Rapid increase of surface water pCO₂ revealed by settling particulate organic matter carbon isotope time series during 2001–2009 in Sagami Bay, Japan

Abstract

Little is known about the rate of increase of coastal seawater pCO₂ (pCO₂^sea), despite its necessity for assessing future oceanic CO₂ uptake capacity. We examined temporal changes in pCO₂^sea in central Sagami Bay during 2001–2009. Weekly pCO₂^sea was reconstructed using time series of particulate organic carbon isotope delta (POC-δ¹³C) of settling particles at 150 m from moored sediment-trap experiments. For pCO₂^sea estimation, an empirical relationship between suspended POC-δ¹³C and aqueous CO₂ concentration from repeat ship observations in 2007–2008 was applied to the trapped POC-δ¹³C. Air–sea CO₂ flux was calculated using the air–sea pCO₂ difference with gas transfer velocity. Estimated Bay pCO₂^sea varied by 190 μatm (mean 294 μatm) and was mostly below atmospheric pCO₂ (pCO₂^air). The mean oceanic CO₂ uptake was 82 mg m⁻² d⁻¹, suggesting that Sagami Bay is an efficient sink for atmospheric CO₂. Meanwhile, carbon sequestration to the mesopelagic layer by particulate carbon export accounted for 60–75% of the CO₂ uptake, with the rest likely removed horizontally via surface water exchange. The pCO₂^sea showed an increasing trend of + 3.9 µatm y⁻¹, approximately twice that of pCO₂^air, and the two converged. Concurrently, a decreasing trend in POC export flux and an increasing trend in nitrogen isotope delta of trapped particles were found. Particularly, a large summer pCO₂^sea increasing rate (+ 4.9 µatm y⁻¹) was observed accompanied by POC concentration decreasing, which resulted in a decrease in CO₂ uptake over time. Long-term summer nutrient depletion and reduced primary production may increase pCO₂^sea in the Bay.

1 Introduction

Since 1850, the ocean has taken up 26% of total anthropogenic CO₂ (Friedlingstein et al. 2022), resulting in surface water pCO₂ (pCO₂^sea) increase and acidification over the past 3 decades. An analysis including millions of pCO₂^sea observations over a 37 year period from 1970 to 2007 (Takahashi et al. 2009) shows that contemporary oceanic water uptakes atmospheric CO₂ at 1.4 to 1.6 PgC y⁻¹, and that pCO₂^sea has increased at a mean rate of + 1.5 μatm y⁻¹, which is comparable to that of atmospheric pCO₂ (pCO₂^air), suggesting that air-sea CO₂ exchange is the primary control on surface water CO₂ regulation in oceanic areas. This rate provides a basis for predicting the future CO₂ uptake capacity of the ocean that defines pCO₂^air levels, but it does not account for coastal CO₂ exchange with much larger variations in both spatial and temporal scales than the open ocean.

Coastal waters receive large inputs of land-driven materials that undergo active biological and chemical transformations there, consequently exchanging carbon with a variety of other pools, including the atmosphere, biota, sediments, and open-ocean waters. There are limited near-shore locations where pCO₂^sea data are available with the high temporal resolution necessary to determine whether it is a source or sink of CO₂, and this imposes an uncertainty on the role of global oceans in atmospheric CO₂ exchange. Nevertheless, Chen et al. (2013) aggregated data from around the world and found that the global release of CO₂ from estuaries was 0.1 PgC y⁻¹ and that the continental shelves absorb 0.4 PgC y⁻¹. This suggested that their combined 0.3 PgC y⁻¹ absorption is equivalent to approximately 20% of the oceanic water CO₂ uptake, while coastal waters represent only 7% of the surface area of oceans. Roobaert et al. (2019), using updated high‐resolution (0.25°) monthly climatology data for pCO₂^sea, showed that the annual CO₂ uptake in coastal regions (excluding estuaries and inland water bodies) was 0.20 PgC y⁻¹, explaining approximately 12% of total oceanic CO₂ uptake. Although there were uncertainties in CO₂ exchange due to large areas that remain under sampled for pCO₂^sea, these results indicate a more efficient CO₂ uptake in the coastal areas than in the open-ocean areas.

In addition, recent studies (Kubo et al. 2017; Tokoro et al. 2021) with dense shipboard observation data sets revealed annual-scale CO₂ unsaturation in Japan's urbanized inner bays where sewage treatment reduces the organic carbon loading, and Kubo et al. (2017) predicted that by 2050, when urbanized coastal areas have expanded, CO₂ release by global estuaries will be less than 1/5 of current levels, resulting in increased total coastal CO₂ absorption. Thus, for future prediction of coastal CO₂ absorption, it is necessary to quantitatively assess the CO₂ exchange distribution over the entire area (including estuaries and inner bays) and to examine the long-term trend of changes in each area. In other words, further expanding pCO₂^sea monitoring to various coastal areas is required.

Moreover, to assess the sustainability of the large coastal CO₂ uptake, it is first essential to understand the mechanisms supporting its efficient carbon sequestration capacity. Although it is generally attributed to active biological activity in coastal environments (Chen and Borges 2009), there is insufficient quantitative knowledge on the fate of CO₂ absorbed: is the photosynthetically fixed organic carbon exported from the surface layer, or does it remain in the surface water or is it horizontally transported to the open ocean? Muller-Karger et al. (2005) concluded that more than 40% of the ocean carbon sequestration via the biological carbon pump (in which POC sinks) to depths below the permanent thermocline may occur along continental margins where the seafloor is shallower than 2000 m and accounts for 13.4% of the total oceanic area. It should be noted, however, that their results were based on the first-order estimates of POC flux calculated using satellite-based net primary production with a single empirical model of POC remineralization in the water column in both continental margin and open ocean. This is because it is very difficult to deduce POC flux distributions from a few measurements (which are conventionally obtained by sediment-trap experiment), especially at continental margins. Given the inhomogeneity of the coastal environment, the uncertainty in POC flux due to its variable attenuation in water would be extremely large. Ultimately, there are few measurement-based studies in coastal area, in which the relationship between local CO₂ exchange and POC transport to the ocean interior can be examined.

Marine-origin particulate organic carbon isotope delta (POC-δ¹³C) is known to record the availability of dissolved CO₂ ([CO_2aq]) when organic matter is produced (O’Leary 1981). This is due to the apparent magnitude of the ¹³C fractionation effect (ε_p: δ¹³C difference between the CO₂ taken up and the organic carbon formed photosynthetically) by ribulose-1,5-bisphosphate carboxylase/oxygenase (RUBISCO), which is regulated by the balance between CO₂ supply and demand (Farquhar et al. 1982). Strictly speaking, algal growth rates and cell size, which affect CO₂ demand and diffusive supply, respectively, also could constrain the ε_p (Laws et al. 1995; Popp et al. 1998), and if carbon-fixing enzymes other than RUBISCO, or active uptake of bicarbonate (with a δ¹³C distinct from CO₂) are involved as part of the carbon concentrating mechanisms (CCM) (Raven 1997; Burkhardt et al. 2001), they may have a significant effect on ε_p (Keller and Morel 1999; Laws et al. 2002). However, the relationship between POC-δ¹³C, ε_p, and [CO_2aq] has been recognized with a certain robustness as it applies to sedimentary POC-δ¹³C to reconstruct paleoenvironmental CO₂ (e.g., Rau et al. 1991a; Jasper and Hayes 1990; Bentaleb and Fontugne 1998).

