A machine-learning reconstruction of sea surface <em>p</em>CO<sub>2</sub> in the North American Atlantic Coastal Ocean Margin from 1993 to 2021

Wu, Zelun; Lu, Wenfang; Roobaert, Alizée; Song, Luping; Yan, Xiao-Hai; Cai, Wei-Jun

doi:https://doi.org/10.5194/essd-2024-309

Preprints

https://doi.org/10.5194/essd-2024-309

Preprints

19 Aug 2024

| 19 Aug 2024

Status: a revised version of this preprint was accepted for the journal ESSD and is expected to appear here in due course.

A machine-learning reconstruction of sea surface pCO₂ in the North American Atlantic Coastal Ocean Margin from 1993 to 2021

Zelun Wu, Wenfang Lu, Alizée Roobaert, Luping Song, Xiao-Hai Yan, and Wei-Jun Cai

Abstract. Insufficient spatiotemporal coverage of partial pressure of CO₂ (pCO₂) observations has hindered precise studies of the coastal carbon cycle along the North American Atlantic Coastal Ocean Margin (NAACOM). Earlier pCO₂-products have encountered difficulties in accurately capturing the heterogeneity of regional variations and decadal trends of pCO₂ in the NAACOM. This study developed a regional reconstructed pCO₂-product for the NAACOM (Reconstructed Coastal Acidification Database-pCO₂, or ReCAD-NAACOM-pCO₂) using a two-step approach combining random forest regression and linear regression. The product provides monthly pCO₂ data at 0.25° spatial resolution from 1993 to 2021, enabling investigation of regional spatial differences, seasonal cycles, and decadal changes in pCO₂. The observation-based reconstruction was trained using Surface Ocean CO₂ Atlas (SOCAT) observations as ground-truth values, with various satellite-derived and reanalysis environmental variables known to control sea surface pCO₂ as model inputs. The product shows high accuracy during the model training, validation, and independent test phases, demonstrating robustness and capability to accurately reconstruct pCO₂ in regions or periods lacking direct observational data in the NAACOM. Compared with all the observation samples from SOCAT, the pCO₂-product yields a determination coefficient of 0.83, a root-mean-square error of 18.64 µatm, and an accumulative uncertainty of 23.83 µatm. The ReCAD-NAACOM-pCO₂ product demonstrates its capability to resolve seasonal cycles, regional-scale variations, and decadal linear trends of pCO₂ along the NAACOM. This new product provides reliable pCO₂ data for more precise studies of coastal carbon dynamics in the NAACOM region. The dataset is publicly accessible at https://doi.org/10.5281/zenodo.11500974 (Wu et al., 2024a) and will be updated regularly.

Received: 21 Jul 2024 – Discussion started: 19 Aug 2024

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.

Download & links

Zelun Wu, Wenfang Lu, Alizée Roobaert, Luping Song, Xiao-Hai Yan, and Wei-Jun Cai

Status: closed

RC1:
'Comment on essd-2024-309', Anonymous Referee #1, 14 Oct 2024

The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2024-309/essd-2024-309-RC1-supplement.pdf

Citation: https://doi.org/10.5194/essd-2024-309-RC1
- AC2: 'Reply on RC1', Zelun Wu, 02 Nov 2024
  
  Dear reviewer,
  Thanks for reviewing this manuscript. We have carefully revised the manuscript based on the constructive and valuable suggestions from you and another reviewer. The manuscript has been extensively improved during this revision. Please see the attached file for details.
  Best,
  Zelun Wu.
  
