Impact of assimilating NOAA VIIRS Aerosol Optical Depth (AOD) observations on global AOD analysis from the Copernicus Atmosphere Monitoring Service (CAMS)

Garrigues, Sebastien; Ades, Melanie; Remy, Samuel; Flemming, Johannes; Kipling, Zak; laszlo, Istvan; Parrington, Mark; Inness, Antje; Ribas, Roberto; Jones, Luke; Engelen, Richard; Peuch, Vincent-Henri

doi:https://doi.org/10.5194/egusphere-2023-398

Preprints

https://doi.org/10.5194/egusphere-2023-398

Preprints

21 Mar 2023

| 21 Mar 2023

Impact of assimilating NOAA VIIRS Aerosol Optical Depth (AOD) observations on global AOD analysis from the Copernicus Atmosphere Monitoring Service (CAMS)

Sebastien Garrigues, Melanie Ades, Samuel Remy, Johannes Flemming, Zak Kipling, Istvan laszlo, Mark Parrington, Antje Inness, Roberto Ribas, Luke Jones, Richard Engelen, and Vincent-Henri Peuch

Abstract. Global monitoring of aerosols is required to analyse the impacts of aerosols on air quality and to understand their role in modulating the climate variability. The Copernicus Atmosphere Monitoring Service (CAMS) provides near real time forecasts and reanalyses of aerosols using the ECMWF Integrated Forecasting System (IFS), constrained by the assimilation of MODIS and PMAp Aerosol Optical Depth (AOD). Given the potential end-of-lifetime of MODIS AOD, implementing new AOD observations in the CAMS operational suite is a priority to ensure the continuity of the CAMS forecast performances. The objective of this work is to test the assimilation of the NOAA VIIRS AOD product from S-NPP and NOAA20 satellites in the IFS model. Simulation experiments assimilating VIIRS on top or in place of MODIS were carried out from June 2021 to November 2021 to evaluate the impacts on the AOD analysis.

For maritime aerosol background, the assimilation of VIIRS and the use of VIIRS from NOAA20 as an anchor reduce the analysis AOD values compared to MODIS-based experiments, in which the analysis values were too high due to the positive bias of MODIS/TERRA over ocean. Over land, the assimilation of VIIRS induces a large increase in the analysis over biomass burning regions where VIIRS shows larger AOD than MODIS. For dust source regions, the analysis is reduced when VIIRS is assimilated on top of or in place of MODIS, particularly over the Sahara, Arabian Peninsula and few places in Asia in the July-August period. The assimilation of VIIRS leads to an overall reduction of the bias in AOD analysis evaluated against AERONET measurements with the largest bias reduction over Europe, desert, and maritime sites.

Received: 04 Mar 2023 – Discussion started: 21 Mar 2023

Download & links

Preprint (PDF, 2007 KB)

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (2007 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

22 Sep 2023

Impact of assimilating NOAA VIIRS aerosol optical depth (AOD) observations on global AOD analysis from the Copernicus Atmosphere Monitoring Service (CAMS)

Sebastien Garrigues, Melanie Ades, Samuel Remy, Johannes Flemming, Zak Kipling, Istvan Laszlo, Mark Parrington, Antje Inness, Roberto Ribas, Luke Jones, Richard Engelen, and Vincent-Henri Peuch

Atmos. Chem. Phys., 23, 10473–10487, https://doi.org/10.5194/acp-23-10473-2023,https://doi.org/10.5194/acp-23-10473-2023, 2023

Short summary

Sebastien Garrigues et al.

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2023-398', Anonymous Referee #1, 12 Apr 2023

The authors describe the assimilation of NOAA-VIIRS AOD in CAMS atmospheric model. This is an important study due to the drifting of MODIS Terra and Aqua satellites that will soon need to be replaced by VIIRS. The assimilation methodology is clearly defined, and the modeling results are clearly presented and explained. Overall, I recommend publication in ACP with two minor suggestions that might be taken into account by the authors: (1) I would encourage the authors to further extend the discussion on the sensitivity of the assimilation method towards specific aerosol types. This could be done for example by the evaluation of model performance at specific events when the dominant atmospheric composition is known (e.g. strong biomass burning or dust outbreaks). (2) I suggest that the discussion in Session 5 (especially 5.2 and 5.3) is based on bias improvement considerations (i.e. where and when the model performs better) rather than on the differences between the model runs.