A moored sediment-trap (MST) experiment was conducted from 2001 to 2009 at a depth of approximately 150 m at station S3 in the central part of Sagami Bay (Fig. 1; seafloor depth: ~ 1500 m) where monthly hydrographical observations were made by the Tokyo University of Marine Science and Technology (TUMSAT) at that time (e.g., Hashimoto et al. 2005; Hashihama et al. 2008; Mitbavkar et al. 2009). This shallow trap experiment was conducted as part of the projects “Profiling Ocean Primary Productivity Study (POPPS)” and “POPPS-2,” which aimed to develop a new autonomous, underwater profiling buoy system and to evaluate the biological pump efficiency of Sagami Bay using the time-series observations from this system combined with satellite observations (e.g., Saino 2007; Fujiki et al. 2008). The repetitive MST deployments over the 9 years, despite some challenges, provided a mostly continuous time series of trapped POC-δ¹³C and POC fluxes, which was well suited for reconstructing the surface pCO₂^sea in the deep bay and examining long-term changes in CO₂ exchange and POC export. Here, we aim to (1) examine a method for estimating [CO_2aq] using trapped particle POC-δ¹³C in Sagami Bay to reconstruct time series of surface pCO₂^sea during the MST experiment, (2) examine the seasonal change and long-term trends of surface pCO₂^sea, and (3) examine the quantitative relationship between CO₂ and POC fluxes.

Fig. 1

(a) Location of the study area, (b) an enlarged view of the study site in Sagami Bay showing the 100, 200, 500, 1000, 1500, and 2000 m isobaths. The star indicates the location of the sediment trap mooring S3, whereas the dot indicates the sampling station in the Bay. The open triangle on Oshima Island indicates the Japan Meteorological Agency weather station. In panel (a), three typical paths of the Kuroshio current are shown: A. typical large meander, B. near-shore non-large meander, and C. offshore non-large meander

2 Materials and methods

2.1 Moored sediment-trap (MST) experiment

The MST deployments were conducted 18 times between January 2001 and July 2009, in which a cylindrical trap (610 mm length/200 mm width) fixed to a cylindrical frame (SMD26S‐6000, Nichiyu Giken Kogyo, NiGK, Co. Ltd., Japan) was deployed at depths of approximately 150 and 600 m at station S3 (35.33°N, 139.33°E; Fig. 1) in central Sagami Bay from the training vessel (T/V) Seiyo-maru belonging to TUMSAT to examine fluxes and chemical properties of settling particles. Before deployment, the collection cup was filled with seawater-based 10% buffered formalin as a preservative. The sampling interval was set to 7 days. Following trap recovery, particle samples were subsampled for microscopic observations and the rest was stored in a refrigerator until further analysis on shore. For isotopic analysis, a quarter of the sample was filtered with Nucleopore polycarbonate filters (Millipore, USA) with a pore size of 0.4 μm and washed with pure water. Particles on the filters were dried at 60 °C for 36 h and weighed to calculate total mass flux, followed by grinding to obtain a homogeneous sample for subsequent chemical analysis.

2.2 Hydrographical observation and sample collections

Samples for POC-δ¹³C of suspended particles in the Bay were collected during eight cruises (KT-07-10, May 2007; KT-07-17, July 2007; KT-07-30, November 2007; KT-08-01, February 2008; KT-08-05, April 2008; KT-08-15, July 2008; KT-08-24, September 2008; and KT-08-29, November 2008) on board the research vessel (R/V) Tansei-maru, belonging to the Japan Agency for Marine-Earth Science and Technology (JAMSTEC). Approximately 20 L of surface water was collected with a bucket at each sampling station shown in Fig. 1 (however, the number of stations varied among cruises, Online Resource 2) and filtered through pre-combusted GF/F filters (Whatman, USA). The filters were kept frozen until analysis on shore. During three cruises, KT-08-15, KT-08-24, and KT-08-29, we conducted short-term, 2 to 4 d surface-tethered drifting sediment-trap (DST) experiments (Knauer et al. 1979) to collect settling particles at shallower depths than the upper MST depth. Cylindrical traps (620 mm length, 75 mm width) were filled with high-salinity filtrated seawater (S = 39, adjusted by addition of NaCl) and deployed on a drifter array in three layers between 40 and 140 m. Particles that had settled to the bottom of the traps were collected on pre-combusted GF/F filters and then frozen immediately after swimmers (i.e., mesozooplankton that entered actively sediment-trap collection bottles) on the filters were removed with tweezers. Direct underway pCO₂ measurements were conducted during three cruises in 2007 (May, July, and November), following the method used in Kayanne et al. (2005). The analyzer was calibrated using CO₂ standard gases (0 and 501 ppm) regularly during the cruises. On the five subsequent cruises in 2008, surface water samples for dissolved inorganic carbon (DIC) concentration and total alkalinity (TA) measurements were collected to estimate pCO₂^sea. Mercury chloride was added to water samples for DIC and TA to prevent biological activity.

2.3 Chemical analyses

Finely powdered MST samples as well as DST and suspended particles collected onto GF/F filters were exposed overnight to hydrochloric acid fumes to remove calcium carbonate (CaCO₃), dried in vacuum, and then pelletized with a tin disk. The POC and particulate nitrogen (PN) concentrations, and their isotope delta (POC-δ¹³C and PN-δ¹⁵N, respectively) in the pellets were measured with an elemental analyzer coupled to a continuous flow isotope-ratio mass spectrometer (EA1110-DELTAplus, Thermo Fisher Scientific, USA). The precision for PN and POC mass fractions analysis was better than 3% and 5%, respectively. The precision of δ¹³C and δ¹⁵N measurements was better than 0.15‰ and 0.2‰, respectively, estimated from repeated measurements of laboratory standards (Amino Standard, SI Science, Japan) along with the samples.

The samples for DIC and TA were analyzed in the laboratory using a flow-through analyzer (Watanabe et al. 2004). DIC and TA were measured to accuracies within 2–3 mol kg⁻¹, based on measurements of certified reference materials from the Scripps Institution of Oceanography (distributed by Prof. Andrew Dickson). The calculated pCO₂ had an accuracy and precision of ~ 10 μatm from TA-DIC evaluated using the method described by Grasshoff et al. (1999, Chapter 8). The direct underway pCO₂ measurement had an accuracy of ~ 5 μatm and precision with a similar range or better. [CO_2aq] was calculated from measured pCO₂ and the solubility equilibrium constant of CO₂ in seawater (Weiss 1974), which is a function of salinity and water temperature.

