the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 25 Nov 2025
Geostationary observations of atmospheric ammonia over East Asia: spatio-temporal variations revealed by three years of FY-4B/GIIRS measurements
Abstract. Satellite observations play a crucial role in quantifying ammonia sources by capturing large-scale variations of atmospheric NH3 concentrations. As the world's first geostationary hyperspectral infrared sounder, the Geostationary Interferometric Infrared Sounder (GIIRS) on board China's FengYun-4 satellite series provides a unique opportunity to monitor the diurnal cycle of NH3. Using NH3 retrievals between July 2022 and June 2025, this study investigates the spatio-temporal variability of NH3 columns over East Asia, with a focus on daytime variations (07:00–19:00 local time) in major agricultural regions. Inter-comparison with polar-orbiting IASI and CrIS data shows that GIIRS NH3 retrievals are consistent in capturing the spatial patterns and temporal dynamics. The NH3 peaks occur between March and July, with the timing shifting from north to south, reflecting regional differences primarily driven by agriculture activities. Validation with ground-based FTIR measurements at Hefei in eastern China demonstrates the accuracy of GIIRS NH3, with a correlation coefficient of 0.77 and an RMSE of 9.67 × 1015 molec/cm2, while reproducing daytime variations observed by FTIR. For major agricultural areas, the NH3 columns generally increase from early morning to late afternoon, reaching 1.10–1.56 times morning levels in summer and spring. Compared with GEOS-CF model simulations, the results reveal pronounced discrepancies in spatial distributions over the Sichuan Basin in southwestern China. These findings highlight the valuable capability of FY-4B/GIIRS in identifying and tracking daytime dynamics of NH3 sources over East Asia, offering new insights beyond current low-Earth orbit (LEO) instruments.
- Preprint
(6326 KB) - Metadata XML
-
Supplement
(5269 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-5699', Anonymous Referee #1, 28 Jan 2026
-
RC2: 'Comment on egusphere-2025-5699', Anonymous Referee #2, 11 Feb 2026
This manuscript by Sheng et al. describes a 3-year ammonia product retrieved over Asia from geostationary thermal IR spectrometer GIIRS aboard the FengYun-4 satellite. The retrieval is based on the FY-GeoAIR algorithm developed by the same group in a 2023 AMT paper (Zeng et al., 2023, https://doi.org/10.5194/amt-16-3693-2023). This work comprehensively compares GIIRS ammonia column amount with temporally matched counterparts from IASI and CrIS. The GIIRS columns are also evaluated against ground-based FTIR measurements in Hefei in east China. Seasonal and diurnal patterns are presented with the focus on a few source regions in north/southwest China and south Asia. The spatiotemporal patterns observed by GIIRS are compared with the MIX emission inventory and GEOS-CF CTM simulation, with significant differences noted. This paper presents quite solid work, and the ammonia retrieval from a geostationary hyperspectral imager is a significant advancement from the prior work. The authors may consider the following comments for further improvements of this manuscript.
The main general comment is about how the retrieval of this product differs from the algorithm paper, Zeng et al. (2023) and related, the consistency of using averaging kernel to quantify vertical sensitivity of the retrieved column. The algorithm paper fully resolves ammonia and water vapor profiles as 11 separate layers, whereas the current manuscript, presumably will serve as a data product paper, scales a single a priori ammonia profile. This implies that there is only one state vector element for ammonia, instead of 11. This appears to be a significant change of the retrieval setting, which should be better justified, evaluated, and augmented by details of the current retrieval setting and how these differ from the algorithm paper. The details of this retrieval are arguably more important than those in the algorithm paper for the users of this product.
