Review of

The NOAA Aerosol Reanalysis version 1.0 (NARA v1.0): Description of the Modeling System and its Evaluation

By Wei et al.

Overview:

The paper presents a global aerosol reanalysis data set for the year 2016 (NARA v1.0) produced by assimilating MODIS AOD in the GEFS aerosol model using the JEDI data assimilation framework. The reanalysis data and a control run (free) without AOD assimilation is evaluated with AOD and Angstroem exponent (AE) observations from Aeronet, surface PM observations derived from openAQ and speciated observations from the IMPROVE network. Further, it is intercompared against the MERRA-2 and CAMS aerosol reanalysis data sets. The authors find improved AOD compared to AERONET observations but no or only smaller improvements for AE and PM.  Overall NARA v1.0 seems closer to the MERRA-2 re-analysis than to CAMS.  

General remarks

The paper gives a reasonable description of the modelling system and the scientific accuracy of the NARA data set.  However, there are further updates required before the paper can be recommended for publication.

• It is an omission of the paper that NARA, CAMS and MERRA-2 are only compared against each other for AOD and regional vertical profiles. Instead, the evaluation against independent observations should be carried for CAMS and MERRA-2 too. The accuracy measure of the three data should be compared in the paper. It is of great interest to the reader to get know which of these data sets is closer to observations.
• The evaluation results should be represented in a more quantitative manner and always for NARA and the FREE run to document the impact of the AOD assimilation.  For example, the pdf plots do not have a colour scale legend and only some of them contain information about bias and correlation.
• The approach for the PM evaluation needs to be described in more detail, in particular the seemingly missing application of any quality control of the openAQ data should be justified. Further, the method to account for vertical stratification and spatial representativeness is not always clear.   
• The abstract should be revised to contain more factual information from the paper.  General remarks that are as such not direct conclusion from the paper should be avoided. 
• The data assimilation procedure seems to include an optimisation of the emissions (see also Fig 1) Please provide more detail on this important aspect from a technical and a scientific perspective.  

Specific comments:

 L 11: once published that is not a manuscript

L 11: “first ever”  may be misleading because there are other aerosol re-analysis

L 24:  There is no clear evidence in the paper that single-wavelength assimilation is the main factor of the limited impact.

L 28:  What is meant here by “climatologies”

L 28:  “In our opinion, such uncertainties may translate to inaccuracies in weather and climate modeling when impacts of aerosols on atmospheric radiation and/or cloud processes are considered.”  The paper does not deal in any way with the impact of aerosol and radiation and cloud processes and it is therefore not a conclusion of this paper. Please avoid statements that are not substantiated by the paper.

L 71: Please check of SOA’s and Nitrates are included in CAMSRA (I believe not)

L 114: openAQ data are know for the lack of  QC. Please expand on that and clarify the data sources that were compiled in the openAQ data set for 2016.  A good solution would be to provide a global map to show the spatial distribution of the PM2.5 observations.  

L 123: Why are gases mentioned here? Did you use the data , for example SO2 ?

L 134: Comment on the importance of representing Nitrates and Secondary Organic Aerosol especially for the accuracy of PM.

L 170-174: Please expand on these  aspects (SPPT and emission updates) and provide more information about their usefulness for the realism of the NARA data set.

L 174: Please show and discuss the modified emissions.  

L 205: Please provide more evidence for this statement “We hope ….”

L 232: Section 5.2 should also include a verification of CAMS-RA and MERRA-2 consistent with the verification of NARA and a comparison of the results

L 304:  Please, clarify how you account for the stratification. This would require making assumptions about the vertical aerosol profiles within the model grid box.  Just using the diagnosed air density, will not achieve that.

L 307: Please provide information about the size distribution of the different aerosol  component and  motivate the PM formulae.

L 358: Please add a comparison of the evaluation results from MERRA-2, CAMS and NARA here.

L 362-365: “These observations …  “ That is a very general statement and not really justified by the findings of the paper as you do not test the impact of NARA aerosol on radiation and clouds. Please include multiple references if you want to make a point here. For example, you mention yourself that Bozzo et al. 2017 (there is also an ACP paper) successfully used CAMSRA to represent aerosol in the ECMWF model.

Figures:

Fig 1:. Please here or in the text, provide more details in particular on the emission updates, the observation operators and the length of the forecast and assimilation window.

Fig 2:  Please add colour legend and basic statistics (bias, R) . Please show AOD in linear scale and not in log scale.  The pdf of AOD does not justify the use of the log scale.  

Fig 3:  Please, clarify if you show a spatial or temporal Correlation coefficient R.

Fig 12: Please add a colour legend, consider showing the plot for different regions and not just a global plot.

Review of “The NOAA Aerosol Reanalysis version 1.0 (NARA v1.0): Description of the Modeling System and its Evaluation

By Wei et al.

Overview:  This manuscript describes an aerosol product generated for the year 2016, referred to as NARA v1.0, using the GEFS aerosol model and NNR AOD product assimilation using a 3Den-Var data assimilation.  The results were compared to NASA’s MERRA-2 and ECMWF’s CAMSRA reanalysis products as well as against the AERONET AOD and surface PM2.5 data.  The 2016 NARA v1.0 output was found to be more consistent with AERONET than the free model run.  The results are also more consistent with MERRA2 than CAMSRA reanalysis.  This isn’t really surprising as the NARA v1.0 setup has more similarities to MERRA-2, using GOCART aerosol and NNR AOD product that is also assimilated for MERRA-2. 

General Remarks:  The main issue I have with this manuscript is referring to the output as a reanalysis when the results are only shown for 1 year.  A reanalysis is typically over a long period of time (10years+) as is the case for the other aerosol reanalysis products that are available.  I would call this an evaluation paper of the performance of the 3Den-Var with the GEFS model.  I think it’s fair to say that you will apply this setup in the future for generation of a reanalysis product, but don’t think you can call it that here. Given that, I think more needs to be done to define what is different here than in the Huang et al. 2023 paper that defined the data assimilation setup and evaluates the analysis results, although for only a month time period versus a year.  I also think there are more details needed in the manuscript, including the data assimilation setup.  More description of observations and reanalysis data used in this paper would also be helpful.  For the observations, was there any QC done prior to using the data? Was there any temporal averaging done in order to make point data comparable to model output?  I do want to highlight that reanalysis products are very important and once you are able to generate that, I think that will be a valuable contribution.

Specific Comments:

• First sentence in the abstract. I see what you are saying here, but I think this could be a bit misleading.  It could come off as you have the first aerosol reanalysis product ever, which is not the case.  Please reword this as “the first version of the aerosol reanalysis for NOAA”, or something like that in order to clarify.
• Page 4, first line: “In this study, we used and designed a specific JEDI-based 3D-EnVar DA configuration to produce the NOAA Aerosol ReAnalysis version 1.0 (NARA v1.0). “ Please elaborate on how this differs from the setup described in Huang et al. 2023.  The difference is not clear and why is the chosen configuration better for a reanalysis?
• Have you also looked at performance of fine mode fraction or fine/coarse AOD?
• I think looking at timeseries that are not monthly-averaged would also be helpful, at AERONET sites for example, to see if your product is able to generate daily variability.  Perhaps you can select reference sites that have a good amount of data and at locations that are representative of big dust/smoke/pollution/sea salt dominated regimes.