Explore the vast range of titles available.

Ocean acidification has profoundly altered the ocean's carbonate chemistry since preindustrial times, with potentially serious consequences for marine life. Yet, no long-term, global observation-based data set exists that allows us to study changes in ocean acidification for all carbonate system parameters over the last few decades. Here, we fill this gap and present a methodologically consistent global data set of all relevant surface ocean parameters, i.e., dissolved inorganic carbon (DIC), total alkalinity (TA), partial pressure of CO2 (pCO2), pH, and the saturation state with respect to mineral CaCO3 (Ω) at a monthly resolution over the period 1985 through 2018 at a spatial resolution of 1∘×1∘. This data set, named OceanSODA-ETHZ, was created by extrapolating in time and space the surface ocean observations of pCO2 (from the Surface Ocean CO2 Atlas, SOCAT) and total alkalinity (TA; from the Global Ocean Data Analysis Project, GLODAP) using the newly developed Geospatial Random Cluster Ensemble Regression (GRaCER) method (code available at https://doi.org/10.5281/zenodo.4455354, Gregor, 2021). This method is based on a two-step (cluster-regression) approach but extends it by considering an ensemble of such cluster regressions, leading to improved robustness. Surface ocean DIC, pH, and Ω were then computed from the globally mapped pCO2 and TA using the thermodynamic equations of the carbonate system. For the open ocean, the cluster-regression method estimates pCO2 and TA with global near-zero biases and root mean squared errors of 12 µatm and 13 µmol kg−1, respectively. Taking into account also the measurement and representation errors, the total uncertainty increases to 14 µatm and 21 µmol kg−1, respectively. We assess the fidelity of the computed parameters by comparing them to direct observations from GLODAP, finding surface ocean pH and DIC global biases of near zero, as well as root mean squared errors of 0.023 and 16 µmol kg−1, respectively. These uncertainties are very comparable to those expected by propagating the total uncertainty from pCO2 and TA through the thermodynamic computations, indicating a robust and conservative assessment of the uncertainties. We illustrate the potential of this new data set by analyzing the climatological mean seasonal cycles of the different parameters of the surface ocean carbonate system, highlighting their commonalities and differences. Further, this data set provides a novel constraint on the global- and basin-scale trends in ocean acidification for all parameters. Concretely, we find for the period 1990 through 2018 global mean trends of 8.6 ± 0.1 µmol kg−1 per decade for DIC, −0.016 ± 0.000 per decade for pH, 16.5 ± 0.1 µatm per decade for pCO2, and −0.07 ± 0.00 per decade for Ω. The OceanSODA-ETHZ data can be downloaded from https://doi.org/10.25921/m5wx-ja34 (Gregor and Gruber, 2020).

Over the last decade, advanced statistical inference and machine learning have been used to fill the gaps in sparse surface oceanCO2 measurements (Rödenbeck et al., 2015). The estimates from these methods have been used to constrain seasonal, interannual and decadal variability in sea–air CO2 fluxes and the drivers of these changes (Landschützer et al., 2015, 2016; Gregor et al., 2018). However, it is also becoming clear that these methods are converging towards a common bias and root mean square error (RMSE) boundary: “the wall”, which suggests that pCO2 estimates are now limited by both data gaps and scale-sensitive observations. Here, we analyse this problem by introducing a new gap-filling method, an ensemble average of six machine-learning models (CSIR-ML6 version 2019a, Council for Scientific and Industrial Research – Machine Learning ensemble with Six members), where each model is constructed with a two-step clustering-regression approach. The ensemble average is then statistically compared to well-established methods. The ensemble average, CSIR-ML6, has an RMSE of 17.16 µatm and bias of 0.89 µatm when compared to a test dataset kept separate from training procedures. However, when validating our estimates with independent datasets, we find that our method improves only incrementally on other gap-filling methods. We investigate the differences between the methods to understand the extent of the limitations of gap-filling estimates ofpCO2. We show that disagreement between methods in the South Atlantic, southeastern Pacific and parts of the Southern Ocean is too large to interpret the interannual variability with confidence. We conclude that improvements in surface ocean pCO2 estimates will likely be incremental with the optimisation of gap-filling methods by (1) the inclusion of additional clustering and regression variables (e.g. eddy kinetic energy), (2) increasing the sampling resolution and (3) successfully incorporatingpCO2 estimates from alternate platforms (e.g. floats, gliders) into existing machine-learning approaches.