2.4 Other data

Sea surface temperature (SST) at station S3 from January 2001 to July 2002 was obtained from the mean of measurements taken at 0–5 m depths by Conductivity, Temperature, and Depth (CTD) profiler with rosette system (ICTD, Falmouth Scientific Inc., USA) during the monthly Seiyo-maru observations, and from then until July 2009 from the new generation sea surface temperature (NGSST) product provided by Tohoku University, Japan, by averaging over the area of 34.92°N–35.08°N, 139.25°E–139.42°E (approximately 18 × 15.5 km) centered on S3. Sea surface salinity (SSS) for the entire duration of trap deployments was obtained from the CTD measurement during monthly observations described above. Daily average wind speed data were provided by the Oshima weather station, Japan Meteorological Agency (JMA), located in the southern part of Sagami Bay (34.78°N, 139.36°E, elevation 74 m; Fig. 1). Daily mean atmospheric pCO₂ data from the inland pCO₂ station at Kisai, Saitama Prefecture (36.08°N, 139.55°E) were provided by The World Data Centre for Greenhouse Gases.

2.5 Calculation of air–sea CO₂ flux

The air-sea CO₂ flux was calculated as follows:

$${\text{CO}}_{{2}} {\text{flux}} = k_{{{\text{CO2}}}} K_{0} \left( {p{\text{CO}}_{{2}}^{{{\text{ sea}}}} - p{\text{CO}}_{{2}}^{{{\text{ air}}}} } \right) = k_{{{\text{CO2}}}} K_{0} \Delta p{\text{CO}}_{{2}} ,$$

(1)

where k_CO2 is the gas exchange coefficient calculated following Wanninkhof (2014). The pCO₂^air indicates atmospheric pCO₂, and ΔpCO₂ is the difference between pCO₂^sea and pCO₂^air. Note that negative values of the flux indicate oceanic CO₂ uptake. The time scale for each parameter and thereby the flux is weekly. The weekly pCO₂^air was derived from water vapor-corrected daily mean data. The weekly wind speed at 10 m above the sea surface, U₁₀, which is required to calculate k_CO2, was obtained from daily wind speed data from the Oshima weather station, corrected for altitude and further corrected using an empirical relationship derived from a comparison with actual sea winds measured at S3 during the cruise. Estimated weekly U₁₀ during the whole sediment-trap experiment period ranged from 3.1 to 10.4 m s⁻¹.

2.6 Calculation of deseasonalized monthly means

In this study, following Takahashi et al. (2006) and Wakita et al. (2021), we removed potential seasonal bias of sampling from the observed and estimated values and examined long-term changes in physical and chemical parameters. For example, the deseasonalized monthly mean of seawater pCO₂, pCO_{2 deseasonalized,} was calculated as follows:

$$p{\text{CO}}_{{\text{2 deseasonalized}}} = p{\text{CO}}_{{\text{2 est}}}^{{{\text{ month}}}} {-}p{\text{CO}}_{{\text{2 mean}}}^{{{\text{ month}}}} + p{\text{CO}}_{{\text{2 mean}}}^{{{\text{ annual}}}} ,$$

(2)

where $p{\text{CO}}_{{\text{2 est}}}^{{{\text{ month}}}}$ is the mean of estimated pCO₂ for each month, and $p{\text{CO}}_{{\text{2 mean}}}^{{{\text{ month}}}}$ and $p{\text{CO}}_{{\text{2 mean}}}^{{{\text{ annual}}}}$ are the monthly and annual mean values for the entire period of data acquisition, respectively. The rate of change of these deseasonalized monthly means that time series was calculated using a linear least-squares method and was considered statistically significant at p < 0.05. The error value for the rate is the standard error of the linear regression slope.

3 Results

3.1 Variations in POC export and POC-δ¹³C

The POC flux ranged from 3.9 to 513 mg m⁻² d⁻¹ with a mean of 53 ± 46 mg m⁻² d⁻¹ for the duration of all trap deployments, with a mean of 53 ± 44 mg m⁻² d⁻¹ for the period 2001–2008 (Table 1, Fig. 2a). Although sporadic spike fluxes occurred, exceptions included continuous high fluxes > 100 mg m⁻² d⁻¹ in April–May and September–December 2006. Excluding high fluxes in the top 5% (> 120 mg m⁻² d⁻¹), seasonal variation occurred with higher and lower fluxes in summer (June–August) and winter (December‒February), respectively. The POC-δ¹³C of trapped particles varied from − 28.0 to − 17.8‰ with a mean of − 21.9 ± 1.5‰ (Fig. 2b). Most of the high POC-δ¹³C values (> − 20‰) including the maximum were found in 2006, particularly during the high flux event from September to December. Shipboard observations conducted on 13 October 2006 revealed that large (> 50 µm) diatoms Pseudo-nitzschia spp. were abundant in the surface water of Sagami Bay at trap location S3. On the same day, the Bureau of Environment, Tokyo Metropolitan Government (https://www.kankyo.metro.tokyo.lg.jp/water/tokyo_bay/red_tide/download.html, in Japanese) determined that these diatoms were the cause of the red tide in the inner part of Tokyo Bay. In addition, surface chlorophyll (Chl) a spreading from within Tokyo Bay to central Sagami Bay was observed in satellite ocean color images (MODIS, Online Resource 1) during the period 16–23 October. These findings suggest that particulate organic matter with high δ¹³C (− 19 to − 16‰: Sukigara and Saino 2005; Sukigara et al. 2022) of planktonic origin within Tokyo Bay probably entered the trap in 2006. In contrast, the very low POC-δ¹³C data (< − 27‰, including the lowest value recorded) from December 2007 to early January 2008 appeared at the same time as a very strong cold wave under a strong to super La Niña event (with the bi-monthly NCEP–NCAR Multivariate ENSO Index of <  − 1.5); the shipboard observation immediately afterward (18 January 2008) reported that the surface layer at S3 was mixed deeply (i.e., beyond 120 m). This deep mixing may have contributed to the formation of particles with low POC-δ¹³C. Overall, POC-δ¹³C showed a seasonal cycle of low and high in winter and summer, respectively, with a mean change of approximately 3.5‰. Excluding the extreme values of POC-δ¹³C in June–December 2006 and December 2007, there was a significant declining trend (− 0.2 ± 0.0‰ y⁻¹, p < 0.001) for the deseasonalized monthly mean of POC-δ¹³C over the period 2001–2008, whereas there was no significant trend for the POC flux. However, when data from the first half of 2006, including outlier POC flux with high δ¹³C, were also excluded, a significant decreasing trend (p = 0.00654) of − 2.2 ± 0.8 mg m⁻² d⁻¹ (y⁻¹) was apparent in the POC flux (Table 2).

