Thea Hatlen Heimdal
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Thea Hatlen Heimdal has a total of 65 co-authors. Here are the top co-authors by number of shared publications:
- Amanda R. Fay
Columbia University (New York, United States of America)
13 shared publications - Galen A. McKinley
Columbia University and Lamont-Doherty Earth Observatory, Palisades, NY, USA
13 shared publications - Abby P. Shaum
Columbia University
7 shared publications Viviana Acquaviva
Columbia University and Lamont-Doherty Earth Observatory Palisades NY USA
7 shared publications- Adrienne J. Sutton
National Oceanic and Atmospheric Administration, Pacific Marine Environmental Laboratory (NOAA/PMEL), 7600 Sand Point Way NE, Seattle, WA 98115, USA
5 shared publications Abby P Shaum
Columbia University and Lamont-Doherty Earth Observatory
3 shared publications
Latest Publications
- Open Access
- 23 July 2026
- Published by IOP Publishing in Machine Learning: Earth
- Vol. 2 (2) , 025005
An update of the LDEO fCO 2 -residual data product: algorithmic choices improve ocean carbon sink estimates
We evaluate the impact of various algorithmic design choices on reconstruction skill and estimated air–sea CO2 flux using the Lamont–Doherty Earth Observatory fCO2-Residual machine learning data product. We reconstruct fCO2 globally over the period 1982–2023 by optimizing the hyperparameter selection strategy to specifically reduce reconstruction bias (ResidualOPT) and/or using ΔfCO2-Residual (subtracting fCO2atm from fCO2ocn) as a target variable (ΔResidualOPT). We compare these approaches to the original fCO2-Residual methodology, which uses fCO2-Residual as a target variable and selects hyperparameters based on reducing mean-squared error (ResidualORG). We find general agreement between the three approaches for the majority of the analysis period, but there are significant differences towards the beginning and end. Compared to the original methodology (ResidualORG), ΔResidualOPT and ResidualOPT show stronger ocean carbon uptake in the 1980s, with a global mean difference over 1982–1987 of 0.6 and 0.4 PgC yr−1, respectively. In 2023, the global mean air–sea CO2 flux for both these approaches is reduced by 0.4 PgC yr−1. The test error metrics show that the highest reconstruction skill is achieved for ΔResidualOPT. We present an additional analysis of reconstruction fidelity for the three approaches by using a testbed of CMIP6 Earth system models to reconstruct fCO2. The testbed analysis is in agreement with the observation-based test error metrics; both updated approaches improve reconstruction skill compared to the original methodology (ResidualORG), but ΔResidualOPT leads to the lowest reconstruction errors for the full analysis period (1982–2023). We conclude that, for the fCO2-Residual method, the impacts of algorithmic design choices (i.e., target variable or hyperparameter selection strategy) on reconstruction skill and estimated air–sea CO2 fluxes are significant. This indicates that it is possible to improve surface ocean fCO2 reconstruction fidelity even without the availability of additional observations.
- 23 July 2026
- Published by IOP Publishing in Machine Learning: Earth
- Vol. 2 (2) , 025005
Open AccessAn update of the LDEO fCO 2 -residual data product: algorithmic choices improve ocean carbon sink estimates
We evaluate the impact of various algorithmic design choices on reconstruction skill and estimated air–sea CO2 flux using the Lamont–Doherty Earth Observatory fCO2-Residual machine learning data product. We reconstruct fCO2 globally over the period 1982–2023 by optimizing the hyperparameter selection strategy to specifically reduce reconstruction bias (ResidualOPT) and/or using ΔfCO2-Residual (subtracting fCO2atm from fCO2ocn) as a target variable (ΔResidualOPT). We compare these approaches to the original fCO2-Residual methodology, which uses fCO2-Residual as a target variable and selects hyperparameters based on reducing mean-squared error (ResidualORG). We find general agreement between the three approaches for the majority of the analysis period, but there are significant differences towards the beginning and end. Compared to the original methodology (ResidualORG), ΔResidualOPT and ResidualOPT show stronger ocean carbon uptake in the 1980s, with a global mean difference over 1982–1987 of 0.6 and 0.4 PgC yr−1, respectively. In 2023, the global mean air–sea CO2 flux for both these approaches is reduced by 0.4 PgC yr−1. The test error metrics show that the highest reconstruction skill is achieved for ΔResidualOPT. We present an additional analysis of reconstruction fidelity for the three approaches by using a testbed of CMIP6 Earth system models to reconstruct fCO2. The testbed analysis is in agreement with the observation-based test error metrics; both updated approaches improve reconstruction skill compared to the original methodology (ResidualORG), but ΔResidualOPT leads to the lowest reconstruction errors for the full analysis period (1982–2023). We conclude that, for the fCO2-Residual method, the impacts of algorithmic design choices (i.e., target variable or hyperparameter selection strategy) on reconstruction skill and estimated air–sea CO2 fluxes are significant. This indicates that it is possible to improve surface ocean fCO2 reconstruction fidelity even without the availability of additional observations.