The Southern Ocean is a complex system yet is sparsely sampled in both space and time. These factors raise questions about the confidence in present sampling strategies and associated machine learning (ML) reconstructions. Previous studies have not yielded a clear understanding of the origin of uncertainties and biases for the reconstructions of the partial pressure of carbon dioxide (pCO2) at the surface ocean (pCO2ocean). We examine these questions through a series of semi-idealized observing system simulation experiments (OSSEs) using a high-resolution (± 10 km) coupled physical and biogeochemical model (NEMO-PISCES, Nucleus for European Modelling of the Ocean, Pelagic Interactions Scheme for Carbon and Ecosystem Studies). Here we choose 1 year of the model sub-domain of 10∘ of latitude (40–50∘ S) by 20∘ of longitude (10∘ W–10∘ E). This domain is crossed by the sub-Antarctic front and thus includes both the sub-Antarctic zone and the polar frontal zone in the south-east Atlantic Ocean, which are the two most sampled sub-regions of the Southern Ocean. We show that while this sub-domain is small relative to the Southern Ocean scales, it is representative of the scales of variability we aim to examine. The OSSEs simulated the observational scales ofpCO2ocean in ways that are comparable to existing ocean CO2 observing platforms (ships, Wave Gliders, carbon floats, Saildrones) in terms of their temporal sampling scales and not necessarily their spatial ones. The pCO2 reconstructions were carried out using a two-member ensemble approach that consisted of two machine learning (ML) methods, (1) the feed-forward neural network and (2) the gradient boosting machines. The baseline data were from the ship-based simulations mimicking ship-based observations from the Surface Ocean CO2 Atlas (SOCAT). For each of the sampling-scale scenarios, we applied the two-member ensemble method to reconstruct the full sub-domainpCO2ocean. The reconstruction skill was then assessed through a statistical comparison of reconstructedpCO2ocean and the model domain mean. The analysis shows that uncertainties and biases forpCO2ocean reconstructions are very sensitive to both the spatial and the temporal scales of pCO2 sampling in the model domain. The four key findings from our investigation are as follows: (1) improving ML-based pCO2 reconstructions in the Southern Ocean requires simultaneous high-resolution observations (<3 d) of the seasonal cycle of the meridional gradients ofpCO2ocean; (2) Saildrones stand out as the optimal platforms to simultaneously address these requirements; (3) Wave Gliders with hourly/daily resolution in pseudo-mooring mode improve on carbon floats (10 d period), which suggests that sampling aliases from the 10 d sampling period might have a greater negative impact on their uncertainties, biases, and reconstruction means; and (4) the present seasonal sampling biases (towards summer) in SOCAT data in the Southern Ocean may be behind a significant winter bias in the reconstructed seasonal cycle of pCO2ocean.

Air–sea flux of carbon dioxide (CO2) is a critical component of the global carbon cycle and the climate system with the ocean removing about a quarter of the CO2 emitted into the atmosphere by human activities over the last decade. A common approach to estimate this net flux of CO2 across the air–sea interface is the use of surface ocean CO2 observations and the computation of the flux through a bulk parameterization approach. Yet, the details for how this is done in order to arrive at a global ocean CO2 uptake estimate vary greatly, enhancing the spread of estimates. Here we introduce the ensemble data product, SeaFlux (Gregor and Fay, 2021, 10.5281/zenodo.5482547​​​​​​​,https://github.com/luke-gregor/pySeaFlux, last access: 9 September 2021​​​​​​​); this resource enables users to harmonize an ensemble of products that interpolate surface ocean CO2 observations to near-global coverage with a common methodology to fill in missing areas in the products. Further, the dataset provides the inputs to calculate fluxes in a consistent manner. Utilizing six global observation-based mapping products (CMEMS-FFNN, CSIR-ML6, JENA-MLS, JMA-MLR, MPI-SOMFFN, NIES-FNN), the SeaFlux ensemble approach adjusts for methodological inconsistencies in flux calculations. We address differences in spatial coverage of the surface ocean CO2 between the mapping products, which ultimately yields an increase in CO2 uptake of up to 17 % for some products. Fluxes are calculated using three wind products (CCMPv2, ERA5, and JRA55). Application of a scaled gas exchange coefficient has a greater impact on the resulting flux than solely the choice of wind product. With these adjustments, we present an ensemble of global surface ocean pCO2 and air–sea carbon flux estimates. This work aims to support the community effort to perform model–data intercomparisons which will help to identify missing fluxes as we strive to close the global carbon budget.

