Dear Madam/Sir,

We appreciate your careful reading of the manuscript and all your corrections which have been included in a new version improving the quality of the text. Following your suggestion we used the impersonal verb tense in the new version to keep consistency. Also, the discrepancies raised along the text have been corrected by re-wording the sentences and/or providing more information to avoid confusion. We included in the text the DAC effect as source of errors in altimetry data and added the corresponding suggested references.

On the other hand, the impact of river runoff on tide gauge series was taken into account in the tide gauge processing since we reject observations showing values threefold larger than the standard deviation of the time series. We added in the new version the following sentence in Section 2.2 specifying this issue: “observations with values larger than three times the standard deviation of the time series were rejected as they could not be representative of ocean sea level changes but local features (e.g., river discharge, Laíz et al., 2013).“   Finally, the Discussion and conclusion section has been carefully checked to avoid the description of the methodology used and provide more discussion on the results supported by the proposed literature.

We appreciate your comments which have been useful improving the manuscript. Below we have responded to each of the specific comments and trust that these clarifications and amendments meet your approval.

The readability is highlighted in the introduction where a lot of ‘new’ corrections or innovations are done but are not directly referenced in this manuscript. For example, what is the internal tide model? This would be directly solved by expanding on these points and adding citations to them directly within the manuscript, although I realise this is also covered by CMEMS-SL-QUID documentation.

We agree with you that we provided barely information about the innovations of the new DUACS L4 products in the text. Following your suggestions, we added more detailed information about the different corrections applied in the new reprocessing in a new version of the manuscript. Namely, for the internal tide correction we added the following sentence:

“(i) a new internal tide correction that allows the prediction of the two main tidal constituents of both diurnal and semidiurnal tidal frequencies has been applied. The solution proposed by Zaron (2019) is used (HRET 8.1 version). This correction reduces the coherent signal characteristic of internal tide and provides a more precise reconstruction of mesoscale eddies. The use of the internal tide correction induces a reduction of internal tide signature on along-track data improving the precision of the resulting L4 gridded product (CMEMS-SL-QUID, 2022).”

I think it is firstly important to emphasise in the manuscript itself, that this product is not optimised for coastal band estimations. I.e. this is not the main objective of the product. There are several processing steps such as retrackers but also corrections such as ionospheric and tidal corrections that could be improved in the altimetry process pipeline which would improve these results even further. This is not the subject of this dataset, but is this the subject of future versions of the dataset?

As the referee says, CMEMS and C3S global L4 gridded product are not optimised for coastal band estimations. Actually, satellite altimetry was originally designed for the open ocean and temporal resolution of sensor was supposed not to be high enough to retrieve accurate sea levels in the coastal zone. That’s the reason why most of the efforts of the international community in the recent past have been focused on the research and development of techniques for coastal altimetry, with substantial support from space agencies such as the European Space Agency (ESA), the Centre National d’Études Spatiales (CNES), and other research institutions. Efforts are aimed at extending the capabilities of current altimeters closer to the coastal zone. This includes the application of improved geophysical corrections, data recovery strategies near the coast using new editing criteria, and high-frequency along-track sampling associated with updated quality control procedures. Concerning the geophysical corrections, one of the major improvements is in the tide models where the tidal component is not part of the observed signal and needs to be removed.

New reprocessings try to get closer to the coastal zone solving the lack of meaningful signals of sea level change in this region due to the typically shallower water, bathymetric gradients, and shoreline shapes, among other things. Actually, in the CMEMS-SL-QUID documentation is stated the following:

“The quality of the global gridded SLA products was estimated by comparison with independent altimeter along-track and tide gauge measurements, with focus on mesoscale signal and coastal signal respectively. The methodology is better discussed in Pujol et al. (2016) and Taburet et al. (2019).“ This provides an idea about how important are sea level retrievals in the coastal zone in the product’s quality.

In any case, we have included in a new version of the text that the global products used are not optimized for the coastal band, this promoting larger errors in this region with respect to the open sea.