- Open Access
- 22 June 2026
- Published by IOP Publishing in Environmental Research Letters
- Vol. 21 (13) , 134011
How well do global ocean approaches constrain local fCO 2 ?
The ocean absorbs 29\% of humanity’s annual anthropogenic carbon dioxide (CO$_2$) emissions, and the future of climate change is strongly dependent on how this sink evolves. Marine Carbon Dioxide Removal (mCDR) approaches to enhance this sink are actively being developed. In the interest of understanding how well state-of-the-art global products and models can help to distinguish mCDR signals from the background ocean carbon sink, we use sparse independent data to quantify their local-scale (1$^{o}$x1$^{o}$, monthly) errors in CO$_2$ fugacity (fCO$_2$), the primary regulator of air-sea CO$_2$ fluxes. Our 2000-2023 analysis uses 5 large regions (``superbiomes") to aggregate data across high-latitude, subtropical and equatorial zones, and we compare to 10 products and 10 models. The observed long-term trend of surface ocean CO$_2$ due to rising atmospheric CO$_2$ is consistently estimated. At the same time, we find substantial fCO$_2$ biases for the ensemble means: $\pm$5 $\mu$atm for the products, and from +1 to +15 $\mu$atm for the models. Across the superbiomes, ensemble-mean unbiased root-mean-square errors (uRMSEs) are 19-37 $\mu$atm for the products and 25-56 $\mu$atm for models. Individual products and models have larger errors. Seasonality is well-correlated to the independent data for the products and for the models in the subtropics, but the largest component of uRMSE also lies at this timescale. Deseasonalized variability has smaller uRMSEs and low correlations to independent data. In summary, local-scale errors in global observation-based products and models for the background ocean carbon sink are more than 1 order of magnitude larger than the local to regional-scale signals likely for open-ocean mCDR ($\sim$1 $\mu$atm). Reducing local-scale uncertainty in the ocean carbon sink will be critical if these products and models will be part of mCDR monitoring, reporting and verification systems. Improving the representation of seasonality should be a key target for improving both observation-based products and models.
- 22 June 2026
- Published by IOP Publishing in Environmental Research Letters
- Vol. 21 (13) , 134011
Open AccessHow well do global ocean approaches constrain local fCO 2 ?
The ocean absorbs 29\% of humanity’s annual anthropogenic carbon dioxide (CO$_2$) emissions, and the future of climate change is strongly dependent on how this sink evolves. Marine Carbon Dioxide Removal (mCDR) approaches to enhance this sink are actively being developed. In the interest of understanding how well state-of-the-art global products and models can help to distinguish mCDR signals from the background ocean carbon sink, we use sparse independent data to quantify their local-scale (1$^{o}$x1$^{o}$, monthly) errors in CO$_2$ fugacity (fCO$_2$), the primary regulator of air-sea CO$_2$ fluxes. Our 2000-2023 analysis uses 5 large regions (``superbiomes") to aggregate data across high-latitude, subtropical and equatorial zones, and we compare to 10 products and 10 models. The observed long-term trend of surface ocean CO$_2$ due to rising atmospheric CO$_2$ is consistently estimated. At the same time, we find substantial fCO$_2$ biases for the ensemble means: $\pm$5 $\mu$atm for the products, and from +1 to +15 $\mu$atm for the models. Across the superbiomes, ensemble-mean unbiased root-mean-square errors (uRMSEs) are 19-37 $\mu$atm for the products and 25-56 $\mu$atm for models. Individual products and models have larger errors. Seasonality is well-correlated to the independent data for the products and for the models in the subtropics, but the largest component of uRMSE also lies at this timescale. Deseasonalized variability has smaller uRMSEs and low correlations to independent data. In summary, local-scale errors in global observation-based products and models for the background ocean carbon sink are more than 1 order of magnitude larger than the local to regional-scale signals likely for open-ocean mCDR ($\sim$1 $\mu$atm). Reducing local-scale uncertainty in the ocean carbon sink will be critical if these products and models will be part of mCDR monitoring, reporting and verification systems. Improving the representation of seasonality should be a key target for improving both observation-based products and models.
