Introduction

Eastern Boundary Upwelling Systems (EBUS) are among the most productive regions in the ocean and contribute 11% to global new production, which refers to biomass largely produced on the basis of nutrients introduced via upwelling and vertical mixing from the deep dark ocean into surface waters1,2,3. The associated assimilation of CO2 through the generation of biomass and its transfer to the deep sea is an integral part of the biological carbon pump4,5 which reduces atmospheric CO2 concentrations through the storage of CO2 as dissolved inorganic carbon (DIC) in the deep ocean. Even though it is widely assumed that the biological carbon pump responds to climate change, the magnitude and even the direction of change is still unknown6,7,8. However, the amount of DIC kept by the biological carbon pump is assumed to be linearly related to the inventory of regenerated nutrients, which are released during the remineralization of biomass in the deep ocean9. They stand in contrast to biologically unused so-called preformed nutrients, whose formation represents a leakage through which the biological carbon pump loses CO2 and makes e.g. the Southern Ocean south of 44°S a natural CO2 source to the atmosphere10,11. This leakage evolves when nutrients that upwell along with DIC are not fully utilized by biological production due to light and iron limitation12,13,14. Nowadays, rising CO2 concentrations in the atmosphere reverse the air-sea fluxes of CO2 and have converted the Southern Ocean into a key site for anthropogenic CO2 uptake with a current estimate of approximately 740 \(\pm\)290 Tg C year−1 15,16. Due to the counteracting effect of natural CO2 released from the biological carbon pump13 of about 400 \(\pm\)180 Tg C year−1 15,16, this results in a CO2 invasion of 340 \(\pm\)110 Tg C year−1 between 1990 and 200916,17,18. However, approximately 27.5% (~110 Tg C year−1) of the natural CO2 release from the biological carbon pump occurs within the Atlantic sector of the Southern Ocean between 44° and 58°S15,16, while the resulting preformed nutrients are transported northwards and subducted beneath warmer and lighter subtropical water masses19,20,21. Loaded with preformed nutrients, these mode waters support primary productivity in upwelling systems at lower latitudes and thus could potentially restore the CO2 uptake efficiency of the biological carbon pump. Although EBUS could therefore act as regional CO2 sinks through the utilization of preformed nutrients and low sea water temperatures which increase the solubility of CO222,23,24, global models suggest that upwelling systems (in particular ones at lower latitudes) act as net CO2 sources to the atmosphere25,26,27. This was also found to be the case in a modelling study of the Benguela Upwelling System (BUS), which is located in the southeast Atlantic Ocean (Fig. 1) and considered the most productive of all EBUS28. The study however suffered from a model that poorly represented the BUS, and an unclearly defined upwelling region27. Other studies within the BUS have not resolved these issues as indicated by estimated CO2 fluxes ranging from −5.1 to 1.54 Tg C year−1 25,26,29, with opposing air-sea fluxes30 noted between the northern (11.5 Tg C year−1)31 and southern (−1.4 to −2.8 Tg C year−1)31,32 upwelling area. It thus remains elusive whether the BUS is a net CO2 sink or CO2 source to the atmosphere. Similar to the Southern Ocean, the BUS also suffers from a sparsity of data, and model simulations and evaluations26,27 were constrained by pCO2 data (partial pressure of CO2) from the Surface Ocean CO2 Atlas (SOCAT)33. So far this data product misses coverage particularly over the northern BUS shelf region, thus curtailing estimates of air-sea gas exchange for the BUS (Fig. 1).

Fig. 1: Maps of pCO2 measurements of research vessel cruises within the Benguela Upwelling System used in this study.
figure 1

a Underlying cruise tracks of the 13 cruises (red) and those embedded in SOCAT v202033 (blue). b Recorded pCO2 measurements (in µatm), normalized to the reference year 2020. c Normalized pCO2 measurements (in µatm) interpolated on a 0.5° grid. Dashed lines mark the area of the northern and southern Benguela Upwelling System (NBUS, SBUS), while contour lines represent the atmospheric pCO2 of the reference year 2020 based on Mauna Loa records (414 µatm). The interpolation was performed after the minimum curvature interpolation method105 implemented in the Generic Mapping Tools (GMT). Country outlines sourced from the GSHHG dataset106 and were plotted with GMT.

Full size image

Here, we address this problem by presenting a compilation of shipboard pCO2 data for the BUS from the extended SOCAT v202033 data base that includes data gathered from 14 cruises throughout the BUS from 2008 to 2019 (see Supplementary Table 1). This revised data set allows for a seasonal-based examination of regional air-sea fluxes of CO2 across the BUS and underlying mechanisms affecting these fluxes, which indicate that the biological carbon pump in the BUS serves as a globally significant CO2 sink.

