The manuscript “Global total ozone recovery trends derived from five merged ozone datasets“ by M. Weber provides an update to a study published by the first author in 2018, with four more years of data added to the five analyzed datasets (four satellite datasets and one dataset comprised of ground-based measurements). A multiple linear regression is applied to annual mean data from the period 1979 to 2020 to determine total column ozone (TCO) trends in different broad latitudinal bands for the period in which concentrations of ozone depleting substances (ODSs) increased in the atmosphere, and for the period after the peak concentrations had been reached. The multiple linear regression includes next to the typical proxies also several dynamical variables (e.g. a proxy for the Brewer Dobson circulation (BDC) or the Antarctic/Arctic Oscillation (AAO/AO)) which is one of the main differences to other trend analyses based on TCO data. The authors find with this method significant positive trends (related to the reduction in ODSs in the atmosphere) for the period 1997-2020 for the near global mean (60S-60N), as well as for the Northern hemisphere mid-latitudes for which the trend is near zero if the dynamical proxies are not included in the regression.

The manuscript is very well written and well structured, mostly the data and methods are explained in enough detail to allow the reader to understand what is going on (in a few cases I found the description slightly too short and I have mentioned them in the details below), and the topic lays clearly within the scope of the ACP journal. There are a few minor things that I commented about below that are easy to fix, but there are two main points that I think need careful adjustment of the manuscript or some additional thought.

I recommend the publication of the manuscript after revisions.

Two main points:

• Attempting an attribution with a multiple linear regression that includes non-orthogonal proxies is tricky. Especially if several proxies include a trend. The hope then is, that the regression is able to separate the trend contribution from the different proxies based on the additional variability the proxies provide. However, it is possible that trends are not assigned correctly to the different proxies which would falsify the signal of the trend that if of interest, in this case here, the trend caused by ODSs and not by changes in dynamical variables. The authors argue that with the addition of the dynamical proxies the variability of the time series’ are matched better by the regression results. There are two points that make me somewhat doubtful of this statement: (1) the pre-1996 trends change clearly with the introduction of the dynamical proxies (Figures 3 and 4) although the main trend signal should be coming from ODS-related changes in this period; (2) the signal from the SH Brewer Dobson circulation proxy in the NH polar regions that cannot really be explained. I think the manuscript needs more discussion of these points to strengthen the claim that the addition of the dynamical proxies can indeed robustly isolate the ODS-related trends. For the first point I raised I would suggest to check the older literature about regression results for the pre-1996 period where dynamical proxies have been used. I have added two references in the comments below that might be worth checking out. And there might even be more that could be checked and where the results could be compared to the pre-1996 ODS-related trends calculated here. For the second point I raised I think it would be helpful to do some sensitivity test to check the robustness of the trend results and the contribution of the individual proxies: (I) not using the trend proxy but JUST the dynamical proxies, how do their contributions change if at all; (II) use some of the dynamical proxies only in the regions where they occur, e.g. AAO only in the SH, AO only in the NH, etc.; how does the contribution of these proxies change (if at all), and how does the ODS-related trend change? I think these sensitivity test will go a long way to show the robustness of the results presented here in this manuscript.
• I think it is really important to clarify throughout the manuscript (including the title!) what kind of trends the authors talk about. Mostly, the trends that are discussed are the trends that are attributed to the reduction of ODSs in the atmosphere WITHOUT any contribution of dynamics to the trend. In many places this is not totally clear since the trends are only called “recovery trends”. However, for me this is the main point of the manuscript and the difference to other studies. It would therefore be extremely important and very helpful if the authors could be more specific in how they name the trends throughout the manuscript (e.g. instead of referring in the abstract in line 11 to “The near global trend of the median of all datasets…” it would be better to be more specific and refer to “The near global ODS-related trends …”, and specifying this in the title like “Global total ozone recovery trends attributed to ODS changes derived from five merged ozone datasets”)