  Citation: https://doi.org/10.5194/essd-2024-309-AC2
RC2:
'Comment on essd-2024-309', Anonymous Referee #2, 17 Oct 2024

Overview: The authors are presenting a new, regional pCO2-product specifically designed for the North American Atlantic coastal region that provides monthly pCO2 at a .25-degree spatial resolution from 1993-2021. The product uses integrated random forest and linear regression methods incorporating observational products in order to generate their monthly reconstructed pCO2-product. This allows for analysis of regional, seasonal, and yearly trends in addition to an uncertainty calculation. The authors find through validation that their product provides high accuracy, improving public access for more precise, higher resolution coastal carbon dynamics in the NAACOM region.
There is great need for products like these to be publicly accessible, and this contributes an important resource to the scientific community. The authors do a nice job of introducing the field, the data gaps, and where this product can contribute to those gaps. I do recommend for publication, following a few edits as outlined below.
One edit I have for the paper regards the interchangeable use of fCO2 and pCO2. In section 2.1 the authors mention that “both are commonly used in oceanographic studies”, which is accurate. However, they are not interchangeably used. At the end of section 2.2 an equation to convert pCO2 to fCO2 is provided, but it’s unclear at what point this conversion is made. Figure 2 shows values in fCO2, but the rest of the figures use pCO2. I recommend a clear statement about conversion with the introduction of fCO2, as well as consistency in the figures (I would convert figure 2 to showing pCO2 or at least have a clear statement on the conversion and reason for fCO2 presentation in the figure caption).
My second edit has to do with the calculation of uncertainty. There is a calculation representing the inputs (u_inputs), but some of the products are in the original resolution and others are linearly interpolated to that resolution. Does the interpolation introduce more error, and is this taken into account? Additionally, I will note that I greatly appreciate the inclusion of an uncertainty calculation and the strength it lends to the product. I would have liked to see it highlighted a bit more in the rest of the paper results—some of the data could also be discussed with uncertainty included, rather than purely keeping the uncertainty in one section at the end of the paper. I think the addition of the uncertainty calculation makes this product stronger, and should be displayed as such.
Finally, Figure 5 and figure 7, and the associated discussions in 3.2 and 3.4, present some extremely interesting data. We compare some of the regional differences and the products effectiveness of capturing broader pCO2 patterns across the North Atlantic coast. I would have loved to had this extrapolated on a little further, and perhaps seen more numbers broken down by region. We can visually look at the figures, but it’s a little hard to assess and I think the paper would be strengthened by expanding this section a little more with increased quantitative results.

Specific Comments:
Figure 1: I felt that the colors of this figure made it difficult to interpret. The way the lines were drawn made the topography difficult to see. For consistency with the other figures in the paper, I would suggest shifting the coastal contour line to being black. Then perhaps make the Gulf stream and Labrador current lines dotted or dashed lines (also, change Gulf Stream’s color if you shift contour to black), so they don’t block as much topography. I would match the labels of the regions to the lines denoting the regions, and finally increase the deviation in the color scale. Right now, it’s not very easy to tell a difference between 800-1000m, and similarly between 0-300m is all about the same tone.
Equations: center the equations in the document
Lines 183-184: The nature of this sentence is implying an interchangeable use of fCO2 and pCO2, which I don’t think is accurate
Line 284+: Authors show an area-mean bias of +0.17, but with the regional breakdown and discussion, I’d be very curious how that bias varies by region. Can we see numbers for the other regions as well?
Figure 5: Similar edit suggestions to figure 1; update contour line to be black and match the colors of regional names to the lines denoting the regions
Figure 6: the error bar denotes one standard deviation of the monthly mean climatology, but didn’t the authors also actually calculate a pCO2 error? Why is that not included in any of the figures?
Line 415: The statement “This uncertainty is deemed reasonable” confused me. Deemed reasonable by who? What metrics are being used? “reasonable” is a very vague term.
Acknowledgments: Make sure to include the SOCAT statement from the website that they ask you to include when you use their product (“The Surface Ocean CO2 Atlas (SOCAT) is an international effort, endorsed by the International Ocean Carbon Coordination Project (IOCCP), the Surface Ocean Lower Atmosphere Study (SOLAS) and the Integrated Marine Biosphere Research (IMBeR) program, to deliver a uniformly quality-controlled surface ocean CO2 database. The many researchers and funding agencies responsible for the collection of data and quality control are thanked for their contributions to SOCAT.”)

