Biological carbon pump affected by CO2 uptake in the Benguela Upwelling System

Eastern Boundary Upwelling Systems (EBUS) are well-known for their high productivity and fishery 34 yields. However, being scarcely sampled and poorly represented in global models, their role as CO 2 35 sources and sinks to the atmosphere remains elusive. Here, we present a compilation of shipboard 36 measurements over the past two decades, showing how the Benguela Upwelling System (BUS) in the 37 southeast Atlantic Ocean acts as a CO 2 source in the north and CO 2 sink in the south. Surface warming of 38 upwelled waters increases the partial pressure of CO 2 (pCO 2 ) and outgassing in both regions, but in the 39 south, the biologically-mediated drawdown of CO 2 exceeds this warming effect. Here, the biological 40 carbon pump owes its stronger impact on pCO 2 to higher shares of upwelling source waters carrying 41 preformed nutrients supplied from the Southern Ocean. Their formation increases pCO 2 in surface waters 42 and counteracts human-induced invasion of CO 2 in the Southern Ocean. However, their utilization in the 43 BUS compensates for over 20% of the CO 2 loss occurring in the Atlantic sector of the Southern Ocean. 44 This emphasizes the role of the BUS as key to improve our understanding of the ocean’s response to 45 climate change and the future evolution of CO 2 in the atmosphere.

Eastern Boundary Upwelling Systems (EBUS) are among the most productive regions in the ocean and 48 contribute 11% to global new production, which refers to biomass largely produced on the basis of 49 nutrients introduced via upwelling and vertical mixing from the deep dark ocean into surface waters 1-3 . 50 The associated assimilation of CO2 through the generation of biomass and its transfer to the deep sea is an 51 integral part of the biological carbon pump 4,5 which reduces atmospheric CO2 concentrations through the 52 storage of CO2 as dissolved inorganic carbon (DIC) in the deep ocean 5 . Even though it is widely assumed 53 that the biological carbon pump responds to climate change, the magnitude and even the direction of 54 change is still unknown 6,7 . However, the amount of DIC kept by the biological carbon pump is assumed to 55 be linearly related to the inventory of regenerated nutrients, which are released during the 56 remineralization of biomass in the deep ocean 8 . They stand in contrast to biologically unused so-called preformed nutrients, whose formation represents a leakage though which the biological carbon pump 58 losses CO2 and turned e.g. the Southern Ocean south of 44°S into a natural CO2 source to the 59 atmosphere 9,10 . This leakage evolves when nutrients that upwell along with DIC at the Antarctic 60 Divergence 11 are not fully utilized by biological production due to light and iron limitation 12,13 . water masses [18][19][20] . Loaded with preformed nutrients, these mode waters support primary productivity in 69 upwelling systems at lower latitudes and thus could potentially restore the CO2 uptake efficiency of the 70 biological carbon pump. Although EBUS could act as regional CO2 sinks through the utilization of 71 preformed nutrients and low sea water temperatures which increase the solubility of CO2 21-23 , global 72 models suggest that upwelling systems (in particular ones at lower latitudes) act as net CO2 sources to the 73 atmosphere 24,25,26 . This was also found to be the case in a modelling study of the Benguela Upwelling 74 System (BUS), which is located in the southeast Atlantic Ocean (Fig. 1) and considered the most 75 productive of all EBUS 27 . The study however suffered from a model that poorly represented the BUS, and 76 an unclearly defined upwelling region 26 . Other studies within the BUS hardly improve these conditions as 77 indicated by estimated CO2 fluxes ranging from -5.1 25 24 Tg C year -1 to 1.54 Tg C year -1 28 . It thus remains 78 elusive whether the BUS is a net CO2 sink or CO2 source to the atmosphere. Similar to the Southern 79 Ocean, the BUS also suffers from a sparsity of data and model