Fig. 2

Time-series data of (a) particulate organic carbon (POC) flux (bar), (b) particulate organic carbon isotope delta (POC-δ¹³C, closed circle), and (c) particulate nitrogen isotope delta (PN-δ¹⁵N, open circle) of trapped particles at 150 m in central Sagami Bay during 2001‒2009. Thick line in panel (a) indicates 6-week moving average of POC flux after excluding the top 5% high fluxes. Dashed lines in panels indicate statistically significant regressions of deseasonalized monthly means (p < 0.05). Note that these regressions were obtained after excluding all data from 2006 and the end of 2007, when extreme values of POC-δ¹³C were found (see text)

3.2 CO₂ dependence of suspended POC-δ¹³C and its application to trapped POC-δ¹³C

Suspended POC-δ¹³C in surface waters of the Bay sampled during 2007 and 2008 varied from − 25.6 to − 15.4‰ (Fig. 3). No clear spatial variation was found during the period when there was a deeper mixed layer depth (MLD) where the water temperature changed by 0.5 °C from the SST (e.g., 60 m at S3 in November 2007). However, during the period with stratified water structures with shallower MLDs of 7–10 m in July in 2007 and 2008, the POC-δ¹³C values were higher along the northern coast of the Bay and lower in the southern part facing the open ocean (Fig. 4e–h). The POC-δ¹³C also varied seasonally with higher and lower values found in summer and winter, respectively (Online Resource 2). Spatio-temporal variation in POC-δ¹³C of 10‰ was strongly inversely correlated with changes in surface water aqueous CO₂ concentration, [CO_2aq] (r = 0.84, p < 0.001, Fig. 3). This relationship is constrained by ¹³C fractionation associated with photosynthetic carbon fixation in the CO₂ supply and demand model (Farquhar et al. 1982): larger ε_p under higher [CO_2aq] condition results in POC production with lower δ¹³C, and vice versa. POC-δ¹³C vs. [CO_2aq] relationships have been used to reconstruct the CO_2aq environment at the time when organic matter was formed, and the slope of our relationship of − 1.04‰ μM⁻¹ was within the previously reported range (− 1.5 to − 0.6‰ μM⁻¹, Rau 1994; Fischer et al. 1998, Bentaleb et al. 1998). The magnitude of temporal change in suspended POC-δ¹³C at S3 was ~ 5‰, which followed well the POC-δ¹³C vs. [CO_2aq] regression derived from the dataset throughout Sagami Bay. This relationship was applied to POC-δ¹³C data for trapped particles obtained during the same period as the cruise observations in 2007–2008 to estimate [CO_2aq] and the corresponding pCO₂^sea when the trapped particles were formed. Those trapped POC-δ¹³C-based estimates were in good agreement with the actual pCO₂^sea measurements at S3 during the observations (RMSE = 28 μatm, n = 8; Online Resource 2).

Fig. 3

Relationship between suspended POC-δ¹³C and aqueous CO₂ concentration ([CO_2aq]) in surface waters collected during 2007–2008 in Sagami Bay. Closed and open circles indicate data from station S3 and others

Fig. 4

Distribution of surface water pCO₂ (pCO₂^sea) and suspended POC-δ¹³C in Sagami Bay in (a, e) July 2007, (b, f) November 2007, (c, g) July 2008, and (d, h) September 2008. Note that pCO₂^sea data in 2007 were obtained by an underway measuring system on board, while those in 2008 were calculated from both measurements of dissolved inorganic carbon and total alkalinity of surface water samples

3.3 Temporal variation of pCO₂ ^sea and CO₂ flux

The pCO₂^sea estimation method using the empirical POC-δ¹³C vs. [CO_2aq] relationship, described in Sect. 3.2, was applied to all trapped POC-δ¹³C data to reconstruct surface pCO₂^sea in the Bay from 2001 to 2009. The CO₂ solubility K₀ used here was derived from the SST and SSS at S3 for each particle collection period.

Estimated pCO₂^sea varied from 186 to 396 μatm (mean 292 ± 34 μatm) and showed seasonality, with lower and higher values in spring (March–May) and autumn (September–November), respectively (Fig. 5a, Table 1). During the 8.5-year period, it was consistently below the pCO₂^air level (361–391 μatm with a mean of 377 ± 7 μatm), with exceptions in September 2007 and September 2008. CO₂ flux varied from − 266 to + 25 mg m⁻² d⁻¹ with a mean of − 88 ± 51 mg m⁻² d⁻¹ (in carbon equivalents; Fig. 5b). Note that given the relative error of the pCO₂^sea estimate of 9% and that of k_CO2 of 20% (Wanninkhof 2014), the propagated error of the CO₂ flux was calculated to be 29%. Higher CO₂ uptake occurred in winter and spring compared to that in other seasons (Table 1). The correlation coefficients of k_CO2, K₀, and air–sea pCO₂ difference (pCO₂^sea – pCO₂^air, hereafter ΔpCO₂) to CO₂ flux were − 0.52, − 0.63, and 0.79 (all with p < 0.001, n = 359), respectively. Except for the second half of 2006 and the end of 2007 when the extreme POC-δ¹³C values (mentioned in Sect. 3.2) were observed, the deseasonalized monthly mean of pCO₂^sea between 2001 and 2008 showed an increasing trend (+ 3.6 ± 0.7 μatm y⁻¹, p < 0.001) attributable to a decrease in POC-δ¹³C. Furthermore, if the data for the first half of 2006 were also excluded, the rate was + 3.9 ± 0.7 μatm y⁻¹ (p < 0.001, Table 2). This is well above the increase of pCO₂^air (+ 2.0 ± 0.1 μatm y⁻¹, p < 0.001), resulting in an upward trend of + 1.9 ± 0.7 μatm y⁻¹ (p = 0.00750) for ΔpCO₂, indicating that the two were converging. Meanwhile, no significant trend for CO₂ flux was detected.

Fig. 5

Time-series data of (a) pCO₂^sea (circle) and atmospheric pCO₂ (pCO₂^air, solid line), and (b) air–sea CO₂ flux (bar) in the central part of Sagami Bay during 2001 to 2009. For CO₂ flux, negative values indicate oceanic CO₂ uptake. Open circles in panel (a) indicate pCO₂^sea calculated from "extreme POC-δ¹³C values" (see text). CO₂ fluxes associated with these extreme δ¹³C are presented as gray bars in panel (b). Dashed lines in panel a indicate statistically significant regressions of deseasonalized monthly means (p < 0.05). Note that these regressions were obtained after excluding all data from 2006 and the end of 2007, when extreme values of POC-δ¹³C were found

Table 1 Annual and seasonal means of atmospheric and seawater pCO₂, their difference (ΔpCO₂), air–sea CO₂ flux and underwater POC flux, trapped particle OC% and PN-δ¹⁵N, and the surface environment parameters, sea surface temperature (SST), and wind speed (U₁₀)

Table 2 The rate of change of deseasonalized monthly parameter means during the entire period (2001–2008) and in each season. Only the rate is presented when the trend is significant

