Reduced Ocean Carbon Sink in the South and Central North Sea (2014–2018) Revealed From FerryBox Observations

Surface seawater carbon dioxide partial pressure (pCO2) in the south‐central North Sea was measured between 2014 and 2018 using FerryBox‐integrated membrane sensors on ships‐of‐opportunity. Average annual pCO2 variability was biologically controlled, with thermal effects modulating its amplitude. Deseasonalized winter trends of seawater pCO2 were positive (4.4 ± 2.0–8.4 ± 2.9 µatm yr−1), biogeochemically driven, stronger than the atmospheric pCO2 trend, and more pronounced than previous analyses. The trends calculated including all deseasonalized monthly averages were even higher (9.7 ± 2.8–12.2 ± 1.4 µatm yr−1). During our investigation, the southern study area became a stronger source and the northern part became a weaker sink for atmospheric carbon. Overall, average sea‐air CO2 flux in our study area, from the Skagerrak to the Southern Bight (53°N), changed from −0.75 ± 0.61 mmol m−2 day−1 in 2014 to +0.20 ± 0.96 mmol m−2 day−1 in 2018.

The variability of the carbonate chemistry of the North Sea, a large temperate shelf sea, has recently been investigated as part of large-scale studies of the Northwestern European Shelf Hartman et al., 2019;Kitidis et al., 2019). Other studies focused on seawater pCO 2 from ship-based observations in the North Sea have either been performed over short periods (Omar et al., 2010) or were restricted to the northern North Sea . The North Sea as a whole is exporting carbon to the North Atlantic through a process known as the "continental shelf pump"-the transfer of the carbon-enriched dense bottom shelf waters to the stratified subsurface open ocean (Thomas et al., 2004;Tsunogai et al., 1999). The North Sea is heterogeneous in its CO 2 saturation state, with northern (southern) regions being undersaturated (oversaturated) (Thomas et al., 2004). Both observations (Clargo et al., 2015) and models (Lorkowski et al., 2012) have suggested an ongoing weakening in the carbon uptake capacity of the North Sea with distinct drivers of the variability in the northern and southern parts. The large spatiotemporal variability means that in order to better understand the mechanisms and improve prediction ability, sustained pCO 2 observations in the North Sea are needed at a higher frequency and coverage than those already existing in Surface Ocean Carbon Atlas (SOCAT, Bakker et al., 2016Bakker et al., , 2020. In this study, the recent spatiotemporal variability of the surface seawater pCO 2 over a large area of the North Sea is analyzed using autonomous ship-of-opportunity (SOOP) sensor observations. We identify subregions based on stratification regimes and calculate their average annual cycles and deseasonalized pCO 2 trends. The relative influence of thermal and non-thermal effects on the overall variability is assessed. Finally, we investigate how the distribution and extent of carbon sink/source regions changed during the 2014-2018 study period and provide estimations on the change in average sea-air CO 2 fluxes.

Data
FerryBoxes are automated instrument packages, usually installed on commercial ships that make use of regular shipping routes to provide surface seawater measurements (Petersen, 2014). The pCO 2 measurements used in this study were taken with 4H-Jena Contros HydroC CO 2 -FT membrane-based instruments (manufacturer provided uncertainty of 1%) integrated with the FerryBoxes running on two cargo vessels, the Lysbris Seaways and the Hafnia Seaways. These two SOOP were regularly transiting the North Sea between 2014 and 2018, as part of Coastal Observing System for Northern and Arctic Seas (Baschek et al., 2017) monitoring efforts (Figure 1). The sensors have been regularly maintained, changed and recalibrated. The raw data were processed according to manufacturer recommendations, correcting for instrument drift (Fietzek et al., 2014). The Lysbris measurements have been evaluated and validated against a traditional showerhead equilibrator instrument in a recent intercalibration study (Macovei et al., 2021-details in the Text S1). The Lysbris data between April and June 2015 featured a strong baseline drift so they were excluded from this analysis (Macovei et al., 2021). Since Hafnia was not sailing during this time, a 2-month gap in the data exists.
