The authors present a matchup database combining various in situ observations with associated satellite remote sensing and reanalysis products. The challenge of combining these datasets which are collected over vastly different spatiotemporal scales is handled by the novel development of spatiotemporal regions of varying size.

This technique can clearly be applied to other parameters, and the dataset has the potential for reuse.

The dataset is a novel combination of other datasets, some of which have never previously been made publicly available.

Data is accessible as described and easily downloaded without the need of registration. A subset of these data was downloaded for inspection. These netcdf-v4 compatible files were inspected using the ncdf4 R library.

the "_FillValue" for the variables consists of 8-bytes which is not a valid type for R and these are coerced to a double precision value, but otherwise no issues. The files are well described in typical netcdf format and adhere to CF conventions.

The files were also tested in a python 3.8 environment using the netCDF4 package.

The dataset is missing earlier years, 1971, 1964-1967 and 1959. I presume this is due to lack of associated in-situ data, but please confirm.

No erroneous data were found and the carbonate system and associated parameters (e.g. nutrients, temperature) are shown to span appropriate ranges and be consists. At least to the extent of my knowledge of the global distribution of these parameters.

The abstract is a little long. It includes reference to an example application which is not appropriate for an abstract and is also "not shown" - line 30. This appears to have been partially copied from the Introduction section. The article appears otherwise well written.

Each sub-source of data is well described, with collection method, calibration information and uncertainties. However, In the contributions is it mentioned that A. Polukhin assessed the possibility of using certain data due to differences in methods. These differences are not well discussed in the input datasets section so I suggest the inclusion of a short paragraph summarising the rationale for this work.

We note the AMT and Kara Sea datasets start over a decade before the publication of the Dickson et al 2007 best practice document which is given as a reference. I presume the authors are simply saying that the methods are basically the same, but it would be good to clarify.

Line 486 - when and by whom will reassess the appropriateness of the maximum ROI scales? community feedback?

The figures provide a good overview of the nature of the dataset and are of generally good quality.

Figure 1 and 2 could be arranged better, the axis labels very small considering the amount of whitespace between panels

General Comments

Over the past couple decades, large datasets of in situ carbonate system measurements (e.g., SOCAT, Bakker et al., 2016) have been compiled and quality controlled, and are frequently used to train models for prediction and/or spatiotemporal gap filling of carbonate chemistry using proxy variables. However, models trained on sparse and irregularly spaced data can suffer from biases that favor more highly sampled regions or time periods. These biases can be partially alleviated by aggregating data points to bins of constant spatial and temporal extent, but binning in this way can still result in sub-optimal data division due to elongation of bins near the poles, unintentional splitting of clustered data, and unaccounted for interactions with coastlines.

To address these shortcomings, Land and coauthors compile and describe a database of in situ surface carbonate chemistry measurements built around regions of interest (ROIs), which are constructed so as to include as many in situ measurements as possible within a maximum timespan of 10 days and a maximum diameter of 100 km. This database — OceanSODA-MDB — also includes spatiotemporal matchups with satellite, model, and reanalysis datasets corresponding to each ROI. Land et al. display the utility of this newly compiled database by re-training a global algorithm from Takahashi et al. (2014) to predict potential alkalinity (PA) from sea surface salinity, displaying a global reduction in the root mean squared error between measured and predicted PA: from 15 to 12 µmol/kg in marine waters and 32 to 23 µmol/kg in coastal waters.

The ROI strategy is a compelling and creative way to group in situ data and match the grouped data with other datasets for algorithm training. The strategy is explained well in this manuscript, and certainly has potential advantages over fixed spatiotemporal binning. OceanSODA-MDB is easily accessible as a series of NetCDF files. It provides co-located satellite, model, reanalysis, and in situ data that can surely be used to re-train various algorithms that are in use today. This manuscript is a valuable contribution to the literature in its current form, but I’ve added a few comments and suggested corrections in the following sections.

Specific Comments

CO₂ system calculations only performed from C_T-A_T:

In lines 112–113 it is stated that carbonate chemistry calculations were only performed when C_T and A_T were available. Are there a significant number of cases for which other carbonate chemistry pairs were available (e.g., C_T-pH, A_T-pCO₂, etc.)? I’d be interested in the reasoning for only making carbonate chemistry calculations with the C_T-A_T pair. Similar accuracy in calculated parameters should be obtainable by pairing either C_T or A_T with either pH or pCO₂ when the measurements are of sufficient quality (e.g., Orr et al., 2018).

Creating the radial in situ data:

I think the explanation of how the data grouping is performed is great. It is clearly a complex process but breaking it down step-by-step in section 3.2 is very helpful. I am curious about the prospect of expanding this methodology to subsurface data and also how easily the methodology might be adapted to form finer resolution ROIs. I hope the authors might consider making their code publicly available at some point in the future to facilitate adaptions of their methodology to other datasets and/or spatiotemporal resolutions.

Updates to OceanSODA-MDB:

It is indicated in section 6 how the MDB could be updated. Are they any concrete plans to make updates to OceanSODA-MDB at regular intervals in the future?

Minor Comments and Technical Corrections

Line 130: The references in this line are repeated.

Line 135: Should this be dataset no. 4? Additionally, I’d be interested in knowing how the Argo data were acquired. A monthly snapshot should have an associated DOI (e.g., doi.org/10.17882/42182#95967 for Sep. 2022). If files were individually downloaded, were individual profile data files used, or interpolated Sprof files? I think this would be very helpful information for anyone who wants to replicate your methodology.

Line 141: “Dickson et al.” should be outside the parentheses here.

Lines 157–161: Is it safe to say that this paragraph and the entries for these datasets in Table 1 should be eliminated since they are added to OceanSODA-MDB along with CODAP-NA?

Lines 275–277: Are discrete pCO₂ measurements treated any differently that underway pCO₂ measurements? I’d image that in many cases discrete pCO₂ data points would spatiotemporally match with those in the SOCAT database, but may be discarded based on the criteria noted here.

Lines 381–382: I don’t think it’s obvious why the advent of Bio-Argo floats would cause the mean number of pH measurements per ROI to decrease rather than increase. I think I can infer: the 10-day sampling cycle of the floats and the fact that only one pH measurement is obtained in the upper 10 meters means one individual ROI is generally created for each float profile? Regardless, since this result seems counterintuitive on first glance, I think it should be explained here.

Figure 16: Should the green line in the legend be T14 fit, rather than TS13?

References: Lauvset et al., 2018 is not in the reference list.

References

Bakker, D.C.E., et al. A multi-decade record of high-quality fCO₂ data in version 3 of the Surface Ocean CO₂ Atlas (SOCAT), Earth System Science Data, 8, 383–413, 2016.

Orr, J.C., Epitalon, J.-M., Dickson, A.G., and Gattuso, J.-P. Routine uncertainty propagation for the marine carbon dioxide system, Marine Chemistry, 207, 84–107, 2018.

Takahashi, T., et al. Climatological distributions of pH, pCO₂, total CO₂, alkalinity, and CaCO₃ saturation in the global surface ocean, and temporal changes at selected locations, Marine Chemistry, 164, 95–125, 2014.

Please see R3 review which was erroneously posted as a reply to reviewer 2.  Apologies for any confusion!