Citation: https://doi.org/10.5194/essd-2024-309-RC2
- AC1: 'Reply on RC2', Zelun Wu, 02 Nov 2024
  
  Dear reviewer,
  Thanks for reviewing this manuscript. We have carefully revised the manuscript based on the constructive and valuable suggestions from you and another reviewer. The manuscript has been extensively improved during this revision. Please see the attached file for details.
  Best,
  Zelun Wu.
  
  Citation: https://doi.org/10.5194/essd-2024-309-AC1
EC1:
'Comment on essd-2024-309', Sabine Schmidt, 25 Oct 2024

Both reviewers agree that your manuscript presents a useful product, i.e. it reconstructs sea surface pCO2 in the North American Atlantic coastal ocean margin from 1993 to 2021. However, they also agree that there is room for improvement.

In particular, reviewer 1 asked for a better explanation of the methodology used to adjust the RFR estimates with an LR. The results should be discussed further, especially the differences between raw data and reconstruction, to help future users of the product.

Reviewer 2 regretted the interchangeable use of fCO2 and pCO2; clarification of the conversion is needed, as well as consistent use in the figures. The introduction of the uncertainty calculation is to be welcomed, but there are a number of issues raised by Reviewer 2 that need to be considered.

There are also a number of minor comments that should be taken into account to produce a quality article and ensure that the reconstructed database is used.

Citation: https://doi.org/10.5194/essd-2024-309-EC1
- AC3:
  'Reply on EC1', Zelun Wu, 02 Nov 2024
  Dear Editor,
  We have carefully revised our manuscript according to the constructive feedback provided by both reviewers. The main revisions include:
  Clarifying the distinction between fCO₂ and pCO₂ throughout the manuscript
  
  Improving the presentation and explanation of Mean Bias Error (MBE) in Section 3.2 and Figure 5 to enhance clarity
  
  Substantially revising Section 3.3 to explain better the discrepancies between product-estimated pCO₂ and SOCAT observations in northern areas, highlighting the impact of limited observational coverage.
  
  Restructuring Section 3.4 and Figure 7 to clearly distinguish between previously documented regional variations and newly identified phenomena revealed by our product
  
  Detailed responses to each reviewer's comments are attached in the replies to the two reviewers.
  We appreciate the time and effort you and the reviewers have invested in helping us improve this work.
  
  Best regards,
  Zelun Wu
  
  Citation: https://doi.org/10.5194/essd-2024-309-AC3

Status: closed

RC1:
'Comment on essd-2024-309', Anonymous Referee #1, 14 Oct 2024

The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2024-309/essd-2024-309-RC1-supplement.pdf

Citation: https://doi.org/10.5194/essd-2024-309-RC1
- AC2: 'Reply on RC1', Zelun Wu, 02 Nov 2024
  
  Dear reviewer,
  Thanks for reviewing this manuscript. We have carefully revised the manuscript based on the constructive and valuable suggestions from you and another reviewer. The manuscript has been extensively improved during this revision. Please see the attached file for details.
  Best,
  Zelun Wu.
  
  Citation: https://doi.org/10.5194/essd-2024-309-AC2
RC2:
'Comment on essd-2024-309', Anonymous Referee #2, 17 Oct 2024