The North Sea was split into regions based on the density stratification regimes defined from a 1996 to 2010 hydrobiogeochemical model simulation (van Leeuwen et al., 2015). The shapefiles are available at https:// www.researchgate.net/publication/309308832_Shapefiles_Shelf-wide_regimes_1996-2010. Nearly 60% of the valid FerryBox pCO 2 measurements in the North Sea were taken in one of these regions of stratification.
Seawater temperature and salinity were measured with Teledyne/Falmouth Scientific Inc. sensors, also integrated within the FerryBox (uncertainties of ±0.1°C and ±0.02 respectively). Since temperature is MACOVEI ET AL.   (van Leeuwen et al., 2015). Small areas belonging to the "Region of freshwater influence" (ROFI, yellow) close to Immingham, UK (star) are separated (ROFI-W, yellow hatched). Points of interest, where data was used for example calculations, are shown with diamonds colorcoded to specific regions. Data from the area contoured in brown were used to construct the diagram in Figure 4. not measured immediately after the intake, ∼0.37°C of warming can be expected to occur in the pipes (Haller et al., 2015). The temperature, salinity, and pCO 2 data are available from the European FerryBox Database (https://ferrydata.hzg.de/) and from the Pangaea Data Publisher (https://doi.org/10.1594/PAN-GAEA.930383). No corrections were applied to the temperature or salinity data.
We used atmospheric CO 2 data from the Mace Head Observatory in Ireland (World Data Centre for Greenhouse Gases, 2020), which we assume to be a lower-bound estimate of the atmospheric value (Derwent et al., 2002). We converted the dry mole fraction to partial pressure according to Humphreys et al. (2019) using the water vapor pressure formula of Alduchov and Eskridge (1996), alongside barometric pressure and dew point temperature in our study area downloaded from the ERA5 reanalysis product (Copernicus Climate Change Service, 2020). We used 10 m wind speed from the same reanalysis product.

Processing and Calculations
Data were first sorted into the five stratification regions based on their geographical position. Investigating the pCO 2 variability revealed big differences between the two main regions of freshwater influence (ROFI), off the western and eastern coasts of the North Sea. While their stratification regime is similar, our biogeochemical investigation required a further separation into ROFI-W and ROFI-E, respectively. Data were temporally averaged for every calendar month within the six resulting regions and mean annual cycles were calculated.
For each region, trends were computed after deseasonalizing by subtracting the 5-year average monthly mean from the individual monthly means, thus obtaining pCO 2 anomalies. This is similar to the Tjiputra et al. (2014) method, without the need to remove regional aliasing. Our regions are small, homogeneous in pCO 2 variability, and the data coverage throughout the study period did not produce spatial biases. Linear regressions against time were applied both for the entire time series as well as just for the winter months, here defined as November-February.
The effects of seawater temperature on pCO 2 variability were investigated by decomposing the annual cycles into an isochemical and isothermal component, similar to Keeling et al. (2004). The change in seawater pCO 2 due to the monthly change in sea surface temperature, that is the isochemical term, can be calculated according to the experimentally derived pCO 2 temperature sensitivity factor of Takahashi et al. (1993) as follows: where SST and pCO 2 are the monthly averages, and i and i−1 denote the current and previous months, respectively. The isothermal term is the difference between the absolute monthly change and the isochemical change term, and mainly represents changes in dissolved inorganic carbon (DIC) concentrations.
The sea to air CO 2 flux was calculated using code available online (https://github.com/mvdh7/co2flux)see Humphreys et al. (2019)-following the parameterization according to Wanninkhof (2014): where k is the gas transfer velocity, calculated using the Schmidt number and the quadratic wind speed, while α is a function of sea surface temperature and salinity.

Regional Variability
Seawater pCO 2 consistently dropped during the spring season and increased during the late spring-summer season. This cyclicity is a result of the combined effect of temperature and biological effects in mid-latitude marine environments (Jiang et al., 2013;Macovei et al., 2020;Omar et al., 2010;Takahashi et al., 2002). The phytoplankton spring bloom produces organic matter by consuming DIC in the surface seawater, which consequently lowers the pCO 2 to an annual minimum, in spite of sea surface temperatures increasing at the same time with an opposite effect on pCO 2 . Summer warming and remineralization of the organic matter subsequently increase the pCO 2 . Variations from this typical cyclicity are likely caused by advective or local effects such as river influence, coastal upwelling, or exchanges with surrounding water masses Frankignoulle & Borges, 2001;Kitidis et al., 2019).
