Sharp and co-authors use a random forest regression approach to map the sea surface pCO2 for the California current system at a high 0.25x0.25 degree resolution. The authors extensively test their approach and compare it with global mapped products from the literature and conclude that their regional approach provides a substantially better representation of the sea surface pCO2, particularly when compared to local measurements.

I believe this is a novel approach and an important analysis and should therefore be published. While it is not particularly surprising that this new local approach outperforms global approaches from L17 and L20, the message is important and clear: Don’t rely on global approaches if you are interested in local features.

I am particularly impressed about the effort the authors put into evaluating their approach. Withholding independent observations or full years really guarantees that the observations seen by their machine learning method are truly independent. This should indeed be the standard – well done.

I only have a few minor comments listed below:

.) I am missing a rationale why random forest regression was chosen? It clearly was a reasonable choice, but I was wondering whether there was a particular motivation to pick this method out of the many available?

.) Uncertainty: Firstly, I believe the authors overestimate the uncertainty. As shown e.g. by Landschutzer et al 2014, the larger scale (or the full region) error is also dependent on autocorrelation features of the error. Say e.g. your gridding uncertainty for the open ocean is 4.8µatm (based on the std of each grid cell). The standard error for the entire region investigated, however, would scale with the number of uncorrelated grid cells, such that the error E=std/sqrt(N) where N is the number of independent grid cells. For an entirely random (independent) sample of grid cells, N might be very large and the error very small (since grid cell errors cancel each other out). For a dependent (correlated) sample of grid cells, N might be small. So, in a nutshell, by not accounting for the N effect, the authors overestimate their uncertainty. This should be stated

.) Additionally, the authors do a lot of effort calculating the uncertainty, but do not always display it in their figures – Figure 2 and B3 should have an error-shading (corresponding to measurement errors – plus gridding errors for the mapped products) as well, so the reader can see whether differences are within the respective uncertainty.

.) Lines 11-12: Not necessarily – other methods (not relying on the pCO2) exist as well

.)Lines 22-23: “alternative global prodcts” – I agree with the sentence but would remove “alternative”. Global products are not an alternative for regional efforts, but target a different research question (e.g. the global ocean CO2 uptake)

.)Lines 32-33 – It should be noted that the 25% refer to the annual uptake

.)Line 66: The Rodenbeck product actually covers the coastal ocean, but due to its coarse resolution it is not considered a coastal product

.)Line 110: It is more relevant to state how many SOCAT observations exist in the study region

.)Line 155 – Table 1: Probably not an issue, but are there any missing chlorophyll values in the study region?

.) Line 198: “most similar pCO2 observations” – what does this mean and how is this determined?

.) Line 365 and following: fair point to use SOCATv4 and SOCATv5, but this only partly compensates for the fact, that the RFR-CCS method still was trained with a newer dataset (and the global estimates with observations from the globe)

.) Lines 389-394: I agree with point 1 but disagree with point 2 (for reasons outlined in Gregor et al 2019). Of course local phenomena will only be better represented if a local reconstruction approach is used, but I have my doubt that exploring new methods will overcome this issue – The authors can easily test this by applying their approach to the full coastal domain and then compare the error statistics in the study region.

A review of

A monthly surface pCO2 product for the California Current Large Marine Ecosystem

J. D. Sharp et al.

General Comments

Sharp and coauthors produce a pCO2 observation-based product for the California Current region, expanding and improving upon available coastal products. Their monthly product extends back to 1998 and is at a 0.25 degree spatial resolution and utilizes a Random Forest Regression machine learning technique. Comparisons to other available products as well as in situ observations are thorough and clear.

The paper is clearly written with an important product for the ocean carbon community. I have a few simple questions to clarify and suggestions for improvement of the flow of the text, but other than that support publication of this description of the available product.

Specific Comments

Overall I feel that the abstract could be improved. For one, it doesn’t include the years that this new product (currently) covers. In my opinion, the first half of the abstract reads more like an introductory section and did not give me the information about the excellent product available, the specific improvements it exhibits over other available products, and it’s potential uses for the community. I suggest editing to expand on those important aspect further.

Also, the sentence that begins on line 36 seems disjointed from the topic of the previous sentences: anthropogenic CO₂, impact of dissolving CO₂ on ocean life, etc which then jumps to calculating CO₂ transferred between ocean and atmosphere needs the partial pressure. Consider revising to improve flow of this first paragraph.