outcomes 25,26 were constrained by pCO2 80 data (partial pressure of CO2) from the Surface Ocean CO2 Atlas (SOCAT) 29 . So far this data product 81 misses coverage particularly over the northern BUS shelf region, thus curtailing estimates of air-sea gas 82 exchange for the BUS (Fig. 1). a seasonal influence in the northern region between ~17°S and ~21°S which diminishes towards the 109 south. The seasonality observed in the north corresponds to seasonal variations in upwelling intensities, 110 which for this region are strongest during the winter 37 . However, when averaged across the whole NBUS, 111 the ~15% difference in seasonal means and high s.d. in pCO2 between spring and summer (485.7 ±113 112 µatm), and autumn and winter (556.03 ±126 µatm), imply that the upwelling-related seasonality is only 113 weakly pronounced. In the SBUS, seasonal mean pCO2s of 378.15 ±65 µatm (spring, summer) and 114 381.91 ±36 µatm (autumn, winter) do not reflect the seasonality of upwelling intensities, which are 115 strongest during the summer 38 . This could be attributed to a lower data coverage in SBUS as compared to 116 NBUS (Supplementary Fig. 1) or to the function of the SBUS as CO2 source and sink as indicated by 117 pCO2 concentrations above and below those in the atmosphere at the coast and further offshore (Fig. 2b). 118 Relative to the mean atmospheric pCO2 of our reference year 2020 (~414 µatm), the seasonal as well as 119 annual mean pCO2s of 487.6 ±2.9 µatm for the NBUS and 379.2 ±1.74 µatm for the SBUS reveal the 120 opposing character of the two subsystems as a CO2 source and sink, respectively. 121 Air-sea CO2 flux estimates 122 In contrast to pCO2, the air-sea CO2 fluxes reveal a pronounced seasonality in both systems as being more 123 than twice as high during upwelling than during non-upwelling seasons. The intensification of wind 124 during the upwelling season is considered the primary driver of this difference which, according to the 125 opposing signs, strengthens the CO2 source and sink functions in the NBUS and SBUS, respectively 126 (Table 1). 127 Integrating the annual mean fluxes over the upwelling area results in a CO2 emission of 14.98 Tg C year -1 128 in the NBUS and a CO2 uptake of -3.37 Tg C year -1 in the SBUS. These area-integrated CO2 fluxes 129 correspond with those given in ref. 31 for the NBUS (11.5 Tg C year -1 ) and SBUS (-1.4 Tg C year -1 ), 130 respectively, but are about 30% and 140% higher, mainly due to the use of a smaller area in the ref. 31 131 south and source region in the north 26 , while modelled CO2 fluxes from the entire BUS between 18°S to 133 28°S amount to 2.25 mol C m -2 year -1 on average over the time period between 1982 and 2015. 134 Comparing our measurements with modelled data by recalculating CO2 fluxes for the same region with an 135 offshore boundary at 800 km, results in a weaker annual CO2 source into the atmosphere of 1.19 mol C m -136 2 year -1 . This discrepancy seems to be caused mainly by a poleward misplacement of the outgassing cell 137 in the applied model framework 26 , which underestimates the CO2 sink behavior of the SBUS. 138 Preformed nutrients as a driver of regional variability in sea surface pCO2 139 As seen for the Oregon upwelling system, the utilization of preformed nitrate (Npref), with concentrations 140 of up to 12 µmol kg -1 , turned the upwelling area into a CO2 sink 21,23 , while the impact of Npref with 141 concentrations of around zero µmol kg -1 were too low to support the sink functionality of the Peruvian In order to quantify the possible impact of preformed nutrients on the air-sea gas exchange, we calculated 151 pCO2 on the basis of upwelling source water mass characteristics as obtained from our data (see Methods 152 section) and sea surface conditions (Table 2) by applying the CO2SYS program 39,40 , which is commonly 153 used for analysing the marine carbonate system 39 . Three different scenarios were taken into account to 154 demonstrate that the chosen source water mass characteristics comply with the obtained pCO2 155 climatology.
