This paper presents an analysis of the interannual variability of dissolved oxygen in the Southern Adriatic Sea, based on analysis of the CMEMS Mediterranean reanalysis for 1999-2021, bias corrected using in situ observations made by BGC-Argo floats. EOF analysis is used to separate the main components of variability, which are then linked to different drivers. The analysis is thorough and the results are novel and interesting, but a few points are worth attention before publication.

I am confused by the use of the terms temporal mode and spatial mode. The description in the methods section implies to me that the EOF analysis has been applied separately to identify variability over (vertical) space and variability over time – this would give a different set of modes in each case. But in the results the two sets are presented together, with the “spatial modes” being used to show the vertical variation seen in the “temporal modes”. Wouldn’t it be more accurate to refer to spatial and temporal aspects of the four modes? Or have I misunderstood the method? I suggest that either a fuller description of the methods or a revision of the language used in the results would help.

I am also uncertain why 2021 is picked out for particular discussion. In Figure 1d and Figure 3 it looks to me like the continuation of a trend that goes back to 2016, rather than an unusual year – why focus on this year in particular? It also appears to have some similarity to 1999 – is there some possibility of a cyclical pattern? I don’t think that a major reworking of the analysis is needed, but I suggest the authors consider a shift in the way they present the 2021 results.  The reference to the entrance of new water masses from the Levantine Basin is interesting – was this the reason for looking at 2021?  It would be good to see more detail than is given in lines 155-157 about how unusual this change is.

Specific points

line 19, 46: I don’t understand “alternate” in the description of the North Ionian Gyre.

line 22-23 “We ascribe the lower content of DO in 2021 to a negative anomaly of the subsurface production in the same year, in agreement with the previous correlation analysis, but not to heat fluxes”: this point made in abstract is not stated explicitly in the main body of the paper (though it is implied).

line 23-25 “we observe the entrance of warmer and exceptionally saltier waters favored by the cyclonic circulation of NIG from 2019 onwards”: cyclonic circulation is also observed in 1999-2005 and most of 2011-2016; no evidence is presented about the temperature and salinity of the water.

line 67-74: regarding the bias correction, was the bias consistent across the three periods where float data was available? Did the bias correction allow for uncertainty in the in situ observations?

line 97: the dimensions of the SAdr box are much smaller than the region from which DO data was taken (i.e. the 0.9 autocorrelation contour). Why was a smaller region used for the forcing indices?

line 119 Fig.3c-e: This should be 3c,e if it’s referring to the temporal modes only or 3c-f if it’s referring to both temporal and spatial modes.

line 129 “the first mode explains the variability (e.g. seasonal) connected with solubility”: I’m not sure why the third mode is not involved too – it has a correlation of 0.51 with heat flux and hence, presumably, to temperature and solubility.

line 167-8 “We do not recognise a clear deoxygenation trend in the subsurface layer”: there seems to be a very clear decline for 2010-2021. Is the point that there is no overall decline in the period 1999-2021 (though there is a rise then a fall)?

line 168-169 “the multiannual variability is characterized by a sort of cyclicity”: I’m not sure what is meant here.

line 170-172: I’m not sure that I understand the point being made here – is it that the SAdr responds quickly to change because of the small water volume and residence time?

Overall I think this is a good paper and I enjoyed reading it. But I advise the authors to go through the text carefully to make sure that their main findings are presented really clearly and backed up by enough information. I also recommend an English language edit – the paper is generally well written but in a few places I was unclear about the meaning.

General Comments

The manuscript investigates the interannual variability of dissolved oxygen (DO) over the last 23 years driven by multiple mechanisms (atmospheric, circulatory, and biological processes) in the southern Adriatic Sea. The aim is to demonstrate the importance of DO as indicator of current changes and environmental status.