Overview: The authors are presenting a new, regional pCO2-product specifically designed for the North American Atlantic coastal region that provides monthly pCO2 at a .25-degree spatial resolution from 1993-2021. The product uses integrated random forest and linear regression methods incorporating observational products in order to generate their monthly reconstructed pCO2-product. This allows for analysis of regional, seasonal, and yearly trends in addition to an uncertainty calculation. The authors find through validation that their product provides high accuracy, improving public access for more precise, higher resolution coastal carbon dynamics in the NAACOM region.
There is great need for products like these to be publicly accessible, and this contributes an important resource to the scientific community. The authors do a nice job of introducing the field, the data gaps, and where this product can contribute to those gaps. I do recommend for publication, following a few edits as outlined below.
One edit I have for the paper regards the interchangeable use of fCO2 and pCO2. In section 2.1 the authors mention that “both are commonly used in oceanographic studies”, which is accurate. However, they are not interchangeably used. At the end of section 2.2 an equation to convert pCO2 to fCO2 is provided, but it’s unclear at what point this conversion is made. Figure 2 shows values in fCO2, but the rest of the figures use pCO2. I recommend a clear statement about conversion with the introduction of fCO2, as well as consistency in the figures (I would convert figure 2 to showing pCO2 or at least have a clear statement on the conversion and reason for fCO2 presentation in the figure caption).
My second edit has to do with the calculation of uncertainty. There is a calculation representing the inputs (u_inputs), but some of the products are in the original resolution and others are linearly interpolated to that resolution. Does the interpolation introduce more error, and is this taken into account? Additionally, I will note that I greatly appreciate the inclusion of an uncertainty calculation and the strength it lends to the product. I would have liked to see it highlighted a bit more in the rest of the paper results—some of the data could also be discussed with uncertainty included, rather than purely keeping the uncertainty in one section at the end of the paper. I think the addition of the uncertainty calculation makes this product stronger, and should be displayed as such.
Finally, Figure 5 and figure 7, and the associated discussions in 3.2 and 3.4, present some extremely interesting data. We compare some of the regional differences and the products effectiveness of capturing broader pCO2 patterns across the North Atlantic coast. I would have loved to had this extrapolated on a little further, and perhaps seen more numbers broken down by region. We can visually look at the figures, but it’s a little hard to assess and I think the paper would be strengthened by expanding this section a little more with increased quantitative results.

Specific Comments:
Figure 1: I felt that the colors of this figure made it difficult to interpret. The way the lines were drawn made the topography difficult to see. For consistency with the other figures in the paper, I would suggest shifting the coastal contour line to being black. Then perhaps make the Gulf stream and Labrador current lines dotted or dashed lines (also, change Gulf Stream’s color if you shift contour to black), so they don’t block as much topography. I would match the labels of the regions to the lines denoting the regions, and finally increase the deviation in the color scale. Right now, it’s not very easy to tell a difference between 800-1000m, and similarly between 0-300m is all about the same tone.
Equations: center the equations in the document
Lines 183-184: The nature of this sentence is implying an interchangeable use of fCO2 and pCO2, which I don’t think is accurate
Line 284+: Authors show an area-mean bias of +0.17, but with the regional breakdown and discussion, I’d be very curious how that bias varies by region. Can we see numbers for the other regions as well?
Figure 5: Similar edit suggestions to figure 1; update contour line to be black and match the colors of regional names to the lines denoting the regions
Figure 6: the error bar denotes one standard deviation of the monthly mean climatology, but didn’t the authors also actually calculate a pCO2 error? Why is that not included in any of the figures?
Line 415: The statement “This uncertainty is deemed reasonable” confused me. Deemed reasonable by who? What metrics are being used? “reasonable” is a very vague term.
Acknowledgments: Make sure to include the SOCAT statement from the website that they ask you to include when you use their product (“The Surface Ocean CO2 Atlas (SOCAT) is an international effort, endorsed by the International Ocean Carbon Coordination Project (IOCCP), the Surface Ocean Lower Atmosphere Study (SOLAS) and the Integrated Marine Biosphere Research (IMBeR) program, to deliver a uniformly quality-controlled surface ocean CO2 database. The many researchers and funding agencies responsible for the collection of data and quality control are thanked for their contributions to SOCAT.”)

Citation: https://doi.org/10.5194/essd-2024-309-RC2
- AC1: 'Reply on RC2', Zelun Wu, 02 Nov 2024
  
  Dear reviewer,
  Thanks for reviewing this manuscript. We have carefully revised the manuscript based on the constructive and valuable suggestions from you and another reviewer. The manuscript has been extensively improved during this revision. Please see the attached file for details.
  Best,
  Zelun Wu.
  
  Citation: https://doi.org/10.5194/essd-2024-309-AC1
EC1:
'Comment on essd-2024-309', Sabine Schmidt, 25 Oct 2024

Both reviewers agree that your manuscript presents a useful product, i.e. it reconstructs sea surface pCO2 in the North American Atlantic coastal ocean margin from 1993 to 2021. However, they also agree that there is room for improvement.

