We use GLODAPv2.2002 potential temperature, DIC, and oxygen bottle measurements. The GLODAP data product has undergone extensive quality control, with adjustments applied when required, in order to remove any biases between cruises . Using DIVA (Data-Interpolating Variational Analysis) gridding techniques , we interpolate the GLODAP bottle data with the same method as onto the GO-SHIP Easy Ocean A05 section. We then map the A05 section onto a regular grid with 651 vertical levels and 670 longitude points, giving a resolution of 10 m depth by 0.1° longitude. The A05 hydrographic section is located at approximately 24.5° N, spanning 80–13° W across the subtropical North Atlantic (Fig. a). The section was sampled in six GO-SHIP cruises, and the collection and analyses of these samples followed standard operating procedures as outlined in the GO-SHIP manuals . The six cruises were conducted between 1992 and 2015 (Table ), with the average profiles and sections of temperature, DIC and oxygen shown in Fig. b–g. We quantify changes in oxygen concentration, potential temperature, and DIC by computing the difference between measurements obtained during the initial cruise in 1992 and those obtained in each subsequent cruise (e.g., 1998 minus 1992, 2004 minus 1992, and so forth).

We define the total change in oxygen as the sum of four distinct changes (Fig. ): (1) change in the excess oxygen (ΔO_e), (2) redistribution of the background field (ΔO_rd), (3) change in the amount of remineralisation (ΔO_rem), and (4) change due to variability in sea surface disequilibrium (ΔO_d). Excess changes are those due to changes in sea surface temperature which alter solubility – this component is the warming-driven deoxygenation. The vertical redistribution of the oxygen already in the water column is a result of local circulation changes, or reorganization of the local oxygen structure, and so will sum to zero globally. It is a result of the local movement or change in the distribution of water masses, and occurs on all time scales associated with local variability in circulation, which is well measured at the A05 section . Our assumption, following , is that these redistributed changes are linked by local stratification in theta-oxygen-DIC space. While we determine the interannual changes in redistribution well, we do not resolve them well on shorter timescales, as they vary on timescales much faster than the nominal GO-SHIP repeat of ∼5 years at this latitude. However, we are interested in the longer-term changes in water mass properties, which are resolved by the ∼5 year frequency.

Changes in remineralisation can be either due to a change in the rate of export of organic matter leading to an altered rate of remineralisation at depth, or to changes in the spatial patterns of remineralisation, i.e. because changes in large scale circulation patterns affect the residence time of water and hence the accumulation of the remineralised oxygen deficit. A change in the remineralisation term at 24° N reflects a change in the accumulated remineralisation signal of that water mass since subduction. The final term, disequilibrium change is due to the change in saturation levels when the water mass was initially subducted, which changes the potential amount of oxygen that can be absorbed by the surface ocean.

These mechanisms are shown schematically in Fig. . By linking each of these mechanisms to simultaneous changes in temperature and DIC, we develop a set of equations that can be solved to calculate the magnitude of each process. The total temperature change is the sum of only excess and redistributed temperature changes, while total DIC change is defined as a sum of the excess, redistributed and remineralisation changes of DIC. Excess DIC change is dominated by the anthropogenic increase , and has an associated temperature change. Isolating this excess change means that we can also calculate the remaining redistribution component of temperature change. Since the redistribution of all three properties is driven by local changes in stratification (or distribution of water masses), we can use the redistribution of temperature to calculate the redistribution of DIC and oxygen. The remaining DIC change, due to the production of DIC during remineralisation, is connected to the associated loss of oxygen. The disequilibrium oxygen term remains weakly constrained as it is not directly linked to another process. A disequilibrium term is not included for DIC because, while disequilibrium does affect surface CO₂ concentrations, this term is instead incorporated in the coefficient relating any excess DIC and temperature changes, described in Sect. .

The total change in potential temperature (ΔT), DIC (ΔC), and oxygen (ΔO) is thus described by the following equations: 1ΔT=ΔTe+ΔTrd,2ΔC=ΔCe+ΔCrd+ΔCrem,3ΔO=ΔOe+ΔOrd+ΔOrem+ΔOd, where the subscript “e”, “rd”, “rem”, and “d” relate to excess, redistribution, remineralisation, and disequilibrium change, respectively. Each of these contributions are currently unknown. However, by defining coefficients that link the individual components of temperature, DIC, and oxygen change, we can form a system of equations that can be solved to obtain the magnitude of each process driving oxygen change. The calculation of these coefficients is described below, along with the final system of equations, while the full derivation is shown in the Appendix. We only look at the upper 2000 m of the water column because the changes in the deeper ocean over the period that the A05 section has been occupied are smaller than the accuracy of the GLODAP data, particularly for DIC.