The study is based on modelled and observational data from the Copernicus Marine Service: the biogeochemical reanalysis for the Mediterranean Sea covering the 1999-2021 period and BGC-Argo float measurements for the 2014-2020 period. DO estimated by BGC-Argo floats are used to bias-correct the modelled DO along the entire reanalysis time series using a quantile mapping technique.

The bias-corrected reanalysis signal is then decomposed using EOF analysis. Then a correlation analysis between the first four EOF temporal modes of variability and key drivers allows evaluating the relative importance of the different drivers to explain DO variability. Finally, year 2021 is compared to the 1999-2020 climatology to identify the main contributor(s) in 2021.

This is a very interesting study which has multiple interests in the context of actual changes. And DO is a good candidate to detect and monitor changes. However, I have some concerns about the foundations of the study. They are detailed below:

1. The ability of the QM method in correcting the bias of modelled DO.

QM covers a variety of methods that do not necessarily have a similar ability to correct model bias for a given domain or a given variable. Some QM methods may even deteriorate the model outputs compared to the non-corrected form. So it is therefore important to choose an appropriate bias correction method.

- Have you compared different QM methods? Does the performance vary significantly from each other? Please describe strengths and weaknesses of the QM method used for this study.

- In the current draft, it is unclear to me how the reanalysis product is corrected with the BGC-Argo profiles. More details are needed on this correction.

- Are there no other in-situ data (other than Copernicus In-Situ TAC) to evaluate the bias-corrected DO time series (before BGC-Argo area) and confirm the ability of the QM method to correct the modelled DO ?

- To validate the QM method, it would be very helpful to provide the Hovmoller diagram of modelled DO (same as Figure 1d) before applying the bias-correction procedure.

- The bias-corrected DO time series (Fig 1d) shows that the deep layers are progressively enriched with oxygen over a period of 7-8 years (1999 to 2006), and then a mechanism re-initialize the oxygen at depth. Another cycle starts again (2007 to 2013), and then a third one seems to start and then perhaps disarmed by the bias-correction. This cyclic deep enrichment in oxygen is not really discussed in the paper. Is it an observed feature ? Is it already present in the non-corrected time series ? Is it the result of the bias-correction ? It could even be interpreted like a drift of the modelled DO, with a relaxation to initial conditions each 7-8 years. A lot of questions can arise from this cyclic enrichment in DO… So more clarification is needed.

2. The necessity to bias-correct the modelled DO

Modelled DO is corrected ? But why is it necessary?

It is difficult to appreciate the bias correction. Need to see the temporal evolution of modelled DO before applying the bias-correction (hovmoller diagram).

Bias-correction modifies the signal and may alter the link between DO and its drivers and thus reduce the correlation between the decomposed DO time series and forcing indexes… In fact, all correlations are quite “low”, at best “moderate”. Can't it come from this bias-correction? Have you tried the same analysis without bias-correction ?

Or maybe, the correction does not modify the signal… but in this case, more clarification is needed in the text.

3. The focus on the interannual variability

The draft states that this study focuses on DO interannual variability related to multiple drivers.

But, 5 over the 6 forcing indexes presented on Figure 2 mainly highlight a seasonal pattern. Seasonal pattern is also a main feature of the 3 first EOF temporal modes. And a major part of the results (section 3.2) aims to characterize the seasonal variability.

If the paper is intended to address DO inter-annual variability and long-term dynamics, shouldn't the seasonal cycle be removed from the signal before starting the analysis ? Have you tried this ?

4. The interpretation of the correlation analysis

The study is based on an analysis of correlation between the first four EOFs temporal modes and the six main forcing in the area. On the 24 correlations calculated, only 4 are greater than 0.5 (with a maximum correlation of 0.68), 15 are less than 0.5 and 5 are not significant. I find the link between EOFs modes and forcing indexes a bit week, and the message sometimes confusing, with the term “significant correlation”. A statistically significant correlation must not be confused with relevant correlation. Please make a clear distinction between the two.

