the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Long term Strengthening of the CO2 Sink and Spatiotemporal pCO2 dynamics in the northern Gulf of Mexico: Insights from a 22 year Satellite based Machine Learning Reconstruction
Abstract. The northern Gulf of Mexico (nGOM) is a river‑dominated marginal sea with strong physical‑biogeochemical variability. We reconstruct sea surface partial pressure of CO2 (pCO2) at 4‑km, 8-day resolution from 2003 to 2024 using a satellite‑based, season‑specific random forest model (independent validation R² = 0.82, RMSE = 27.6 μatm). The climatological pCO2 distribution exhibits a sharp coastal‑to‑offshore gradient: river‑influenced coastal waters (SSS < 33) have persistently low pCO2 with high spatial variability, while offshore waters (SSS > 33) have higher pCO2 with weaker heterogeneity and lower seasonal amplitude. The nGOM acts as a net CO2 sink for atmospheric, largely concentrated in the river‑influenced plume region due to riverine nutrient‑stimulated biological uptake. Seasonal pCO2 variation is dominantly controlled by temperature but counteracted by spring‑summer biological drawdown (reducing pCO2) and autumn‑winter vertical mixing with CO2‑rich deeper water (raising pCO2). Interannual pCO2 variability is dominantly affected by year-to-year changes in river discharge and nutrient loading, with higher discharge leading to lower pCO2 via enhanced biological uptake. On a decadal timescale, sea surface pCO2 increased at a rate of 0.50 ± 0.20 μatm yr-1, much slower than atmospheric pCO2 (2.13 ± 0.04 μatm yr-1), leading to a strengthening oceanic CO2 sink with the sea-to-air flux becoming more negative at −0.41 ± 0.06 mmol C m-2 d-1 yr-1. Furthermore, a decreasing frequency of easterly winds has reduced the westward transport of the Mississippi River plume, causing a higher pCO2 increasing rate on the western Texas‑Louisiana shelf.
- Preprint
(2981 KB) - Metadata XML
-
Supplement
(944 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-3034', S.E. Lohrenz, 25 Jul 2026
-
RC2: 'Comment on egusphere-2026-3034', Anonymous Referee #2, 11 Aug 2026
General comments:
The manuscript presents a method to reconstruct surface ocean pCO2 in the northern Gulf of Mexico (nGOM) using Machine learning. A 22 year, spatially resolved, time series of nGOM surface pCO2 is created. The method itself seems robust, filling the gaps in the observational dataset using satellite observations. The dataset is then analyzed for spatial and seasonal patterns. Most of these findings have been previously described using direct observations or models and it is therefore unclear how this new method improve our understanding of surface pCO2 in the region. Without new results the manuscript becomes a methodological paper, which is useful, but would be best published elsewhere.
The decadal trends are probably the most interesting result from this study. The manuscript should focus more on these trends. Thermal and non-thermal trends are interesting and provide additional knowledge for nGOM. However, the mechanisms to explain these trends are currently lacking.
Finally, the format chosen by the authors, mixing Results and Discussion, does not help. This format often results in a lack of discussion, which is the case in the manuscript. The discussion is largely missing. Splitting Results and Discussion section would force the authors to clarify the results and to put them in perspective, which would likely greatly improve the manuscript.Specific comments:
L101: What do you mean by true values?
L157: Can you provide some information on the use of the internal validation dataset for hyperparameter tuning?
L196: Is this pCO2,mean as well?
L208: The model results deserve a bit more than 1 paragraph discussion.
L222: Additional validation: Please provide more discussion of Figure S1 and include individual statistics.
L236-253: This is already known, how does this new high resolution dataset improve our understanding of surface pCO2 in the nGOM region?
Figure 3: It would be interesting to have a map of the spatial errors associated with these maps.
L264-276: It would be better to focus on pCO2 and relate it to the other variables, instead of describing what is already known first.
L270-276: This would fit better in a proper Discussion section.
L281: Indeed it was already shown by Huang et al 2015, so what is the contribution from the new model?
Figure 4:
- Can you tighten the space between panels and increase the font size?
- Panel b: Practical Scale Salinity does not have units, see https://www.jstor.org/stable/43924646 Please update all figures/text accordingly.
- Panel e: Supposedly there is a significant temporal variability that is not shown here.
- Panel g: Is this calculated from the mean of spatially/temporally varying ∆pCO2 (as it should) or as the difference of the 2 lines in panel 4e? Visually it seems to be the latter, which I don't think is the right way to calculate ∆pCO2.
L376-380: This is not shown, you should stick to the results or discuss based on the literature.
L396-398: These trends are not statistically significant and may be due to the time window chosen for the study. If there are indeed trends, please reference the corresponding studies.
L402-404: This is an overstatement and doesn't seem to match the citations. Boyce et al is a global study, whereas Rabalais et al mentions the accumulation of nutrients as a reason for Chl increase.
