Review of "The GERB Obs4MIPs Radiative Flux Dataset: A new tool for climate model evaluation" by Russell et al. 2024 ESSD

A valuable and unique geostationary satellite dataset observing the diurnal cycle of Earth's radiation budget over the Africa-Atlantic hemisphere has been reprocessed and formatted to optimise and aid exploitation by climate modelling centres. The manuscript carefully outlines the rational, decisions necessary for dealing with data gaps and issues that users of the dataset need to consider in best using the product. An illustration of its use in evaluating climate model simulations is presented. The analysis is reasonable and the work will help in promoting this useful dataset to the community. 

Overall, my opinion is that a more focused descrpition of the data, highlighting strengths and weaknesses and presenting quantitative assessment of the uncertainty would improve the impact of this dataset. Although interesting, I am not sure the illustration of the dataset's use in evaluating a climate model simulation is central to a dataset description paper and would benefit from a more detailed exploration elsewhere. However, if the authors prefer this more lengthy discussion, I do not see any scientific reason to object and I consider that this work is suitable for publication with only minor modifications listed below.

1) Title - I would not describe GERB Obs4MIP as a tool? How about something like: 

"The GERB Obs4MIP: a dataset for evaluating diurnal and monthly variation in top of atmosphere radiative fluxes in climate models."

2) L8 - since GERB data has been used in model evaluation for 20 years it is not new. What is new is the Obs4MIP aspect ("newly reprocessed" may be more appropriate). 

3) L11 "approximately" since it is not a square 60S-60N, 60W-60W domain

4) L19 ":" --> "," (or we could compromise with a ";")!

5) L20 improved fidelity, relative to GC3.1

6) L27 Since outgoing longwave is used later it may be worth stating here e.g. "emitted thermal infrared (outgoing longwave) and solar reflected..."

7) L36 also Allan et al. (2011), Examination of long-wave radiative bias in general circulation models over North Africa during May–July. Q.J.R. Meteorol. Soc., 137: 1179-1192. https://doi.org/10.1002/qj.717

8) L55 add a line pointing to the fact that the HR product is a resolution enhanced version of the GERB broadband radiative fluxes using SEVIRI narrow band measurements.

9) L63 "are" --> "were"

10) L72 why is 70 degrees zenith the cut-off and not slightly more or less?

11) L74 "proceed without prejudice" is not really clear what is meant (sounds like legal speak). Is one observation representative of a 100x100km area? Presumably there are mostly lots of values per 1x1 degree box.

12) L76 incoming solar flux from what dataset or algorithm?

13) Figure 3 - the colour bar seems to bear no resemblance to the maps. Also this figure could be designed in a way to maximise the size of the panels (or reduce the size of the plot and all the dead space)

14) L187 "exiting" --> "existing"

15) L200 this paragraph could be reduced to the first line plus "(compare Figures 5 and 3)".

16) L205 these lines are difficult to understand. Can they be written more clearly? There is a lot of detail in this section that may be unnecessary for users of the dataset so another option is to more briefly note issues and refer to prior papers if the reader is interested.

17) Figures 6/7 - could these be designed to fit better on the page and maximise the size of the panels? Does the line at 25oS relate to a dead or damaged pixel?

18) L251 - of course it is feasible to fill missing GERB data with GERB like - there is no need to hypothesise, just quantify the expected error and assess whether this is tolerable for the designed usage of the dataset. A key line to emphasise is that the strategy is to "fill missing GERB fluxes with the corresponding GERB-like fluxes, adjusted by the GERB/GERB-like ratio calculated at the monthly hourly mean temporal and

1⁰ spatial scale." and now the associated errors are quantified.

