. S1–S6 in the Supplement).

Because NH3 is mainly concentrated in the lower troposphere, the retrieval state vector only adjusts NH3 in 11 atmospheric layers from the surface to 200 hPa. The reported NH3 total column is calculated as the sum of the retrieved surface-to-200 hPa NH3 column and the fixed a priori contribution above 200 hPa. Since the NH3 abundance above 200 hPa is generally small and is not independently constrained by the spectra, the spatial and temporal variability discussed in this study is dominated by the retrieved surface-to-200 hPa component.

An optimal state vector for minimizing discrepancies between forward model-simulated spectra and sensor-measured spectra is derived by the optimal estimation framework. In practice, the retrieval algorithm outputs the state vector that minimizes the cost function defined as: 1J(x)=χ2=y-F(x,b)TSε-1y-F(x,b)+(x-xa)TSa-1(x-xa). Here, y denotes the satellite-measured radiance within the NH₃ retrieval band. F is the forward model that produces simulated radiance for the retrieval process. x is the retrieval state vector, including the NH3 profile-scaling factor, the H2O profile-scaling factor, scaling factors for interfering trace gases, surface skin temperature, atmospheric temperature scaling, and four surface-emissivity Legendre polynomial coefficients. xa denotes the a priori vector. b stands for a set of fixed non-retrieved auxiliary parameters. Sa is the a priori covariance matrix corresponding to the state vector. Sε represents the measurement error covariance matrix, treated as a diagonal matrix derived from spectral noise estimates. Retrieval outputs include the NH3 column, posteriori column retrieval uncertainty estimation, and column averaging kernel (AVK). The column AVK quantitatively represents the response of total NH3 column to variations in the partial column at each layer, which is important for cross-comparison and model assimilation of satellite retrievals. A more detailed description of column AVK is provided in Sheng et al. (2026).

As an illustrative demonstration of the global sampling capability of the LEO constellation, we first examine representative 7 d mean NH3 column distributions in January, April, July, and October 2024, specifically 12–18 January, 12–18 April, 17–23 July, and 1–7 October. These weekly maps are intended to demonstrate the global spatial coverage and broad seasonal contrast captured by the constellation, rather than to provide a climatological seasonal mean. We first retrieve global daily maps of NH3 columns, re-grid them to a 0.5° × 0.5° spatial resolution, and then calculate 7 d averages for these representative periods. A quality control process is applied to ensure the reliability of satellite retrievals. Observations that fail to achieve convergence within 10 iterations are excluded from the dataset. We then implement a series of post-filters, retaining only those retrieval results that satisfy all the following criteria simultaneously. First, the retrieved NH3 columns are required to be positive. Second, the absolute difference in surface skin temperature between the a priori and retrieved values is less than 10 K. Third, the retrieval error of the column does not exceed 300 %. Fourth, the surface AVK is greater than 0.1. Fifth, the absolute thermal contrast (TC), defined as the temperature difference between the surface and the lowest atmospheric layer, is larger than 3 K. The absolute TC > 3 K and surface AVK > 0.1 criteria used here are intended for the global spatial-distribution analysis, where the objective is to illustrate the broad coverage and spatial patterns retrieved by the LEO constellation while retaining sufficient sampling density. These criteria remove retrievals with weak thermal contrast and limited near-surface sensitivity, but they are less restrictive than the criteria used later for the regional hotspot diurnal analysis. The effect of different TC and AVK thresholds on spatial distributions, diurnal cycles, and data retention is further evaluated and presented in Figs. S7–S10. Sixth, a threshold for surface emissivity at 8.3 µm is set at 0.9, with only results meeting this threshold retained. This step avoids abnormally high NH3 column concentrations over desert regions, following the method of Clarisse et al. (2019). The surface emissivity data used for this filter are obtained from the Combined ASTER and MODIS Emissivity over Land Database Monthly Global 0.05° V003, which provides global monthly emissivity data at a 0.05° spatial resolution.

