See attached file

The authors reconstructed 0.25 degree monthly full carbonate system variables during the period 1985-2021 based on surface ocean observation data. Distributions of pCO2 were reconstructed based on the machine learning method established by the authors (Trang-Chau et al. 2022) and those of TA were based on the LIAR method (Carter et al. 2016; 2018). While few reconstructions of full carbonate system variables are available at this moment, a comprehensive understanding of global surface ocean pH distributions is essential for monitoring ocean acidification, which is related to the SDG indicator 14.3.1. This study can enhance researches on the global carbon cycle as well as provide critical information to policymakers and stakeholders. I think this study has sufficient value to be published in this journal, but major concerns listed below should be addressed appropriately. I would like to encourage the authors to improve the study and revise the manuscript for better understanding.

General comment

The concept of this study itself is not novel, and the assessment of uncertainty in the reconstructed fields and the validation of the method become important. The authors derived uncertainty distributions in reconstructed parameters from the spread of 100 model ensemble. They also demonstrated the validity of the method by comparing the result of this study with observation data that were not used for learning and those of time-series points. The time-series used in this study are biasedly located in the subtropical region, so comparing their data with the results of this study does not seem a good indicator of uncertainty. For validation of this method, the authors must take a comparison with other reconstruction(s) into account, if needed.

The authors use external SST and SSS instead of those incorporated in the datasets in the learning process. This seems unusual because the oceanographic condition represented by temperature and salinity considerably affects the ocean biogeochemistry in the observed area. If the authors think the use of external SST/SSS to be essential, they must demonstrate that the impact of differences between external SST/SSS and those in the datasets is negligible.

In addition, the manuscript seems to contain unnecessary sentences and be lengthened. Shortening the manuscript will increase readability.

Specific comment

1. Introduction

This section seems too long and needs to be shortened.

L71

not only “extrapolate” but also “interpolate”.

L91 Table 1 and Appendix A

Table 1 only shows six carbonate system variables and is not necessary. Reference to them in the text is enough. In the same context, Appendix A is also unnecessary because it only contains general explanations of carbonate system variables as written in, e.g., Dickson et al. 2007.

L95-131

All or a part of these explanations had better be transferred to the beginning of the “3 Reconstruction method” section.

L135

Which were used, sea surface height anomaly (SLA) or sea surface dynamic height (SLA+MDT)? Please clarify.

L155

SOCAT’s full name was already mentioned in L66.

L160-165

Using global 0.25 deg binned data derived from SOCAT cruise data is a usual way, even though using 1 deg binned data does not significantly affect the result.

L213

It should be clarified how you dealt with longitude and latitude parameters.

Table 4

This table contains RMSDs and coefficients of determination, and the name “skill score” is not appropriate. RMSDs of r025 are not significantly different from those of r100 according to Table 4, and therefore the authors should not emphasize an improvement of the prediction skill. The results only show that a fine-scale reconstruction was achieved with no adverse effect.

Fig 2-4

The results from the two methods, r100 and r025, have almost the same structure. Please explain the reason why the authors focused on the comparison of them.

L374 Fig. 3

RMSD for the Sea of Japan is suppressed by using data in the subtropical regions (Tsushima warm current area and Kuroshio area) which generally can be estimated more easily. The RMSD must be calculated from data restricted north of the subtropical front.

L403-414

The discrepancy between the estimated and observed pCO2 not only originated from the timescale but also from the method itself. The method cannot express short-term phenomena inherently because it used external SST and SSS instead of those incorporated in the datasets in the learning process.

5.2 Total alkalinity and dissolved inorganic carbon

In the method of this study, discrepancies in estimated and observed DIC were initially derived from pCO2 and TA estimation and propagated via carbonate system calculations. The discussion on uncertainty should be written along with such a concept.

L467-469

The authors attributed a large σ of DYFAMED estimates to a limited number of observations in the Mediterranean, but GLODAPv2 includes alkalinity measurement data in the Mediterranean. Schneider et al. 2007 successfully derived the salinity-alkalinity relationship. The discrepancy in DYFAMED seems to be attributable to salinity discrepancy only.

L539-540

I think that SDG indicator 14.3.1, “Average marine acidity (pH) measured at agreed suite of representative sampling stations”, is worth mentioning here. Global mean pH based on observation can be a proxy for the indicator. In addition, it is also valuable information that the global mean pH becomes 8.0 with one decimal place, not 8.1 often said.

6 Conclusion and discussion

This section had better be titled “Summary”. It does not seem to include discussion.

Carter et al. 2016, Locally interpolated alkalinity regression for global alkalinity estimation, Limnol. Oceanogr. Methods, 14, 268–277, https://doi.org/https://doi.org/10.1002/lom3.10087.

Carter et al. 2018, Updated methods for global locally interpolated estimation of alkalinity, pH, and nitrate, Limnology and Oceanography: Methods, 16, 119–131, https://doi.org/https://doi.org/10.1002/lom3.10232, 2018.

Chau et al. 2022, A seamless ensemble-based reconstruction of surface ocean pCO2 and air–sea CO2 fluxes over the global coastal and open oceans, Biogeosciences, 19, https://doi.org/10.5194/bg-19-1087-2022

Dickson et al (Eds.) 2007, Guide to Best Practices for Ocean CO2 Measurements. PICES Special Publication 3, 191 pp.

Schneider et al. 2007, Alkalinity of the Mediterranean Sea, Geophys. Res. Lett., 34, L15608, doi:10.1029/2006GL028842.