19) L259 1 degree "latitude/longitude" (throughout)

20) Figure 8 - make text bigger. It's not obvious what the use of this is since the decision to scale GERB-like with the ratio seems obvious (and the mean bias of about zero is by design)

21) L270 - the improvement in agreement between GERB and GERB-like after essentially removing the bias is obvious by design. This sentence can be removed. Section 2.5 seems to be the main analysis following the rational for the approach, which could be more concisely presented as the method

22) Figure 9 - it's not clear to me what the benefit of showing all the unfilled results as well as all the filled results. If the idea is to compare Figure 9 and Figure 4 then an unfilled vs filled line could be presented on the same figure. Are all times lumped together? In this case massive errors due to missing parts of the diurnal cycle will be introduced won't it (which would obviously not be considered in practice)?

23) L306 AMIP needs to be explained

24) Section 3.1 - although important and well described, this is a bit of a gear change from describing the dataset. Is all this information necessary?

25) Figure 10/11 are very qualitative and it is rather difficult to link colours on the bar to values in the plots. A plot of differences would be more informative. Why are coincident model years not uesed (CMIP6 amip simulations usually end in 2014)? I am not sure the Figure 12 results are very relevant to the GERB data description; a reference could suffice. Figure 13/14 seem much more relevant - a legend to denote model simulation is needed and perhaps a thick mean (or median) model value would be useful. Showing albedo may reveal the diurnal cycle of stratocumulus better than RSW (which is more dependent on the insolation). Figure 14 - if the idea is to compare the model versions, it would be more useful to have the two mean lines (and perhaps shading for range) in the same plot.

26) L352 - it could be mentioned that these issues are not fully solved in higher resolution simulations e.g. Watters et al. (2021) J. Clim doi: 10.1175/JCLI-D-20-0966.1.

27) Table 1 - how much is positional/regime error and how much is cloud property error? How have biases changed since earlier analysis (e.g. for the NWP version comparison in Allan et al. (2007) QJRMetS clouds were instead too bright due to too much water)?

28) L442 the model seems to overestimate OLR for both months - does cirrus outflow or water vapour contribute?

29) L475 - it is still not clear what this factor of 10 reduction in uncertainty means. Is this just for simplistic averaging, where parts of the diurnal cycle are missing and which would not be undertaken in a serious analysis? Or is it referring to missing days?

30) L478 Some clear statements about product uncertainty and it's recommended use could strengthen the conclusions

31) L479 some illustrative comparison with model simulations are useful for introducing the dataset though I think the work presented here is more deserving of a separate, more detailed investigation.

32) Acknowledgements - the work of the GERB team could be mentioned in the acknowledgments and more specifically the contribution to initial discussions of the strategy for dealing with sunglint led by Jo Futyan and the terminator led by me could be stated.

Richard Allan

The authors present the GERB Obs4MIPs Radiative Flux Dataset, a 1x1 degree satellite-based product providing monthly means of the diurnal cycle of TOA reflected shortwave and TOA outgoing longwave fluxes. The manuscript describes the methodology used to aggregate spatially and temporally GERB observations and to gap-fill the numerous gaps existing in the GERB HR observations. A dataset like this would be very valuable to (i) monitor the TOA energy budget, (ii) assess the quality of products based on polar-orbiting satellites by taking advantage of the better sampling of the diurnal cycle provided by geostationary satellites, and (iii) assess the quality of global climate models (as done by the authors in Section 3). However, before being published in ESSD, the authors should address two main points. First, they should clarify how their dataset improves and complements the existing TOA satellite-based products. A comparison of GERB Obs4MIP against state-of-the-art products (e.g., CERES) would be highly recommended. Second, they should provide a clearer description of the methodology, splitting between methodology and results, so manuscript readers and dataset users can find more easily all the methodological steps used to produce the dataset.

Major comments

-Section 1 – Introduction. I would suggest including a summary of the existing satellite-based products for RSW and OLR (e.g., CERES, NASA GEWES SRB, CLARA). It is true that most of the existing products are based on polar-orbiting instruments that focus on daily and monthly fluxes. However, products such as CERES SYN and NASA GEWEX SRB are also able to represent the diurnal TOA cycle (3-hourly resolution). The authors should highlight how their GERB Obs4MIP can complement and improve these products.