In the discussion of how the gridded observations is created for this product (Line 135), I would be interested to see how the mean and standard deviation values vary between platforms. Could that be included in the manuscript or is it available through SOCAT somewhere?

The discussion of your machine learning method is thorough and clear. I find the discussion of three strategies for assessing skill of your model (Section 2.6) very interesting and specifically find the third strategy most exciting as local seasonality is a topic tough to dig into with coastal locations where the seasonality would be enhanced but observations tough to come by. I suggest highlighting the findings related to these tests. I find the results shown in Figure 2 (and associated supplementary figures) to be the most striking and perhaps the most clear take-away from the paper.

For Figure 4, is the N value (i.e. the number of dots on each plot) the same and could you include that number on the plot. These types of plots are sometimes hard to glean that information from since so many of the points are overlaid. From its current presentation it doesn’t look like the N value would be the same for all 4 subplots.

For Figure 5, the moorings line has the largest standard deviation/shading. I was wondering if that uncertainty bound is also calculated from the monthly mean values, as would be the case for the products being shown as comparisons (for L20, L17, and RFR-CCS-clim). I wanted to ensure it wasn’t simply that difference that is causing the larger spread in the moorings observations vs the products.

Section 3.5 has a clear discussion of uncertainty calculations and the three main sources of uncertainty considered by the authors. With a stated uncertainty of 43.6uatm in the coastal regions for this product, how do the authors suggest users go forward with that information and utilize this product given that large uncertainty range? Also, this ties into the Figure 8a plot where no uncertainty range is provided for the RFR-CCS seasonal cycle. I would imagine that given an uncertainty that large, that the spread on the resulting flux estimates would be significant. I would suggest changing the plot to include the individual estimates from the products included in the SeaFlux product (rather than the shading showing the range) and then include a shading for the uncertainty associated with the RFR-CCS product. Obviously, each of the products included in SeaFlux would have their own associated uncertainties, similar to those described here for RFR-CCS, but would be good to see regardless. Overall though, I find the comparisons discussed in Section 3.7 to be strong, clear, and decisive.

Overall/Next step Comments

I finished the paper wishing that there was more discussion on how to extrapolate this work to all coastal regions, or at least eastern upwelling regions with similar physics. Is there “sufficient” data anywhere else in the coastal regions to do such a product? Would the authors expect that the relationships from the RFR would be similar in other coastal regimes? Or would other machine learning methods perhaps be better utilized in areas with less data density? Your conclusions does touch on this somewhat, suggesting that the neural network approach use an increased number of clusters for the coastal ocean areas, but I wonder if you could comment more on this. Could SeaFlux be updated with this (and other) regional coastal products to help fill in missing areas omitted by the open ocean products? Or will the open-ocean products gradually start expanding their coverage to coastal regions with improved machine learning techniques catered specifically to coastal as you have done here?

general comments

The manuscript applies a machine-learning technique to a surface pCO2 data set in order to fill the gaps of the data set. The resulting data product is assessed at multiple levels, against field data and similar data products: its uncertainties are determined, its vulnerabilities are discussed, it is compared to other data products, its features are described, as well as its effect on the flux calculations.

Overall, the article is of very high quality. The subject is current and of high importance in our current state of the climate. It is the result of a great amount of work, well thought out and very clearly presented. 

The methodology is very sound. the authors properly selected, processed and referenced their data. The machine learning method is clearly described. Their evaluation of the model selected is thorough and proper, as detailed above.

The visualization of the data and the product is sufficient and clear for the most part (see one specific comment). 

I also think that the separation of material between the main text and the appendices is appropriate.

Lastly, the manuscript is well written, clear and concise and does a good job at explaining complex concepts. 

I only have a few minor comments, I leave it up to the authors whether they want to take the few suggestions below into consideration. 

specific comments

- On line 117, the authors mention that they back calculated the pCO2 from fCO2 values. I am wondering why since the correction is small and would not affect the air-sea difference?  

- around line 120, the authors mention the fact that SST and SSS are not really surface values, due to intake depths. At the same time, I think it is also important to note that these data have not been QC’d.

- The authors assess the effect of data availability on the model (section 2.6) and the effect of sporadic sampling on the coastal flux estimates. It would have been interesting to assess the sensitivity of the model to entire cruises, which would give a realistic idea of cruise data dependance. 

- In figure 8, the comparison of the SeaFlux and RFR-CCS products would be better served by plotting the difference between the 2. As I said, just a suggestion.

technical corrections

- line 516: replace “Also show..” by “Also shown…’