All the long-term pCO 2 monthly averages in the permanently and seasonally stratified regions are below 400 µatm (Figures 2a and 2b). The spring drawdown is evident in these regions as the phytoplankton bloom is favored by stratification . The annual cycle in the permanently mixed and intermittently stratified regions does not display such a clear spring decrease (Figures 2c and 2d). Furthermore, these regions are situated in the south of our study area and show higher long-term pCO 2 monthly averages between around 370 and 460 µatm, consistent with past knowledge of the southern North Sea being an overall carbon source (Bozec et al., 2005;Prowe et al., 2009).
The intra-annual variability of seawater pCO 2 in the regions of freshwater influence (Figures 2e and 2f) Figure 1 with the additional separation into light and dark orange for region of freshwater influence (ROFI)-E and ROFI-W, respectively.
The lowest monthly averages were recorded off the Netherlands coast after the spring bloom of 2014 with values below 200 µatm. The ROFI-E region, influenced by the European continental runoff from rivers such as the Scheldt, Rhine, Weser or Elbe (Burt et al., 2016), is the only one where the multi-year monthly averages for April and May are below 300 µatm. Excess total alkalinity, brought in by river inputs, intertidal sediment pore water exchanges and bottom sediment biogeochemical interactions (Voynova et al., 2019), is known to increase the buffer capacity and strengthen the spring pCO 2 drawdown in this area (Omar et al., 2010). On the other hand, the ROFI-W region had consistently higher pCO 2 values and a less regular annual cycle. This is likely a result of the inputs of very high pCO 2 water from the Humber river which drains agricultural and industrial areas resulting in CO 2 production through estuarine microbial degradation of organic matter in the presence of high nutrient concentrations (Jarvie et al., 1997). Since the ships sailed to the estuarine port of Immingham, the measurements were influenced by the mixing of North Sea water with the very high concentration river end-member (Volta et al., 2016). This shows that while these two regions experience similar stratification, they differ in their biogeochemical characteristics.
The annual ranges of pCO 2 found in this study are consistent with past North Sea observations Thomas et al., 2004). We investigated the influence of thermal and non-thermal effects on seasonality. The isochemical term is characterized by a summer maximum (July in all regions except ROFI-W), driven by the sea surface temperature maximum. The isothermal term usually features a minimum in the spring and a maximum in the fall, suggesting the main drivers of DIC changes are biological production during the spring bloom and remineralization of the organic matter later in the year. The annual isochemical amplitude is only about 60% of the isothermal annual amplitude, except in the region of intermittent stratification, where the two terms are approximately equal. Thus, the seasonal cycle of pCO 2 is dominated by chemical/biological variability, with temperature effects moderating the amplitude, such as when warm late-spring temperatures counteract the pCO 2 decrease caused by the phytoplankton bloom and cold winter temperatures dampen the pCO 2 rise in surface waters due to remineralization. Other observations at higher latitudes have confirmed DIC variability as the main driving factor for pCO 2 (Olsen et al., 2008).

Trends
Using all available monthly anomalies, we found statistically significant trends of increasing surface water pCO 2 , ranging from 9.7 ± 2.8 µatm yr −1 in ROFI-W to 12.2 ± 1.4 µatm yr −1 in ROFI-E (Figure 3). Winter trends were in general smaller, ranging from 4.4 ± 2.0 µatm yr −1 in the region of permanent stratification to 8.4 ± 2.9 µatm yr −1 in ROFI-W, all statistically significant at the 0.05 significance level. Only using winter months minimizes the effect of primary production (Fröb et al., 2019) and, in the case of our time series, lowers the trends by excluding the strong spring drawdown at the start of our study period and the high spring and summer values toward the end. The year-round and winter trends in ROFI-W were not statistically distinct, but this region's unusual annual cycle indicates that biological influence cannot be avoided in trend calculations by simply using the November-February anomalies.