-Line 73 - temporal averaging: Why are the temporal averages not weighted as done for the spatial averages? The 15-minute instantaneous GERB observations are not perfectly aligned with the hourly UTC intervals. Moreover, the exact timestamp of the retrieval changes between pixels following the GERB scanning cycle. First, this temporal mismatch should be described in the manuscript. Second, due to this mismatch, authors should consider 5 GERB observations (instead of 4) and weight the observations at the edge of the hourly interval accordingly.

-Line 75 - temporal averaging: explain better the albedo conversion. I assume that incoming solar radiation is calculated twice. First, at the GERB retrieval timestamp (to transform GERB instantaneous RSW observations into instantaneous albedo), and then, at the center of the UTC hourly interval (to transform hourly albedo averages into hourly RSW averages). I would suggest including the corresponding equations to clarify this part of the methodology.

-Line 75 – spatial averages: Why was the albedo conversion not applied to the spatial averages? The same bias mentioned in line 77 for the temporal average could be introduced in the temporal averages if RSW instantaneous observations are systematically missing at some parts of the 1x1 degree pixel, due to the change of RSW with latitude.

-(missing) methods section. The current version of the manuscript presents a sequential structure that mixes methodology paragraphs with results paragraphs. For instance, the averaging and gap-filling processes are currently described in two different sections (section 2.1 and section 2.4), while sections 2.2 and 2.3 contain methods and results regarding the impact of missing data. I consider that the readability of the manuscript could improve by splitting methods and results. The methods section could be further split into “dataset production” (a kind of ATBD) and “dataset evaluation” (e.g., the impact of missing data before and after gap-filling). A specific section on the “dataset production” containing all the methodological steps (including an extended version of Fig 1 diagram, which currently only focuses on the averaging process) would be highly valuable for potential users.

-I would also suggest adding a specific section or sub-section listing all the attributes of the final product (e.g., spatial and temporal resolution, spatial and temporal coverage, data format, data layers available in the final product, etc.).

-Dataset evaluation: The manuscript would significantly improve if the authors included a validation of their dataset against an external TOA satellite-based product. The obvious choice would be using CERES products. This will not only allow benchmarking the new dataset against state-of-the-art products but also having an independent reference to quantify the improvement obtained with some of the methodological steps proposed by the authors (gap-filling, GERB/GERB-like ratios)

Minor comments

-Line 84: “Hence, twilight and night-time RSW HR fluxes are not included in the averaging to the daily hourly scale if the central solar zenith angle is less than 85 but are used to replace grid-box values when the central solar zenith angle is equal to or exceeds 85” Could you clarify this sentence? Regarding the first part of the sentence, does it mean that 9km pixels with SZA > 85 are not used in the spatial average if the SZA at the center of the 1x1 pixel is less than 85 degrees?

-Line 194: “empirical narrowband to broadband conversion” Please, include either the equation with the coefficients or a reference to a document describing this conversion.

-Figures 2 & 5: Is there any reason to use these unevenly spaced categories (0, 1-5, 6-22, >22) for the number of missing days? If so, explain it. Otherwise, I would suggest using an evenly-spaced color palette or a continuous color palette

-Figures 3, 4, 6, 7, 8, 9: The panel number is not seen very well. Please, take it out of the panel and increase the font size (and/or use bold text).

-Figure 3: I would suggest using a diverging color palette centered around 0. Otherwise, it is difficult to interpret the differences. I would also suggest describing the four realizations in the figure caption.

-Figure 4 & 8. Add a horizontal line (in the background) to better interpret the bias plots.

-Figure 6 & 7. As for Figure 3, use a diverging color palette centered around 1

-Section 2.5: Could you clarify if GERB-like fluxes are used (a) only to replace fully missing 1x1 degree averages, or (b) also to replace missing 9km GERB HR observations before the spatial averaging?

-Discuss the challenges to extend this methodology to GERB instruments onboard other MSG satellites, and if you have any plans to undertake this project in the near future.