Daniel Ford and coauthors describe a study in which three different biological parameters — chlorophyll a (Chl a), net primary production (NPP), and net community production (NCP) — were tested as predictors in neural networks to estimate the partial pressure of CO2 in the surface ocean (pCO2(sw)) in the South Atlantic Ocean. Fields of pCO2(sw) generated by these three neural networks were compared to each other, as well as to fields generated by two additional neural networks that did not include biological predictors, a recently published global surface pCO2(sw) product (Watson et al., 2020), and in situ literature values of pCO2(sw). Also, a perturbation study was carried out to quantify the potential for improvements to pCO2(sw) predictions from each of the three neural networks with biological predictor parameters.
The authors conclude that the approach that includes NCP as a biological predictor provides the most accurate values of pCO2(sw) in equatorial upwelling regions and in the Amazon plume region. They demonstrate this result by comparing climatologies generated by the neural networks to in situ buoy measurements, values of pCO2(sw) reported in the literature, and climatologies generated by separate neural networks without biological predictors. They also conclude that the approach that includes NCP as a biological predictor has the greatest capacity for improvement to its performance as uncertainties are reduced.
The authors have responded well to the reviewers’ comments, resulting in improvements to the presentation and discussion of their results. The modifications made to Figs. 1, 3, and 5 are especially helpful to the manuscript. I do support the publication of this work, as the implications are both important and interesting. Nevertheless, additional comments and editorial corrections are listed in the following section, which I hope will lead to further improvement to the manuscript.
Specific Comments and Technical Corrections:
Lines 9–10: Recommend revising to “As a part of this process…”
Line 14 (and elsewhere): Recommend revising to “…which biological proxy produces the most accurate fields of pCO2(sw).”
Line 18: Add missing period after “parameters”
Line 20: Recommend revising to “…this region appears to be a sink for CO2”
Line 45: Recommend revising to “Where NCP is positive…” to match the structure of the following sentence.
Line 64: Recommend revising to “This dynamic biogeochemical variability in conjunction with…” or something more descriptive than just “This”
Line 69: eliminate errant “a” between “alongside” and “two”
Lines 97–100: This paragraph seems unnecessary until reading in section 2.6 that the PIRATA buoy data are flagged E. I’d either mention the PIRATA data here, or just remove this paragraph. There is no mention of dataset quality flags in the preceding paragraph, so there is not necessarily a reason for the reader to assume that flag E data weren’t also downloaded along with the core SOCAT data.
Line 115: Recommend revising to “These satellite algorithms were shown to be the most accurate…”
Line 116: Change “accounting” to “accounted”
Lines 151–158: Although it is explained here, I was initially confused as to exactly which parameters are used in training each of the NNs. A table may be helpful in clarifying this. Most importantly, that SA-FNNNO-BIO-2 and Watson et al. (2020) are the only NNs that use salinity and mixed layer depth as predictors.
Lines 268–270: This sentence is a bit confusing at the moment. One suggested revision here: “This showed that a reduction in pCO2(sw) RMSD of 36% was achieved by eliminating satellite NCP uncertainties, 34% by eliminating satellite NPP uncertainties, and 19% by eliminating satellite Chl a uncertainties.”
Figure 3: Unfortunately, with the helpful addition of new data to these plots, this figure has become very difficult to interpret (at least given the quality of image I have). This could perhaps be remedied by simply reshaping the panels: an elongated y-axis might help emphasize the distinctions between individual lines. Another option may be to adjust the color palette selection. Or, to split this into two separate figures, showing the climatology from SA-FNNNCP in both.
Line 286: Change “climatology” to “climatologies”
Line 363: “…indicated however, that elevated pCO2(sw) at ~430 uatm exist…” During what time of the year is this elevated pCO2(sw) occurring? Year-round?
Line 364: “The PIRATA buoy pCO2(sw) observations (Fig. 3a) clear highlight the difference between these years…” It’s not clear to me how the monthly climatology in Fig. 3a highlights a difference between the years. Are PIRATA observations only available from 2008 to 2011, during which time Bruto et al. indicate higher pCO2(sw)?
Line 472: Change “reduced” to “eliminated” or “reduced to zero”
Line 475: Recommend revising to “…and two neural networks that do not use…”
Line 478–479: Add to the end of this sentence “occurred” or “was observed”