4 Discussion

4.1 POC-δ¹³C-based pCO₂ ^sea estimation

The POC-δ¹³C signal with an inverse linear trend with [CO_2aq] has been used to estimate paleoenvironmental pCO₂ from sedimentary POC-δ¹³C records (Rau et al. 1991a; Jasper and Hayes 1990; Bentaleb and Fontugne 1998). For application of it in the Southern Ocean, where it may have contributed to glacial–interglacial changes in the control of atmospheric CO₂ (e.g., Francois et al. 1997), various relationships between POC-δ¹³C and [CO_2aq] from this region have been reported with slopes (i.e., the sensitivity of POC-δ¹³C to [CO_2aq]) of − 1.1 to − 0.6‰ μM⁻¹ (Rau et al. 1989, 1991b; François et al. 1993; Kennedy and Robertson 1995; Dehairs et al. 1997; Bentaleb et al. 1998; Lourey et al. 2004). Even steeper slopes (to − 1.5‰ μM⁻¹) of the regression have been found in the Northeast Atlantic (Rau et al. 1992) and the tropical–subtropical Atlantic (Fischer et al. 1998). Notably, there was a large difference in the range and variation of [CO_2aq] among datasets; however, the slope is likely to be steeper in regions where biological activity is the dominant control on [CO_2aq] than it is in regions with temperature (solubility)-controlled [CO_2aq]. The regression with a slope of − 1.5‰ μM⁻¹ presented by Rau et al. (1992) was obtained from time-course observations during a phytoplankton bloom in the NABE (North Atlantic Bloom Experiment) site, whereas the global (open-ocean) data compilation-based relationship by Rau (1994) showed a slope of − 0.63‰ μM⁻¹. The slope of our empirical regression for Sagami Bay data was − 1.04‰ μM⁻¹ (Fig. 3), which is in the middle of the previously reported values. This is probably due to the different features of the deep bay in a temperate region: active biological processes occur as are typical of coastal waters; meanwhile, the upper layer condition of the Bay, which has an annual SST cycle of that varies by 12 °C, is strongly influenced by open water (i.e., Kuroshio warm water; Hinata et al. 2003). The CO₂ dependence of POC-δ¹³C is possibly disturbed by algal growth rates and cell size as well as the presence of CCMs (Laws et al. 1995, 2002; Popp et al. 1998; Keller and Morel 1999). However, the strong correlation (r² = 0.71) between POC-δ¹³C and [CO_2aq] suggests that the compound effects of these changes were eventually minor, at least for surface water POM in Sagami Bay.

Suspended POC-δ¹³C in the euphotic zone is assumed to primarily reflect the δ¹³C signal of autotrophs (i.e., phytoplankton), whereas trapped POC-δ¹³C could be influenced by heterotrophic processes (trophic transfer and microbial degradation; DeNiro and Epstein 1978; Fry 1988; Aita et al. 2011, etc.). However, the 2–4 day surface-tethered DST experiments at S3 in July and November 2008 showed POC-δ¹³C of the settling particles at 40–140 m with − 22.1 ± 0.6‰ and − 22.8 ± 0.2‰, respectively, which were comparable to the surface water suspended POC-δ¹³C (with a difference by ~ 0.4‰), although POC-δ¹³C of DST particles in September (− 21.5 ± 0.1‰) deviated positively from suspended POC-δ¹³C by 1.1‰ (Online Resource 3). Moreover, the mean difference in POC-δ¹³C from the MST particles collected near the DST deployment period was − 0.2 ± 1.0‰. For a thorough comparison between MST, DST, and suspended particles, the differences in sampling period (MST: 7 days, DST: 2–4 days) as well as the time scale in which suspended POC-δ¹³C changes should be considered. However, overall consistency of POC-δ¹³C among them suggested that heterotrophic δ¹³C modification to trapped particles was small. Even if it had occurred, particles trapped at 150 m would have had a short time to undergo microbial degradation as they sunk from the euphotic layer. In fact, the mean of the trapped particulate organic carbon-to-nitrogen (C:N) mole ratio was 8.1 ± 2.2, which is substantially lower than that of deep layer trapped particles in coastal regions: mean C:N of 9.2 at 750 m at Tokyo Bay mouth (Sukigara and Saino 2005); 8.9–13.7 at ~ 1300 m off Mauritania, Atlantic Ocean (Fischer et al. 2021). The increases in trapped particle C:N with depth have been revealed by multi-layer MST deployments in oceanic time-series stations: e.g., BATS (31°40N, 63°10W), 7.8 ± 1.4 at 500 m and 9.7 ± 1.4 at 3,200 m (Conte et al. 2001); K2 (47°N, 160°E), 8.2 ± 2.7 at 500 m, 9.9 ± 3.8 at 4810 m (Mino et al. 2016; JAMSTEC K2S1 database, https://ebcrpa.jamstec.go.jp/k2s1/en/index.html). Given that a C:N ratio increases with the degree of degradation in which N-rich organic matter is consumed preferentially (Waples and Sloan 1980), the trapped POM by our shallow MST at Sagami Bay was likely composed of relatively fresh material.

Thus, nearly continuous collection of trapped POC-δ¹³C data with less heterotrophic δ¹³C modification and a 1-week resolution is of great importance as it was used to estimate pCO₂^sea in the central part of the deep bay where monitoring facilities are difficult to construct. The eventual estimation accuracy of 28 μatm is much larger than the pCO₂^sea measuring precision of approximately 5 μatm (Roobaert et al. 2019) and cannot contribute to a global database such as SOCAT (Bakker et al 2016). However, it is sufficient to evaluate changes in pCO₂^sea with a seasonal amplitude (maximum minus minimum) of ~ 150 μatm. Moreover, it is smaller than the increase of the deseasonalized pCO₂^sea over 8.5 years (i.e., + 3.9 μatm y⁻¹ × 8.5 years = 33.2 μatm), and thus, our detected increase rate of pCO₂^sea is substantial.

4.2 Evaluation of CO₂ uptake in Sagami Bay

Estimated surface pCO₂^sea in central Sagami Bay was below pCO₂^air for most of the observation period, with atmospheric CO₂ absorbed at a mean flux of − 82.1 mg m⁻² d⁻¹ or − 30.0 g m⁻² y⁻¹ (Fig. 5b). The mean CO₂ uptake flux for the entire Sagami Bay region was likely higher, because the pCO₂^sea measurements at S3 during stratified periods (July and September) were higher than those at near-shore stations on the north side of the Bay where active biological CO₂ consumption likely occurred (Fig. 4a–d; Online Resource 2). Nevertheless, the annual CO₂ uptake flux in central Sagami Bay is still 6.1 times higher than the global mean of climatological values (based on a vast volume of measured data) of − 4.9 g m⁻² y⁻¹ and 3.0 times larger than the Northern Hemisphere temperate (14°N–50°N) average of − 10.1 g m⁻² y⁻¹ (Takahashi et al. 2009). Moreover, it is comparable to, or slightly lower than, the considerably large CO₂ uptake of − 34 to − 39 g m⁻² y⁻¹ (Kubo et al. 2017; Tokoro et al. 2021) of the adjacent Tokyo Bay. Note that the k_CO2 scaling factors used in these flux calculations were almost the same: 0.251 for this study and Tokoro et al. (2021) and 0.26 for Takahashi et al. (2009). This adequate comparison suggests that Sagami Bay is an efficient sink for atmospheric CO₂.

