| 27 Apr 2026
Global aerosol composition constraints from simultaneous data assimilation of satellite AOD and trace gas observations
Abstract. The integration of satellite aerosol optical depth (AOD) and trace gas observations using data assimilation has the potential to improve our understanding of aerosol composition. This study evaluates these synergistic effects through combined constraints on total aerosols by AOD and on secondary aerosol formation by trace gases. The simultaneous data assimilation (DA) of NO2, SO2, CO, and HNO3 from OMI, TROPOMI, MOPITT, and MLS, together with AOD from MODIS and VIIRS, improved aerosol analyses in most cases compared to conventional DA runs that separately assimilate AOD or trace gases satellite observations. Validation against independent surface observations of sulfate, nitrate, and ammonium (SNA), and PM2.5 showed improved agreements by 6–98 % compared to the conventional DA runs and the control simulation without any data assimilation. Notably, the reduction in PM2.5 model biases exceeded that achieved by the conventional DA of AOD by 56 % in Northeast Asia. These improvements were achieved by reduced SO2 and soil dust emissions by 30 % and 60 % globally and increased NOx and carbonaceous aerosol emissions by 30 % and 15 %. The simultaneous DA provides even larger reductions in SNA and AOD biases by up to 25 % and 48 % respectively, when the current generation instruments (TROPOMI and VIIRS) is used, instead of the previous generation instruments (OMI and MODIS). This coupled aerosol and trace gas DA framework offers significant advantages for improving global aerosol composition analyses, informing policy decisions with co-benefits for air quality and climate, and optimizing the use of the current satellite observing network.
Competing interests: One of the co-authors is a member of the editorial board of Atmospheric Chemistry and Physics.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
Overview:
The study presents a comprehensive comparison of different satellite retrieval assimilation strategies utilising the CHASER V4.0 Chemical Transport Model (CTM) with a Kalman Filter approach. This data assimilation (DA) method is specifically designed to simultaneously adjust both atmospheric concentrations and emissions, thereby enabling a more accurate representation of atmospheric composition.
The analysis focuses on two principal groups of assimilated observations: gas-phase retrievals and aerosol optical depth (AOD) measurements. The gas-phase retrievals include tropospheric column measurements of nitrogen dioxide (NO2), total column sulphur dioxide (SO2) from the Ozone Monitoring Instrument (OMI), carbon monoxide (CO) from MOPITT, and stratospheric profiles for ozone (O3) and nitric acid (HNO3) from the Microwave Limb Sounder (MLS). In addition, AOD measurements are incorporated using data from MODIS.
The findings indicate that the combined assimilation of trace-gas and AOD observations yields the most robust results. This outcome is validated through evaluations using AERONET AOD readings, surface particulate matter (PM) data from Europe, North America, and China, and results from the KROKOS aircraft campaign. Furthermore, the authors explore the impact of assimilating data from TROPOMI and VIIRS in place of OMI and MODIS, providing additional insight into the effectiveness of different satellite retrieval combinations.
General comments:
The paper demonstrates the efficacy of a state-of-the-art data assimilation (DA) system for multiple satellite retrievals of atmospheric composition, capable of constraining both emission fluxes and atmospheric concentrations. The system exhibits robustness, as incorporating additional and complementary satellite observations enhances the realism of concentration field analyses and subsequent forecasts.
The authors observe that their DA system tends to adjust emissions more than concentrations, warranting further discussion. Specifically, since the operation of any DA system is influenced by the selection of prior uncertainties or perturbations during propagation within the Kalman filter framework, it would be beneficial to explore how these parameter choices affect the relative adjustments to emissions and concentrations. A tabulated summary detailing the assumed or typical calculated uncertainties for emission fluxes and mass mixing ratios would provide valuable clarity.
Reporting on posterior emissions should be contextualized alongside prior uncertainties. Additionally, because both concentrations and emissions are subject to modification, updates to emissions alone cannot offer a comprehensive commentary on emission fluxes; issues related to emissions may still manifest as concentration changes.
