The paper describes an updated and improved version of the Trajectory Mapped Ozonesonde Dataset (TOST), which provides gridded ozone profile data from the 1970s until 2020. Overall this appears to be a good dataset. The paper is overall OK and suitable for ACP (or even better ESSD?). However, there are a number of issues that should be addressed before publication.

The paper is very long and contains a lot of redundant information in text and plots. Further down I suggest a number of way to simplify Figures. I strongly suggest to also shorten the corresponding text and to shorten and focus the conclusions section. 

It appears the TOST data set uses a 5°x 5°x 1km latitude x longitude x altitude grid (e.g. lines 26, 183, 184). However, it is not really clear what the provided time coordinate is. From lines 189 and 190 it appears that one time coordinate might be 12 monthly means, for each of the 5 decades 1970 to 1979, 1980 to 1989, ..., 2010 to 2019. Another time coordinate seems to be 52 annual means for each of the years 1970 to 2021. This should be clarified in a few places, especially in Abstract and Conclusions. Also, it begs the question, why the data-set is not simply provided as 12 monthly means for each of the 52 years.

To me, the paper contains way to many similar plots and panels. This makes it very hard for a reader. If there is no significant difference between seasons, decades, ... just show one plot / panel. See e.g. my comment on Fig. 5 below. Additional plots could go to the supplement, but even there: If there is no significant difference between seasons, decades, ... just show one plot / panel. The goal of the paper should be to clearly bring out the major messages, not to overwhelm and confuse the reader with redundant information.

I am quite confused by the various relative and absolute measures used for differences in the validation part of the paper. Sometimes the authors seem to use mean relative difference (RD), sometimes bias (=absolute mean difference?) , sometimes root mean square differences (RMS, absolute or relative?), sometimes root mean square differences of the mean (RMS/sqrt(N), absolute or relative?). I think this should be clarified, and if at all possible simplified and unified.

One such confusing example is Table S2, where I have no clue in what units the various quantities are given. I assume RMS is in ppbv, which is kind-of meaningless because ~400 ppbv would be a huge 400% uncertainty in the troposphere, and a reasonable 10% uncertainty in the stratosphere. I also assume that bias is in ppbv (absolute difference), and is essentially the same as RD (which seems to be relative difference in %). If relative and absolute difference ar given (RD and bias?), why are not also relative and absolute RMS given? In Figure 2, there is a sensible separation between tropospheric, stratospheric and intermediate ozone regimes. Why is that not done here in Table S2?

Line 363 and following: What is RMSE? Not defined. I assume it is root mean square error. How is that different from RMS difference? 

Line 460, 461: What is NRMSE? Needs to be defined. It seems to be the same as relative root mean square error / difference. Why is it not in %? In most other places relative differences and relative uncertainties are in % (and absolute ones in ppbv). Please define better and make consistent, e.g. always give RD and RMS in % and ppbv.

Figure 2: I find the vertical bars for R quite confusing. I would much prefer a third set of symbols / lines. I assume that each dot corresponds to one latitude-longitude-altitude grid-cell and one annual mean? Should probably be stated somewhere.

Figure 3: Why not also give numbers for the spread / width of the distributions, e.g. full-width at half maximum, or 1 standard deviation? I assume that the underlying data points are one latitude-longitude-altitude grid-cell and twelve calender months? Should probably be stated somewhere.

Figure 5: I don't see any clear or significant differences between the top four panels, or between the bottom four panels. Therefore, I strongly suggest to just have one panel showing SAGE - TOST (all seasons, years), and one panel showing MLS - TOST (all seasons, years). It would, however, be helpful to also plot the relative RMS differences.

Figure 6: There is a lot of redundancy between Fig. 6 and Fig. 5. The single profile panels of Fig. 6 contain more or less the same information as Fig. 5 (especially if my suggested reduction is done). The main additional information in Fig. 6 is the seasonal variation (which is clearly visible for MLS). Maybe there is no need for Fig. 5, or the single profile panels of Fig 6. could be dropped?

Figure 7, Figure S3: Again, I don't see the need for four panels, as I don't see a significant difference between the panels. On the other hand the split between < 50 ppbv and 50 to 150 ppbv seems very artificial here. It seems to me that just one panel that includes all data from 0 to 150 ppbv would be enough and more sensible here. 

Line 441 and following: SE/mean is that not simply the relative RMS/sqrt(N) (in %). Another example where a more consistent nomenclature and use of relative and absolute differences would be helpful.

Figure 9: unless there is a large and significant seasonal variation: two rows might be enough. However, I would like to see a third column with relative RMS (in %, without the 1/sqrt(N)). I guess this third column  would carry comparable information as Fig. 10? These RMS numbers should be compared with estimates of ozone sonde uncertainty, e.g. those given be Tarasick et al. 2016, 2021.

Figure 10: Would not Fig. 10 and this entire uncertainty discussion (section 3.5 and Figs. 9 and 10) fit much more logically directly after Figs. 3 and 4 and section 3.1, which also compares Traj-Derived with Sonde?? Is NRMSE not the same as relative RMS? Should it not also be given in %. Should panel a.) not also have altitude on the vertical coordinate, like all the other plots? These RMS numbers and the profile in panel a.) should be compared with estimates of ozone sonde uncertainty profiles, e.g. those given be Tarasick et al. 2016, 2021.

 

Lines 503 to 505: This is important and needs to appear prominently also in the conclusions, and in the introduction (e.g. after line 91). We don't need another "tropical ozone hole" paper and consequent rebuttal like Chipperfield et al. 2022.