The mean POC flux at 150 m, i.e., the amount of carbon exported from the surface layer to deep water, was 18.1 g m⁻² y⁻¹ (= 50 mg m⁻² d⁻¹), corresponding to 60% of the CO₂ uptake at the sea surface. In general, shallow MST experiments raise concerns regarding particle under-trapping due to hydrodynamic effects (Buesseler et al. 2007). Honda et al. (2015) reported that POC flux into cylindrical DSTs was on average three-to-four times higher than those into conical MSTs when comparing flux data at 200 m from both DST and MST experiments conducted in time-series stations of subarctic K2 and subtropical S1 in the western North Pacific. However, the POC flux vertical attenuation derived from our multi-layer (40−140 m) DSTs flux data at S3 was consistent with the MST flux data at 150−160 m (Online Resource 3). This consistency was partly attributable to our MST having a cylindrical mouth like DSTs though its aspect ratio (A, length/width) of 3.05 was lower than the DST’s A of 8.27. Under calm conditions, like when our DST experiments were conducted, surface-tethered DSTs have a comparable particle trapping efficiency to the neutrally buoyant sediment traps that are hardly affected by hydrodynamic biases (Buesseler et al. 2000). When these are considered, our MST-derived POC flux at Sagami Bay was likely close to the true flux, and far from undervalued.

Unfortunately, there were no data for particulate inorganic carbon (PIC) flux associated with CaCO₃ in the trapped particles. In the surface layer of S3 in central Sagami Bay, the Haptophytes-3 group (primarily coccolithophytes) was present with a minor contribution to total Chl a whenever the influence of Tokyo Bay water was small (Hashihama et al. 2008). Based on the particle fluxes from deep sediment traps in Sagami Bay reported by Masuzawa et al. (2003), the trapped POC:PIC flux ratio at ~ 1200 m was 66:15.8. Even if this ratio was applied to our POC fluxes at a shallower depth (a poor assumption, given that organic carbon in settling is more readily degraded than CaCO₃), the total carbon flux would still be 22.4 g m⁻² y⁻¹, only 75% of CO₂ uptake. Dissolved organic carbon could also be exported vertically, however, since it depends primarily on seawater sinking, i.e., downwelling (Walsh et al. 1992), we do not consider its contribution. Rather, DOC would be removed by horizontal transport. If steady-state conditions hold with respect to the surface layer carbon budget, then 25–40% of the CO₂ absorbed was probably removed by horizontal water exchange with waters from outside of the Bay.

The upper environment of Sagami Bay (~ 250 m) is influenced by the Kuroshio warm water intrusion through the Oshima west and east channels, which are closely related to the Kuroshio path rather than by the inflow of Tokyo Bay water. For example, when the Kuroshio takes a typical large meander (Fig. 1a), Kuroshio warm water strongly intrudes into the Bay through the west channel and then flows out through the east channel when southward to southwestward winds dominate in winter (Hinata et al. 2003). This flow may transport surface water containing abundant absorbed CO₂ in winter out of the Bay. Moreover, if the temperature of the inflowing Kuroshio water drops as it flows into the Bay, it would lower pCO₂^sea by increasing gas solubility, thus absorbing atmospheric CO₂ before it flows out of the Bay. This theory should be tested in the future. Although our estimated CO₂ flux had a relative error of 29%, the results may indicate that downward POC flux in the water was smaller than the atmospheric CO₂ influx to the ocean. This suggests that CO₂ absorbed in a given coastal area is not necessarily exported as particulate carbon to the mesopelagic layer; its fate is important to consider when evaluating atmospheric CO₂ sequestration.

4.3 Seasonal changes in pCO₂ ^sea and CO₂ uptake

CO₂ uptake was approximately twice as high in winter and spring than that in summer and autumn (Table 1). This is primarily because pCO₂^sea was lower in winter and spring, accompanied by a higher gas exchange coefficient due to higher wind speeds in winter. The seasonal variation in pCO₂^sea, with minimums in March–April, maximums in August–October, and an amplitude of 190 μatm, was sinusoidal and likely followed the annual cycle of 12 °C SST fluctuations with a time lag of ~ 1 month (Fig. 6a). This is because, in addition to the major effect of SST that determines gas solubility, changes in DIC (depending on vertical and horizontal DIC supply and biological DIC consumption) affect pCO₂^sea. Here, following Takahashi et al. (2002), we separated seasonal pCO₂^sea with changes in SST and DIC using the following equations:

$$p{\text{CO}}_{{2}}^{{{\text{ sea}}}} \;{\text{at}}\;T_{{{\text{obs}}}} = \left( {p{\text{CO}}_{{2}}^{{{\text{ sea}}}} } \right)_{{{\text{mean}}}} {\text{exp}}\left[ {0.0{423}\left( {T_{{{\text{obs}}}} - T_{{{\text{mean}}}} } \right)} \right],$$

(3)

$$p{\text{CO}}_{{2}}^{{{\text{ sea}}}} \;{\text{at}}\;T_{{{\text{mean}}}} = \left( {p{\text{CO}}_{{2}}^{{{\text{ sea}}}} } \right)_{{{\text{obs}}}} {\text{exp}}\left[ {0.0{423}\left( {T_{{{\text{mean}}}} - T_{{{\text{obs}}}} } \right)} \right],$$

(4)

where T is the SST and the subscripts “obs” and “mean” indicate the observed and annual mean values, respectively. A (pCO₂^sea)_mean of 296 µatm and T_mean of 20.6 °C were used. The experimentally determined temperature sensitivity factor, ∂lnpCO₂^sea/∂T = 0.0423 °C⁻¹ (Takahashi et al. 1993), was used. Here, we used the pCO₂^sea, which linearly excludes the increasing trend of + 3.9 ± 0.7 μatm y⁻¹ described above, with respect to that on 1 January 2005. The pCO₂^sea in Eqs. 3 and 4 were referred to as isochemical and isothermal pCO₂^sea, respectively (Keeling et al. 2004), and these seasonal amplitudes indicate the magnitudes of the effect of SST and DIC changes, which were calculated to be 155 and 144 μatm, respectively (Fig. 6b). The biological pCO₂^sea drawdown expected from a seasonal decrease of dissolved inorganic nitrogen (DIN) of 10–11 μM in surface water at S3 (Fig. 6c) with a Redfield C:N ratio of 106:15, Revelle factor of 10 (Sabine et al. 2004), and winter typical DIC of 2065 μM (measured in February 2008), was 129–142 μatm, which is in close agreement with the calculated seasonal amplitude of isothermal pCO₂^sea of 144 μatm. This supports the validity of the equation.