At 110, it is stated that the annual and semi-annual frequencies are not included in the tidal correction as they are included in the altimetry data. What does this exactly mean? In the CMEMS-SL-QUID documentation, this is not referred to at all? I assume the authors refer to the SA and SSA tides? This is again why I am in favour of more descriptions to the data creation being used within this manuscript.

The annual and semi-annual frequencies are not removed from the tide gauge time series during their processing prior to the comparison with altimetry because these frequencies are included in the altimetry data so they are needed in order to have comparable time series to altimetry.  We guess that there has been a misunderstanding. Actually, in the text is written:

“correction of oceanic tidal effects by filtering tidal components. We used the u-tide software (Codiga, 2011). The annual and semiannual frequencies are kept in the tidal residuals since they are included in the altimetry data.

These frequencies are, as mentioned by the referee, the SA and SSA. Such frequencies are mainly driven by steric effect, which is captured by the altimetry measurements. It is not specified in the QUID of the product but it can be clearly seen when plotting altimetry time series at a given grid point.

Can you provide a brief description as to how the “valid data pairs” are determined? I.e. what are the criteria here? And why would this differ relative to the DT2018?

Once altimetry and tide gauge time series are processed by applying the different corrections described in the text, we apply the following procedure:

We collocate both datasets in time and space. It means that, for each tide gauge site we identify the altimetry grid points within a radius of 1 degree and extract the altimetry time series at each grid point for the time period covered by the tide gauge time series. We only use tide gauge series with a length larger than three years to allow statistical significance. Then we compute the Pearson linear correlation between each altimetry time series and the tide gauge record, and select the most correlated altimetry time series (this usually does not correspond to the closest grid point to the tide gauge due to the problems of altimetry solving the signal close to coast). We remove outliers (data values larger than 3 times the standard deviation of the time series) from the collocated time series (altimetry and tide gauges) at each tide gauge site. These outliers can be associated with errors in the altimetry products and/or tide gauge records. It is expected that DT2021 products will have a low number of outliers due to the improved corrections applied to altimetry with regards to the DT2018 processing and thus, a larger number of valid data pairs (altimetry-tide gauge) will be available for the statistical analysis.

We repeat this process for the 213 tide gauge sites used in this study and then we construct a single tide gauge record by concatenating the time series of each single tide gauge site. The same applies to the most correlated altimetry time series identified at each tide gauge site. We compute the statistics shown in the different tables in the text from these concatenated series. Also, a bootstrap method is used to estimate the accuracy of the results.

In Taburet et al 2019, the DT2018 dataset only uses data up to 2015. Is this the case in the dataset used here? If so, how is this adjusted for when comparing their resultant variances with the tide gauges? Are the same data periods used for both DT2018 and DT2021 for the tide gauges? If so, the results of DT2018 would probably be better when comparing apples with apples in terms of time-series length. This is simply because the data in DT2018 will not be able to see the variability shown in the data past 2015 (until 2020) particularly when you include the annual and semi-annual frequencies in the tide gauges. So the tide gauges should be restricted, in terms of total variance estimations to the same period as the respective products.

In Taburet et al. (2019) the DT2018 dataset covers the time period spanning from 1993 to 2017. Here, the datasets used for both the DT2021 and DT2018 reprocessings cover a common time period spanning from January 1993 to May 2020. This time gap was chosen due the present availability of the DT2018 products. It is stated in the last part of the paragraph of section 2.1 as follows:

“The time period investigated common to both DT2021 and DT2018 reprocessings spans from 1 January 1993 to 31 May 2020 due to the presently availability of DUACS DT2018 products”

On the other hand, the tide gauge time series were interpolated in time to the altimetry measurements so they cover the same temporal lag than altimetry data. The time period analysed for each tide gauge record can be seen in Table A2 in Appendix A. Notice that there are tide gauge sites covering the whole altimetry time period (Jan 1993 – May 2020) but others not. In these cases,  altimetry time series covering the same time period that these tide gauge sites were used to keep consistency.