Citation: https://doi.org/10.5194/egusphere-2023-398-RC1
- AC1:
  'Reply on RC1', Sebastien Garrigues, 19 Jun 2023
  We would like to thank reviewer-1 for the interest shown in our paper and the valuable feedbacks that will help us to improve the quality of the paper.
  We agree with the two following suggested changes:
  I would encourage the authors to further extend the discussion on the sensitivity of the assimilation method towards specific aerosol types. This could be done for example by the evaluation of model performance at specific events when the dominant atmospheric composition is known (e.g. strong biomass burning or dust outbreaks). »
  =>Figure 10 provides the evaluation of the AOD analysis from each experiment against AERONET data for distinct groups of sites influenced by specific aerosol species. Fig. 10a shows the evaluation for desert sites which are mainly influenced by dust and Fig. 10b presents the evaluation for maritime sites mainly which are influenced by sea salt. In the revision, we added a similar regional plot for 47 West US sites which were influenced by the intense biomass burning events from mid-August to end of September 2020. All experiments have large negative and positive biases over this period compared to June-July and October-November. The AOD related to these extreme fires were poorly estimated by both MODIS and VIIRS, probably because of large cloud contamination and representativity uncertainties in the aerosol models. We added these new results in the revised version : p 9 line 225 ; p 12, line 308-312 and p 22, line 504.
  
  « I suggest that the discussion in Session 5 (especially 5.2 and 5.3) is based on bias improvement considerations (i.e. where and when the model performs better) rather than on the differences between the model runs. »
  
  =>We modified the discussion in Section 5.2 to better emphasize where and when the assimilation of VIIRS leads to more accurate AOD analysis.
  
  Citation: https://doi.org/10.5194/egusphere-2023-398-AC1
RC2:
'Comment on egusphere-2023-398', Anonymous Referee #2, 18 Apr 2023

In this manuscript, the authors present a study showing the impact of assimilating VIIRS AOD observations alone and jointly with MODIS AOD in CAMS. The paper is well written and is scientifically interesting, regarding that MODIS instruments are going to retire eventually. I recommend accepting it for publication.
I have minor comments:
Abstract l19-20: I think the dates of your experiments need to be checked: The year is 2020 in your following text while in the conclusion it is 2021-2022.
L23-24: “Over land, the assimilation of VIIRS induces a large increase in the analysis over biomass burning regions where VIIRS shows larger AOD than MODIS.”: You may want to provide reasons on why VIIRS has larger AOD than MODIS?
L109-114: Any references about VIIRS NOAA algorithm should be provided.
L155: You are assimilating total AOD at 550nm, however AOD distribution is non-gaussian. Do you take it into account in your assimilation system? Do you perform any quality control of your data using filters?
Section 3.3: You may want to revise your title to include VIIRS as well.

Citation: https://doi.org/10.5194/egusphere-2023-398-RC2
- AC2:
  'Reply on RC2', Sebastien Garrigues, 19 Jun 2023
  We would like to thank reviewer-2 for the interest shown in our paper and the valuable feedbacks that will improve the quality of the paper. Below, we provide the answers to the comments.
  