- Open Access
- 20 May 2026
- Published by Wiley
Reconciling ocean carbon uptake estimates: a multi-model assessment of surface ocean CO2 reconstruction skill within the SOCOMv2 initiative
- 20 May 2026
- Published by Wiley
Open AccessReconciling ocean carbon uptake estimates: a multi-model assessment of surface ocean CO2 reconstruction skill within the SOCOMv2 initiative
- Open Access
- 22 March 2026
- Published by California Digital Library (CDL) in EarthArXiv
How well do global ocean approaches constrain local pCO2?
The ocean absorbs 29% of humanity’s annual anthropogenic carbon dioxide (CO2) emissions, and the future of climate change is strongly dependent on how this sink evolves. Marine Carbon Dioxide Removal (mCDR) approaches to enhance this sink are actively being developed. In the interest of understanding how well state-of-the-art global products and models can help to distinguish mCDR signals from the background ocean carbon sink, we use sparse independent data to quantify their local-scale (1ox1o, monthly) errors. Our analysis is for 2000-2023 and uses 5 large regions (“superbiomes”) to aggregate sparse data across high-latitude, subtropical and equatorial zones, and we compare to an ensemble of 10 products and 10 models. The observed long-term trend of surface ocean CO2 due to rising atmospheric CO2 concentrations is consistently estimated. At the same time, we find substantial biases for the ensemble means: ±5µatm for the products, and ranging from +1 to +15µatm for the models. Across the superbiomes, ensemble-mean unbiased root-mean-square-errors (uRMSEs) are 19-37µatm for the products and 25-56µatm for models. Seasonality is generally well-correlated to the independent data for the products and for the models in the subtropics, but the largest component of uRMSE also lies at this timescale. Deseasonalized variability has smaller uRMSEs and low correlations to independent data. In summary, local-scale, monthly errors in global observation-based products and models for the background ocean carbon sink are more than 1 order of magnitude larger than the local to regional-scale signals likely for open-ocean mCDR (∼1µatm). Reducing local-scale uncertainty in the ocean carbon sink will be critical if these products and models will be part of mCDR monitoring, reporting and verification systems. Improving the representation of seasonality should be a key target for improving both observation-based products and models.
- 22 March 2026
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessHow well do global ocean approaches constrain local pCO2?
The ocean absorbs 29% of humanity’s annual anthropogenic carbon dioxide (CO2) emissions, and the future of climate change is strongly dependent on how this sink evolves. Marine Carbon Dioxide Removal (mCDR) approaches to enhance this sink are actively being developed. In the interest of understanding how well state-of-the-art global products and models can help to distinguish mCDR signals from the background ocean carbon sink, we use sparse independent data to quantify their local-scale (1ox1o, monthly) errors. Our analysis is for 2000-2023 and uses 5 large regions (“superbiomes”) to aggregate sparse data across high-latitude, subtropical and equatorial zones, and we compare to an ensemble of 10 products and 10 models. The observed long-term trend of surface ocean CO2 due to rising atmospheric CO2 concentrations is consistently estimated. At the same time, we find substantial biases for the ensemble means: ±5µatm for the products, and ranging from +1 to +15µatm for the models. Across the superbiomes, ensemble-mean unbiased root-mean-square-errors (uRMSEs) are 19-37µatm for the products and 25-56µatm for models. Seasonality is generally well-correlated to the independent data for the products and for the models in the subtropics, but the largest component of uRMSE also lies at this timescale. Deseasonalized variability has smaller uRMSEs and low correlations to independent data. In summary, local-scale, monthly errors in global observation-based products and models for the background ocean carbon sink are more than 1 order of magnitude larger than the local to regional-scale signals likely for open-ocean mCDR (∼1µatm). Reducing local-scale uncertainty in the ocean carbon sink will be critical if these products and models will be part of mCDR monitoring, reporting and verification systems. Improving the representation of seasonality should be a key target for improving both observation-based products and models.