Results and discussion

Sea surface pCO2 characteristics

In order to overcome the problem of unclearly defined upwelling regions that led to strong discrepancies in the definition of the BUSs offshore boundary26,31,34, we specified the upwelling zone’s spatial extent by considering the average cross-shelf distribution of sea surface pCO2 (Fig. 2a, b) for the northern and southern subsystems (NBUS, SBUS). The latitudinal boundary between the NBUS and SBUS is formed by the Lüderitz cell (~26°S)35,36,37 within the Lüderitz upwelling region (24°S–28°S)38,39 that is subject to perennial coastal upwelling. In general, upwelling systems exhibit highest pCO2 in the nearshore region due to the upwelling of carbon-rich waters, and a decreasing trend offshore due to degassing and the biologically-mediated carbon uptake within the offshore-advecting upwelled water34,40,41. By assuming that a decrease in pCO2 variability marks a decreasing influence of upwelling on pCO2, we determined the upwelling zone’s boundary as the distance from shore when the standard deviation (s.d.) decreases persistently to below \(\pm\)30 µatm in both subsystems.

Fig. 2: Spatial and temporal variability of sea surface pCO2.
figure 2

a Average cross-shelf distribution of all pCO2 measurements (in µatm) of the northern part of the Benguela Upwelling System (NBUS) with distance to the coast (in km). The grey line marks the offshore boundary of the upwelling zone. Values were normalized to the reference year 2020, and averaged over intervals of 10 km coastal distance. b As in a, except for the southern part (SBUS). c Average latitudinal pCO2 concentrations across the NBUS and SBUS regions, for austral spring and summer (blue) and austral autumn and winter (red). d Histogram of pCO2 within the NBUS for both seasons. e As in d, except for the SBUS. The dashed grey line in subplot a, b and c represents the atmospheric pCO2 of the reference year 2020 based on Mauna Loa records (414 µatm), while the shaded areas represent the standard deviation.

Full size image

In the NBUS, this results in an upwelling boundary at 340 km offshore, which lies within the lower range of other studies that considered an offshore extension from 300 to 800 km27,28,42,43. In the SBUS, the decreasing variability in pCO2 suggests a closer boundary at 200 km distance to shore. In comparison to the NBUS, the much lower intensity of upwelling observed in the SBUS44,45 may explain the discrepancy observed in the offshore extent of elevated pCO2 values between the subsystems. Considering additionally the latitudinal extent of the SBUS (26°S to 34°S) and NBUS (16°S to 26°S) results in upwelling areas of 177.600 km2 and 377.400 km2, respectively.

Relative to the mean atmospheric pCO2 of our reference year 2020 (~414 µatm), the annual mean pCO2s of 492.3 \(\pm\)115.82 µatm for the NBUS and 383.9 \(\pm\)53.73 µatm for the SBUS reveal the opposing character of the two subsystems as a regional CO2 source and sink, respectively. The uncertainty in the mean annual pCO2 could thereby be attributed to the variability of pCO2 in close proximity to the coast due to the impact of upwelling and the seasonality in upwelling intensities. To estimate the latter, pCO2 data were averaged across the mean offshore extent and latitudes of the upwelling areas for the austral spring and summer (September-March), and austral autumn and winter seasons (April-August, Fig. 2c). The resulting plot of pCO2 variability with latitude indicates a seasonal influence in the northern region between ~17°S and ~21°S which diminishes towards the south. The observed seasonality with enhanced pCO2 between April and August in the north corresponds to seasonal variations in upwelling intensities, which for this region are strongest during this time of the year46. However, when averaged across the whole NBUS and considering the standard deviation (Fig. 2c), the ~17% difference in seasonal means (481.0 \(\pm\)117 µatm, September-March versus 560.7 \(\pm\)66 µatm, April-August) imply that the upwelling-related seasonality is only weakly pronounced. In the SBUS, seasonal mean pCO2s of 381.7 \(\pm\)36 µatm (September- March) and 388.5 \(\pm\)44 µatm (April-August) do not reflect the seasonality of upwelling intensities, which are strongest during the austral spring and summer (September- March)47. This could be attributed to the balancing between initial outgassing and biologically-mediated CO2 uptake under intensified upwelling conditions, and minor effects of vertical water mass transports and biology on the pCO2 during the non-upwelling season. Although our data mirrors the upwelling-related variability in pCO2 for each subsystem during both seasons (Fig. 2d, e), there are more measurements available for the austral spring and summer season (September-March) (see also Supplementary Fig. 1 and 2). Hence, assumptions on seasonal differences and annual estimates for a given subsystem should be treated carefully. Overall, in line with annual means, the seasonal pCO2 estimates display the opposing behaviour of the NBUS and SBUS as a regional CO2 source and sink, respectively.

