General comments:

Tuinder et al. provide a first full quality assessment of the GOME-2A/B/C Level-3 ozone profile datasets developed within EUMETSAT’s Atmospheric Composition Satellite Application Facility, making use of ozonesonde, lidar, FTIR, and microwave radiometer observations as reference data. This work fits the scope of Atmospheric Measurement Techniques, but requires clarifications in terms of terminology, methodology, and reference data selection before it can be considered for publication. Moreover, this work at several instances lacks positioning within the existing literature.

Specific comments:

Line 4-5: Add that also the gridded uncertainty or covariance matrix is stored, unless this is not the case. The user community would at least expect it.

References are missing for several statements made in the first two paragraphs of the introduction, and for the Metop satellites and GOME-2 instruments in lines 41-44.

The importance of tropospheric ozone (monitoring) is only mentioned in line 40, almost as a side note. The author could better motivate their tropospheric ozone column assessment.

If the authors aim at demonstrating the importance of the GOME-2 instrument for atmospheric ozone monitoring in lines 49-62, it seems remarkable that RAL’s GOME-2 ozone profile retrieval within ESA’s Ozone CCI is missing.

In Section 2, references seem to be missing for the offset on line 89, Optimal Estimation on line 92, an assimilated ozone field on line 108 (different from the climatology, which is referred to earlier), and the remarks in Table 1.

Section 2.3 on averaging kernels is too vague in terms of terminology and methodology to be clear. The authors frequently mix (also throughout the text) “averaging kernels” with the “averaging kernel matrix” that combines all retrieval-layer-specific averaging kernels (as the AK matrix rows, uncommonly called “weight(s) curves” with Fig. 1). The authors should be more specific in lines 119-121: What does “the averaging kernel [matrix] cannot be inverted” mean, as one could in theory easily use a pseudo-inverse? In addition, why can a reference data set with a high vertical resolution easily be compared to a satellite data set, and not the other way around? The averaging kernel smoothing approach that is applied later in the text should be explained with reference to Eq. (1). The authors should add to that what happens to the lower-resolution FTIR data, which are the result of a sort of retrieval themselves and hence come with averaging kernels as well. Is AK smoothing applied on the satellite data or bi-directionally then, as suggested in the literature?

Lines 128-129: Do you mean that each retrieved pixel is divided into 10 by 10 km sub-pixels who’s centres need to be within a grid cell to be included? This is not fully clear.

Line 147 refers to the retrieval (co)variance (matrix?) and error (profile?), while these are not introduced in Section 2.

Lines 185-186 and Figure 2: Only the AK matrix dispersion (in terms of interpercentiles) at a specific location is shown, while upon constructing averaged averaging kernels for entire latitude bands, one would rather be interested in the AK matrix dispersion within each entire band. Showing this for all 18 bands might be too much for the main text, but should be considered as a supplement, especially given the known non-triviality of applying mean averaging kernels, as discussed in the work of von Clarmann and Glatthor (https://doi.org/10.5194/amt-12-5155-2019), which the authors should introduce in their analysis.

Fig. 3: Plot titles, legends, and axis labels should be made more brief and clear. The plotting colours do not seem to correspond to what is indicated in the legend, assuming that M01 to M03 corresponds to Metop-A to Metop-C, respectively. It should be clarified in the text how the retrieval degrees of freedom are obtained and why two modes appear in the DFS of GOME-2A before 2013.

Given the coarse (effective) resolution of FTIR observations, one would expect Table 3 to include FTIR observation specifics.

Section 4.1: Only ~20 ozonesonde sites are used for validation, while the EVDC (also hosted at NILU, including WOUDC and NADIR data) covers ~60 ozonesonde stations providing data in the period under study. Moreover, within the TOAR-II HEGIFTOM working group, the data of ~40 ozonesonde stations has been homogenised for long-term analyses. The authors should hence justify or correct for their (limited) ozonesonde data selection.

Lines 268-269: The authors’ phrasing is too brief to be clear here. What does it mean to apply Eq. 1 if that does not contain a reference profile? The term (also throughout the text) “(AK) retrieved ground-based” data is very confusing. Some ground-based reference data are sort of retrievals by themselves. What the authors refer to are “ground-based data corrected for vertical smoothing difference errors” which is commonly abbreviated as “smoothed ground-based data” but not “retrieved”.

Sections 5 and 6: It is unclear why the authors leave a gap in the vertical extent of their analysis, as seen from the unmatched altitude ranges in the second and third columns of Table 4. Moreover, Sections 5 and 6 do no longer refer to FTIR data. Where are they in the analysis? Is this only as part of Table 5, based on all reference stations shown in Fig. 4? The authors should provide an indication of the consistency of the validation results between the different reference data sources. Finally, the authors should discuss their findings with respect to existing GOME-2 ozone profile retrieval products and uncertainties, both their own Level-2 input as third-party Level-2/3 data.

Fig. 5: The plot layout in Figs. 6 and 7 is much more clear than in 5, which seems to have pairwise identical plots in the first column, and the L3 a priori profiles (green lines) missing.