In particular, reviewer 1 asked for a better explanation of the methodology used to adjust the RFR estimates with an LR. The results should be discussed further, especially the differences between raw data and reconstruction, to help future users of the product.

Reviewer 2 regretted the interchangeable use of fCO2 and pCO2; clarification of the conversion is needed, as well as consistent use in the figures. The introduction of the uncertainty calculation is to be welcomed, but there are a number of issues raised by Reviewer 2 that need to be considered.

There are also a number of minor comments that should be taken into account to produce a quality article and ensure that the reconstructed database is used.

Citation: https://doi.org/10.5194/essd-2024-309-EC1
- AC3:
  'Reply on EC1', Zelun Wu, 02 Nov 2024
  Dear Editor,
  We have carefully revised our manuscript according to the constructive feedback provided by both reviewers. The main revisions include:
  Clarifying the distinction between fCO₂ and pCO₂ throughout the manuscript
  
  Improving the presentation and explanation of Mean Bias Error (MBE) in Section 3.2 and Figure 5 to enhance clarity
  
  Substantially revising Section 3.3 to explain better the discrepancies between product-estimated pCO₂ and SOCAT observations in northern areas, highlighting the impact of limited observational coverage.
  
  Restructuring Section 3.4 and Figure 7 to clearly distinguish between previously documented regional variations and newly identified phenomena revealed by our product
  
  Detailed responses to each reviewer's comments are attached in the replies to the two reviewers.
  We appreciate the time and effort you and the reviewers have invested in helping us improve this work.
  
  Best regards,
  Zelun Wu
  
  Citation: https://doi.org/10.5194/essd-2024-309-AC3

Zelun Wu, Wenfang Lu, Alizée Roobaert, Luping Song, Xiao-Hai Yan, and Wei-Jun Cai

Data sets

A Reconstructed Coastal Acidification Database (ReCAD) pCO2 data product for the North American Atlantic Coastal Ocean Margins Zelun Wu, Wenfang Lu, Alizée Roobaert, Luping Song, Xiao-Hai Yan, and Wei-Jun Cai https://doi.org/10.5281/zenodo.11500974

Model code and software

Python and MATLAB code used to reconstruct the data product Zelun Wu https://github.com/zelunwu/ReCAD_product_v1

Zelun Wu, Wenfang Lu, Alizée Roobaert, Luping Song, Xiao-Hai Yan, and Wei-Jun Cai

Viewed

Total article views: 636 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
521	98	17	636	7	8

HTML: 521
PDF: 98
XML: 17
Total: 636
BibTeX: 7
EndNote: 8

Views and downloads (calculated since 19 Aug 2024)

Month	HTML	PDF	XML	Total
Aug 2024	171	35	7	213
Sep 2024	135	12	2	149
Oct 2024	128	31	5	164
Nov 2024	87	20	3	110

Cumulative views and downloads (calculated since 19 Aug 2024)

Month	HTML	PDF	XML	Total
Aug 2024	171	35	7	213
Sep 2024	135	12	2	149
Oct 2024	128	31	5	164
Nov 2024	87	20	3	110

Viewed (geographical distribution)

Total article views: 621 (including HTML, PDF, and XML) Thereof 621 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 23 Nov 2024

Short summary

This study addresses the lack of comprehensive sea surface CO₂ data in North American Atlantic coastal regions by developing a new pCO₂-product (ReCAD-NAACOM-pCO₂). Using machine learning and environmental data, it reconstructs sea surface CO₂ levels from 1993–2021. The product accurately captures seasonal cycles, regional variations, and long-term trends, outperforming earlier attempts. It provides crucial data for studying coastal carbon dynamics and climate change impacts.


Total:	0
HTML:	0
PDF:	0
XML:	0

A machine-learning reconstruction of sea surface pCO2 in the North American Atlantic Coastal Ocean Margin from 1993 to 2021

Data sets

Model code and software

Viewed

Viewed (geographical distribution)

A machine-learning reconstruction of sea surface pCO₂ in the North American Atlantic Coastal Ocean Margin from 1993 to 2021