In your results, what are the main causes for the differences between DT2018 and DT2021? Is it related to processing technique or is it related to more satellite altimetry [i.e. better spatial coverage] or is it related to different corrections used? Again, a table of which altimeters and which corrections are used in DT2018 and DT2021 would make it easy to contrast this. Motivated again by line 195 - 196.

The differences observed between DT2021 and DT2018 processing versions are mostly related to the new standards and updated geophysical corrections applied to the DUACS DT2021 reprocessing compared to the previous DT2018 version. Also, we observed a larger improvement of the CMEMS DT2021 product (all satellite) with respect to the previous reprocessing than that observed for the C3S (two satellite) product (improvement 60% lower). This is due to the different mapping parameters used for the CMEMS and C3S products. It is explained in the text in the following sentence:

“This fact could be explained by differences in the mapping parameters used for the two products: DT2021 mapping parameters (i.e., spatial and temporal correlation scales, a priori errors on the measurements) are evolved in CMEMS products (CMEMS QUID, 2022) with the objective to better retrieve mesoscale signals, whilst no evolution of the mapping parameter was implemented in C3S DT2021 product (C3S PUG, 2022).”

To avoid confusion about the corrections and improvements of the new reprocessing respect to the former one, we have included in the new version an appendix showing a table with the corrections applied to both reprocessings.

The distance to the nearest satellite grid point estimation is not clear. Why would the gridded product have different ‘valid’ points? The grid spatial resolution hasn’t changed, right? What would make the gridded product have valid or invalid points?

In this work we do not consider the nearest satellite grid point to tide gauge sites but the altimetry point exhibiting a largest Pearson linear correlation with the tide gauge time series within a radius of 1 degree around the tide gauge site (see response to a previous comment). Distance (km) between the tide gauge location and the altimetry grid point is computed by applying the formula to estimate distances on a coordinate plane. Altimetry grid points in the coastal band have valid or invalid data (different QC values assigned to each grid point) according to the corrections applied. This makes, for instance that one grid point could be valid in the all satellites product and invalid in the two satellite one. This is the reason why we obtain difference mean distances between altimetry and tide gauges for the different altimetry products investigated. Apart of this, we remove outliers (data values larger than 3 times the standard deviation of the time series) from the collocated time series (altimetry and tide gauges) at each tide gauge site during the processing. These outliers can be associated with errors in the altimetry products and/or tide gauge records. Therefore, depending on the corrections applied to the different products, altimetry values will be close or not to the tide gauge ones and thus will be removed accordingly. This makes that the final number of data pairs at a given tide gauge site will be different for the different altimetry products used.

What are the reasons for the spatial differences discussed in 197 - 204? Is this processing or altimetry correction based? Or data length? Could some of the errors be related to the processing differences between the tide gauge and the altimetry data? The biggest suggestion to test and potentially improve your validations would be to correct the tide gauges themselves with the FES2014 model to be consistent with the altimetry processing. This is because the utide correcting of the tide gauge is more ‘accurately’ removing the tides than what the models are able to for the altimetry (and in fact utide uses a lot more constituents than FES2014). So when using the FES2014 correction for the gauges, your altimetry and tide gauges would be more consistent and have the same ‘error’.

Lines 197 – 204 describe the spatial differences between consistency (altimetry – tide gauges) computed for all satellites datasets from DT2021 and DT2018 reprocessings. This paragraph does not describes consistency (and thus errors) between altimetry and tide gauges. Tide gauge time series in both computations are the same and also the processing applied to both datasets to perform the inter-comparisons. Thus, the observed differences in Figure 2 (showing an overall improvement of DT2021 product) are strictly due to the different corrections applied to both DT2021 and DT2018 reprocessings.