A related issue is the averaging kernel and the vertical sensitivity of the product. In general, this manuscript may benefit from more discussion and presentation on this topic, as the diurnal coverage and diverse spatial coverage likely lead to large variations of vertical sensitivity. Namely, the lack of interesting signals in some time/place may be due to weak sensitivity to near-surface concentration. By switching from profile to single scaling factor retrieval, this manuscript seems to (partially) move from the optimal estimation view of averaging kernel, as in Zeng et al. 2023 and CrIS algorithm papers, to the DOAS view of averaging kernel as in the IASI v4 paper (Clarisse et al., 2023, https://doi.org/10.5194/amt-16-5009-2023). This could lead to profound confusion without detailed clarification. The “column averaging kernel” discussed in Section 2.1.1 and Fig. S1 seem to be the DOAS AVK, not the optimal estimation AVK. The actual “column averaging kernel”, in optimal estimation, is also a vector calculated from the averaging kernel matrix (e.g., see Eq. 8 in Connor et al., 2008, https://doi.org/10.1029/2006JD008336). The text definition in lines 151-153 of this manuscript seems to match the optimal estimation column averaging kernel, not the DOAS one shown in Fig. S1a. Figure S1 actually shows both DOAS (panel a) and optimal estimation AVKs (panel c) without clear distinction. The authors, which include developers of both IASI and CrIS algorithms, are encouraged to disentangle this issue.
Secondarily, the authors may think about the consistency of GIIRS, IASI, and CrIS filtering for the intercomparisons. Each comparison point, e.g., in Figs. 3 and 4, are averaged from a number of soundings for each instrument. Uncoordinated filtering schemes according to each instrument’s recommendation may lead to relative biases in sampling and sampling-induced differences even when they actually measure the same value at the same sounding location/time. For example, the GIIRS data filters negative values (line 158), but IASI allows negative (line 183). Ideally, the strictness of filtering should be comparable, or even better, the soundings are matched at individual overpass level.
Last, the authors may consider enhancing the comparison with a CTM (GEOS-CF is already presented over a short time period). It seems convincing that GIIRS/IASI/CrIS/FTIR agree reasonably well, with caveats noted in the previous two main comments. However, limited data-model comparisons (Fig. 6 and Fig. 15) indicate that the CTM world is very far away from observations. To the degrees feasible, the authors are encouraged to trade some of the similar maps that take large space (Figs. 2-4, 7, 12-14, S3, S16-18, S20-22) with maps that highlight the spatial/temporal differences between measured and simulated columns.
Minor comments:
Line 38: it is unclear how the timing shifts from north to south.
Lines 55-56: unclear how ammonia can influence source and sink of methane in any significant way. It does not appear in the reference.
Line 63: probably inaccurate to name Asia as a hot “spot”.
Line 93: EDGAR should be defined at first appearance.
Line 108: check if the ammonia, not the CO algorithm paper, should be cited here.
Line 117: MIX should be defined.
Figure 1: the ABCD boxes seem to be arbitrary and do not correspond to particular agricultural regions. Suggest some text on how they are selected.
Figure 1: showing a bottom up inventory might be a missed opportunity to present an all-time, high resolution map of the entire GIIRS data set. Consider labeling the source regions that appear in the main text (lines 261-263, 508-509) but only in supplement figure (Fig. S3).
Lines 155-158: do these criteria effectively remove non-detects like the CrIS product (lines 197-199)? Is it fair to say that GIIRS and CrIS exclude non-detects but IASI include them? How does that influence the instrument comparison?
Line 210: fix “showing broadly consistent with”
Figure 3: the axis limits should show negative values. Since the retrieval operates in linear space, instead of log space like CrIS, negative values are meaningful results. It is questionable to remove them, which will bias the mean value high.
Figure 6: presumably these are ammonia column enhancements calculated by subtracting the regional background fitted by Eq. 1, i.e., these are enhancements, not just GIIRS columns. The comparison between column (mol/m2) with emission is largely qualitative than quantitative, because the seasonality of column could be quite different from column-derived emissions (see Figs. 10-11 in Li et al. 2026, https://doi.org/10.5194/acp-26-703-2026). Does the enhancement calculation go one step further from column to emission? If so, it could be used to justify section 2.5 and Fig. 6.