Fig. 6

Annual composite time series of (a) seawater pCO₂^sea (circle), atmospheric pCO₂^air (dot), and sea surface temperature (SST, cross), (b) pCO₂^sea values normalized to the mean annual temperature of 20.1 °C (isothermal pCO₂^sea; closed circle) and the mean annual pCO₂^sea values corrected for temperature changes (isochemical pCO₂^sea; open circle), and (c) trapped PN-δ¹⁵N and dissolved inorganic nitrogen (DIN) concentration in the surface waters at S3. The pCO₂^sea and pCO₂^air, SST, and trapped PN-δ¹⁵N were obtained during 2001–2009, while DIN concentrations were observed during 2001–2007. Note that the linearly de-trended pCO₂^sea values for the long-term change are presented in panel a and were used to calculate both isochemical pCO₂^sea and isothermal pCO₂^sea in panel b (see text). The PN-δ¹⁵N in panel c was also de-trended

The isochemical pCO₂^sea reached a minimum of 220–230 μatm in February–March owing to increased solubility with winter cooling, and then increased to ~ 390 μatm until August owing to the effects of seasonal warming. In contrast, isothermal pCO₂^sea had a maximum of 386 μatm in January–February owing to DIC enrichment associated with winter mixing, and then gradually decreased from March, 1 month before water temperature increases, to < 230 μatm in July when nutrients were depleted, primarily owing to biological consumption (i.e., phytoplankton CO₂ uptake). The accompanying decrease in DIN and summer depletion can also be seen as a gradual rise in trapped PN-δ¹⁵N during that period (Fig. 6c), which is because the remaining nitrate δ¹⁵N rises due to the ¹⁵N fractionation effect associated with nitrate uptake by phytoplankton (Miyake and Wada 1967). The active organic matter production resulted in high POC export flux in spring and summer (Table 1). Subsequently, isothermal pCO₂^sea increased again from October, probably due to DIC supplied by mixing. In fact, surface water DIN of more than 4 μM was found occasionally in October–November (Fig. 6c). The combined effects of these temperature and biological DIC changes determine the seasonal variation in pCO₂^sea. As in the temperate open ocean, the temperature effect exceeds the DIC effect (but only slightly) in Sagami Bay. However, the biological CO₂ drawdown effect in the Bay (144 μatm) is large on a global scale as effects exceeding 140 μatm are primarily found only in productive high-latitude oceans north of 40°N, and only below 50 μatm in the vast tropical and subtropical areas (Takahashi et al. 2002).

4.4 Long-term trends in pCO₂ ^sea and CO₂ uptake in Sagami Bay

Although there was no significant trend in the deseasonalized CO₂ flux in Sagami Bay, there was a significant upward trend in ΔpCO₂, the primary source of variation in CO₂ flux, of + 1.9 μatm y⁻¹ (i.e., pCO₂^air and pCO₂^sea converged by 1.9 μatm each year). If maintained, this is likely to lead to a decrease in CO₂ uptake in the Bay in the long term. The increase rate of pCO₂^sea (+ 3.9 μatm y⁻¹), which was approximately twice that of pCO₂ ^air (+ 2.0 μatm y⁻¹), is quite large compared to those in other ocean regions: the global open-ocean average during 1970–2007 was + 1.5 μatm y⁻¹ (Takahashi et al. 2009), whereas the highest basin-scale rate was approximately + 2.1 μatm y⁻¹ in the Antarctic Ocean. At stationary sites in the western subarctic North Pacific (47°N, 160°E and 44°N, 155°E), the rate was + 2.1 μatm y⁻¹ during 1999–2016 (Wakita et al. 2017), and for meridional transects from low to mid-latitudes (3°N–34°N, 137°E), the rate was + 0.8 and + 1.5 μatm y⁻¹ during the periods 1984–1997 and 1999–2009, respectively (Midorikawa et al. 2012). Even when restricted to the near-shore side north of the Kuroshio current (31°N–34°N, 137°E), the rate was + 1.5 μatm y⁻¹ during 1994–2008 (Ishii et al. 2011). Although there were some regional differences in pCO₂^sea increase rates for open waters, they have been considered comparable to the increase rate of atmospheric pCO₂, suggesting that air-sea CO₂ exchange was the primary control on surface water CO₂ regulation. In contrast, high increase rates have been reported for inner bay waters in Japan as well as Sagami Bay: + 5.8 μatm y⁻¹ in Tokyo Bay and + 3.2 μatm y⁻¹ in Ise Bay (Tokoro et al. 2021). There must be some cause for these remarkably faster increases of bay water pCO₂^sea compared to that of pCO₂^air, which is discussed below.

Such faster pCO₂^sea increasing trends have been found at several coastal stations located on the northwestern European shelf (3.0–3.5 μatm y⁻¹, Gac et al. 2021) and in the broad southeastern US coastal waters (3.0–3.2 μatm y⁻¹ on the middle and outer shelves, 3.2–3.7 μatm y⁻¹ in the coastal zone and on inner shelf, Reimer et al. 2017). Gac et al. (2021) suggested that 31–37% of the pCO₂^sea trend over the 2008–2020 period in the southern Western English Channel was caused by the effect of SST rise due to global warming (4.23% increase in pCO₂^sea per ℃, Takahashi et al. 1993). Reimer et al. (2017) also attributed the 26-year trend of pCO₂^sea increasing on the middle-to-outer shelf partly to increased SST, while it in the inner zone was due to the laterally advected high pCO₂^sea and organic matter-rich waters from land. However, such a thermal effect did not contribute to the pCO₂^sea increasing trend in Sagami Bay, because no significant trend is found in our SST dataset (2001–2009) at S3. Rather, the bay-wide average annual SST from another source (Tokyo Regional Headquarters, JMA; https://www.data.jma.go.jp/tokyo/shosai/umi/kaiyou/sst/index.html, in Japanese) shows a significant downward trend (− 0.10 ± 0.02 ℃ y⁻¹) over the same period. This is not a trend unique to Sagami Bay, but rather a decadal variability with a maximum in 2000 and a minimum in 2010 observed over a wide area around Japan (including the East China Sea and the Sea of Japan as well as the western North Pacific off the south coast of Honshu with Kuroshio), which was influenced by variations in the East Asian winter monsoon (Yoshita et al. 2020). In any case, faster increase rate of Sagami Bay pCO₂^sea should be caused by local non-thermal processes.