Resolving and understanding the drivers of variability of CO2 in the Southern Ocean and its potential climate feedback is one of the major scientific challenges of the ocean-climate community. Here we use a regional approach on empirical estimates of pCO2 to understand the role that seasonal variability has in long-term CO2 changes in the Southern Ocean. Machine learning has become the preferred empirical modelling tool to interpolate time- and location-restricted ship measurements of pCO2. In this study we use an ensemble of three machine-learning products: support vector regression (SVR) and random forest regression (RFR) from Gregor et al. (2017), and the self-organising-map feed-forward neural network (SOM-FFN) method from Landschützer et al. (2016). The interpolated estimates ofΔpCO2 are separated into nine regions in the Southern Ocean defined by basin (Indian, Pacific, and Atlantic) and biomes (as defined by Fay and McKinley, 2014a). The regional approach shows that, while there is good agreement in the overall trend of the products, there are periods and regions where the confidence in estimated ΔpCO2 is low due to disagreement between the products. The regional breakdown of the data highlighted the seasonal decoupling of the modes for summer and winter interannual variability. Winter interannual variability had a longer mode of variability compared to summer, which varied on a 4–6-year timescale. We separate the analysis of the ΔpCO2 and its drivers into summer and winter. We find that understanding the variability of ΔpCO2 and its drivers on shorter timescales is critical to resolving the long-term variability of ΔpCO2. Results show that ΔpCO2 is rarely driven by thermodynamics during winter, but rather by mixing and stratification due to the stronger correlation of ΔpCO2 variability with mixed layer depth. Summer pCO2 variability is consistent with chlorophyll a variability, where higher concentrations of chlorophyll a correspond with lower pCO2 concentrations. In regions of low chlorophyll a concentrations, wind stress and sea surface temperature emerged as stronger drivers of ΔpCO2. In summary we propose that sub-decadal variability is explained by summer drivers, while winter variability contributes to the long-term changes associated with the SAM. This approach is a useful framework to assess the drivers of ΔpCO2 but would greatly benefit from improved estimates of ΔpCO2 and a longer time series.

The Southern Ocean accounts for 40 % of oceanic CO2 uptake, but the estimates are bound by large uncertainties due to a paucity in observations. Gap-filling empirical methods have been used to good effect to approximatepCO2 from satellite observable variables in other parts of the ocean, but many of these methods are not in agreement in the Southern Ocean. In this study we propose two additional methods that perform well in the Southern Ocean: support vector regression (SVR) and random forest regression (RFR). The methods are used to estimate ΔpCO2 in the Southern Ocean based on SOCAT v3, achieving similar trends to the SOM-FFN method by. Results show that the SOM-FFN and RFR approaches have RMSEs of similar magnitude (14.84 and 16.45 µatm, where 1 atm = 101 325 Pa) where the SVR method has a larger RMSE (24.40 µatm). However, the larger errors for SVR and RFR are, in part, due to an increase in coastal observations from SOCAT v2 to v3, where the SOM-FFN method used v2 data. The success of both SOM-FFN and RFR depends on the ability to adapt to different modes of variability. The SOM-FFN achieves this by having independent regression models for each cluster, while this flexibility is intrinsic to the RFR method. Analyses of the estimates shows that the SVR and RFR's respective sensitivity and robustness to outliers define the outcome significantly. Further analyses on the methods were performed by using a synthetic dataset to assess the following: which method (RFR or SVR) has the best performance? What is the effect of using time, latitude and longitude as proxy variables on ΔpCO2? What is the impact of the sampling bias in the SOCAT v3 dataset on the estimates? We find that while RFR is indeed better than SVR, the ensemble of the two methods outperforms either one, due to complementary strengths and weaknesses of the methods. Results also show that for the RFR and SVR implementations, it is better to include coordinates as proxy variables as RMSE scores are lowered and the phasing of the seasonal cycle is more accurate. Lastly, we show that there is only a weak bias due to undersampling. The synthetic data provide a useful framework to test methods in regions of sparse data coverage and show potential as a useful tool to evaluate methods in future studies.