Air-sea CO2 flux estimates

In contrast to pCO2, air-sea CO2 fluxes reveal a pronounced seasonality in both systems as being more than twice as high during upwelling than during non-upwelling seasons. The intensification of wind during the upwelling season is considered the primary driver of this difference which, given the opposite signs of the flux in the two regions, strengthens the CO2 source and sink functions in the NBUS and SBUS, respectively (Supplementary Table 2). Integrating the annual mean fluxes over the upwelling area and considering their uncertainties results in a CO2 emission of 15.64 (−2.95–73.16) Tg C year−1 in the NBUS and a CO2 uptake of −2.94 (−4.5–3.58) Tg C year−1 in the SBUS. Taking additionally the different gas transfer velocity parameterizations into account (see Methods section, Supplementary Table 3), annual mean CO2 fluxes can increase in both subsystems by ~71% to 26.69 Tg C year−1 (NBUS) and −5.03 Tg C year−1 (SBUS), underlining the profound difference in the sink and source character of the SBUS and NBUS, respectively. However, in comparison to our initial area-integrated CO2 fluxes given in ref. 31. for the NBUS (11.5 Tg C year−1) and SBUS (−1.4 Tg C year−1), our respective estimates of 15.64 (NBUS) and −2.94 (SBUS) Tg C year−1 are about 40 and 110% higher, mainly due to the use of a smaller area in the ref. 31. study, but potentially also due to less ship-board measurements that were used to calculate the CO2 fluxes. Similar to these data-based estimates, modelled carbon fluxes also indicate a CO2 sink region in the south and source region in the north27, while modelled CO2 fluxes from the BUS between 18°S to 28°S amount to ~24 Tg C year−1 on average over the time period between 1982 and 2015. Comparing our measurements with modelled data by recalculating CO2 fluxes for the same region with the same offshore boundary at 800 km, results in a weaker annual CO2 source into the atmosphere of ~8 Tg C year−1. This discrepancy seems to be caused mainly by a poleward misplacement of the outgassing cell in the applied model framework27, which underestimates the CO2 sink behaviour of the SBUS.

Nutrients as a driver of regional variability in sea surface pCO2

In order to differentiate between effects of surface-warming and the biologically-mediated drawdown of CO2 on the air-sea gas exchange, we followed a commonly applied bottom-up approach to estimate new and export production, assuming nutrients of upwelled source waters to be consumed at the surface and transformed into organic matter (new production) that is subsequently exported out of the euphotic zone (export production)1,3,48.

Before we used this approach, it was validated in two steps. In a first step we assumed surface-warming and the biologically-mediated drawdown in nearshore regions to largely be negligible and the high measured pCO2 to be an immediate consequence of the cold, DIC and nutrient-enriched source waters that are introduced into the surface layer. Hence, we first calculated the potential pCO2 in freshly upwelled waters based on temperature, salinity, as well as TA, DIC and nutrient concentrations of the upwelling source water masses (ESACW and SACW, Supplementary Table 4) and compared the resulting pCO2 with those measured in the nearshore regions.

In a second step, surface warming and effects of nutrient utilization on pCO2 in surface waters were estimated by taking into account that phytoplankton consumes upwelled nutrients to fix DIC into biomass, as also indicated by data showing nutrient depletion of offshore flowing upwelled waters at distances of about 180-200 km to the coast49,50. The Redfield ratio (106:16) is, in turn, used to translate nutrient utilization into DIC consumption and the associated release of total alkalinity (see e.g1,48.). By subtracting the amount of DIC which is transformed into organic matter from the original source water DIC concentration, and adding the released TA to the original source water TA, we can calculate the pCO2 in upwelled water after upwelled nutrients have been consumed and exported. During the offshore flow, upwelled waters warm up as indicated by the difference between temperatures of the source waters and the sea’s surface temperature. Hence, by using sea surface temperatures and salinities instead of those of the source water, we can further consider the warming of upwelled waters and its effect on pCO2.

The exercise implies, in line with our data (Fig. 2a, b), a high pCO2 in the freshly upwelled water near the coast that decreases towards the outer boundaries of the subsystems where nutrients are consumed (Fig. 3). The mean calculated pCO2 in the freshly upwelled water (first step: NBUS: 842–1110 µatm, SBUS: 596–795 µatm) falls within the upper range of the nearshore recorded pCO2 (NBUS: 294–1012 µatm, SBUS: 334–610 µatm). This implies that biologically-mediated drawdown of CO2 occurs simultaneously with coastal upwelling as indicated by e.g., high satellite-derived chlorophyll concentrations at a narrow belt along the nearshore region51,52,53. At the outer boundaries where nutrients are already consumed, measured values fall, in turn, well within the ranges of calculated ones (second step: NBUS: 319–423 µatm, SBUS: 310–426 µatm). The correspondence between the measured and calculated pCO2 within both subsystems (Fig. 3) places confidence in our bottom-up approach and validates its use for estimating effects of surface warming and the biologically-mediated drawdown of CO2 on the pCO2 in surface waters. To further resolve the latter effects on a spatio-temporal scale, we reconstructed the non-thermally controlled pCO2. Therefore, we calculated DIC based on our pCO2 climatology which includes annual and seasonal gridded fields on pCO2, SST and SSS, as well as the SSS and TA correlation (see Methods for a detailed description). To eliminate the surface warming effect, pCO2 was recalculated by using DIC and TA but instead of using SST and SSS, source water salinities and temperatures were used. Hence, the non-thermally controlled pCO2 indicates the pCO2 which one would have expected in case upwelling waters would not have been warmed at the surface.