On the other hand, as the referee mentions,  FES2014 model is used to apply a tidal correction to DT2021 and DT2018 reprocessings. FES2014 model is not only used to correct the main diurnal (O1, K1) and semi-diurnal (M2, S2) tidal constituents but 34 ones including linear (K1, M2, N2, O1, P1, Q1, S1, S2, K2, 2N2, EPS2, J1, L2, T2, La2, Mu2, Nu2, R2), non-linear (M3, M4, M6, M8, MKS2, MN4, MS4, N4, S4) and  long-period (MSf, Mf, Mm, MSqm, Mtm, Sa, Ssa) components (Lyard et al., 2021, https://doi.org/10.5194/os-17-615-2021). Thus, it is similar to the correction applied by the utide software.

Have the authors compared the results in terms of linear trends to that of other products? There are other institutions that produce global and regional (particularly in the North Sea and Baltic Sea) estimations of trends using differing techniques. This would be a nice value add to this manuscript.

We thank the referee for the suggestion. We have included in the new version of the text a comparison with linear trends estimated in the Baltic Sea (where most of the long-term tide gauge stations are located) by Agha-Karimi et al., 2021, https://doi.org/10.3389/fmars.2021.702512. These authors reported a mean overestimation of trends from altimetry of 1 mm/year for the time period spanning between 1993 and 2020 when comparing with tide gauges, this being quite similar to the overestimation obtained here (1.2 mm/year) in the region. We added the following sentence:

“trends computed from DT2021 products are on average around 1.2 mm year^-1 larger than those obtained from tide gauges. Similar overestimations in altimetry mean trends were reported by Agha-Karimi et al. (2021) in the Baltic Sea for datasets covering the time period spanning between 1993 and 2020. These discrepancies could be attributed to the heterogeneous distribution of both datasets and also the crustal land uplift due to postglacial rebound resulting from the last glacial age affecting the Baltic Basin, where most of the tide gauge stations are located. This translates in altimetry measurements being not accurate enough in the coastal zone.”

Line 354 onwards in the Conclusion, maybe this is misunderstood, but these statements don’t match the results presented in the appendix. E.g. 3.14 mm/yr - 1.96 mm/yr = 1.18 mm/yr not 1.43 mm/yr? Also, in the Appendix, the results for linear trends are considerably better for the DT2018 2 sats? My direct calculations based on the table itself don’t match, i.e. I get these differences: 1.177, 1.166, 1.216, 0.889 mm/yr respectively for DT2021_all, DT2021_two, DT2021_all, DT2021_two relative to tide gauges.

The referee did not misunderstand the text. The values provided in Table A2 (last row) and discussed in the text correspond to the mean linear trend estimated for the altimetry products and also tide gauges in the region. However, the values reported later in the text correspond to the analysis of the differences in trends between altimetry and tide gauges. The values provided by the referee for the DT2021 all satellites product are a good example:  we obtained a mean valued for satellite of 3.14 mm/year, and for tide gauges of 1.96 mm/year. This provides a difference of  1.18 mm/year (~ 1.2 mm/year). This difference (altimetry overestimation) in explained in terms of the heterogeneous distribution of the datasets and other processes affecting the region (see response to the previous comment). However, the aforementioned computation relates to the spatial distribution of trends, not the differences in trend between altimetry  and tide gauges. If we compute the difference at each tide gauge site and then compute the mean value of such differences we obtain a mean value of the differences of 1.43 mm/year, which is discussed in the text  and showed in Figure 7. We think that this is the most appropriate way to assess the differences in trend: to perform the computation at each tide gauge site and the provide the mean value, instead of computing the mean value in the basin and then provide the difference of such mean value.

On the other hand, the referee is right respect to the results reported for the two satellites product from the DT2018 processing version: we obtained results 0.5 mm/year closer to tide gauges on average from this reprocessing with respect to the new one. This can be seen in Figure 5 lower panel. Thus, it seems that the C3S DT2018 processing performs better than the new one on the retrieval of long-term sea level when comparing with altimetry in the region investigated.

The authors refer a couple of times to an Abstract (Faugere et al 2022), but this is not an actual reference for the dataset.

Faugère et al. (2022) is not a reference of the dataset in the sense of a report or paper but this presentation showed for the first time the new DT2021 reprocessing and first results so we think that can be used as a valid reference for the new reprocessing.