Excess changes in oxygen and DIC are both linked to excess changes in temperature, via different mechanisms. Changes in surface heat fluxes lead to changes in solubility, and therefore oxygen concentration, while an almost linear relationship exists between ocean heat and carbon content, both on local and global scales , due to the oceanic uptake of heat and carbon as a response to anthropogenic CO₂ emissions.

Excess oxygen changes are due to temperature-driven change in solubility, a relationship that is close to linear over the observed temperature ranges on the A05 section (Fig. a). This relationship is described by the coefficient η, the gradient of the temperature-oxygen solubility curve, assuming 100 % saturation and using the temperature during the 1992 cruise as the initial temperature of the curve. Within the A05 section, the coefficient has a range of -0.33 to -0.12 °C (µmol/kg)⁻¹, and increases with depth. The effect of oxygen under-/over-saturation is not incorporated in this coefficient since the saturation level is not known. Instead, adjustments for saturation effects are addressed through the inclusion of disequilibrium change as a distinct oxygen change term. The relationship between excess oxygen and temperature change is described by Eq. (): 4ΔTe=ηΔOe.

To describe the relationship between excess temperature and excess DIC, we use a temperature-dependent coefficient, α, . Equation () defines the relationship between the excess temperature and excess DIC terms. This coefficient was approximated as uniform via Eq. (), following . The coefficient α varies between 0.010 and 0.024 °C (µmol/kg)⁻¹ within the A05 section in 1992, and increases with depth (Fig. b), similar to η. We use the initial (1992) values for α, however temporal variability in the coefficient is minimal over these timescales, with a mean difference on the order of 10⁻⁶ µmol kg⁻¹. Since α describes the observed relationship between the excess component of DIC and temperature, it also takes into account any impact that the change in CO₂ saturation has on the surface flux. Separating this effect would be non-trivial and add extra unknowns into the equation, so it is simply included as part of the excess term and no DIC disequilibrium term is introduced: 5α=0.025-0.0005T,6ΔTe=αΔCe.

The redistribution coefficients βC and βO relate vertical redistribution driven changes in temperature to redistribution driven changes in DIC and oxygen, at a given geographical location. Each coefficient is approximated by the ratio of stratification between temperature and DIC/oxygen, following : 7βC=∂T∂z∂C∂z-1,8βO=∂T∂z∂O∂z-1.

Within the A05 section, βC has a mean value of -0.043 °C (µmol/kg)⁻¹ (Fig. c), and is primarily negative in the upper 1000 m as temperature decreases with depth while DIC increases. The coefficient then transitions to positive at ∼1000 m as DIC concentration reaches a maximum and begins to decrease slightly. βO has a mean of -0.048 °C (µmol/kg)⁻¹ (Fig. d), and is negative in the upper 100 m as temperature and oxygen are relatively constant. The coefficient then shifts to positive as both the oxygen concentration and temperature decreases, until a depth of around 800 m, when oxygen reaches a minimum and begins to decrease again, driving a negative βO. At this depth, as the coefficient changes sign, redistribution is effectively zero. The relationship between redistributed changes in temperature, DIC, and oxygen can then be expressed by the following equations: 9ΔTrd=βCΔCrd10ΔTrd=βOΔOrd.

Once the redistributed oxygen term is calculated via Eq. (), it can be used to compute the temperature and DIC redistribution via Eqs. () and ().

We define the relationship between changes in remineralisation in oxygen and those in DIC using the Redfield ratio of 106C : -138O , generally considered to be the approximate average ratio of change during remineralisation. The change in remineralised oxygen is therefore described in Eq. (), where γ=-106138≈-0.77: 11ΔCrem=γΔOrem. The ratio γ between oxygen and DIC is tightly constrained based on interior tracer fields, and while there will be some spatial changes in this ratio, within the Atlantic, the remineralisation ratio between oxygen and carbon can be considered constant with depth . While variation in the Redfield ratio was tested, changing the value by as much as 50 % has little impact on the overall findings (Fig. ).