Fig. 8: Why is the tropospheric column not included? Please provide an indication of the corresponding relative differences. Given that L3 data are considered, global maps of differences over the entire time series could be very insightful too.

Data references: The data availability statement should also include the ground-based reference data. FTIR data are not mentioned in the author contributions. Only the NDACC network is acknowledged, while other network sources appear in the text.

Technical corrections:

Throughout the text, acronyms should be spelled out upon first usage, e.g., SAF, GOME, FTIR, DU, and STDEV in the abstract.

Abstract: Repetition of lines 13 and 18.

Line 15: “In the lower…”

Line 36: “ter-molecular interactions” looks like a typo.

Adding references to the subsequent manuscript sections in the last paragraph of the introduction would improve readability.

Line 87: “ozone partial column density values (in DU)” contains a contradiction between the quantity and the units.

Use appropriate quantities and symbols in Eqs. 2, 6, 7, 8, 9, 10, 11, and 13.

“seen” in Table 2 probably refers to the number of sub-pixels being included?

Where only layer numbers are mentioned, like in line 174, please add the corresponding altitude or pressure. The coloured legend in the central panel of Fig. 1 is insufficiently clear to leave it to that.

Quantify the “relatively small differences” of line 182.

Caption of Fig. 1: Does “at 45° N” mean for the band going from 40° to 50° N?

It would be helpful to add the diagonals in the colour plots of Figs. 1 and 2 to guide the eye.

Line 223: Wrong usage of tildes.

Line 240: “profile” instead of “spectrum”?

Lines 254-255: Do you mean comparing mean reference data with the Level-3 grid cell that overlaps with the station location?

Captions of Figs. 6 and 7: Explain the meaning of r in the second column.

Line 357: In “use of merged Metop GOME-2 datasets” it is unclear whether these already exist to the community.

References somehow do not appear in alphabetical order.

General comments

Olaf et al. present a long-term gridded ozone profile data record derived from the GOME-2 A/B/C series, covering the period 2007–2024, along with a detailed description of the methodology and validation. This work is highly valuable for end users transforming Level-2 swath data into Level-3 gridded products. The scope of the study is well aligned with AMT, and the manuscript is generally well written. However, the paper should address the following aspects before it can be accepted for publication.

Specific comments

1. L298 : Please better specify about “a seasonal dependency present in the dataset” and this was also the case in the level 2 product”.
2. L313 said “30-50 km range, where the application of the averaging kernel is essential due to the limited vertical sensitivity of nadir-viewing instrument”. In general, the nadir-UV observations better perform for retrieving stratospheric profiles than tropospheric profiles (OMI/Liu et al 2010, TROPOMI/Keppen et al. 2024), due to the Rayleigh scattering-induced vertical sensitively. Liu et al. (2010) also show the less importance of applying averaging kernels for evaluating the stratospheric column ozone.

-Keppens et al.: 5 years of Sentinel-5P TROPOMI operational ozone profiling and geophysical validation using ozonesonde and lidar ground-based networks, Atmos. Meas. Tech., 17, 3969–3993, https://doi.org/10.5194/amt-17-3969-2024, 2024.

-Liu, X et al. Validation of Ozone Monitoring Instrument (OMI) ozone profiles and stratospheric ozone columns with Microwave Limb Sounder (MLS) measurements, Atmos. Chem. Phys., 10, 2539–2549, https://doi.org/10.5194/acp-10-2539-2010, 2010.

1. Figure 6. The correlation between L3 and ozonesonde is negative for column amount of 40-50 km, consistently from GOME A/B/C. The time-series clearly show some shift. This result should be deeper addressed. I believe, it could be rooted from satellite a priori data.
2. We believe that a L3 product intended for long-term applications should primarily be evaluated in terms of its capability to reproduce robust trends, rather than focusing on systematic mean biases. Please include a quantitative assessment of trend consistency between the ozonesonde observations and the L3 product, to see how well the L3 product reproduces ozonesonde-derived trends.
3. This work reports validation statistics based on high-resolution reference data that have been convolved to the GOME vertical resolution. While this approach is useful for assessing other retrieval errors excluding the smoothing error, the resulting statistics do not represent the actual accuracy and precision experienced by end users. The authors should therefore primarily provide validation metrics without vertical smoothing, and additionally (as supplementary information) report validation metrics with vertical smoothing.

 Technical comments

L216: please revise “4 % Stubi et al. (2008)”

L223: Please revise 13 hPa, 30-32 km.

L241: Please revise (Garcia et al. (2022))

L294 Corresponding levels derived from ozonesonde data, following (Table 4) ==> “ozonesonde data” for improving readability.

L294 The averaging kernels are applied to high-resolution ozone profiles for this evaluation.

L295. Table 5 summaries the validation statistics as function of latitude and vertical regimes specified in Table 5 for GOME-2 A/B/C, respectively. In addition, Figure 5 supplements the validation results from the mid-latitude stations in terms of time-series approach.  

L319. Inconsistence of terminology between “retrieved ground-based” (L319) and sat adj (Figure 6-7). And, “retrieved” is not proper.