  “Abstract l19-20: I think the dates of your experiments need to be checked: The year is 2020 in your following text while in the conclusion it is 2021-2022.”
  =>We agree and we corrected the dates in the conclusion.
  “L23-24: “Over land, the assimilation of VIIRS induces a large increase in the analysis over biomass burning regions where VIIRS shows larger AOD than MODIS.”: You may want to provide reasons on why VIIRS has larger AOD than MODIS?”
  =>VIIRS and MODIS exploit distinct aerosol models, which can explain part of the larger VIIRS AOD values. Tao et al., 2017 showed that the dust scattering properties are overestimated in the MODIS deep blue retrieval algorithm which results in a negative AOD bias over desert regions. Besides, VIIRS algorithm is applying dynamically varying aerosol models during the retrieval: it thus potentially carries more information than the MODIS product which relies on prescribed and fixed in space and time aerosol models. Another source of explanation is cloud contamination. The use of heavy smoke test in the VIIRS algorithm leads to a better discrimination between aerosols and clouds and thus reduced cloud classification commission errors compared to MODIS. We better emphasized these explanations in the revised manuscript p2, line 24 and p28, lines 470-472 and lines 500-504
  “L109-114: Any references about VIIRS NOAA algorithm should be provided.”
  =>The NOAA VIIRS AOD retrieval algorithm is described in Laszlo, I. and Liu, H.: EPS Aerosol Optical Depth (AOD) Algorithm Theoretical Basis Document, NOAA-NESDISSTAR,Center for Satellite Applications and Research, https://www.star.nesdis.noaa.gov/jpss/documents/ATBD/ATBD_EPS_Aerosol_AOD_v3.4.pdf (last access: 9 November 2022), 2020.. We added this reference in the product description Section 2 (line 94)
  
  “L155: You are assimilating total AOD at 550nm, however AOD distribution is non-gaussian. Do you take it into account in your assimilation system? Do you perform any quality control of your data using filters?”
  =>In the IFS assimilation system, both background and satellite observation error probability distribution functions are assumed to be Gaussian. We added this statement in line 140 page 6.
  Distinct quality controls are applied to the observation data at different steps of the assimilation system:
  Prior to their use in IFS: the observation data are filtered by selecting only best quality retrievals according to the data provider recommendations. In CAMS, MODIS DT retrievals associated with a quality assessment (QA) equal to three over land and larger or equal to one over ocean are selected. DB retrievals associated with QA larger or equal to two are used to gap-fill DT over land. For PMAp and VIIRS, best quality retrievals are selected.
  
  Within IFS, a first filter consists in removing the observations that are far from the model first guess value. Then, a variational quality control, based on a Bayesian formalism, consists in reducing the analysis weight given to the observations which show large departure with the model first guess but still fall within an acceptable distance from the model (Andersson and Janarvinen (1999).
  
  We introduced a paragraph dedicated to the quality control in the revised manuscript to better emphasize these aspects (page 7, line 170)
  “Section 3.3: You may want to revise your title to include VIIRS as well.”
  =>We changed the title to include VIIRS as suggested.
  
  Citation: https://doi.org/10.5194/egusphere-2023-398-AC2

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2023-398', Anonymous Referee #1, 12 Apr 2023

The authors describe the assimilation of NOAA-VIIRS AOD in CAMS atmospheric model. This is an important study due to the drifting of MODIS Terra and Aqua satellites that will soon need to be replaced by VIIRS. The assimilation methodology is clearly defined, and the modeling results are clearly presented and explained. Overall, I recommend publication in ACP with two minor suggestions that might be taken into account by the authors: (1) I would encourage the authors to further extend the discussion on the sensitivity of the assimilation method towards specific aerosol types. This could be done for example by the evaluation of model performance at specific events when the dominant atmospheric composition is known (e.g. strong biomass burning or dust outbreaks). (2) I suggest that the discussion in Session 5 (especially 5.2 and 5.3) is based on bias improvement considerations (i.e. where and when the model performs better) rather than on the differences between the model runs.

Citation: https://doi.org/10.5194/egusphere-2023-398-RC1
- AC1:
  'Reply on RC1', Sebastien Garrigues, 19 Jun 2023
  We would like to thank reviewer-1 for the interest shown in our paper and the valuable feedbacks that will help us to improve the quality of the paper.
  We agree with the two following suggested changes:
  I would encourage the authors to further extend the discussion on the sensitivity of the assimilation method towards specific aerosol types. This could be done for example by the evaluation of model performance at specific events when the dominant atmospheric composition is known (e.g. strong biomass burning or dust outbreaks). »
  =>Figure 10 provides the evaluation of the AOD analysis from each experiment against AERONET data for distinct groups of sites influenced by specific aerosol species. Fig. 10a shows the evaluation for desert sites which are mainly influenced by dust and Fig. 10b presents the evaluation for maritime sites mainly which are influenced by sea salt. In the revision, we added a similar regional plot for 47 West US sites which were influenced by the intense biomass burning events from mid-August to end of September 2020. All experiments have large negative and positive biases over this period compared to June-July and October-November. The AOD related to these extreme fires were poorly estimated by both MODIS and VIIRS, probably because of large cloud contamination and representativity uncertainties in the aerosol models. We added these new results in the revised version : p 9 line 225 ; p 12, line 308-312 and p 22, line 504.
  