Fig. 3: Thermal and biological controls on sea surface pCO2 based on CO2SYS simulations.
figure 3

Measured (grey) and modelled (blue) sea surface pCO2 (in µatm) during coastal upwelling and nitrate (N) consumption through biologically-mediated CO2 uptake at the offshore boundary, based on CO2SYS96,97 calculations using source water mass characteristics (Supplementary Table 4) for the a northern (NBUS) and b southern Benguela Upwelling System (SBUS). The simulated pCO2 at the offshore boundary includes the effect of surface warming on pCO2. The grey dashed line represents the atmospheric pCO2 of the reference year 2020 based on Mauna Loa records (414 µatm). Sea surface pCO2 concentrations below (above) the atmospheric level indicate a source (sink) of atmospheric CO2. The uncertainties in measured pCO2 are presented as the standard deviation, whereas uncertainties in the modelled pCO2 are based on the standard error of the average source water mass characteristics.

Full size image

As a result, the non-thermally controlled pCO2 as depicted in Fig. 4 mirrors spatial and temporal trends seen in the measured pCO2 (Fig. 2), but remains below the measured pCO2. Hereby, the non-thermally controlled pCO2 falls below the atmospheric pCO2 at the outer boundaries, indicating that the consumption of upwelled nutrients would have turned both subsystems into CO2 sinks to the atmosphere if not overcompensated by surface warming in the NBUS (Fig. 4a, b). Even though the latitudinal averages express this during the austral spring and summer season (September-March), the austral autumn and winter season (April-August) does not reflect this trend (Fig. 4c, d) due to sample biases. As discussed before, less measurements were available from the offshore region from this time period so that the available nearshore data dominate the mean. At these sites, the biological carbon pump is still quite inefficient as upwelled nutrients have not fully been consumed, causing the regenerated DIC which upwells along with the nutrients to increase pCO2 in surface waters (Fig. 4a). In this regard, the availability of nutrients, in addition to changes in the Redfield carbon to nutrient ratio, largely control the potential biologically-mediated CO2 uptake. Since in the BUS, the Redfield carbon to nutrient ratio is assumed to be constant49,54,55,56, its impact on biologically-mediated CO2 uptake is neglected in the following discussion. The drawdown of CO2 by the biological consumption of nutrients plays, in turn, different roles in the marine carbon cycle: The regenerated nutrient consumption balances the input of regenerated CO2 from below the euphotic zone whereas the utilization of preformed nutrients compensates for CO2 outgassing during their formation at higher latitudes. Hence, changes in the regenerated nutrient consumption do not affect the net flux of CO2 across the air-sea interface, because of associated variations in the supply of regenerated nutrients and CO2, while the utilization of preformed nutrients can affect the net CO2 uptake by compensating CO2 losses from the Southern Ocean57,58,59 (see Methods). Hence, it is the relative strength of the warming-driven increase of pCO2 and consumption of preformed nutrients that finally controls the regional CO2 sink and source functions of the SBUS and NBUS.

Fig. 4: Spatio-temporal variability of the non-thermally controlled sea surface pCO2 based on CO2SYS simulations.
figure 4

Average cross-shelf distribution of measured (red) and non-thermally controlled pCO2 (blue) (in µatm) with distance to the coast (in km) based on CO2SYS96,97 calculations using sea surface data and source water mass characteristics (Supplementary Table 4) for the a northern (NBUS) and b southern Benguela Upwelling System (SBUS). Values were averaged over intervals of 10 km coastal distance. c Average latitudinal pCO2 concentrations (measured and non-thermally controlled) across the NBUS and SBUS regions during austral spring and summer. d As in c, except for austral autumn and winter. The grey lines in each subplot represent the atmospheric pCO2 of the reference year 2020 based on Mauna Loa records (414 µatm). Sea surface pCO2 concentrations below (above) the atmospheric level indicate a source (sink) of atmospheric CO2. The uncertainties in measured and non-thermally controlled pCO2 are presented as the standard deviation.

Full size image

To assess the CO2 uptake by the utilization of preformed nitrate (Npref) in the SBUS and NBUS, we followed the bottom-up principle to calculate new/export production rates as mentioned earlier1,60. However, instead of using the total amount of nitrate, we only consider the preformed nitrate concentration within the upwelling source waters, which have been compiled from different cruises (see Supplementary Table 4). Hence, Npref of 6.44 ± 0.61 and 7.44 ± 0.36 µmol kg−1, with an upwelling volume of 0.9 Sverdrup (Sv) for the NBUS45,60 and 0.4 Sv for the SBUS45, amount to new production rates driven by the utilization of Npref of 14.5 ± 1.4 and 7.5 ± 0.4 Tg C year−1, respectively.