Resolving and understanding the drivers of variability of CO2 in the Southern Ocean and its potential climate feedback is one of the major scientific challenges of the ocean-climate community. Here we use a regional approach on empirical estimates of pCO2 to understand the role that seasonal variability has in long-term CO2 changes in the Southern Ocean. Machine learning has become the preferred empirical modelling tool to interpolate time- and location-restricted ship measurements of pCO2. In this study we use an ensemble of three machine-learning products: support vector regression (SVR) and random forest regression (RFR) from Gregor et al. (2017), and the self-organising-map feed-forward neural network (SOM-FFN) method from Landschützer et al. (2016). The interpolated estimates of ΔpCO2 are separated into nine regions in the Southern Ocean defined by basin (Indian, Pacific, and Atlantic) and biomes (as defined by Fay and McKinley, 2014a). The regional approach shows that, while there is good agreement in the overall trend of the products, there are periods and regions where the confidence in estimated ΔpCO2 is low due to disagreement between the products. The regional breakdown of the data highlighted the seasonal decoupling of the modes for summer and winter interannual variability. Winter interannual variability had a longer mode of variability compared to summer, which varied on a 4–6-year timescale. We separate the analysis of the ΔpCO2 and its drivers into summer and winter. We find that understanding the variability of ΔpCO2 and its drivers on shorter timescales is critical to resolving the long-term variability of ΔpCO2. Results show that ΔpCO2 is rarely driven by thermodynamics during winter, but rather by mixing and stratification due to the stronger correlation of ΔpCO2 variability with mixed layer depth. Summer pCO2 variability is consistent with chlorophyll a variability, where higher concentrations of chlorophyll a correspond with lower pCO2 concentrations. In regions of low chlorophyll a concentrations, wind stress and sea surface temperature emerged as stronger drivers of ΔpCO2. In summary we propose that sub-decadal variability is explained by summer drivers, while winter variability contributes to the long-term changes associated with the SAM. This approach is a useful framework to assess the drivers of ΔpCO2 but would greatly benefit from improved estimates of ΔpCO2 and a longer time series.

The Southern Ocean accounts for 40 % of oceanic CO2 uptake, but the estimates are bound by large uncertainties due to a paucity in observations. Gap-filling empirical methods have been used to good effect to approximate pCO2 from satellite observable variables in other parts of the ocean, but many of these methods are not in agreement in the Southern Ocean. In this study we propose two additional methods that perform well in the Southern Ocean: support vector regression (SVR) and random forest regression (RFR). The methods are used to estimate ΔpCO2 in the Southern Ocean based on SOCAT v3, achieving similar trends to the SOM-FFN method by Landschützer et al. (2014). Results show that the SOM-FFN and RFR approaches have RMSEs of similar magnitude (14.84 and 16.45 µatm, where 1 atm  =  101 325 Pa) where the SVR method has a larger RMSE (24.40 µatm). However, the larger errors for SVR and RFR are, in part, due to an increase in coastal observations from SOCAT v2 to v3, where the SOM-FFN method used v2 data. The success of both SOM-FFN and RFR depends on the ability to adapt to different modes of variability. The SOM-FFN achieves this by having independent regression models for each cluster, while this flexibility is intrinsic to the RFR method. Analyses of the estimates shows that the SVR and RFR's respective sensitivity and robustness to outliers define the outcome significantly. Further analyses on the methods were performed by using a synthetic dataset to assess the following: which method (RFR or SVR) has the best performance? What is the effect of using time, latitude and longitude as proxy variables on ΔpCO2? What is the impact of the sampling bias in the SOCAT v3 dataset on the estimates? We find that while RFR is indeed better than SVR, the ensemble of the two methods outperforms either one, due to complementary strengths and weaknesses of the methods. Results also show that for the RFR and SVR implementations, it is better to include coordinates as proxy variables as RMSE scores are lowered and the phasing of the seasonal cycle is more accurate. Lastly, we show that there is only a weak bias due to undersampling. The synthetic data provide a useful framework to test methods in regions of sparse data coverage and show potential as a useful tool to evaluate methods in future studies.

The Southern Ocean is a major region of ocean carbon uptake, but its future changes remain uncertain under climate change. Here we show the projected shift in the Southern Ocean CO 2 sink using a suite of Earth System Models, revealing changes in the mechanism, position and seasonality of the carbon uptake. The region of dominant CO 2 uptake shifts from the Subtropical to the Antarctic region under the high-emission scenario. The warming-driven sea-ice melt, increased ocean stratification, mixed layer shoaling, and a weaker vertical carbon gradient is projected to together reduce the winter de-gassing in the future, which will trigger the switch from mixing-driven outgassing to solubility-driven uptake in the Antarctic region during the winter season. The future Southern Ocean carbon sink will be poleward-shifted, operating in a hybrid mode between biologically-driven summertime and solubility-driven wintertime uptake with further amplification of biologically-driven uptake due to the increasing Revelle Factor.