Reconstruction and Multiyear Prediction of Global Air-sea CO<sub>2</sub> Flux using NOAA GFDL&rsquo;s SPEAR Ensemble

Liu, Xiao; Dunne, John P.; Jia, Liwei; Stock, Charles A.; Yang, Xiaosong; Harrison, Matthew J.; Zhang, Liping; Rosati, Anthony

doi:10.5194/egusphere-2026-4194

Preprints

https://doi.org/10.5194/egusphere-2026-4194

Preprints

22 Jul 2026

| 22 Jul 2026

Status: this preprint is open for discussion and under review for Earth System Dynamics (ESD).

Reconstruction and Multiyear Prediction of Global Air-sea CO₂ Flux using NOAA GFDL’s SPEAR Ensemble

Xiao Liu, John P. Dunne, Liwei Jia, Charles A. Stock, Xiaosong Yang, Matthew J. Harrison, Liping Zhang, and Anthony Rosati

Abstract. The ocean acts as a critical carbon sink, but its efficiency in absorbing anthropogenic CO₂ varies significantly over multiyear to decadal timescales. Accurately predicting this variability is essential for anticipating atmospheric CO₂ growth and establishing the unperturbed baselines necessary for verifying climate mitigation efforts, such as marine Carbon Dioxide Removal (mCDR) activities. This study introduces a fully coupled physical-biogeochemical prediction framework, which integrates the NOAA GFDL Seamless System for Prediction and EArth System Research (SPEAR) with the COBALTv3 ocean biogeochemical model. We conducted ensembles of uninitialized historical simulations, data-assimilative reconstructions, and retrospective initialized predictions of global air-sea CO₂ flux, evaluating their skill against observation-constrained products for a recent 40-year period (1984–2023).

The uninitialized ensemble skillfully predicts the amplitude of the observed historical increases, but struggles to resolve multiyear and decadal variability. We show that initialization significantly improves prediction skill. Globally, skill is enhanced for lead times up to two years, extending up to five years in specific higher-latitude regions. Through Average Predictability Time (APT) analysis, we isolated distinct physical drivers of the dominant predictability. We find that skillful predictions up to two years are primarily governed by the El Niño–Southern Oscillation (ENSO) and its impact on tropical upwelling. Moreover, we identified a multidecadal, potentially predictable signal linked to long-term changes in Eastern Boundary Current upwelling, as well as Southern Ocean mixed layer depth and sea surface temperature. However, verifying this long-term potential predictability remains fundamentally constrained by the sparsity of multidecadal observational records in these remote or nearshore regions. This underscores the critical need for sustained, optimized ocean observations and an improved understanding of the uncertainties associated with existing observational data.

Received: 15 Jul 2026 – Discussion started: 22 Jul 2026

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 paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 6659 KB)

Supplement (606 KB)

Download & links

Xiao Liu, John P. Dunne, Liwei Jia, Charles A. Stock, Xiaosong Yang, Matthew J. Harrison, Liping Zhang, and Anthony Rosati

Status: open (until 01 Oct 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-4194', Anonymous Referee #1, 24 Aug 2026 reply
General comments:
This manuscript presents a new prediction system based on NOAA GFDL’s SPEAR physical model coupled to COBALTv3 ocean biogeochemical model. The authors evaluate historical reconstructions and initialized retrospective predictions of air-sea CO₂ flux over 1984–2023. They report that initialization improves global prediction skill of air-sea CO₂ flux for approximately two years, primarily through ENSO-related variability, and identify a multidecadal predictable component associated with Eastern Boundary Current upwelling and Southern Ocean mixed layer depth and sea surface temperature.
The development of a coupled prediction capability enabling carbon cycle variations is valuable, and the comparison of predictive skill against observation-based reconstructions and the model’s own reconstruction provides further implications of predictive skill triggered by reference data uncertainty. This study requires a lot of work on setting up the prediction system, running ensemble of simulations, and conducting analyses. This manuscript is well designed and clearly written. A few aspects listed below need to be further clarified.
The authors claimed in the abstract and conclusion of separating natural variability from anthropogenic variability for predictions. It is unclear how the separation was done. Besides CO₂ forcing, aerosol forcing was also introduced, a simple linear trend might not lead to a clear separation of anthropogenic effects.

I wonder why the predictive skill of air-sea CO₂ fluxes presented here is in general shorter than the previous studies (Lovenduski et al. 2019; Li et al. 2019; Ilyina et al. 2021…) revealed. The longer-predictive skill (>3 years) was attributed to non-thermal processes such as the circulation and biological processes, therefore, initialization of the ocean interior is critical for maintaining higher skill. The authors explained why they use the current assimilation method practically, but do the predictions have the potential to improve by adding more ocean interior data?