Alternatively, this part of the new production can also be estimated by using previously published new production rates30,31,48 and the contribution of preformed nutrients to the total nutrient concentrations as derived from our water mass characteristics (see Supplementary Table 4). The published new production rates fall within a comparatively wide range of 68–245 T C year−1 for the NBUS and 13.5–42 Tg C year−1 for the SBUS30,31,48, which in sum covers published new production rates of 241 Tg C year−1 derived for the entire BUS from 16°S to 34°S1. In addition to the contributions of preformed nutrients to the total nutrient concentrations of 24 ± 2% in the NBUS and 38 ± 2% in the SBUS, this amounts to a CO2 uptake by the utilization of Npref of 16.3–58.8 Tg C year−1 for the NBUS, and 5.1–16.0 Tg C year−1 for the SBUS. Hereby, despite of the inherent methodology that could be held responsible for the discrepancy in the estimated new production rates, they provide lower and upper cases to assess the magnitude of the effect preformed nutrient consumption may hold in the BUS. Hence, given these lower (14.5 + 7.5 = 22) and upper (58.8 + 16.0 = 74.8) estimates, a new production driven by the utilization of Npref of ~22–75 Tg C year−1 implies that the biological carbon pump in the BUS countervails 20 up to 68% of the CO2 release from the biological carbon pump within the Atlantic sector of the Southern Ocean between 44° and 58°S of ~110 Tg C year−115,16 (Fig. 5). These results emphasize the role of the BUS as a significant hub for restoring the CO2 uptake efficiency of the biological carbon pump. However, its strength may be impacted by global change processes, such as the response of upwelling intensities to climatic changes, as well as by human impacts through fishery practices. While the first is difficult to estimate due to e.g. the poor representation of the BUS in numerical models27 and ongoing changes in the ecosystem structure61,62, the latter refers e.g., to emissions of aqueous CO2 caused by disturbances to the seafloor through bottom trawling. In relation to the CO2 uptake of the biological carbon pump (~22–75 Tg C year−1), a release of sedimentary carbon through bottom trawling of around 5 Tg C year−1 as estimated for the BUS63 could lead to significant disruptions of the system, with as yet unknown consequences for the biological carbon pump that will need to be addressed in future studies.

Fig. 5: Schematic overview on the coupling between Southern Ocean CO2 loss and CO2 uptake in the Benguela Upwelling System through the biological carbon pump.
figure 5

CO2 fluxes (grey arrows) of the biological carbon pump at the air-sea interface and subsurface, regulated by the effects of preformed-based new production in the Benguela Upwelling System and natural outgassing of CO2 in the Atlantic sector of the Southern Ocean (110 Tg C year−1 15,16). The transport of water masses from Southern Ocean across the South Atlantic (blue arrows) shapes the supply of preformed nutrient inventories of the upwelling source water masses (SACW South Atlantic Central Water, ESACW Eastern South Atlantic Central Water) of the northern and southern Benguela Upwelling System. The figure was generated using Inkscape107.

Full size image

Methods

Study region

The Benguela Upwelling System (BUS) stretches across the north of Cape Frio from ~15°S to Cape Agulhas (~35°S), while being bound by warm waters of the Angola and Agulhas current to its northern and southern ends, respectively47. Coastal upwelling is thereby controlled by south-easterly winds emanating from the interplay between the South Atlantic Anticyclone (SAA) and the continental low pressure trough, causing the emergence of distinct upwelling cells along the shoreline46. The strongest upwelling cell is located at Lüderitz (~26°S) and separates the northern (NBUS) from the southern (SBUS) upwelling region35,36,37. One of the main upwelled waters is South Atlantic Central Water (SACW), which represents a Sub-Antarctic Mode Water that itself is a mixture of Antarctic Intermediate Water and Subtropical Mode Water57,58,59. These water masses subduct beneath warmer subtropical surface waters north of the Sub-Antarctic Front around 36°S–54°S and are transported eastward as SACW across the South Atlantic into the Cape basin64,65,66. Here, SACW converges with the Agulhas water from the Indian Ocean to form ESACW that enters the SBUS from the south67,68,69. The majority of the SACW circumvents the SBUS along the Benguela Current and enters the BUS from the north via the Angola-Benguela Frontal Zone (ABFZ) as a poleward undercurrent, which, unlike ESACW, is nutrient-enriched and oxygen-depleted70.