As shown in Figs. 2 and 3, broad seasonal contrasts in the representative weekly global NH3 maps are evident across the four selected periods. In January, elevated NH3 columns are concentrated in the Indo–Gangetic Plain and Central Africa, which is primarily associated with ongoing agricultural activities such as winter fertilizer application and extensive biomass burning for land clearance in these regions. In April, high NH3 loading persists in the Indo–Gangetic Plain, East China and Central Africa, driven by intensified spring planting and corresponding fertilizer use in agricultural hotspots. Southeast Asia also exhibits increased NH3 levels during this month, a pattern attributed to widespread slash-and-burn agriculture practices that are common in the regional dry season to prepare farmland. Additionally, scattered high NH3 columns are observed in parts of North America, likely linked to localized agricultural operations and small-scale biomass burning. In July, high NH3 columns remain concentrated in the Indo–Gangetic Plain, the North China Plain, Central Africa and Western Europe. The sustained high levels in these areas correlate with summer agricultural activities and elevated temperatures that enhance emissions (Ding et al., 2024). In particular, large-scale wildfires during the summer in North America exert a broad regional impact, leading to significantly elevated NH3 columns across extensive regions as a result of biomass burning releasing substantial amounts of NH3 (Lutsch et al., 2019). Within smoke plumes, gas-phase NH3 can rapidly partition into particulate ammonium, especially in fresh, dense plumes injected into colder, higher-altitude air where ammonium nitrate formation is favored. Together with plume transport and dilution, this partitioning may modulate the observed NH3 columns as the smoke evolves (Lindaas et al., 2021). October still sees detectable high NH3 columns over the Indo–Gangetic Plain, where post-harvest agricultural residue burning and late-season fertilizer application maintain elevated emissions. In South America, elevated NH3 levels are mostly attributed to combined effects of fertilizer usage for autumn crops and biomass burning, as documented by Luo et al. (2022). The seasonal patterns reflect the close linkage between NH3 distribution and region-specific anthropogenic activities, alongside seasonal natural processes. These spatial patterns exhibit general consistency with results from IASI (e.g., Van Damme et al., 2015) and CrIS (e.g., Shephard et al., 2020).

We note that these weekly global maps may still be influenced by short-term episodic events, such as biomass burning. Therefore, they are used here primarily as representative examples of the constellation-derived global NH3 spatial distributions. The more quantitative regional diurnal and seasonal analyses in Sect. 3.2 are based on full-month averages using all available observations in each selected month.

To conduct an in-depth analysis of NH3 variations, we select three major emission hotspots located in the North China Plain (37° N, 118° E), the Indo–Gangetic Plain (28° N, 78° E), and the Central United States (36.5° N, 100.5° W), respectively. These hotspots are characterized by intensive agricultural activities and high livestock densities, consistent with the high-emission areas for their substantial NH3 column abundances and distinct diurnal and seasonal dynamics. For each hotspot, we extract data within a 2.5° × 2.5° grid centered at the specified coordinates, covering six overpass times per day. A 2.5° × 2.5° box is selected to balance spatial representativeness and sample size. Because the analysis separates retrievals by satellite overpass time and month and further applies strict quality filters, a smaller box would often lead to insufficient valid retrievals, especially during nighttime or low-sensitivity conditions. The selected box size provides enough samples for stable monthly means while still focusing on the core NH3 hotspot regions. The monthly mean is calculated only when the number of filtered retrievals within the 2.5° × 2.5° box exceeds 10 during that month. TC affects the retrieval accuracy (Bauduin et al., 2017), and higher absolute TC contributes to better retrievals (Clarisse et al., 2010). When TC is close to zero, the NH3 spectral signal becomes weak and the retrieval sensitivity decreases, whereas a larger absolute TC generally enhances the spectral signal and improves the retrieval sensitivity. To ensure high-quality NH3 column retrievals, we apply stricter sensitivity filters than those used for the global spatial-distribution maps in Sect. 3.1. Specifically, the same basic quality-control criteria are used, but the TC and AVK thresholds are tightened from absolute TC > 3 K and surface layer AVK value > 0.1 to absolute TC > 5 K and surface layer AVK value > 0.3. These stricter criteria are adopted to reduce the influence of local-time-dependent retrieval sensitivity on the inferred diurnal cycles, because TC itself has a pronounced diurnal variation.

The diurnal cycles of full-month averaged NH3 columns for the three hotspots in each selected month are presented in Fig. 4, with the IASI NH3 ANNI V4.0 product overlaid for comparison. Also shown are the monthly mean planetary boundary layer heights (BLH) from ERA5 reanalysis. All three hotspots exhibit consistent seasonal differences in diurnal variability, a pattern that is consistent with the findings of Clarisse et al. (2021) from GIIRS observations and aligns with broader research on NH3 dynamics (Tevlin et al., 2017). Warmer months (April–September) show distinct day-night contrasts, while winter months (December–February) display minimal diurnal fluctuations.