Minor comments:

• Line 10: “… is indeed on slowly recovering…” – remove the “on”
• Line 16: data from which phase of CCMI? Please specify.
• Line 71: It is not clear in this section what the spatial coverage of the described datasets is. I assume 90S-90N since also polar regions are analyzed. Please add this information to the dataset descriptions.
• Line 72: “ground-based” instead of “ground”
• Line 78: “ground based Brewers, …” - remove the “ground based” since it is already mentioned at the beginning of the sentence.
• Line 80: Add also here the information from which phase of the CCMI project simulations were analyzed.
• Line 129: It is not clear how and by whom the ground-based dataset was updated. The references for the dataset are relatively old, therefore if would be good to add a few words on how the dataset was updated to the year 2020.
• Line 135: The word “belt” is used here, although it is only explained in the following sentence what exactly is meant by it. This should be switched to make it clearer for the reader what is meant with “belt”.
• Line 154: the data were bias corrected. It would be nice to give here a range of biases that needed to be adjusted. I understand that the biases can be different for the broad latitude bands and datasets, but some kind of number/range would be nice here.
• Line 169: “applies” should be “apply”
• Line 175: “.” is missing after the parenthesis.
• Line 176: The year 1996 is the time for maximum EESC concentrations for which region of the globe? Tropics? Everything beside the polar regions?
• Line 177: It would be good to give the exact latitude ranges here which define the polar regions.
• Line 190: The end of the sentence is slightly misleading. I would add “for these years” before “were calculated” to clarify that only for the years with too many missing data no annual means were calculated.
• Line 226: What about the pre-1996 trends? Did they stay very similar to W18 as well?
• Line 248: “agree” instead of “agrees”
• Line 255-257: It might be nice to add here a table with the trends reported from W18 and calculated here. It would provide a nice overview of things that changed and things that stayed roughly the same (just for the multi-observational median, not each individual dataset)
• Line 269: “ground-based“ instead of “ground”?
• Line 285: Are there any studies that report on trends pre-1996 based on regression methods that use also dynamical proxies? There is one looking at ozone soundings at Payerne (Weiss et al., JGR, Vol. 106, D19, 22685-22694, 2001), and one looking at individual TCO station measurements (Maeder et al., 2007, https://doi.org/10.1029/2006JD007694) but there might be even more analyzing total column ozone data with dynamical proxies. As mentioned above, I think it would be helpful to provide an estimate how well the ODS-related trends compare with earlier findings for the pre-1996 period since they did change quite a bit with the introduction of the dynamical proxies.
• Line 305/306: Couldn’t this signal be a spurious regression result where the attribution did not work properly between the trend proxy and the dynamical proxies also including a trend? I think some sensitivity test (as mentioned above) would be helpful here to test the robustness of this signal.
• Line 316: “.” missing after the parenthesis.
• Line 331: “have” instead of “has”
• Line 366-368. This sentence seems somehow out of place here. I think it needs a little more explanation and detail.

Review of 'Global total ozone recovery trends derived from five merged ozone datasets',

Weber et al., ACPD, 2021

1. Short resume

Weber et al present a comprehensive analysis of trends in total ozone, focusing primarily on the period since the turnaround in ozone-depleting substances.  This is an update and extension of earlier work published in 2018.  In contrast to latter publication, the authors now claim the detection of increases ($0.4\%$/decade) in near-global (60S--60N) total ozone since 1996, with high confidence ($>$3--4$\sigma$).  Positive trends over broad mid-latitude region in both hemispheres (35N--60N and 35S--60S), about 0.5--0.7$\%$/decade, are significant as well although close to the 2 sigma detection threshold.  