L406: Li et al (2022): Is this for NGOM or for the whole Gulf? This statement is misleading because it mixes processes that control open water chl and the Mississippi plume chl.
L408-409: The change in chl is highly variable though and may not be in sync with the change in SST. Indeed, the trend in NpCO2 is not significant.
L409-411: If you want to show this effect you need a more convincing figure.
Figure 9:
- Panel g: This is much more informative than Figure 4g and questions the need for Figure 4, particularly that panel.
L436: This is almost the same rate as the atmospheric pCO2, can you comment? How did Kealoha interpret this trend in the northwestern shelf?
L448: How significant is this trend? Can the units be interpreted like this: the dark blue regions correspond to 0.4% per year, which over the ~20 years cover by the study would be a ~8% decrease in the frequency of easterly winds?
L449: Note that for this to be true the change in easterly winds need to be calculated over the productive season only.
L454-456: You need to provide better supportive data/figure to demonstrate this.
L457-467: There is weak support for these conclusions.
Figure 10: Be careful with overinterpretation: these panels don't provide information about the significance of these trends.
L510: Can you provide the model code on GitHub? I wonder if sharing reconstructed data only is enough to meet BG standards.
Figure S3d: There seem to be sharp changes in derived pCO2, presumably associated with the switch between models. Can you comment on that? The other variables have much smoother transitions in the seasonal cycle.
Technical comments, typos:
L72: seasonality in air-sea CO2 exchange
L87: ...these limitations and achieve a continuous, satellite remote sensing... Something is missing in this sentence.
L92: made a significant
L100: suggestion: In this area, the true values of the decadal trend had large uncertainties...
L102: this study attempt to develop
L106: Our strategy is to train dedicated
L132: Replace "gravel and detritus" by "dissolved and detrital organic matter"
Figure 1f: Training
L277: remove "clear", this terminology is very subjective.
Figure 5: See earlier comment about psu.
Figure 9:
- Same comment as earlier about the size of the font
- Panel c: psu: see earlier comment
Figure 10: Panel e: psu: see earlier comment
- L496: Acknowledgements (according to BG template)
Figure S2: Can you provide the animation?
Citation: https://doi.org/10.5194/egusphere-2026-3034-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 123 | 40 | 16 | 179 | 24 | 12 | 17 |
- HTML: 123
- PDF: 40
- XML: 16
- Total: 179
- Supplement: 24
- BibTeX: 12
- EndNote: 17
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of “Long-term Strengthening of the CO2 Sink and Spatiotemporal pCO2 dynamics in the northern Gulf of Mexico: Insights from a 22-year Satellite-based Machine Learning Reconstruction” by Jiang et al.
This manuscript applies a seasonally specific random forest model to a long term pCO2 dataset to provide a 22 year simulation of variations in pCO2 in the northern Gulf of Mexico. Their findings support the view that the northern Gulf of Mexico is a net sink for atmospheric CO2, driven to a large extent by the river-influenced region. The authors also examined both thermal and non-thermal effects on pCO2.
The manuscript is well-written and provides a comprehensive literature review of prior related work in the northern Gulf. Methods are clearly described. The effort builds on and goes beyond prior work in developing a more extensive time-series and examining controlling factors on pCO2. The authors provide an insightful analysis of the drivers of pCO2 variations. Findings were related to seasonal and long-term SST variations, and they also identify important trends in changes in biological drivers (e.g., increasing chlorophyll) and physical factors mediated by vertical mixing and plume transport.
It would be useful if the authors also discussed how their CO2 flux rates compared their rates to those of prior work such as that cited in the paragraphs of the introduction on lines 69-101. Perhaps a table comparing rates from different studies could be included?
Overall, this is a useful and important contribution and, provided the authors can address this and additional minor comments below, I recommend publication.
Specific comments:
Lines 106-108: Explain how transitions between the seasonal submodels occur. How to avoid abrupt transitions in the simulations?
Line 132: The ‘g’ in ‘adg’ stands for ‘Gelbstoff’, which a German word for ‘yellow substance’, and refers to the dissolved organic matter fraction of absorption. So the term ‘adg’ refers to the dissolved-detrital fraction of absorption. Please correct.
Lines 156-157: The in situ pCO2 data was randomly divided into three subsets: training, internal validation and external validation. How representative was the external validation subset to different regions of the Gulf? It would also be helpful to have more explanation of how the internal validation subset was used for the hyperparameter tuning, and how it usage differed from the external subset as the results for the two were very similar.
Lines 161-172: As mentioned above, how are the transitions between seasons managed in order to avoid discontinuities in the time-series?
Figure 7: This figure is especially small and it would be helpful if the plots and text could be enlarged slightly.
Figure 9: Text is very difficult to read in this figure, particularly where it overlaps with data plots.