The scale used to classify correlations is not really appropriated. It is somewhat exaggerated to mention a correlation of r=-0.61 as “strongly correlated”.

A reasonable classification would be:     r < 0.5 à  low correlation

                                                                       0.5 < r < 0.7 à moderately correlated

                                                                       0.7 < r < 0.9 à highly/strongly correlated

Maybe, scatter plots (EOF temporal mode vs forcing index) would be helpful to appreciate the relationship between the 2 signals.

Specific Comments

L14: “we used DO modelled by the latest Copernicus Marine biogeochemical reanalysis”, please add “for the Mediterranean Sea”.

L68: why is it necessary to correct the reanalysis? please add a few words

L77: what does "auto correlation" mean ?

L85: Is Pearson correlation used ?

L97: why is the SAdr box used to average forcing (41.6-42.1°N  / 17.6-18.1°E) different from the SAdr area used to average modelled DO (area of autocorrelation 0.9) ?

L102-103: why has the vorticity been filtered ? why other forcing indexes have not been filtered ? please clarify

L117: change “significantly correlated…” to “statistically significant but moderate correlation with the heat flux (r=0.56) and low correlation with the subsurface chlorophyll concentration (r=0.43)”

L123: “strongly correlated” for r=-0.61 and r=0.68 is somewhat exaggerated…  please change to “moderately correlated”

L139-141: I would not say that Fig 1d shows 2 distinct periods 2005-2006 and 2012-2014, but rather 2 periods of 7-8 years each with progressive oxygenation of the deep layers (1999-2006 and 2007-2013). Is this the third mode that explains this deep oxygenation ? Which mechanism re-initialize the oxygen content at depth in 2007 and 2014 ?  Is it confirm by analysis of DO advection from northern Adriatic sea ? here a time series of DO advection would confirm the hypothesis.

L142: The analysis performed on detrended DO time series provides pretty similar results.

But did you try to perform the analysis on de-seasonalized DO time series ?

L151-152: I am sorry, I am not able to see the anomaly (lower than average) in subsurface chlorophyll in Fig 2d… Could you give more details please?

L152: please add the reference to Fig. 2d at the end of the sentence.

L153-155: 2021 anomaly is an important result of the paper, and a central aspect for the Ocean State Report. I think that figure(s) is/are missing to support the text of Section 3.3. For exemple, a time serie of Temperature, salinity and oxygen content at entrance of the SAdr trough the Otranto Strait would confirm the regime shift mentioned in the text, with the entrance of new water masses (warmer, saltier and less oxygenated) well reproduced by the physical and biogeochemical models.  And the reasons why circulation has changed could be developed further.

L169-170: “the cyclicity” is briefly mentioned here while it is an obvious feature in figure 1. More discussion about this point is necessary.

Figure 1a: please add the bathymetry.

Figure 2: is Product 3 the forcing of Product 1 ? and are Products 4, 5, 6 used as atmospheric forcing or data assimilation in Product 3 ? please mention the links between the products.

Table 2: please change “Not significant correlations…” to “Not statistically significant correlations…”

GENERAL COMMENTS

In this study, the authors combine various datasets (in situ and remote observations, and numerical simulations) and various variables (physical and biogeochemical data) to analyse the dissolved oxygen variability over 1999-2021 in the Southern Adriatic Sea through an EOF decomposition. In order to estimate the contribution of a set of drivers, the correlations between four first modes of variability and the drivers are computed. The study is interesting and the authors show that the dissolved oxygen is an relevant indicator of multiple drivers of the marine ecosystem.

The manuscript is very nice example of ocean data integration and provides a new relevant indicator for the Copernicus Ocean State Report. The manuscript is relatively well written although some corrections are required (but I am not English native). Indications about the datasets and computations of some derived variables are strongly missing but the authors could easily add them.  Precisions on methodology are also required. The figures appear with low quality and most of them need improvements. I would recommend the publication of this manuscript after some clarifications and improvements (see my comments below).