The dynamical proces terms (Arctic and Antarctic Oscillation, Brewer-Dobson circulation) in the regression model play a central role in this detection, especially at northern mid-latitudes.  The authors deliberately chose not to detrend the dynamical terms prior to regression, in order to account for any long-term changes in AO, AAO and BDC.  In doing so, they find that trends become less negative before 1996 and more positive since 1996 across large regions of the low- and mid-latitudes.  This more complete attribution results in a higher significance of the trends, especially in the northern hemisphere where the $2\sigma$ detection threshold was passed.  Hence, the authors conclude that dynamical changes appear to counterbalance the recovery of ozone in the mid-latitude NH.

The authors furthermore explain the positive recovery trend of total ozone as a result of changes in ozone-depleting substances.  Indeed, the ratio of the rate of increase and decrease in ODS concentrations is consistent with the rate of depletion and recovery of total ozone across all 5$^\circ$ latitude bands between 60S--60N.

2. Recommendation

This paper provides an important update to previous assessments of long-term changes in total ozone.  It is very well written and accessible to a large scientific audience.  The methodology is sound and the presented results support the claims made by the authors.  I highly recommend publication of this work in ACP if my remarks below have been addressed.

3. Major comments

Ordered in order of appearance in the text.

3.1 Extension of GOME-type backwards in time (Sect. 2.7)

I understand the importance of covering a suffiencly long period, but is this backwards extension for GOME-type data records still needed now that more than two solar cycles have been completed since 1995?  Doesn't this break the independence between SBUV and GOME-type estimates?  By how muc does the negative trend in the SBUV period influence the recovery trend estimates during the GOME-type period?   

Have you tested the sensitivity of the resulting trend to the choice of NASA COH or NASA MOD, and without the extension?  

Avoiding data gaps is important but preserving data quality / stability is perhaps even more important under high aerosol backgrounds.  Could you elaborate why gaps are more important or, if that is not the case, comment about the stability of both SBUV records after Pinatubo?

3.2 No reference to how trend errors are estimated (Sect. 3)

Many trend estmimates (Fig. 3) are close to the $2\sigma$ threshold.  The computation of MLR coefficient uncertainties therefore deserves some attention, this is missing right now.  Please explain how MLR parameter errors are computed or refer to relevant publications.

Somewhat related to this, was there any consideration of including reported measurement errors in the regression?  

3.3 Annual time series

p.7, l.180: Could you motivate the choice for analysis of annual mean time series instead of monthly mean data?  Is there an impact on the trend estimates and their significance?  Please refer to relevant publications.

3.4 Robustness of attribution to dynamical processes (Sect. 5)

Previous work by the authors (Weber et al, 2018) also considered terms for dynamical processes in the MLR.  At the time, however, no significant positive trends were detected (Fig. 9).

It would be enlightening to discuss whether the four additional years of data have truly helped to attribute ozone changes more robustly to dynamical changes.  Or, whether it is plausible that the current attribution is subject to geophysical variability (and measurement uncertainty).

4. Minor comments

p.1, l.12-13: Near-global trend values disagree with quoted values in Section 4.  Please revise.

p.3, l.82: "Annual mean timeseries of all five merged datasets are in very good agreement".  Somewhat subjective, please add a number.

p.5, l.132: The evolution in satellite quality has been described adequately.  This is missing in the WOUDC section.  Surely, there must have been progress in the calibration of these instruments or the coherence of the network since the work by Fioletov in 2008.  If so, could you update this section accordingly?

p.6, l.142-143: "[...] can be estimated with a precision comparable with satellite-based data sets ($\sim$1$\%$)."  A reference would be appropriate.

p.6, l.150: Remove "from the past into the future" as the statement "between 1960 and 2100" is more than sufficient.

p.6, l.154-156: I am sorry, I did not get the point of "The multi-dataset mean was then added back to each dataset, such that all bias corrected timeseries are provided in units of the total column amounts (W18). However, the trend results derived from them are identical to those derived using anomaly timeseries."  Could this be clarified a bit better for the non-expert?

p.6, l.154: "to the mean".  The 1998-2008 mean at the global or local level?