Long-term changing trends were also apparent in other parameters for trapped particles (where deseasonalized monthly means were used, Table 2): POC flux showed a decreasing trend of − 2.2 mg m⁻² d⁻¹ (y⁻¹) and PN-δ¹⁵N an increasing trend of + 0.2‰ y⁻¹ (p < 0.001, Fig. 2c). The interannual increase in PN-δ¹⁵N implies a decrease in nitrogenous nutrient availability (e.g., Sukigara and Saino 2006; Mino et al. 2020), which may have reduced primary productivity and led to a downward trend in POC exports to depth. Notably, the magnitude of this decreasing trend in POC flux is comparable to the decrease in the Bay CO₂ uptake by 1.7 mg m⁻² d⁻¹ per year presumed from the rate of change of CO₂ flux relative to ΔpCO₂ (0.89 mg m⁻² d⁻¹ (μatm⁻¹), Fig. 7) and the ΔpCO₂ increasing trend (+ 1.9 μatm y⁻¹). This may imply a link between the two (the supposed "biological pump"), although the POC export can only account for approximately 60% of the CO₂ uptake.

Fig. 7

Scatter plots between deseasonalized monthly means: air–sea CO₂ flux versus (a) ΔpCO₂, (b) U₁₀, and (c) SST. Open circles indicate annual means in 2001–2008 (except 2006). The solid line in panels a and b indicates a linear regression

Significant pCO₂^sea increase rate in each season was found except in autumn, with a particularly large increase rate of + 4.9 μatm y⁻¹ for summer pCO₂^sea (p < 0.001, Table 2). This increase, along with the effect of reduced wind speed, caused a large, statistically significant decrease (p = 0.0211) in CO₂ uptake at a rate of + 4.1 mg m⁻² d⁻¹ (y⁻¹). There was no significant long-term trend in POC flux; however, a decreasing trend in summer organic carbon concentration (OC%, − 0.52% y⁻¹, p = 0.0131) of trapped particles may also indicate reduced organic matter production. Although not detectable during the summer, the upward trend in PN-δ¹⁵N in both spring and autumn that suggests long-term nutrient depletion in the stratified upper waters, and the resulting reduced organic matter production was likely driving the high summer pCO₂^sea increase. This is also supported by the summer POC:CO₂ flux ratio of 1.07, which is highest among seasons, implying a more direct link between both carbon fluxes.

The question remains regarding the reason for reduced nutrient availability. It should be noted that nutrient data from monthly cruise observations between January 2000 and October 2007 did not show a clear downward trend in summer concentrations. This discrepancy with our speculation should be addressed in further studies, perhaps using available data not only from S3 but from other sites in the Bay to determine whether our proposed decreasing trend in nutrient availability and increasing pCO₂^sea is a Bay-wide or localized occurrence. This is because, as seen in Fig. 4, there was a non-uniform pCO₂^sea distribution in the stratified surface water during July and September, unlike during November with well-mixed column. As these are greatly influenced by the Kuroshio warm water intrusion and the associated changes in the current field in the Bay, the type of water mass distribution in the surface layer in summer may determine the CO₂ uptake ability of the central part or the Bay as a whole. As described above, Kuroshio warm water intrusion is sensitive to changes in the Kuroshio path, and therefore, it would not be surprising if a long-term trend in the frequency of inflow of Kuroshio water (likely with high pCO₂^sea and less nutrients) was found. If the contribution of Kuroshio warm water to the upper layer of the bay were to increase, there should be an upward trend in SST at S3, but this may be offset by the above-mentioned decadal SST decrease of waters around Japan, including Kuroshio warm water, during the period. For these analyses, the Bay-wide surface-current field based on high-frequency oceanic radar data would be useful.

5 Conclusions

Sediment trap experiments conducted at 150 m during 2001–2009 in Sagami Bay provide time-series data on POC settling flux and its δ¹³C that were little modified heterotrophically. We applied the empirical inverse relationship between surface suspended POC-δ¹³C and [CO₂] derived from multiple cruise observations to trapped POC-δ¹³C and successfully reconstructed a weekly time series of pCO₂^sea with 28 µatm accuracy over 8.5 years. Using time series of ΔpCO₂, k_CO2, and K₀, the air-sea CO₂ fluxes were also calculated, and their seasonal and long-term changes were discussed.

The estimated pCO₂^sea in the Bay varied by 190 μatm with a mean of 296 μatm, showing an annual cycle with minimum and maximum values in spring and autumn, respectively. Except for a few data points, pCO₂^sea was always below the pCO₂^air level, resulting in an oceanic CO₂ uptake flux with a mean of 82 mg m⁻² d⁻¹. This CO₂ flux is very large on a global scale, suggesting that Sagami Bay is an efficient sink for atmospheric CO₂. In contrast, the mean vertical POC export flux was 50 mg m⁻² d⁻¹, accounting for only 60% of the CO₂ uptake. Although there is 29% uncertainty in the estimated CO₂ flux, it is possible that some of the absorbed CO₂ was removed horizontally from the Bay via surface water exchange (e.g., associated with the Kuroshio warm water intrusion). If true, this process should be studied in more detail for evaluations of future coastal CO₂ sequestration capacity.

CO₂ uptake in winter and spring is approximately twice that in summer and autumn, primarily owing to low pCO₂^sea in winter and spring, accompanied by larger gas exchange coefficients from higher wind speeds. This seasonal variation in pCO₂^sea can be explained by the effects of temperature change via gas solubility and the effect of DIC change. In general, DIC changes depending on winter enrichment due to vertical mixing and biological consumption from spring onward. Following the method of Takahashi et al. (2002), the effects of temperature and DIC changes were separated. The calculated temperature effect on pCO₂^sea of 155 μatm was slightly larger than the DIC effect of 144 μatm. However, the impact of DIC change was large on a global scale, implying a strong pCO₂^sea regulating function of active biological activity in coastal environments.

Long-term trends in parameters were examined using time series of deseasonalized anomalies. During the observation period, there was no significant trend in CO₂ uptake in the Bay; however, the ΔpCO₂ tended to be higher (i.e., pCO₂^air and pCO₂^sea converged) by 1.9 μatm each year. This was primarily due to the pCO₂^sea having an increase rate of + 3.9 μatm y⁻¹, approximately twice that of pCO₂^air. This large increase rate of pCO₂^sea was accompanied by a concurrent decreasing trend in POC flux of − 2.2 mg m⁻² d⁻¹ (y⁻¹) and an increasing trend in PN-δ¹⁵N of + 0.2‰ y⁻¹. Significant pCO₂^sea increase rates were found except in autumn. In particular, a remarkably high increase rate of + 4.9 μatm y⁻¹ was found in summer, with a decrease in CO₂ uptake by 4.1 mg m⁻² d⁻¹ (y⁻¹) over time. In summer with a high POC:CO₂ flux ratio of > 1, the long-term decrease in nutrient availability and the consequent reduction in organic matter production are likely driving the high pCO₂^sea increase rate. However, we should consider whether this long-term trend in the upper layer is a Bay-wide or local occurrence, taking into account the current field and water mass distribution in the Bay. Regardless, it is necessary to monitor the pCO₂^sea in Sagami Bay to assess whether its faster increase over pCO₂^air as revealed in this study will continue in the future and reduce CO₂ absorption in the long term. From the perspective of data comparison, if possible, this should be conducted by a method based on sediment-trap experiments and stable isotope measurements, as in this study.