In an initial step, we estimated pCO2 of SACW and ESACW when being upwelled by using the 157 respective in situ source water mass characteristics such as temperature (SWT), salinity (SWS), DIC and 158 total alkalinity (TA) within isopycnal ranges of sq 26.5 -27.0 (Table 2). In a next step, we quantified the 159 effect of biologically-mediated CO2 uptake at the surface by assuming the complete consumption of all 160 nutrients in the cold upwelled water. Considering a constant Redfield-C:P ratio of 117:1 41 , the impact of 161 biology can be estimated by subtracting the amount of P-associated carbon (P*117) from the source 162 water's initial DIC concentration. Additionally, we consider its associated effect on TA using a factor of 163 ~0.12 42 . This results in pCO2 decreasing below atmospheric concentrations, to levels of ~335 µatm in the 164 NBUS and ~284 µatm in the SBUS. In a third step, the warming of upwelled water was considered by 165 using SST instead of the in situ SWT of the source waters. This indicates a temperature increase of 166 approximately 7°C and led to enhancements in pCO2 by 102 µatm and 97 µatm in the NBUS and SBUS, 167 respectively. The resulting pCO2 of ~437 µatm (NBUS) and ~381 µatm (SBUS) resemble our estimated 168 annual mean pCO2 of the NBUS (487.6 ±2.9 µatm) and SBUS (379.2 ±1.74 µatm). This gives 169 confidence in our outlined source water mass properties and the obtained CO2 climatology, and suggests 170 that deviations from the used classical Redfield-C:P ratio of 117:1, e.g., due to anoxic processes in bottom 171 waters and surface sediments on the NBUS shelf, are of minor importance on the regional scale in the 172 BUS 43-45 . More importantly, it shows that, in contrast to the NBUS, the biologically-mediated drawdown 173 of CO2 overcompensates for the warming effect in SBUS, explaining the opposing CO2 flux directions in 174 the two subsystems. In comparision to SCAW, higher Ppref concentrations of ESACW explain, in turn, the 175 stronger biologically-mediated drawdown of CO2 in the SBUS. 176 To further quantify this effect, we subdivided the biological impact by individually considering the 177 decrease in pCO2 through the use of regenerated versus preformed nutrients ( Supplementary Fig. 2). 178 Compared to a pCO2 of ~437 (NBUS) and ~381 (SBUS) µatm after complete nutrient consumption, the 179 utilization of regenerated nutrients merely results in a pCO2 of ~650 and ~662 µatm in the NBUS and 180 32.8% (650 -437 = 213 µatm) in the NBUS and 42.5% (662 -381 = 281 µatm) in the SBUS, which 182 resembles the mean share of Ppref to the total P of 29% in the NBUS and 47% in SBUS (Table 2). 183 In addition, we assessed the contribution of Ppref to the potential amount of CO2 transfer into the ocean 184 interior on the basis of new production rates from the BUS that were calculated e.g., by multiplying the 185 volume of upwelling waters 1,46 with corresponding nutrient inventories of the source water masses 31 . The 186 resulting estimates for the NBUS ranged between 180 and 650 g C m -2 year -1 31 which, given our defined 187 upwelling area of 377.400 km 2 , amounts to 68 -245 Tg C year -1 . With a mean share of Ppref to the total P 188 of ~29% in the NBUS (Table 2), the utilization of Ppref results in a new production of around 19.7 -71.1 189 Tg C year -1 . Considering the volume of upwelled waters in the SBUS 36 and corresponding nutrient 190 inventories of ESACW (Table 2) Table  237 1). Additional quality-controlled measurements on sea surface fCO2 from the Surface Ocean CO2 Atlas 238 (SOCAT) v2020 29 were converted into pCO2 67 and embedded into our analysis. All pCO2 measurements 239 were normalized to a reference year (2020) 68 by using a mean yearly change rate in seawater pCO2 of 1.5 240 µatm year -1 and multiplying it with respective observations for both upwelling regions (NBUS, SBUS). 