For the North China Plain (Fig. 4a), FY-3E, FY-3F, CrIS, and IASI data collectively reveal that warm-season NH3 columns follow a clear diurnal pattern with a peak between 10:00 and 13:30 LST, and nighttime values approximately 50 %–60 % lower than daytime maxima. The peak corresponds to intensified agricultural activities (e.g., fertilization and livestock management) and temperature-driven volatilization from soils and animal waste, mechanisms identified by Ernst and Massey (1960) and Sutton et al. (2013) as primary drivers of daytime NH3 emission enhancement. Higher BLH during the day facilitates vertical diffusion of NH3 and contributes to the observed column dynamics Saylor et al., 2010). In contrast, winter NH3 columns over the North China Plain show negligible diurnal variation, which is probably due to the fact that cold-season emissions are limited by low temperatures and reduced agricultural activity (Kuang et al., 2020; Clarisse et al., 2021).

The Indo–Gangetic Plain (Fig. 4b) exhibits the highest overall NH3 columns among the three hotspots, with warm-season daytime columns nearly double those of the North China Plain. The lower NH3 columns over the North China Plain relative to the Indo–Gangetic Plain may partly reflect higher levels of NOx and SO2 (Clarisse et al., 2021; Ding et al., 2024), which can be oxidized to nitric and sulfuric acids and promote the conversion of gaseous NH3 to particulate ammonium, thereby shortening the effective lifetime of gas-phase NH3 (Liu et al., 2018; Wang et al., 2020). The most pronounced warm-season diurnal amplitude appears in the Indo–Gangetic Plain, with daytime columns up to three times higher than nighttime values. This pattern reflects the intensive irrigated agriculture and livestock-related emissions in the region, together with the strong temperature dependence of NH3 volatilization (Hempel et al., 2016; Perrone, 2020; Wang et al., 2020). Winter diurnal variations here are similarly weak, though NH3 columns remain higher than those in the other two regions due to persistent agricultural activity in milder winter conditions (Saraswati et al., 2018).

For the Central US (Fig. 4c), the diurnal variation pattern is broadly consistent with the Asian regions but with a smaller amplitude. Warm-season NH3 columns peak between 11:00 and 14:00 LST, with daytime values approximately 1.5–2 times higher than nighttime levels, while winter columns show little diurnal fluctuation. This consistency supports the interpretation that temperature, BLH dynamics, and diurnally varying emissions (e.g., agricultural activities) are common drivers of NH3 diurnal cycles across major high-emission regions. This interpretation is consistent with the analysis of NH3 measurements in the southeastern US by Saylor et al. (2010). The smaller amplitude compared to the Indo–Gangetic Plain likely reflects moderate temperature variations and balanced NH3 emission and deposition processes, as reported by Warner et al. (2017) in satellite observations of North American agricultural regions. Li et al. (2026) further emphasize that local deposition in vegetation-dense areas offsets large-scale transport, explaining the muted diurnal amplitude relative to Asian regions.

The spatial and seasonal dynamics of NH3 columns are further illustrated in Figs. 5–7 for the North China Plain, Indo–Gangetic Plain, Central US, respectively, retrieved from FY-3E/HIRAS-II, FY-3F/HIRAS-II and CrIS. Figure A1 further illustrates monthly mean NH3 column variations for these three regions, confirming consistent time series across the six daily satellite overpasses and coherent seasonal trends. All three regions exhibit a consistent seasonal peak in NH3 columns during spring to summer (March–August), with the lowest values in winter (December–February). For the North China Plain (Fig. 5), all retrievals at six distinct times show that NH3 columns peak in June, with spatial coverage expanding across the entire plain during warm months, reflecting widespread agricultural activities (e.g., wheat fertilization and livestock waste management) and favorable meteorological conditions for emission (Zhang et al., 2010; Zhan et al., 2021). In contrast, winter maps display sparse, localized high-NH3 patches, consistent with reduced agricultural activity and lower temperature-driven volatilization (Schjoerring et al., 1998).