Upwelling intensities in both subsystems exhibit a seasonal pattern which is mainly driven by temporal shifts of the SAA. As the SAA moves north-westward, the NBUS experiences maximum upwelling intensities during austral winter (June-August), while leading to a dominating westerly wind regime which weakens upwelling in the SBUS. With the south-eastward displacement of the SAA, upwelling in the SBUS is mainly confined to the summer season (September-March)38,47,71. Additionally, the seasonality in upwelling is more pronounced in the SBUS as compared to the NBUS due to the perennial upwelling-favourable winds that reign in the northern region46. This is also reflected in primary productivity which in the SBUS is twice as large in the summer than winter72. Furthermore, studies have shown a decrease and intensification of upwelling in the NBUS and SBUS, respectively, over the past decade in response to global warming44,52 and a southward shift of the SAA73. The associated increase in sea water temperatures in the NBUS62 could potentially be a cause of the prominent decrease in zooplankton size spectra in the BUS61. Nevertheless, the productivity has remained consistently elevated, and still supports high fishery yields in both subsystems with sardine and horse mackerel as main target species in the SBUS and NBUS, respectively74,75,76.

Sea surface pCO2 data collection

For the analysis and quantification of air-sea gas exchanges within the BUS, continuous underway measurements were carried out as described in ref. study31 between 2008 and 2019 (Supplementary Table 1). Hereby, various instruments were utilized to analyse the sea surface partial pressure of CO2 and were fed with seawater by the vessel’s circular pumps or autonomous pump systems, drawing water from ~5-7 m from the vessel’s moon pool or at its bows. The devices comprise of a LI-7000 CO2/H20 analyser (Licor Biosciences), which measured pCO2 by equilibration of surface water with air and the detection of the equilibrium concentration of CO2 in air during RV Maria S. Merian cruise 7/2. A Pro-Oceanus Systems Inc. PSI CO2-ProTM was applied during RV Meteor cruise 76/2, using gas equilibration and infra-red detection of the CO2 gas stream while being linked to the FerryBox flow-through system. During the remaining cruises, an underway Carbon Dioxide Analyzer (SUNDANS, Marianda) equipped with the infra-red sensor LI-7000 was applied to measure the mole fraction of CO2 in seawater (xCO2), which was converted to pCO2 using underway records of atmospheric pressure, Sea Surface Temperatures and the equilibrator temperature of the SUNDANS system. Additional quality-controlled measurements on sea surface fCO2 from the Surface Ocean CO2 Atlas (SOCAT) v202033 were converted into pCO277 and embedded into our analysis. All pCO2 measurements were normalized to a reference year (2020)78 by using a mean yearly change rate in seawater pCO2 of 1.5 µatm year−1 for observations up to year 1992 and 1.9 µatm year−1 for the more recent time period according to the updated oceanic pCO2 trend of Takahashi et al.79 and multiplying it with respective observations for both upwelling regions (NBUS, SBUS). Overall, the extended data set on pCO2 records used in this study is homogeneously distributed in space as it covers the shelf and coastal areas along the continental margin off Namibia and South Africa, while spanning a timeframe from 1986 to 2020 with over 250 000 data points within the area from approximately 5°E to 18.7°E, and 16°S to 34.5°S. All normalized measurements were spatially interpolated on a 0.1° x 0.1° grid using ordinary kriging as a geostatistical technique, which allows us to perform an error propagation of the gridding procedure and to account for the spatial autocorrelation of pCO2. The uncertainty in average estimates thereby provides an outline on the strong variability of pCO2 that can be found in coastal upwelling settings. In addition, our pCO2 climatology offers an updated view on CO2 sources and sinks in comparison to previous pCO2 climatologies77,80 which were merely based on the SOCAT dataset that largely misses pCO2 recordings in the NBUS coastal region (Fig. 1a). To reduce the uncertainty in the gridding process, we only applied ordinary kriging to those grid cells with sufficient data coverage. Ordinary kriging was performed using the R automap package81, which incorporates geostatistical routines from gstat82. In a first step, we computed a sample variogram to outline the spatial correlation of observations as a function of distance, and added a model to fit the spatial variation of pCO2 (Supplementary Fig. 3). In a next step, predictions were made for the designated grid cells by including kriging weights that were formed on the basis of the fitted model and pCO2 measurements within the surrounding neighbourhood, while corresponding error estimates were provided as variances (v) and standard deviations (\(\sigma\) = \(\sqrt{v}\)) for each grid cell. As the performance of ordinary kriging is time intensive, we only used observations within a maximum of 0.5° spherical distance from the prediction location to speed up the process, without noting any influence on the resulting predictions or error estimates. The total mean uncertainty (SE) of pCO2 for the NBUS and SBUS is given as the standard error, which results from a combination of measurement error, spatial variance and the gridding procedure. It is derived by dividing the sum squares propagation by the square root of its degrees of freedom (Eq. 1), which represents the number of grid cells (N). To account for spatial autocorrelation, we corrected N by replacing it with the effective number of grid cells (Neff) following Landschützer et al.83

$${{{{{\rm{SE}}}}}}=\frac{\sqrt{{\sum }_{i=1}^{N}{\sigma }_{i}^{2}}}{\sqrt{{N}_{{{{{{\rm{eff}}}}}}}}}$$
(1)

We estimated Neff by using autocorrelation length scales for pCO2 that were derived from the variogram as the distance (range) where the spatial dependence (semi-variance) levels off (Supplementary Fig. 3). By calculating the point-to-point distance of each grid cell within both upwelling regions, we could determine the effective number of grid cells outside of the autocorrelation length scale. We estimated short length scales in the range of around 0.01° to 1.4° spherical distances, which, in kilometres, resembles those found in coastal regions (~50 km) as a result of the heterogeneity of the water masses and physical turbulence caused by upwelling84.