p.6, l.165: See comment below, the second term in Eq. 1 should be $b_1 (t-t_0)$.

p.6, l.166: "coefficients $b_1$ and [...]"  This is inconsistent with the notation in Eq 1.  Sign of first trend term (t0-t) implies that positive b1 values represent a decline in ozone.  Please change this. The factors $X_1(t)$ and $X_2(t)$ define the decline/recovery periods.

p.6, Eq.2 and 3: Figure 1 suggests that the "recovery" period starts in 1996, so the turnaround is defined as $t_0=1996$.  If this is correct, then the notation in Eq. 2 and 3 should be changed to $X_1(t)=1$ for $t < t_0$ and $X_2(t)=1$ if $t \leq t_0$ (and vice versa for $X_i=0$).  The trend model is not continuous at $t_0$, hence $<$ or $\leq$ do make a difference.

p.7, l.185-187: Is there any particular reason why you haven't used GloSSAC v2 (Kovilakam et al., 2020)?  

p.8, Table 2: EHF is missing from this list.  Where can it be downloaded?

p.8, Eq.4: "BDCn" and "BDCs" should be explained in the text.

p.8, l.208: "the linear trend terms best approximate EESC related trends".  Can a match between ozone trend and EESC expectations really validate the choice of terms in the MLR?  There is risk of a circle reasoning here.  If the improved agreement with EESC expectations is motivating the choice of terms in the MLR model then you can't use this same agreement again to conclude a causal relation between trend and EESC.

p.8, l.215-216: This phrase is not entirely clear on whether or not you use the detrended proxy.  This choice is so central to this paper that it must be very clearly stated.

p.9, Fig. 1: $\chi^2$ is the sum of "the squared differences median timeseries minus MLR" 

p.9, l.219: "MLR prediction after fitting" would be clearer than "MLR result from applying".

p.9, l.220: To me, "after 1996" suggests 1996 is not included.  What about replacing "after 1996" by "since 1996" throughout the manuscript?

p.9, l.224: "recovery from reductions in ODS" would be more clear on the effect of ODS on ozone.

p.11, l.260: Replace "from applying" by "when applying"?

p.11, l.260: It is somewhat unexpected to regress a "super"merged timeseries rather than average the trends from individual records.  What is the rationale?  Also, the sample size is just $N=3$, for 1979-1995, so won't the "super"merge-then-regress method lead to more uncertainty in the MLR parameters than the regress-then-average approach?

p.12, Table 3 (caption): The periods in the caption are inconsistent with information in Figs 1 and 2. The first trend period stops in 1995, the second starts in 1996.  Hence, it should be $1979-1995$ and $1996-2000$.

p.12, Table 3: For each latitude belt, the occurrences of "mean/median trend $>1996$" should be $\geq1996$, in order to be in line with Fig. 1 and 2.

p.12, Table 3: The error notation was confusing for me, I haven't seen this specific notation very often.  For instance, what does $-1.9(13)" mean?  Is it $-1.9\pm0.13$ or $-1.9\pm1.3$ or ...?  I find an explicit notation such as "+0.4\pm0.2" much more effective.  I recommend using this throughout this table and also the manuscript.

p.12, l.265: "One notable change from W18 is that the tropical trends during the ODS rising phase are now more negative (down to $-1\%/decade) while before they were mainly close to zero. This may be caused by the additional proxy terms used in this study".  The pre-1996 data have been available for a very long time now.  Has this effect never been looked into before?  If so, please refer to relevant work.

p.12, l.270: Please replace the "maybe" (conditional) by an "is" (certainty).  Trend uncertainty scales with $n^(-3/2}$ (e.g., Weatherhead et al., 2000) so the eight more years in the recovery period already lead to $\sim45\%$ smaller trend error.  This seems not too far from the observed factor $2$ reduction of the error in Table 3 and Fig. 3.