MAJOR COMMENTS

Period of bias-correction

Could the authors explain why applying a bias correction using profiling floats over 2014.-2020? Why not over the whole period? Why excluding the last year? Which is the impact of such temporal sub-sampling in bias-correction on the bias-corrected reanalysis time series? In addition, fig. 1c shows several long temporal gaps? Which impacts?

Analysis and results

In the introduction, the authors state that they analyse the DO interannual variability (l.60). But in section 3., they also dedicate a sub section in the 2021 anomaly. Why? Because of the OSR7 that also focuses on 2021 event? Or this year has a specificity highlighted thanks to the analyses over 1999-2021? Please, clarify.

Correlation

• I don’t agree with the level of correlation in general. For example, 0.5 is not a high correlation. Please rephrase/moderate/modify your statements (l.117, l.123).
• Could you detail the method used (and provide reference) for testing the significance of the correlation coefficient?

Datasets

All products have to be described (briefly) in section 2., not only prod1 and prod2. For each driver of the study, indicate which product is used and how the indicator is computed (in particular, MLD, which criteria is used). In addition, it is important to provide to complete references and DOIs (as indicated in the Copernicus Marine Service website, how to cite: https://help.marine.copernicus.eu/en/articles/4444611-how-to-cite-or-reference-copernicus-marine-products-and-services).  For the dataset, do not write “the latest” in the text. I recommend to provide the complete references and to indicate the date of access in the references associated with dataset [access Month Day, Year]. Please find below the information:

Prod1: The Mediterranean Sea biogeochemical reanalysis at 1/24° of horizontal resolution and daily temporal resolution (Cossarini et al., 2021, Teruzzi et al., 2021)

• Cossarini, G., Feudale, L., Teruzzi, A., Bolzon, G., Coidessa, G., Solidoro C., Amadio, C., Lazzari, P., Brosich, A., Di Biagio, V., and Salon, S., 2021. High-resolution reanalysis of the Mediterranean Sea biogeochemistry (1999-2019). Frontiers in Marine Science.
• Teruzzi, A., Feudale, L., Bolzon, G., Lazzari, P., Salon, S., Di Biagio, V., Coidessa, G., & Cossarini, G. (2021). Mediterranean Sea Biogeochemical Reanalysis INTERIM (CMEMS MED-Biogeochemistry, MedBFM3i system) (Version 1) Data set. Copernicus Monitoring Environment Marine Service (CMEMS)  https://doi.org/10.25423/CMCC/MEDSEA_MULTIYEAR_BGC_006_008_MEDBFM3I (accessed November 17, 2022)

Prod2: The Mediterranean Sea in situ quality-controlled observations, distributed by Copernicus In Situ TAC

https://doi.org/10.48670/moi-00044

Prod 3: The Mediterranean Sea physical reanalysis at 1/24° of horizontal resolution and daily temporal resolution (Escudier et al., 2021)

Escudier, R., Clementi, E., Omar, M., Cipollone, A., Pistoia, J., Aydogdu, A., Drudi, M., Grandi, A., Lyubartsev, V., Lecci, R., Cretí, S., Masina, S., Coppini, G., & Pinardi, N. (2020). Mediterranean Sea Physical Reanalysis (CMEMS MED-Currents) (Version 1) Data set. Copernicus Monitoring Environment Marine Service (CMEMS). https://doi.org/10.25423/CMCC/MEDSEA_MULTIYEAR_PHY_006_004 (accessed November 17, 2022)

Prod4 and prod 5:Reprocessed and near real-time altimeter satellite gridded Sea Level  Anomalies (SLA) computed with respect to a twenty-year 1993, 2012 mean (prod4 and prod5, respectively).  The product gives additional variables (i.e. Absolute Dynamic Topography and geostrophic currents).