Carbon flux calculation and its uncertainties

Differences in the partial pressure of carbon between the sea surface (\({{{{{{\rm{pCO}}}}}}}_{2,{{{{{\rm{sw}}}}}}}\)) and atmosphere (\({{{{{{\rm{pCO}}}}}}}_{2,{{{{{\rm{at}}}}}}}\)) were used to determine carbon flux rates (\({{{{{{\rm{FCO}}}}}}}_{2}\)) using Eq. (2):

$${{{{{{\rm{FCO}}}}}}}_{2}={K}_{0} * k * ({{{{{{\rm{pCO}}}}}}}_{2,{{{{{\rm{sw}}}}}}}-{{{{{{\rm{pCO}}}}}}}_{2,{{{{{\rm{at}}}}}}})$$
(2)

where \({K}_{0}\) is the solubility coefficient of CO285 and \(k\) represents gas transfer velocity of CO286. The gas transfer velocity \(k\) was calculated following Eq. (3):

$$k=0.251 * {u}^{2} * {\left(\frac{{{{{{\rm{Sc}}}}}}}{660}\right)}^{-0.5}\,$$
(3)

\({{{{{\rm{Sc}}}}}}\) is the Schmidt number of CO2 in seawater, \(660\) represents \({{{{{\rm{Sc}}}}}}\) at 20 °C water temperature and \(u\) refers to the wind speed (m/s) at 10 m above the sea surface. Additional data on sea surface temperature (SST) (°C) and salinity (PSU) were thereby required for the determination of \({Sc}\) using the parameterization after Wanninkhof 86. Data on wind speed, SST and salinity were based on shipboard measurements (Supplementary Table 1), which were spatially interpolated on a 0.1° x 0.1° grid based on the ordinary kriging procedures as outlined in the previous section for pCO2. The flux calculation was then performed using the average sea surface pCO2, wind speed, SST and salinity of the kriging predictions within the defined NBUS and SBUS region. Seasonal flux estimates were calculated using kriged and averaged pCO2, wind speed, SST and salinity data collected during spring and summer (September till March), and austral autumn and winter (April till August), respectively. The total mean uncertainty of the individual parameters was estimated using Eq. 1. In two cases of the NBUS austral summer season, the variograms depict long autocorrelation length scales, resulting in a relatively low effective number of grid cells. For these cases, we used the uncorrected number of grid cells N to calculate the total mean uncertainty (entries marked with * in Supplementary Table 2).

To address another source of uncertainty in the air-sea gas exchange, we further calculated the gas transfer velocity k and subsequent fluxes on the basis of other formulations for k given by Wanninkhof 87 (hereafter KW92), Wanninkhof & McGillis88, Nightingale et al.89, McGillis et al.90, as well as Ho et al.91 with differing wind speed dependencies (Supplementary Table 3). The resulting flux values based on KW9287 and Wanninkhof (hereafter KW14)86 thereby represent the highest and lowest estimates, respectively, and differ on average by ~70% in both subsystems. The parameterization after KW1486 foresees the use of a wind speed product with high temporal resolution (6-hour), with estimated flux values resembling those derived from previous formulations of k, which were based e.g., on dual tracer methods conducted in the North Sea (Nightingale et al.89) and Southern Ocean (Ho et al.91). The KW92 parameterization is thereby based on an outdated 14C inventory for the global ocean, while Wanninkhof & McGillis88 assumed a cubic instead of the commonly used quadric dependency between gas transfer and wind speed92.

Water column sampling and analysis

The analysis of water mass characteristics and biogeochemical settings in the BUS was based on data gathered during the various cruises that were partially embedded into the analysis of carbon fluxes (Supplementary Table 1). In addition, we added data from the Global Ocean Data Analysis Project version 2.2020 (GLODAPv2_2020) and data collected during our most recent cruise with RV Sonne (SO285), which took place from 20th August to 2nd November 2021. The sampling was performed with a CTD/Rosette sampler at stations covering on- and offshore areas of both subsystems (Supplementary Fig. 4), allowing a direct comparison of both upwelling zones. We collected CTD profiles of temperature, salinity and oxygen, and defined the upwelling SACW and ESACW source waters by using the potential temperature (\(\theta\)) definition provided by ref. 93. (Eqs. 4 and 5):

$${{{{{\rm{SACW}}}}}}:\,\theta=8.56 * {{{{{\rm{Salinity}}}}}}-289.08$$
(4)
$${{{{{\rm{ESACW}}}}}}:\,\theta=9.44 * {{{{{\rm{Salinity}}}}}}-319.03$$
(5)