A methodological limitation—either inherent to the method or its description—appears to be the influence of Aerosol Optical Depth (AOD) assimilation on primary and secondary aerosol component concentrations. While all components affect AOD, only the surface fluxes of primary dust (AE > 0.5) and carbonaceous aerosol (AE < 0.5) seem modifiable. Clarification on the distribution among the three dust bins, and the consideration of sea salt bin fluxes, is necessary. It remains uncertain why the AOD assimilation is designed to only correct dust and carbonaceous aerosol. Greater detail regarding the specific impacts of AOD assimilation on individual aerosol concentrations and their respective fluxes is required.
Evaluation using in-situ data is incomplete. Since NO2, SO2, and O3 measurements are assimilated, the analyses should also encompass surface air quality( NO2, SO2, and O3) observations from European, North American, and Chinese networks, similar to the approach used for PM2.5.
The evaluation appears to focus primarily on analysis of the concentration. Including an assessment of multi-day forecasts with optimized emissions could further motivate the significance of emission inversion.
In the evaluation section, performance statistics are frequently presented both narratively and in tables, resulting in unnecessary duplication. Retaining the tables while consolidating textual references is advisable.
The comparison between OMI/MODIS and TropOMI/VIIRS could be moved to supplementary materials, as it does not fundamentally enhance the primary discussion on combined AOD and gas retrieval DA and inversions. Results indicate both improvements and degradations with the newer observations, which should be clearly communicated.
Chapter 6 (Discussion and Summary) needs to be streamlined. Conclusions—including findings and future directions—are very brief, and topics such as insufficient constraints on vertical profiles could be addressed. The ensemble’s limited capacity as such to provide sufficiently large emission uncertainty is another issue worth discussing. On the other hand, there is no need for an extensive overview of alternative or forthcoming satellite products (Section 6.3) in this papaer; new satellites should only be mentioned if they address concerns directly raised in the manuscript. Similarly, the model performance section (6.2) can be condensed. The brief “Climate and Health” section (6.4) may be omitted, as it contributes little new information specific to this paper.
Specific comments:
L8 clarify “conventual”
L51 The CAMS re-analysis (Inness et al. 2019, Flemming et al. 2017) assimilate trace gases and AOD simultaneously, although not including the update of emissions.
L74 Sea salt is part of the model but seems to be not included in the DA framework. Why ?
L 107 Please clarify if x constrains both the concentrations and the emissions. Please provide more information about your framework to optimise emissions
L 117 Please explain the self-consistency test here
L 127 Does this mean the ensemble systematically underestimated the emissions uncertainty? What is the motivation of the minimum values and where doe the number come from?
L 131 Why is sea salt no considered here ?
L 153 Please check for duplication of the information in section 2.2 and table 1. Having the info in the table should be sufficient
L 263 Table 2 also contain the info about MLS.
L 263 Please modify Fig 1 to clarify that only sea salt and OC are optimized.
L 285 Please compare also against the surface NO2, SO2 and O3 observations from the AQ networks (as for PM2.5)
L 301 Please specify the location of the KORUS campaign
L 324 It remains unclear how you attribute the concentration analysis changes to emissions modification and direct concentration updates (see general comment)
L333 Why is this the case. What choices have been made to suppress the impact.
L 341 Please clarify how this finding was derived. The attribution seems a bit ad hoc.
L 351. Please clarify this explanation. Say more clearly that Aero DA improved R and RMSE but slightly degraded the biases.
L 356 Please discuss in this section the inability of the DA system to modify the profile shape – or the opposite.
L 458 Please clarify how this was identified? .
L 485 Please discuss here more clearly that your experiments only cover the summer period. Has this taken into account in the comparison of the emissions totals?
L 570 The summary should also contain the OMI/MODIS vs TropOMI/VIIRS comparison
For section 5 and 6 – please see my general comment. They should be completely reviewed.
Figure 5: Please add time and location of the profile observations
Figure 6: Why are the PM 2.5 no circles but grid points ?
Table 2: Mention the MLS observations