- Open Access
- 14 December 2025
- Published by California Digital Library (CDL) in EarthArXiv
An update of the LDEO fCO2-Residual method: algorithmic choices improve ocean carbon sink estimates
We evaluate the impact of various algorithmic design choices on reconstruction skill and estimated air-sea CO2 flux using the fCO2-Residual machine learning (ML) method (Bennington et al., 2022a) to reconstruct surface ocean fCO2. We reconstruct fCO2 globally over the period 1982-2023 by optimizing the hyperparameter selection process (ResidualOPT) and/or using ΔfCO2-Residual (subtracting fCO2atm from fCO2ocn) as a target variable (ΔResidualOPT). We compare these reconstructions to the original fCO2-Residual approach (ResidualORG) from Bennington et al. (2022a). We find general agreement between the three approaches for the majority of the analysis period, but there are significant differences towards the beginning and end. Compared to ResidualORG, ΔResidualOPT and ResidualOPT show stronger ocean carbon uptake in the 1980s, with a global mean difference over 1982-1987 of 0.6 and 0.4 PgC/yr, respectively. In 2023, the global mean air-sea CO2 flux for both approaches is reduced by 0.4 PgC/yr. The test error metrics show that the highest reconstruction skill is achieved for ΔResidualOPT. We present an additional analysis of reconstruction fidelity of the fCO2-Residual method by using a testbed of CMIP6 Earth System Models to reconstruct fCO2. The testbed analysis is in agreement with the observation-based test error metrics; both approaches improve reconstruction skill compared to ResidualORG, but ΔResidualOPT leads to the lowest reconstruction errors for the full analysis period (1982-2023). We conclude that, for the fCO2-Residual method, the impacts of algorithmic design choices (i.e., hyperparameters or target variable) on reconstruction skill and estimated air-sea CO2 fluxes are significant, and it is possible to improve reconstruction fidelity even without the availability of additional observations.
- 14 December 2025
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessAn update of the LDEO fCO2-Residual method: algorithmic choices improve ocean carbon sink estimates
We evaluate the impact of various algorithmic design choices on reconstruction skill and estimated air-sea CO2 flux using the fCO2-Residual machine learning (ML) method (Bennington et al., 2022a) to reconstruct surface ocean fCO2. We reconstruct fCO2 globally over the period 1982-2023 by optimizing the hyperparameter selection process (ResidualOPT) and/or using ΔfCO2-Residual (subtracting fCO2atm from fCO2ocn) as a target variable (ΔResidualOPT). We compare these reconstructions to the original fCO2-Residual approach (ResidualORG) from Bennington et al. (2022a). We find general agreement between the three approaches for the majority of the analysis period, but there are significant differences towards the beginning and end. Compared to ResidualORG, ΔResidualOPT and ResidualOPT show stronger ocean carbon uptake in the 1980s, with a global mean difference over 1982-1987 of 0.6 and 0.4 PgC/yr, respectively. In 2023, the global mean air-sea CO2 flux for both approaches is reduced by 0.4 PgC/yr. The test error metrics show that the highest reconstruction skill is achieved for ΔResidualOPT. We present an additional analysis of reconstruction fidelity of the fCO2-Residual method by using a testbed of CMIP6 Earth System Models to reconstruct fCO2. The testbed analysis is in agreement with the observation-based test error metrics; both approaches improve reconstruction skill compared to ResidualORG, but ΔResidualOPT leads to the lowest reconstruction errors for the full analysis period (1982-2023). We conclude that, for the fCO2-Residual method, the impacts of algorithmic design choices (i.e., hyperparameters or target variable) on reconstruction skill and estimated air-sea CO2 fluxes are significant, and it is possible to improve reconstruction fidelity even without the availability of additional observations.