The analysis of dissolved inorganic nutrients (phosphate P, nitrate N) was carried out as outlined in ref. 49., with samples from the CTD/Rosette being filtered through disposable syringe filters (0.45 µm) after sampling, filled in pre-rinsed 50 ml PE bottles that were subsequently measured on-board or kept frozen at −20 °C until being analysed in the shore-based laboratory after the expedition. The measurements were performed with a continuous-flow injection system (Skalar SAN plus System) according to methods outlined by Grasshoff et al.94. Furthermore, the calculation of preformed and regenerated nutrients (\({P,\, N}_{{pref}}\) and \({P,\, N}_{{reg}}\) respectively) was performed following ref. 95. by including the apparent oxygen utilization (AOU) as well as the oxidation ratios \({R}_{P:{O}_{2}}\) = 1/138 (phosphate) and \({R}_{N:{O}_{2}}\) = 16/138 (nitrate), using Eqs. (6) and (7):

$${P,\, N}_{{reg}}={R}_{P,N:{O}_{2}}\,{{{{{\rm{AOU}}}}}}$$
(6)
$${P,\, N}_{{pref}}=P,\, N-{P,\, N}_{{reg}}$$
(7)

For comparative reasons, we chose the traditional Redfield oxidation ratio to calculate preformed and regenerated nutrient concentrations. For the analysis of total alkalinity (TA) and dissolved inorganic carbon (DIC), samples were collected in 250 ml borosilicate bottles using silicone tubes (Tygon). The bottles were rinsed twice, filled from the bottom to avoid bubbles and analysed on board using the VINDTA 3 C system (Marianda, Kiel, Germany). For TA analysis, the samples were titrated with a fixed volume of hydrochloric acid (HCl) by equal increments of HCl (0.1 N HCl). The analysis of DIC was performed using the coulometric method (Coulometer CM 5015) after CO2 was extracted out of the water sample. During cruise M153, the analysis of DIC was performed with a cavity ringdown spectrometer (Picarro G2201-I, 1510CFIDS2047_v1.0) attached to a Liaison A0301 and an AutoMate Prep device. Both the Picarro and VINDTA 3 C systems were calibrated using Certified Reference Material provided by A. Dickson (Scripps Institution of Oceanography, La Jolla, CA, USA) for quality assurance.

(Non-) thermal component analysis of sea surface pCO2

We reconstructed the non-thermally controlled pCO2 based on our pCO2 climatology, which includes annual and seasonal gridded fields on pCO2, SST and SSS as derived from ordinary kriging interpolations (see previous sections). We thereby applied the CO2SYS96,97 programme, using sea surface TA and DIC concentrations as input parameters. Due to a lack of shipboard underway measurements for TA and DIC that are needed to resolve the carbonate system, we first reconstructed sea surface TA by leaning on the TA-salinity relationship, which refers to the control of surface TA by freshwater addition or removal that can be mirrored through changes in salinity98,99. We therefore performed a linear regression analysis based on TA and salinity gathered from underway shipboard records during cruise SO285, and CTD surface profiles from all cruises listed in Supplementary Table 4. Based on the slope of the linear regression line (Supplementary Fig. 5), we then determined the corresponding TA for each grid cell that contained spatio-temporal interpolated salinity values. Next, we reconstructed sea surface DIC with CO2SYS96,97 for each grid cell that contained values for TA and the corresponding pCO2, sea surface temperature and salinity. For the performance of CO2SYS96,97 calculations, we used the dissociation constants of Mehrbach et al.100 as refit by Dickson and Millero101 and the borate dissociation constant of Dickson102. In a final step, we recalculated sea surface pCO2 by applying CO2SYS96,97 with all available grid cell values of TA and DIC within the NBUS and SBUS for each season (annual, spring-summer, autumn-winter). By using the respective temperatures and salinities from the average source water mass characteristics (Supplementary Table 4) instead of those from the sea surface, we excluded the warming of upwelled waters during their offshore flow. The surface warming effect on the air-sea gas exchange can ultimately be elaborated by comparing the recalculated, so-called non-thermal pCO2 with the one derived from the actual shipboard measurements. Hereby it should be noted that the number of grid cells between the non-thermal and measured pCO2 can differ due to the difference in the availability of pCO2 measurements and salinity records (see n in Supplementary Table 2) that were used for CO2SYS96,97 calculations. In case of the SBUS autumn-winter season, this resulted in a relatively high standard deviation of the non-thermal pCO2 compared to the measured pCO2, as there were fewer values available to determine the latitudinal averages (Fig. 4c).

Statistical information

The uncertainty in average parameter calculations of pCO2, CO2 exchange coefficients and fluxes (Supplementary Table 2), as well as in biogeochemical characteristics of source waters (Supplementary Table 4) is presented as the standard error (s.e.), together with the number of values (n) used for the average calculation.