  « I suggest that the discussion in Session 5 (especially 5.2 and 5.3) is based on bias improvement considerations (i.e. where and when the model performs better) rather than on the differences between the model runs. »
  
  =>We modified the discussion in Section 5.2 to better emphasize where and when the assimilation of VIIRS leads to more accurate AOD analysis.
  
  Citation: https://doi.org/10.5194/egusphere-2023-398-AC1
RC2:
'Comment on egusphere-2023-398', Anonymous Referee #2, 18 Apr 2023

In this manuscript, the authors present a study showing the impact of assimilating VIIRS AOD observations alone and jointly with MODIS AOD in CAMS. The paper is well written and is scientifically interesting, regarding that MODIS instruments are going to retire eventually. I recommend accepting it for publication.
I have minor comments:
Abstract l19-20: I think the dates of your experiments need to be checked: The year is 2020 in your following text while in the conclusion it is 2021-2022.
L23-24: “Over land, the assimilation of VIIRS induces a large increase in the analysis over biomass burning regions where VIIRS shows larger AOD than MODIS.”: You may want to provide reasons on why VIIRS has larger AOD than MODIS?
L109-114: Any references about VIIRS NOAA algorithm should be provided.
L155: You are assimilating total AOD at 550nm, however AOD distribution is non-gaussian. Do you take it into account in your assimilation system? Do you perform any quality control of your data using filters?
Section 3.3: You may want to revise your title to include VIIRS as well.

Citation: https://doi.org/10.5194/egusphere-2023-398-RC2
- AC2:
  'Reply on RC2', Sebastien Garrigues, 19 Jun 2023
  We would like to thank reviewer-2 for the interest shown in our paper and the valuable feedbacks that will improve the quality of the paper. Below, we provide the answers to the comments.
  
  “Abstract l19-20: I think the dates of your experiments need to be checked: The year is 2020 in your following text while in the conclusion it is 2021-2022.”
  =>We agree and we corrected the dates in the conclusion.
  “L23-24: “Over land, the assimilation of VIIRS induces a large increase in the analysis over biomass burning regions where VIIRS shows larger AOD than MODIS.”: You may want to provide reasons on why VIIRS has larger AOD than MODIS?”
  =>VIIRS and MODIS exploit distinct aerosol models, which can explain part of the larger VIIRS AOD values. Tao et al., 2017 showed that the dust scattering properties are overestimated in the MODIS deep blue retrieval algorithm which results in a negative AOD bias over desert regions. Besides, VIIRS algorithm is applying dynamically varying aerosol models during the retrieval: it thus potentially carries more information than the MODIS product which relies on prescribed and fixed in space and time aerosol models. Another source of explanation is cloud contamination. The use of heavy smoke test in the VIIRS algorithm leads to a better discrimination between aerosols and clouds and thus reduced cloud classification commission errors compared to MODIS. We better emphasized these explanations in the revised manuscript p2, line 24 and p28, lines 470-472 and lines 500-504
  “L109-114: Any references about VIIRS NOAA algorithm should be provided.”
  =>The NOAA VIIRS AOD retrieval algorithm is described in Laszlo, I. and Liu, H.: EPS Aerosol Optical Depth (AOD) Algorithm Theoretical Basis Document, NOAA-NESDISSTAR,Center for Satellite Applications and Research, https://www.star.nesdis.noaa.gov/jpss/documents/ATBD/ATBD_EPS_Aerosol_AOD_v3.4.pdf (last access: 9 November 2022), 2020.. We added this reference in the product description Section 2 (line 94)
  