Line 79: should be "lower stratosphere". Above 25 km the lifetime of ozone becomes shorter.

Line 263: should "tropospheric" not be deleted here? Otherwise, why not do stratospheric as well here?

Line 343: I don't see a comparable or better performance of MLS here, unless you mean smaller RMS / error bars, which are barely visible. In this context, see my suggestion above for Fig. 3, to add the RMS profiles to the plots, and to reduce the number of panels.

Line 390: 3d should proably be 7d

You already have some excellent suggestions from RC1.  My issues mainly focus on the methodology and future of these ozone data sets.

Michael Prather

This paper documents and presents a long-term global, gridded harmonized ozone data set for the troposphere and lower stratosphere (TOST) that is readily amenable for developing model metrics and studies of trends and interannual variability.   It is very well written for the  most part and will be a valuable addition to a broad community studying atmospheric chemistry and transport, global air pollution, and global change.  The core datasets are the ozonesondes, and MOZAIC/IAGOS is used for validation – a great choice for well calibrated, highest resolution possible atmospheric composition measurements.  The updated TOST-v2 is a great product.  Yet, this is a disappointing paper in merely repeating the TOST-1 protocol without much thought as to the use of the data in modern models.  At this point it needs to be published as is (with some minor noted corrections) but with the added recognition/recommendations of how to do it better.

To me, the obvious question here is: why not include the IAGOS data as a source for TOST? It seems like you are wasting a major resource by using it only for validation.  I am not asking you to create TOST-3 for this paper, but at least you could discuss this at the beginning.  Are there fundamental problems with this? or just too much work for now (that is OK).

Abst. “of 5º × 5º × 1 km (latitude, longitude, and altitude” sound nice but it is missing two important quantities:  (1) is “altitude” really just altitude (km above the surface) or is it “pressure altitude” ?  be specific (log p, or US STD atmos p like flight levels); (2) time is critical here, what is the resolution and method of averaging?    I see in L164 that you used monthly averages, please state this up front.

Oh, now I see in L204 (“The resulting ozone fields are given in two altitude coordinates (altitude above sea level and altitude above ground level) for users’ convenience”) that you are using geometric altitude.  This is really problematic since the altitude of the land surface depend heavily on the resolution of the model you (and your users) are using.  I think these are possibly the worst possible vertical coordinates you could use, especially for the 6-26 km region where the results are most reliable.  The use of altitude requires one to know the temperature profile, which is seriously problematic since any model profile may NOT be what you use and there fore cannot be compared.  If you are using a fixed T profile, then just provide the data set in pressure coordinates. 

I think the data set must really be in pressure coordinates to be useful to any 3D model.    This you can and should fix.

Overall big problem and opportunity – may be insurmountable, but should be recognized. Spatio-temporal averaging destroys the ozone structure anywhere near the tropopause.  It is clear that this data set does not resolve tropopause ridges-troughs nor strat-trop folds – therefore the averaging of mole fraction ozone means that stratospheric ozone dominates the abundance well into the troposphere.  You simply average the ozone mole fraction in your large cells over the month.  It would be great to produce a more nuanced data set that considers the natural variability in ozone.  Specifically, why not give 10-25-50-75-90 %iles, that way one can test the high resolution (no serious models are running % deg resolution anymore), high-frequency simulations.  These statistics would help identify the frequency of strat-vs-trop, etc. and make model comparisons with the coarse resolution you use more informative.  I think you should be more expansive in diagnosis.

L61:  The satellite data indeed have trouble with the troposphere (except with product involving cloud slicing or OMI-MLS as in Ziemke et al).  I am even worried that MLS and SAGE may have difficulties in the UT/LS give the resolution you cite. 

L77-79:  The argument for ozone being inert for 4 days along the trajectory is reasonable for the UT/LS, but the out-of-date Jacob (1999) paper you use here is simply wrong for the lower troposphere.  Look at the regions of intense ozone loss (>5 ppb/day) in the ATom transects (Prather, Guo, Zhu 2023, doi: 10.5194/essd-15-3299-2023) or the 3-5 day perturbation lifetime of surface ozone pollution in Prather & Zhu (2024, Lifetimes and timescales of tropospheric ozone, Elementa, doi: 10.1525/elementa.2023.00112).  I do not think you can easily do anything (or even should do anything) about this for your TOST-2 product, but there should be a recognition of the potential error.

L169:  The new HYSPLIT may be numerically accurate but the NCAR/NCEP wind fields seem totally out of date – the vertical resolution (17 layers from 0 to 32 km = 2 km at best near the tropopause) can hardly resolve vertical motions in the UT/LS.  Why not use more modern fields like ERA-5 or MERRA-2?  It makes the paper look lazy, you updated the sondes, but just ran with the old parts of TOST-1.  I know you cannot fix this, but it should be recognized as a problem (like the minimal use of IAGOS observations) that should be upgraded in TOST-3.

L272:  I was going to congratulate the authors on their correct use of nmol/mol as the measure of ozone abundance and then I hit the incorrect use of ‘ppbv’ (“RMS of 21.1 ppbv, and higher bias (2.9 ppbv) and”).  The ‘by volume’ should have been scoured out of this community by now but many prominent colleagues continue to abuse this.  The ‘volume’ is not mole fraction since virial corrections would need to be applied, and most all measurements calibrate to dry air mole fraction. 

L475:  You really should be comparing tropospheric O3 column (DU or mean ppb) with Ziemke et al’s work.  The whole paper is well referenced within its limitations (noted above), but you simply must compare the features in Figures 8 and later with Ziemke’s work.

L555:  Again, note that this is monthly averaged.