For the Indo–Gangetic Plain, Fig. 6 shows that the region maintains the highest NH3 columns year-round, with spatial hotspots concentrated in the central plain during both day and night. However, daytime maps exhibit broader coverage extending to the plain's peripheries, while nighttime maps show more concentrated hotspots near emission sources. For the central US, Fig. 7 reveals the lowest overall NH3 columns among the three regions, with daytime spatial patterns spreading across the Corn Belt and Great Plains, and nighttime patterns contracting to smaller agricultural clusters.

The comparison between daytime and nighttime spatial patterns underscores the critical role of BLH in shaping NH3 distribution. Daytime maps display broader spatial coverage because higher BLH (typically 1–3 km during warm-season days) facilitates vertical and horizontal diffusion of NH3 away from emission sources. In contrast, nighttime maps show more localized NH3 columns, as lower BLH (often < 500 m at night) traps emissions near the surface (Walker et al., 2006; Wang et al., 2019). This pattern is especially pronounced over the North China Plain, where Fig. 5b–d show contiguous high-NH3 zones across the plain, while Fig. 5a, e and f limit high columns to central agricultural districts. For the Indo–Gangetic Plain, nighttime spatial maps also reveal persistent NH3 in the residual layer, an effect linked to BLH collapse in the evening, which traps NH3 at higher altitudes for several hours (Lonsdale et al., 2017; Kuang et al., 2020). This residual layer NH3 explains why some nighttime retrievals show broader coverage than expected, even with low BLH, and aligns with the observation by Clarisse et al. (2021) that NH3 vertical profile assumptions can impact nighttime retrieval accuracy.

To assess the impact of the filters on the retrieved NH3 diurnal variation, we perform an additional sensitivity analysis using three filtering configurations: a loose filter with surface AVK > 0.1 and absolute TC > 3 K, the current hotspot filter with surface layer AVK value > 0.3 and absolute TC > 5 K, and a stricter filter with surface layer AVK value > 0.4 and absolute TC > 7 K. The remaining quality-control criteria are kept consistent among the three configurations. The resulting spatial distributions over the North China Plain and Indo–Gangetic Plain and the corresponding diurnal cycles are shown in the Supplement. The total numbers of quality-controlled retrievals retained for each satellite under the different filtering configurations are also reported in the subpanels. The three filtering configurations lead to different data-retention rates, as expected, but the major NH3 hotspot patterns remain spatially similar and the regional diurnal cycles are not substantially distorted. This indicates that the main diurnal features discussed here are not strongly affected by the selected absolute TC and surface AVK thresholds, although the absolute column values and sampling density remain sensitive to filtering strength.

To evaluate the consistency of NH3 column retrievals from the polar-orbiting constellation, we perform spatiotemporally collocated cross-comparisons between FY-3E/HIRAS-II, FY-3F/HIRAS-II, CrIS, and geostationary FY-4B/GIIRS retrievals. The comparisons are conducted over the Indo–Gangetic Plain (15–35° N, 75–95° E) and the North China Plain (25–45° N, 110–130° E), where strong NH3 signals are frequently observed and FY-4B/GIIRS provides geostationary coverage. A polar-orbiting retrieval and a FY-4B/GIIRS retrieval are considered collocated when their longitude and latitude differences are both less than 0.5° and their observation time difference is less than 0.5 h. These thresholds are selected as a compromise between retaining sufficient matched samples and minimizing spatial and temporal representativeness differences.

The comparisons in Fig. 8 also show broad consistency across the matched local overpass periods sampled by the three polar-orbiting sensors. Here, the observational periods refer to the local-time periods of the polar-orbiting observations collocated with FY-4B/GIIRS, namely CrIS at 01:30/13:30 LST, FY-3E/HIRAS-II at 05:30/17:30 LST, and FY-3F/HIRAS-II at 10:00/22:00 LST. The agreement across these matched local-time periods indicates that the constellation retrievals are generally consistent with FY-4B/GIIRS over both nighttime/dawn and daytime/dusk observations, although the agreement is typically stronger during daytime because of larger thermal contrast and higher retrieval sensitivity.