p.13, l.274: "The expected tropical recovery [...]".  Estimated mid-lat NH recovery trends are too small compared to EESC prediction as well.  

p.15, l.320: "NH total ozone has been steadily declining..." conflicts with the first phrase of this paragraph "stable ozone levels in NH since 2000".  Please clarify the text.

p.15, l.324: "with larger springtime polar ozone losses"?

p.15, l.325: Remove "recent" from "A recent downward trend".  Perhaps you meant that this was recently reported?  Ball et al report a continuous decline since the 1980s, not a recent decline.

p.18, l.332: Quoted recovery trend value ($11\%$/decade) conflicts with that in Figure 7 ($12\%$/decade).  Please correct. 

p.19, Table 4: Same comment on error notation as in Table 3 (p.12).

p.19, l.367: The Gaudel paper is about differences between tropospheric ozone data records.  So probably not the best reference when the message is about consistency between tropo/strato/total ozone.

5. Technical corrections

p.1, l.10: Remove "on" from "[...] is indeed on slowly [...]".

p.1, l.12: Remove "in absolute numbers".

p.1, l.15: Add "-" to "chemistry-climate models".

p.2, l.30: Typo "stratosphere".

p.2, l.38: Remove "agreement" from "Montreal Protocol agreement".

p.3, l.75: Replace "in large part" by e.g. "largely".

p.3, l.79: Replace "Observations Zénithales" by "Observation Zénithale".

p.4, l.87: Replace "are processed using the same V8.7 retrieval algorithm" by e.g. "are retrieved using the same V8.7 algorithm".

p.4, l.108: Type "[...] shift to an equivalent [...]".

p.5, l.130-132: Double occurrence of ground-based.  First one could be removed, e.g. "The WOUDC zonal mean ...".

p.7, l.175: Add "." after "W18)".

p.7, l.189: Replace "there are not sufficient number of months" by e.g. "there are not enough months" or "there is not a sufficient number of months".

p.7, l.194: Replace MLR "equation" by MLR "model"?

p.8, l.212: Remove "the possibility", as it is a bit redundant.

p.8, l.212: Replace "MLR results" by "MLR fit residuals" perhaps?  This is a bit clearer as the MLR parameter estimates are MLR results as well.

p.9, l.218: "five bias-corrected" instead of "bias-corrected five".

p.11, l.242-243: Maybe you forgot to remove the newline between paragraphs?

p.11, l.251: Add a $+" sign to the quoted values at start of this line.

p.11, l.256: Remove ' after "timeseries".

p.11, l.261: Add "/decade " after $+0.5\%$.

p.12, Table 3 (caption): Remove "and" from caption "[...] in bold have an absolute [...]"

p.12, Table 3 (caption): Add "prediction" at the end of "and mod$_i$ the MLR".

p.12, Table 3: Add $+$ to trend value $\geq$1996 for median time series near-global.

p.12, Table 3: The quoted $r^2$ value for WOUDC in 20S-20N band is single digit (0.7), should be double (0.70). 

p.13, l.276: Remove "on" from "elucidate further on".

p.13, l.285: Type "Fig. 4a" should be "Fig. 4".

p.15, Fig.5 (caption): There is a missing word in "Negative values an anti-correlation [...]".

p.15, l.311: Add "s" to "chemical effect"?.

p.15, l.316: Add full stop at end of phrase.

p.17, Fig.7 (caption): Capitalise "See".

p.18, l.331: "Earlier signs of ozone recovery have been", should be plural. 

p.18, l.331: Add "," in between "Now with".

p.18, l.332-333: "During September, the Antarctic ozone hole usually grows and [...]".

p.18, l.340: Remove "as shown in Fig. 7".  A bit redundant, you already refered to the figure in the previous phrase.

p.18, l.344: Replace "globally" by "global"?

p.18, l.352: Add "," in between "tropics recovery".

p.18, l.354: Add "," in between "Arctic large".

p.19, l.363: "chemistry-climate models".