The trends were calculated by grouping the data into relatively large provinces. In order to test if the strong increases are observed on smaller geographical scales, we retraced our analysis first for data around five points of interest, chosen in the middle of the regions defined by van Leeuwen et al. (2015), where pCO 2 data availability was high (see Figure 1), and second by grouping all available data into 1° × 1° boxes. Similar strong positive trends were found for both of these methods (details in the Text S1). This suggests that the Central North Sea experienced a ubiquitous rise in surface seawater pCO 2 during the 2014-2018 period.
An important factor influencing seawater pCO 2 is seawater temperature. An increase in temperature can increase the pCO 2 without any chemical composition change (Takahashi et al., 2002). We checked if the observed positive pCO 2 trends are related to an increase in surface seawater temperature. We found statistically significant negative trends for FerryBox sea surface temperature in three of the six regions, the strongest decline being −0.3 ± 0.1°C yr −1 in the region of permanent stratification (details in the Text S1). This is surprising, since the general trend in the North Sea until 2006 was of warming (Lorkowski et al., 2012). However, temperature data from the independent ERA5 reanalysis product over the same period also show negative, albeit not statistically significant, trends (Copernicus Climate Change Service, 2020). The isochemical terms show no trend during the study period, suggesting that the pCO 2 trends are not driven by temperature changes.
The choice of start and end dates for short time series is important. Our study however took place during a period when the winter North Atlantic Oscillation (NAO) index was always in a positive phase (National Center for Atmospheric Research, 2020). The North Sea carbonate system is known to respond to NAO changes (Kuhn et al., 2010;Legge et al., 2020;Salt et al., 2013), but since there was no change in the NAO phase, we cannot determine to what extent the trends presented here are a result of a shift in large-scale climatic variability. Since it is likely that changes in biogeochemistry have strongly influenced our pCO 2 observations, the reduced nutrient inputs to the coastal North Sea in recent decades could be influencing the trends we identified, particularly in ROFI regions (German Environment Agency, 2020; van Beusekom et al., 2019).
The calculated seawater pCO 2 trends in this study are some of the highest reported in the literature for the North Sea, where long-term trends were found to be statistically insignificant or approximately parallel to the atmospheric trend (Laruelle et al., 2018;Omar et al., 2019). Similarly, in the neighboring subpolar North Atlantic, long-term trends did not exceed the atmospheric rise (Fröb et   both in the North Sea-a rise of 26-43 µatm between 2001 and 2005 for summer temperature-normalized data (Salt et al., 2013;Thomas et al., 2007)-and in the subpolar North Atlantic-between 5.8 and 7.2 µatm yr −1 for winter 2001-2008 data (Metzl et al., 2010) and as high as 14.1 µatm yr −1 for summer 1993-1997 data (Leseurre et al., 2020). Since other investigations in this region suggested different pCO 2 trends, we compared our data with analogous measurements available in the most recent SOCAT release . We concluded that our accelerated trends are determined by the fortuitous choice of start and end dates as well as the much higher data resolution, allowing the identification of statistically significant trends in regions where data are otherwise scarce (details in the Text S1). While the long-term observational trends in the North Sea analyzed by Becker et al. (2021) were similar to, or below the currently reported atmospheric trend of 2.3 µatm yr −1 (Friedlingstein et al., 2019), their shortest length trends calculated over the most recent years were higher (2-3.5 µatm yr −1 ), confirming the recent accelerated rise in seawater pCO 2 . Since all our seawater pCO 2 trends are stronger than the atmospheric trend over the 2014-2018 study period, we expect a weakening of the carbon uptake capacity of the North Sea.