  “L155: You are assimilating total AOD at 550nm, however AOD distribution is non-gaussian. Do you take it into account in your assimilation system? Do you perform any quality control of your data using filters?”
  =>In the IFS assimilation system, both background and satellite observation error probability distribution functions are assumed to be Gaussian. We added this statement in line 140 page 6.
  Distinct quality controls are applied to the observation data at different steps of the assimilation system:
  Prior to their use in IFS: the observation data are filtered by selecting only best quality retrievals according to the data provider recommendations. In CAMS, MODIS DT retrievals associated with a quality assessment (QA) equal to three over land and larger or equal to one over ocean are selected. DB retrievals associated with QA larger or equal to two are used to gap-fill DT over land. For PMAp and VIIRS, best quality retrievals are selected.
  
  Within IFS, a first filter consists in removing the observations that are far from the model first guess value. Then, a variational quality control, based on a Bayesian formalism, consists in reducing the analysis weight given to the observations which show large departure with the model first guess but still fall within an acceptable distance from the model (Andersson and Janarvinen (1999).
  
  We introduced a paragraph dedicated to the quality control in the revised manuscript to better emphasize these aspects (page 7, line 170)
  “Section 3.3: You may want to revise your title to include VIIRS as well.”
  =>We changed the title to include VIIRS as suggested.
  
  Citation: https://doi.org/10.5194/egusphere-2023-398-AC2

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Sebastien Garrigues on behalf of the Authors (19 Jun 2023) Author's response Author's tracked changes Manuscript

ED: Publish as is (11 Jul 2023) by Stelios Kazadzis

AR by Sebastien Garrigues on behalf of the Authors (04 Aug 2023)

Journal article(s) based on this preprint

22 Sep 2023

Impact of assimilating NOAA VIIRS aerosol optical depth (AOD) observations on global AOD analysis from the Copernicus Atmosphere Monitoring Service (CAMS)

Sebastien Garrigues, Melanie Ades, Samuel Remy, Johannes Flemming, Zak Kipling, Istvan Laszlo, Mark Parrington, Antje Inness, Roberto Ribas, Luke Jones, Richard Engelen, and Vincent-Henri Peuch

Atmos. Chem. Phys., 23, 10473–10487, https://doi.org/10.5194/acp-23-10473-2023,https://doi.org/10.5194/acp-23-10473-2023, 2023

Short summary

Sebastien Garrigues et al.

Viewed

Total article views: 510 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
368	123	19	510	7	7

HTML: 368
PDF: 123
XML: 19
Total: 510
BibTeX: 7
EndNote: 7

Views and downloads (calculated since 21 Mar 2023)

Month	HTML	PDF	XML	Total
Mar 2023	161	44	6	211
Apr 2023	80	21	4	105
May 2023	17	13	1	31
Jun 2023	42	16	5	63
Jul 2023	23	11	1	35
Aug 2023	22	9	0	31
Sep 2023	23	9	2	34

Cumulative views and downloads (calculated since 21 Mar 2023)

Month	HTML	PDF	XML	Total
Mar 2023	161	44	6	211
Apr 2023	80	21	4	105
May 2023	17	13	1	31
Jun 2023	42	16	5	63
Jul 2023	23	11	1	35
Aug 2023	22	9	0	31
Sep 2023	23	9	2	34

Viewed (geographical distribution)

Total article views: 513 (including HTML, PDF, and XML) Thereof 513 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 22 Sep 2023

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (2007 KB)
Metadata XML

Short summary

The Copernicus Atmosphere Monitoring Service (CAMS) provides global monitoring of aerosols using the ECMWF forecast model constrained by the assimilation of satellite Aerosol Optical Depth (AOD). This work aims at testing the assimilation of the NOAA VIIRS AOD product in the ECMWF model. It shows that the introduction of VIIRS in the CAMS data assimilation system enhances the accuracy of the aerosol analysis particularly over Europe, desert and maritime sites.


Total:	0
HTML:	0
PDF:	0
XML:	0