We further perform seasonal comparisons between each polar-orbiting sensor and FY-4B/GIIRS using the same collocation and quality-filtering criteria. These results are shown in Fig. A2. The seasonal regression slopes mostly fall within the range of 0.7–1.1. The slopes slightly below unity in the annual comparison should not be interpreted as clear evidence of a persistent systematic bias in either FY-4B/GIIRS or the polar-orbiting retrievals. Instead, the remaining differences are likely due to their differences in retrieval sensitivity, collocation representativeness, sampling footprints, viewing geometry, and vertical sensitivity. These factors are particularly important for NH3 because of its short atmospheric lifetime and strong spatial heterogeneity.

We also compare the polar-orbiting satellite retrievals with ground-based FTIR NH3 column measurements at Hefei, China (Wang et al., 2022). Satellite retrievals within a 1° × 1° spatial window and a 1 h temporal window around the FTIR observations are averaged and compared with the corresponding measurements. The comparison is shown in Fig. A3. The satellite retrievals show reasonable consistency with the FTIR observations, although the number of collocated samples is limited, with slopes of 0.87 ± 0.22, 1.03 ± 0.08, and 1.13 ± 0.18 for FY-3E/HIRAS-II, FY-3F/HIRAS-II, and CrIS, respectively. The corresponding RMSE values range from 7.80 × 10¹⁵ to 1.11 × 10¹⁶ molec.cm-2. Overall, the annual and seasonal FY-4B/GIIRS comparisons, together with the Hefei FTIR comparison, support the broad consistency of the LEO constellation retrievals in capturing major NH3 spatial and temporal variability.

Figure 9 presents the diurnal and seasonal dynamics of surface AVK, TC, and column retrieval uncertainty over three representative hotspots, located in the North China Plain (37° N, 118° E), the Indo–Gangetic Plain (28° N, 78° E), and the Central US (36.5° N, 100.5° W). The statistics are derived from FY-3E/HIRAS-II, FY-3F/HIRAS-II, and CrIS retrievals within 2.5° × 2.5° boxes and are shown as monthly averages for January, April, July, and October across six local overpass times. To clarify the robustness of these monthly averages, the number of valid retrievals used for each region, month, and overpass time is provided in Table S2. Monthly means are shown only when more than 10 quality-controlled retrievals are available.

TC exhibits a clear diurnal pattern across all regions. It generally reaches positive maxima during midday overpasses at 10:00–13:30 LST, with peak values of about 10–14 K over the North China Plain and about 17–20 K over the Indo–Gangetic Plain and the Central US in warm months. Nighttime and early-morning TC values are often negative, especially over the Indo–Gangetic Plain and the Central US in January, April, and October, while the North China Plain shows weaker negative TC in January and positive TC throughout July and October. These seasonal contrasts indicate that warm-season daytime observations generally provide strong positive TC, whereas nighttime observations often occur under negative TC conditions

The surface layer value of column AVK, which quantifies the observational information content related to near-surface NH3, also shows pronounced diurnal and seasonal variability. Midday overpasses generally provide useful sensitivity to near-surface NH3 variations, with surface-layer column AVK values mostly ranging from about 0.5 to 0.9 at 10:00–13:30 LST, although the peak surface layer column AVK does not always occur at midday. The Indo–Gangetic Plain exhibits pronounced diurnal swings, with July midday surface layer column AVK reaching 0.6, while the Central US shows moderate values of 0.57–0.70 at 10:00–13:30 LST but higher values during some nighttime overpasses. Overall, warm-season daytime conditions generally provide favorable observational constraints for NH3 retrievals, although the surface layer value of column AVK can be high under negative TC conditions.

Column retrieval uncertainty generally decreases under conditions with larger absolute TC and higher surface AVK, but this relationship is not purely inverse and should not be interpreted as strictly monotonic. In particular, strongly negative TC can also enhance thermal infrared sensitivity because absorption-like spectral features may become emission-like features, as shown for lower-tropospheric pollution retrievals by Boynard et al. (2014). Therefore, TC should be interpreted together with AVK and retrieval uncertainty, rather than as a standalone monotonic sensitivity indicator.

Daytime bins often contain more valid retrievals than nighttime or dusk/evening bins, but the pattern is not uniform across regions and months (Table S2). Some bins have relatively sparse sampling, such as the Indo–Gangetic Plain in July at 05:30 and 22:00 LST and the Central US in October at 17:30 LST, whereas other daytime bins contain more than 1000 valid retrievals. These differences mainly reflect the combined effects of cloud screening and sensitivity-related quality filters.