Indeed, the extensive seawater pCO 2 increase in the North Sea has led to changes in the difference between seawater and atmospheric pCO 2 (ΔpCO 2 - Figure 4a) and implicitly the carbon uptake capacity of large regions in this shelf sea. The strong undersaturation in the spring observed at the start of our study period weakened, while the summer oversaturation in the southern North Sea strengthened. The divide between the summertime under-and oversaturated regions has shifted to higher latitudes during this time period. The spatio-temporal distribution of the CO 2 flux (Figure 4b) roughly matches that of ΔpCO 2 . The average sea to air CO 2 flux in our study area changed from −0.75 ± 0.61 mmol m −2 day −1 in 2014 to −0.07 ± 0.71 mmol m −2 day −1 in 2017 and +0.20 ± 0.96 mmol m −2 day −1 in 2018 (details in the Text S1). These recent estimates suggest a long-term weakening of the North Sea carbon uptake capacity when compared to the start of the century: −4.08 mmol m −2 day −1 in 2001/2002 (Meyer et al., 2018) or a climatological estimate of −8.5 mmol m −2 day −1 for the year 2000 (Takahashi et al., 2009). However, our estimates do not cover the northwestern part of the North Sea, which is known to be a stronger CO 2 sink. Integrating over the ∼2.2 × 10 11 m 2 surface area of our study region, we estimate that 0.72 Tg C were taken up here through sea-air gas exchange in 2014, while 0.19 Tg C were released into the atmosphere in 2018. Were these trends to continue, the North Sea could lose its annually averaged carbon sink status. On the other hand, negative feedback effects, such as the generation of alkalinity, could buffer the carbonate system and potentially dampen the current trends   The data were grouped into 1° latitude × 1 month bins and zonally averaged over the 3° longitude ( Figure 1, brown outline) before plotting. No seawater data were available between April and June 2015, indicated by the vertical gray band. The figures were produced with DIVA gridding in Ocean Data View (Schlitzer, 2020).
would also improve the North Sea's carbon uptake capacity (Holt & Proctor, 2008). It is unclear however if these processes will intensify in the future.

Conclusions
High resolution (spatial and temporal) pCO 2 measurements, such as the ones in this study, are particularly important in dynamic and heterogeneous coastal environments . Our observations revealed increasing North Sea seawater pCO 2 anomalies since 2014, a trend of CO 2 increase larger than the one in the atmosphere. This has consequences for regional air-sea gas exchange and surface seawater biogeochemistry, and here we showed a reduced efficiency in the carbon uptake efficiency. Increasing pCO 2 can lead to decreasing pH (Doney et al., 2009) and the potential seawater acidification could inhibit the metabolism of some phytoplankton species (Gao et al., 2019), affect nitrogen cycling (Blackford & Gilbert, 2007), or reduce future ability of CO 2 uptake (Humphreys et al., 2020).
We provided the most recent in situ pCO 2 observations in the south and central North Sea, which revealed an accelerated, non-thermally-driven increase in seawater pCO 2 compared to previous investigations. This data set is shorter than the 12 years timeframe suggested by Keller et al. (2014) as necessary to calculate seawater pCO 2 trends from large scale models, and much shorter than the 25 years timeframe found by McKinley et al. (2011) as necessary to discern long-term trends in observations from decadal variability. Furthermore, significant interannual variability is observed in the SOCAT data from the North Sea. However, the FerryBox project is ongoing and improving and the data set will be developed in the future (Petersen & Colijn, 2017). The continuation of these time series will enable revisiting this topic to ensure the calculated trends are not an artifact of decadal variability.

Data Availability Statement
The data are available from the European FerryBox Database (https://ferrydata.hzg.de/) and the Pangaea repository (https://doi.org/10.1594/PANGAEA.930383). The Surface Ocean CO 2 Atlas (SOCAT) is an international effort, endorsed by the International Ocean Carbon Coordination Project (IOCCP), the Surface Ocean Lower Atmosphere Study (SOLAS) and the Integrated Marine Biosphere Research (IMBeR) program, to deliver a uniformly quality-controlled surface ocean CO 2 database. well as the captains, officers and crews of Lysbris Seaways and Hafnia Seaways. Our sincere gratitude goes toward the FerryBox group engineers who regularly serviced the instruments. Thank you to Sonja van Leeuwen who provided the stratification regions shapefiles and to David Kaiser for useful discussions. The many researchers and funding agencies responsible for the collection of data and quality control are thanked for their contributions to SOCAT. FerryBox activities were partly supported by the EU project JERICO-NEXT (Grant agreement 654410), as well as funding from the Helmholtz Association. The authors thank the two anonymous reviewers for their constructive comments that improved the manuscript. Open access funding enabled and organized by Projekt DEAL.