- Open Access
- 12 December 2025
- Published by Wiley
Quantifying errors in ocean carbon reconstructions in the context of marine carbon dioxide removal (mCDR)
The Intergovernmental Panel on Climate suggests that carbon dioxide removal (CDR) will be needed to stabilize global temperatures at 4 μatm in the Caribbean and the Gulf, and >10 μatm within the Bering Sea. These reconstruction errors of background pCO2 values are significantly larger than far-field mCDR signals simulated by ocean models in these regions. By simulating the addition of globally distributed floats with no pCO2 bias, or an increase in global ship observations of more than an order of magnitude, the lowest reconstruction errors are achieved regionally within 30°S and 30°N in the open ocean, but errors are still larger than expected far-field mCDR signals. We conclude that further efforts to reduce errors in background surface ocean pCO2 are needed to support validation of the models with which mCDR additionality may be quantified.
- 12 December 2025
- Published by Wiley
Open AccessQuantifying errors in ocean carbon reconstructions in the context of marine carbon dioxide removal (mCDR)
The Intergovernmental Panel on Climate suggests that carbon dioxide removal (CDR) will be needed to stabilize global temperatures at 4 μatm in the Caribbean and the Gulf, and >10 μatm within the Bering Sea. These reconstruction errors of background pCO2 values are significantly larger than far-field mCDR signals simulated by ocean models in these regions. By simulating the addition of globally distributed floats with no pCO2 bias, or an increase in global ship observations of more than an order of magnitude, the lowest reconstruction errors are achieved regionally within 30°S and 30°N in the open ocean, but errors are still larger than expected far-field mCDR signals. We conclude that further efforts to reduce errors in background surface ocean pCO2 are needed to support validation of the models with which mCDR additionality may be quantified.
- Open Access
- 10 October 2025
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 52 (19)
Sensitivity of Ocean Carbon Sink Estimates to Rare Observations
We evaluate the impact of new data in the annual release of SOCATv2024 on air‐sea carbon dioxide (CO 2 ) flux reconstructions. Using the p CO 2 ‐Residual and LDEO‐HPD methods, we reconstruct surface ocean f CO 2 based on SOCATv2023 and SOCATv2024. For both products we find strong agreement in global mean air‐sea CO 2 fluxes for 1990–2017, but a notable divergence beginning in 2018. Reconstructions based on SOCATv2024 estimate substantially less ocean carbon uptake compared to those based on SOCATv2023, with a global mean difference of 0.25–0.27 PgCyr −1 over 2018–2022. We find that flux differences are due primarily to large changes in the Southern Ocean where data from 63 cruises was added for SOCATv2024. Analysis with a model‐testbed indicates that these data, if unbiased, should substantially increase reconstruction fidelity. We conclude that there is high sensitivity of air‐sea CO 2 flux reconstructions to the addition of data in poorly‐sampled regions.
- 10 October 2025
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 52 (19)
Open AccessSensitivity of Ocean Carbon Sink Estimates to Rare Observations
We evaluate the impact of new data in the annual release of SOCATv2024 on air‐sea carbon dioxide (CO 2 ) flux reconstructions. Using the p CO 2 ‐Residual and LDEO‐HPD methods, we reconstruct surface ocean f CO 2 based on SOCATv2023 and SOCATv2024. For both products we find strong agreement in global mean air‐sea CO 2 fluxes for 1990–2017, but a notable divergence beginning in 2018. Reconstructions based on SOCATv2024 estimate substantially less ocean carbon uptake compared to those based on SOCATv2023, with a global mean difference of 0.25–0.27 PgCyr −1 over 2018–2022. We find that flux differences are due primarily to large changes in the Southern Ocean where data from 63 cruises was added for SOCATv2024. Analysis with a model‐testbed indicates that these data, if unbiased, should substantially increase reconstruction fidelity. We conclude that there is high sensitivity of air‐sea CO 2 flux reconstructions to the addition of data in poorly‐sampled regions.