241 All normalized measurements were distributed on a 0.1° x 0.1° grid, with each value representing the 242 average of all observations falling into the same grid cell. Overall, the extended data set on pCO2 records 243 used in this study covers the shelf and coastal areas along the continental margin off Namibia and South 244 Africa, while spanning a timeframe from 1986 to 2020 with over 250 000 data points within the area from 245 approximately 5°E to 18.7°E, and 16°S to 34.5°S. 246 Carbon flux calculation 247 Differences in the partial pressure of carbon between the sea surface ( %,'( ) and atmosphere ( %,)* ) 248 were used to determine carbon flux rates ( % ) using equation (1): 249 where . is the solubility coefficient of CO2 69 and represents gas transfer velocity of CO2 70 . The gas 251 transfer velocity was calculated following equation (2) is the Schmidt number of CO2 in seawater, 660 represents at 20°C water temperature and refers 254 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 using the parameterization after 256 Wanninkhof et al. 70 . Data on wind speed, SST and salinity were based on shipboard measurements 257 (Supplementary Table 1), which were distributed on a 0.1° x 0.1° grid and averaged over each grid cell. 258 The flux calculation was performed using the average sea surface pCO2, wind speed, SST and salinity of 259 all grid cells within the defined NBUS and SBUS region. Seasonal flux estimates were calculated using 260 gridded and averaged pCO2, wind speed, SST and salinity data collected during spring and summer 261 (September till March), and austral autumn and winter (April till August), respectively. 262

Water column sampling and analysis 263
The analysis of water mass characteristics and biogeochemical settings in the BUS was based on data 264 gathered during RV Meteor cruise M153, which covered on-and offshore areas of both subsystems and 265 hence allowed a direct comparison of both upwelling zones for one given timeframe. We collected CTD 266 profiles of temperature, salinity and oxygen, and defined the upwelling SACW and ESACW source 267 waters by using the definition provided by ref. 71 . The analysis of dissolved inorganic nutrients (phosphate 268 P, nitrate N) was carried out as outlined in ref. 44,72 . Furthermore, the calculation of preformed and 269 regenerated nutrients ( , GHIJ and , HIK respectively) was performed following ref. 58  affirming an accuracy of ±12 µmol kg -1 and ±4.3 µmol kg -1 for DIC and TA, respectively. 284

Statistical information 285
The uncertainty in average parameter calculations of pCO2, CO2 exchange coefficients and fluxes (Table  286 1) as well as biogeochemical characteristics of source waters (Table 2)  General Biology Vol.7 (Prentice-Hall, Inc., New York, 1942).   Table 1: CO2 exchange coefficients and flux rates for the northern and southern Benguela 516 Upwelling System. 517 Annual and seasonal mean air-sea CO2 gas exchange rates, with positive flux values indicating CO2 518 outgassing. Each CO2 exchange coefficient represents the average value calculated on the basis of 519 shipboard data gathered during the 14 cruises and those embedded into SOCAT v2020 29 , which we 520  Properties of source water masses presented as average parameters within isopycnal ranges of sq: 26.5 -527 27.0 within the NBUS and SBUS, based on CTD casts and nutrient analysis during RV Meteor cruise 528 M153. Sea surface temperature and salinity (<5 m water depth) are based on annual averages as estimated 529 from shipboard data gathered during the 14 cruises, and those provided by SOCAT v2020 29 . The effect on 530 sea surface pCO2 is estimated using CO2SYS 39,40 with input parameters on total alkalinity (TA), dissolved 531 inorganic carbon (DIC), source water and sea surface temperature (SWT, SST), salinity (SWS, SSS), total 532 and preformed phosphate (Ptotal, Ppref). Uncertainties are given as the standard error, with n number of 533 observations used for average calculations. 534 535 Units in *1 µmol kg -1 , *2 °C, *3 PSU.   Figure 1 Maps of pCO2 measurements of research vessel cruises within the Benguela Upwelling System used in this study. a Underlying cruise tracks of the 14 cruises (red) and those embedded in SOCAT v2020 29 (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. 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 method74 implemented in the Generic Mapping Tools (GMT).

Figure 2
Spatial and temporal variability of sea surface pCO2. a Average cross-shelf distribution of all pCO2 measurements (in μatm) of the northern part of the BUS (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 concentrations8 across the NBUS and SBUS regions, for austral spring and summer (blue) and austral autumn and winter (red). The dashed grey line in each subplot represents the atmospheric pCO2 of the reference year 2020 based on Mauna Loa records (414 μatm).
The shaded areas in each subplot represent the standard deviation.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.