Line 327: please elaborate what “data discrepancies” are.
Figure 7: I wonder if it is more effective to present the difference maps (e.g., IASI subtracting GIIRS). A general comment about figures is that there are so many maps that are very similar and hard to identify informative spatial pattern.
Lines 348 and 352: the Pearson correlation coefficient only depends on x and y data, not on a regression model. Should a coefficient of determination (R2) used here?
Line 360: unclear how can GRIIS be “most” reliable across all daytime hours?
Line 436: clarify what “fixed locations” means.
Line 492: should be Figs. S19-22.
Figure S1: the caption is very confusing and should be revisited. It does not seem to match each panel, with AVK notations mixed as pointed in the first main comment.
Figure S4: the titles of panels do not seem to match the caption. Please double check.
Figure S8: are the green dots all GIIRS observations? Please clarify in the caption.
Figure S11: please provide the spatial coverage.
Citation: https://doi.org/10.5194/egusphere-2025-5699-RC2
Data sets
FengYun-4B/GIIRS FYGeoAIR NH3 retrievals from July 2022 to June 2025 Zeng, Z.-C https://doi.org/10.5281/zenodo.17193848
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 386 | 183 | 24 | 593 | 55 | 25 | 21 |
- HTML: 386
- PDF: 183
- XML: 24
- Total: 593
- Supplement: 55
- BibTeX: 25
- EndNote: 21
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study introduces the first multi-year dataset of geostationary ammonia observations from the FengYun-4B satellite. A comprehensive evaluation of the dataset is presented through comparison with polar orbiting satellites, as well as ground-based FTIR observations. The peak months of ammonia over East Asia are also examined, revealing an interesting pattern related to agricultural emissions. The uniqueness of the data is highlighted by investigating diurnal ammonia variations from early morning to afternoon. Large discrepancies between observations and modelling, as revealed by the data, show that modelling still needs to improve its ability to capture diurnal ammonia variations. Overall, the paper is well structured and clearly written. The retrieval data are unique, and the findings are novel. Given that the IRS, which overlooks Europe and Africa, has recently been launched, and that more geostationary hyperspectral sounders may be in the planning stage, this study is timely. The paper fits the scope of ACP and can be published once the following minor issues have been addressed.
(1) In the introduction section, atmospheric chemistry modelling, such as GEOSchem, has been widely used to study ammonia at global and regional scales. Several studies related to improving NH₃ modelling are suggested for addition.
(2) Figure 1. Why do the observation domains have curved edges? Perhaps the domain has been filtered by viewing zenith angle. Please clarify.
(3) L166. Using the TC threshold may affect the results. Here, 5K is used. Please describe how much data can be retained after this filtering. Perhaps a histogram of the data numbers for different TCs could be made.
(4) Figure 2: The spatial maps for 7–9 h and 17–19 h show some unrealistic edges that look odd. Are these due to the different observation hours? Please explain and see if this can be mitigated.
(5) The agreement in Figure 5 looks very good. In Southeast Asia, there are strong biomass burning emissions every spring. How might this affect the peak? Currently, I see that the majority of the region peaks in March and April.
(6) The MIX inventory data in L315-316 can differ greatly from the observations. Could you provide one or two more references that use the MIX inventory? How reliable is this inventory?
(7) The number of data points in Figure 8 varies considerably between satellites, especially for CrIS, which has many fewer data points. Please explain.
(8) In Section 3.3, the Sichuan Basin is an interesting case, as the topographic effect can be significant. L454-458 does not provide much background information on this region. I suggest adding some references to studies of NH₃ in this region. What is the current understanding of ammonia emissions over this region from polar orbiting satellites?
(9) L530: FY-4C/GIIRS was launched in late 2025. This can be reflected in the rephrases. Could you also provide more details on FY-4C/GIIRS and explain what improvements it offers over 4A and 4B GIIRS?
(10) Please also provide a description of the IRS onboard the MTG, which was launched in 2025.