Once the coefficients are calculated for each individual gridpoint along the A05 section, they can be input into Eqs. (4), (6) and (9)–(11), and represent a system of simultaneous equations in the form W1Ax=W2b, which is solved via a weighted least squares fit approach to obtain the individual drivers of oxygen change at each point. The magnitude of the oxygen change terms can then be added back into the equations relating oxygen change to temperature and DIC changes to compute the magnitude of each driver of temperature and DIC change at each point along the A05 section 12W1ηβo00ηαβoβcγ01111ΔOeΔOrdΔOremΔOd=W2ΔTΔCΔO. As Eq. () is under-determined, an infinite number of solutions exist, and so we regularise the system with an initial estimate of the solution and its estimated variance . These weightings are listed in Table . The b vector consisting of total changes in temperature, DIC and oxygen, is weighted by the inconsistency in each variable between GLODAP cruises ; this weighting determines the possible magnitude of the residual (i.e. the mismatch between the total oxygen change and the sum of the driving terms). Temperature and oxygen weightings are derived from the inconsistency between cruises within the Atlantic, while the global inconsistency in the GLODAP data product is used for DIC, as the Atlantic inconsistency is too high due to the influence of coastal ocean data.

The initial estimate and weighting for the x vector, which consists of the changes in each individual oxygen process, are assigned as the estimated mean, and variance of each component, respectively, when estimates of these values exist in the literature. In instances where no previous information is available, we assign an initial estimate of zero and estimate the spatial variability from previous literature. A weighting and initial estimate for excess oxygen change is approximated from excess temperature change along the A05 section via Eq. (). Weightings for remineralisation and disequilibrium changes are estimated from spatial variability available in the literature . We assume variability in the redistributed term to be similar to that of the total oxygen change, and thus take the variance in the total change as the redistributed weighting.

The observations show an overall decline in total oxygen concentration in the upper 1000 m of the A05 section, which is strongest at the surface (Fig. a–e), with the trend becoming more pronounced from 2010 onwards. The exception to this is in the eastern side of the section, which shows an oxygen increase from around 200–1000 m eastward of 30° W, likely related to the influence of Mediterranean water and denitrification. Below 1000 m, within the uNADW layer, there is no obvious trend, with areas of both positive and negative oxygen change. At depths of 1000–2000 m, an increase in oxygen between 40 and 65° W is observed in most years, except 1992 to 2010, when this region instead experiences a strong decrease.

To better understand the drivers of the average property changes across the A05 section, we focus on the region between 30–70° W (black box in Fig. a). This allows us to focus on the central region that has more robust trends in the upper 1000 m, with clear deoxygenation over time, while removing the impact of regional oxygen increases occurring in the eastern and western boundaries of the section. The average over the central region emphasises the negative trend in total oxygen change in the upper 1000 m of the water column (Fig. a), with deoxygenation reaching a maximum rate of -13.1 µmol kg⁻¹ between 1992 and 2010 at ∼600 m depth. As the average oxygen concentration decreases, the observations simultaneously show a strong DIC increase in the upper ocean (Fig. g), reaching a maximum change of 32.4 µmol kg⁻¹ from 1992 to 2015 at ∼300 m. However, there is no consistently increasing trend in the average temperature of the section over the same period (Fig. l).

There is variation in the driving mechanisms of oxygen, DIC and temperature change at different depths. Averaging over depth ranges also shows high standard deviation at depths below 1000 m (Table ). In the depth range 150–500 m total deoxygenation is dominated by change in remineralisation, while the DIC increase is dominated by excess change, and a slight warming is driven by excess change. The dominance of one process is less evident in the depth range 500–1000 m. Remineralised oxygen change still makes up the majority of the deoxygenation, while for DIC, the remineralisation term becomes more significant, with an average of 3.42 µmol kg⁻¹ compared to a total DIC change of 8.20 µmol kg⁻¹. Within this depth range, the average temperature change is less than the standard deviation, with both excess change and redistribution driving a similar size increase.