The direct quantification of predictive skill is in general clear, but the Average Predictability Time (APT) analysis and how they relate to different climate mode needs further clarification. Somehow it is mixed of natural variability and anthropogenic variability. It is not straightforward to understand why the APT decomposition is based on original time series and the correlations are calculated with detrended time series. In addition, what does a correlation between a component of predictability time analysis with some physical or biogeochemical variable tell?

The predictive skills are attributed to the ENSO, EBC and SO variations through the high correlation with PrC3 and PrC2 from APT analysis. However, the explained variances of PrCs are in general very low: “PrC2 only explain 3.6% of the total variance”; “PrC3 and PrC4, together explaining 7.2% of the variance”. How to interpretate the predictions with a trivial variance explained? Are the attribution and relationship robust?

ENSO itself has predictive skill on seasonal time frame. How would the relatively short predictive skill of ENSO maintain a predictive skill of air-sea CO₂ fluxes of 2 years and longer?

Specific comments:
L62: “of how the”->”of how effective the”

L373: “4i”->”4”

L385-387: Figure 3 capture is inconsistent with the figures shown below

L395: “(e-j)” -> “(f-j)”

Figure 5: The numbers shown are inconsistent with Fig. 2 & 3, in which the predictive skill are shorter than 5 years in most areas. There are more places with 5 year predictive skill in Fig. 5 than those shown in Fig. 2f and Fig. 3f.

L545-546: “Note that this ENSO association explains more variance than PrC2, despite shorter APTs.” This sentence is unclear.

L561-567: the font here is different than the other text.

L573-575: would extended reconstructions and initialized simulations help in this regard on understanding predictability with multidecadal time scale?

Reply
Citation: https://doi.org/10.5194/egusphere-2026-4194-RC1
RC2: 'Comment on egusphere-2026-4194', Anonymous Referee #2, 21 Sep 2026 reply

This manuscript presents multiyear predictions of global air-sea CO2 fluxes by integrating the NOAA GFDL SPEAR model with the COBALT ocean biogeochemical model. The manuscript is well structured, provides a detailed methodology, and offers a balanced discussion on the limitation and future directions of carbon flux prediction using Earth system models. I only have minor comments, which are shown below:

1. I’m wondering whether the first leading predictable component accounts for the majority of the total variance, potentially affecting the extraction of 2nd and 3rd components. Would the ENSO-like patterns emerge more prominently if the analysis were conducted after removing the linear trend? Would the multi decadal pattern more closely resemble standard modes such as the PDO or AMO?

2. Are observed aerosols and CO₂ from the target years used after initialization?

3. There are some errors in writing. Examples are as follows:

- Line 275: The equation should be numbered (4).

- Line 367: Fig. 1b --> Fig. 1e

- Figure 3 caption: The current figure shows six panels, but the caption has the description of eight panels.

- Figure 4 caption: (e-j) --> (f-j).

- Figure 6c,f,i,l: The color description in the legend is opposite to the caption.

- Figure 7-8 caption: The time-series panels are (a,c,e), rather than (a,b,c).

Reply

Citation: https://doi.org/10.5194/egusphere-2026-4194-RC2

Xiao Liu, John P. Dunne, Liwei Jia, Charles A. Stock, Xiaosong Yang, Matthew J. Harrison, Liping Zhang, and Anthony Rosati

Supplement

https://doi.org/10.5194/egusphere-2026-4194-supplement

Xiao Liu, John P. Dunne, Liwei Jia, Charles A. Stock, Xiaosong Yang, Matthew J. Harrison, Liping Zhang, and Anthony Rosati

Viewed

Total article views: 243 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
168	58	17	243	29	30	26

HTML: 168
PDF: 58
XML: 17
Total: 243
Supplement: 29
BibTeX: 30
EndNote: 26

Views and downloads (calculated since 22 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	73	12	10	95
Aug 2026	42	16	2	60
Sep 2026	53	30	5	88

Cumulative views and downloads (calculated since 22 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	73	12	10	95
Aug 2026	42	16	2	60
Sep 2026	53	30	5	88

Viewed (geographical distribution)

Total article views: 221 (including HTML, PDF, and XML) Thereof 221 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 23 Sep 2026

Short summary

This study uses NOAA GFDL’s SPEAR model to improve predictions of how the ocean absorbs human emitted CO₂. While standard models capture long-term trends, they struggle with yearly fluctuations. Researchers found that "initializing" the model with observed data significantly boosts their prediction accuracy for up to two years. This framework helps separate the predictions of natural variability from human-driven carbon changes, a vital step for verifying future marine carbon removal efforts.


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

Reconstruction and Multiyear Prediction of Global Air-sea CO2 Flux using NOAA GFDL’s SPEAR Ensemble

Supplement

Viewed

Viewed (geographical distribution)

Reconstruction and Multiyear Prediction of Global Air-sea CO₂ Flux using NOAA GFDL’s SPEAR Ensemble