- Open Access
- 4 August 2025
- Published by Wiley
Quantifying errors in ocean carbon reconstructions in the context of marine carbon dioxide removal (mCDR)
The Intergovernmental Panel on Climate Change (IPCC) suggests that carbon dioxide removal (CDR) will be needed to stabilize global atmospheric temperatures at 4 μatm in the Caribbean Sea and the Gulf and >10 μatm within the Bering Sea, which is larger than expected mCDR signals (<2 μatm). By simulating the addition of globally distributed floats with no pCO2 bias, increasing the 2020-2021 coverage from 0.5% to 0.7%, reconstruction errors ≤2 μatm are achieved regionally within 30°S and 30°N in the open ocean, but not in the three study areas. We conclude that current estimates of the background ocean carbon sink must be significantly improved before mCDR additionality can be accurately quantified.
- 4 August 2025
- Published by Wiley
Open AccessQuantifying errors in ocean carbon reconstructions in the context of marine carbon dioxide removal (mCDR)
The Intergovernmental Panel on Climate Change (IPCC) suggests that carbon dioxide removal (CDR) will be needed to stabilize global atmospheric temperatures at 4 μatm in the Caribbean Sea and the Gulf and >10 μatm within the Bering Sea, which is larger than expected mCDR signals (<2 μatm). By simulating the addition of globally distributed floats with no pCO2 bias, increasing the 2020-2021 coverage from 0.5% to 0.7%, reconstruction errors ≤2 μatm are achieved regionally within 30°S and 30°N in the open ocean, but not in the three study areas. We conclude that current estimates of the background ocean carbon sink must be significantly improved before mCDR additionality can be accurately quantified.
- Open Access
- 30 July 2025
- Published by IOP Publishing in Machine Learning: Earth
- Vol. 1 (1) , 015003
Targeting bias in algorithm optimization improves reconstructions of surface ocean pCO2
- 30 July 2025
- Published by IOP Publishing in Machine Learning: Earth
- Vol. 1 (1) , 015003
Open AccessTargeting bias in algorithm optimization improves reconstructions of surface ocean pCO2
- Open Access
- 21 July 2025
- Published by Wiley
Sensitivity of ocean carbon sink estimates to rare observations
We evaluate the impact of new data in the annual release of SOCATv2024 on air-sea CO2 flux reconstructions. Using the pCO2-Residual and LDEO-HPD methods, we reconstruct surface ocean fCO2 based on SOCATv2023 and SOCATv2024. For both products we find strong agreement in global mean air-sea CO2 fluxes for 1990--2017, but a notable divergence beginning in 2018. Reconstructions based on SOCATv2024 estimate substantially less ocean carbon uptake compared to those based on SOCATv2023, with a global mean difference of 0.25--0.27 PgCyr-1 over 2018--2022. We find that flux differences are due primarily to large changes in the Southern Ocean where data from 63 cruises were added for SOCATv2024. Analysis with a model-testbed indicates that these data, if unbiased, should substantially increase reconstruction fidelity. We conclude that there is high sensitivity of air-sea CO2 flux reconstructions to the addition of data in poorly-sampled regions.
- 21 July 2025
- Published by Wiley
Open AccessSensitivity of ocean carbon sink estimates to rare observations
We evaluate the impact of new data in the annual release of SOCATv2024 on air-sea CO2 flux reconstructions. Using the pCO2-Residual and LDEO-HPD methods, we reconstruct surface ocean fCO2 based on SOCATv2023 and SOCATv2024. For both products we find strong agreement in global mean air-sea CO2 fluxes for 1990--2017, but a notable divergence beginning in 2018. Reconstructions based on SOCATv2024 estimate substantially less ocean carbon uptake compared to those based on SOCATv2023, with a global mean difference of 0.25--0.27 PgCyr-1 over 2018--2022. We find that flux differences are due primarily to large changes in the Southern Ocean where data from 63 cruises were added for SOCATv2024. Analysis with a model-testbed indicates that these data, if unbiased, should substantially increase reconstruction fidelity. We conclude that there is high sensitivity of air-sea CO2 flux reconstructions to the addition of data in poorly-sampled regions.