Spatial patterns in the total oxygen change (Fig. a–e) resemble those of the remineralisation change (Fig. p–t), with consumption of oxygen via remineralisation generally increasing over time in the upper 1000 m, particularly within the NACW layers. The majority of remineralisation-related oxygen change is negative in the central part of the section (approximately 30–70° W), with only a slightly positive average change below 1000 m, within the uNADW, which is strongest between 1992 and 2004. The increasing total oxygen seen in the east of the section is also due to remineralisation changes. When averaged over the entire central region of the A05 section, remineralisation change is strongly correlated with total oxygen change (r2=0.93, p=0.008), with the maximum remineralisation decrease also occurring at a depth of ∼600 m, and reaching a maximum of -7.4 µmol kg⁻¹ by 2015 (Fig. d).

Excess oxygen changes exhibit the clearest temporal trend, with increased deoxygenation since 1992, particularly in the upper 1000 m, within the NACW (Fig. f–j). Despite its smaller magnitude in comparison to remineralisation change, the term exhibits the largest trend over time, owing to increasing excess temperature changes (Fig. m) and their impact on oxygen solubility. When averaged over the central region, the largest excess changes occur at ∼300 m, with the maximum deoxygenation at this depth trebling between 1992 to 1998 and 1992 to 2015, when it reaches -1.50 µmol kg⁻¹ (Fig. b). Below 1000 m, excess oxygen change is negligible in the uNADW, given the magnitude of excess temperature change generally at this depth remains below ±0.05 °C, and hence drives very little deoxygenation (Fig. m).

Redistributed changes in oxygen are less consistent with time than the excess signal (Fig. c), reflecting local circulation patterns with no apparent trends. The lines of near-zero redistribution close to the surface and at a depth of around 800 m (Fig. k–o) are simply due to the change in sign of the βo coefficient as it switches from positive to negative, and have no correlation with any particular water masses. At this depth, redistribution has no effect on oxygen because the depth gradient is zero. Patterns of disequilibrium oxygen change are relatively similar to those of the total oxygen change, but with considerably lower absolute values (Fig. u–y). When averaged across the central region of the section, the maximum magnitude is only -2.37 µmol kg⁻¹ at depths of ∼600 m between 1992 and 2010 (Fig. e). Since there is no direct link between the disequilibrium change and any of the DIC or temperature terms, the term is weakly constrained, making analysis more challenging. The similarity between the total change is likely because it is a residual term, and so will have a larger magnitude when the total change is greater.

Figure  shows the remineralisation and excess change, the two terms with clear trends, over time. Averaging the changes over the upper 1000 m emphasises the relationship between total oxygen change and the remineralisation component. In both time series, maximum deoxygenation occurs in 2010, before decreasing slightly again. Solubility-driven deoxygenation has a much smaller but fairly constant decrease across the time series in the upper 1000 m. At depths of 1000–2000 m (Fig. b), there is no obvious trend in total oxygen change across the section, however, the total change and remineralisation change are still highly correlated, with the same patterns of increasing and decreasing oxygen over time.

The temperature change in the central region of the A05 section is dominated by redistribution (Fig. n). While no clear trends are observed in the redistributed component, the excess component exhibits a steadily increasing trend over time, particularly pronounced in the upper 1000 m (Fig. m). This leads to decreased solubility and the observed trends in excess oxygen decrease in the upper ocean.

DIC change is dominated by the excess term in the upper 1000 m (Fig. g), due to increasing anthropogenic CO₂ levels. Trends in excess DIC changes mirror those in excess temperature since they are linearly related via the coefficient α, which is constant in time. Excess DIC reaches a maximum change of 21.5 µmol kg⁻¹, occurring between 1992 and 2015 at a depth of ∼270 m (Fig. h). Due to the ratio of oxygen consumption and DIC production during remineralisation, we also see a clear temporal trend in remineralisation DIC change, with increasing production over time. Remineralisation change peaks at a depth of ∼600 m, reaching around 5.7 µmol kg⁻¹. At deeper depths, DIC change becomes close to zero and in some cases slightly negative, but this negative change is generally smaller than the internal accuracy of the GLODAP data product. As with temperature, no distinct temporal patterns are discernible in redistributed DIC changes. The decomposition shows that while the anthropogenic increase in CO₂ fluxes dominate, remineralisation changes are also responsible for a significant amount of the increased DIC in the upper ocean. Excess DIC change in the section is strongly correlated with the total DIC change (r2=0.72, p=0.07), particularly in the upper 1000 m (r2=0.95, p=0.004) but the correlation between the remineralisation change and the total change is also high (r2=0.65, p=0.09, increasing to r2=0.91, p=0.01 for just the upper 1000 m).