Prod4: DOI (product): https://doi.org/10.48670/moi-00141 (accessed November 17, 2022)

Prod 5: DOI (product): https://doi.org/10.48670/moi-00142 (accessed November 17, 2022)

Prod6: Global climate and weather analysis (ERA5, Hersbach et al., 2018) with ¼º of horizontal resolution and hourly/monthly (?) temporal resolution.

Figures

All the figures have low resolution and too small characters (xlabel, ylabel and colorbar). Please improve and enlarge the characters of all the figures. 

MINOR COMMENTS

In my opinion, the abstract is too much detailed (in particular with percentage of variance, correlation numbers). 

In all the manuscript:

• Replace all “associated to“ by “associated with”
• Replace “Levantine/Modified Atlantic Waters” by “Levantine and Modified Atlantic Waters”
• I would write in situ (in italic, and without “–“)
• Avoid “/” in the text. To be replaced by a word.
• In the text, do not indicate “first/second/etc column”. Only reference to the figure

Abstract

 l.14: Precise Mediterranean Sea reanalysis

l.15: satellite chlorophyll concentration

l.15: I would prefer profiling floats that Argo floats

l.16 and 20: of total variance

l.23: biological production?

l.31: I would replace “i.e.” by “such as”

Introduction

• Add the reference to IPPC 2021?
• 38: I would delete “Despite being a marginal Sea”. Not necessary and “despite” introduce a negative aspect…
• 2^nd paragraph: the authors could reorganize the geographical description: first the Adriatic Sea, secondly the Southern part.
• 53:” Marine Strategy Framework Directives (MSFD)” rather than “sensu MFSD”
• 53: refer to IPPC 2021?
• 54: replace “Marine Strategy Framework Directives” by “ MSFD”
• 57: replace “i.e.” by “distributed by the”
• 58: reanalysis in the Mediterranean Sea
• 58: I would replace “at 1/24º horizontal resolution” by “high-resolution” (such precision is given in section 2)

Data and method

See my previous comment concerning the dataset descriptions (references and DOIs) + all datasets have to be introduced.

• 65: reanalysis in the Mediterranean Sea
• 65: replace “in” by “over”
• 66 I would replace “in the time” by “over the”
• 67: replace CMEMS by Copernicus
• I suggest to rewrite the beginning of this section, combining l.64-66 and 67-69 (to avoid repetitions)
• 70: what is the reference to ThemeXX et al.?
• 71: past/future: past and future
• 89: delete “the”
• 90 and 96: at surface and in subsurface, (removing level)
• 94 Levantine Water should be Levantine Intermediate Water (LIW)
• 103: could you explain why 13 months?
• 110: “vertical pattern” rather than “spatial pattern”

Results

- I suggest to delete the 3.1 Subtitle and move this paragraph in section 3.2. Then, Section 3.2 will become 3.1, and 3.3 will become3.2.

- l.115: replace “almost 50%” by “48.9%”

- l-119: 20%, be precise here, indicate 19.7 and 17.7 % respectively. In general, be more precise in section 3. results

-l.123-125: such results raise the questions about the dependency between the drivers. I may be not a problem. Do you have an opinion about that?

-l.138: multiple modes

- l. 140: where the seasonal signal is strong, you could remove it to better highlight and quantify the interannual variability, no? Have you test it?

2021 Anomaly

l.152: r=?

1. 152 “seems to be connected”… coincide with? With what it is written in the literature?
2. 159: remove “latest” (see my comment above)

l.160: rephrase

l.161: long-term

l.165: that we conducted ?

Figures

See my comments above.

Fig 3: add the 0-value line in all panels (a,c,e,g). In legend replace “spatial” by “vertical”, and “,8.4%” by “and 8.4%”.

Fig 4: legend: “climatological mean […] in 1999-2020 “ could be replace by “Mean over 1999-2020” and “reference period” by “1999-2020 period”.

Table 1: see my previous comments

Table 2: EOFs of DO