Ship-based sampling was conducted from 5 to 27 June 2017 on the R/V Tan Kah Kee in the SCS basin (Fig. 1 and Table 1) under the umbrella of the CHOICE-C II project (Carbon Cycle in the South China Sea: budget, controls and global implications). We visited two mega stations (SEATS and SS1) and nine regular stations during the cruise. The in situ observation at station SEATS was conducted before a typhoon (Merbok) which potentially affected the biogeochemistry of the region, and the remaining stations were visited after the typhoon (listed in Table 1). To examine the spatial variability of 234Th, we sampled four closely clustered stations (H01, H06, H08 and H11) around station SS1. Seawater samples were collected using 12 or 10 L Niskin bottles attached to a Sea-Bird 911 conductivity–temperature–depth (CTD) profiler.

At the mega stations, high-vertical-resolution water samples were taken at a depth interval of 10 m within the Ez. For regular stations, lower-resolution (5, 25, 50, 75 and 100 m) samples were collected. Seawater volumes of 4 and 8 L were collected for total 234Th and particulate 234Th/POC analysis, respectively. Samples were collected using acid-cleaned 4 L fluorinated bottles and filtered onto quartz microfiber (QMA) filters (25 mm diameter, 1.0 µm pore size). An amount of 500 mL of seawater was also collected for nutrient analysis from the Niskin bottles. Ancillary parameters, including potential temperature, salinity and fluorescence, were accessed using a Sea-Bird CTD sensor. We calibrated the sensor-derived fluorescence with the Chl a concentrations from discrete samples using the equation Chl a (mg m-3) = 0.855 × fluorescence (R2=0.87, n=139; Fig. S1 in the Supplement).

In addition, we deployed an array of floating sediment traps for 72 h at 50, 100 and 200 m at both mega stations, SEATS and SS1, to collect sinking particles during the survey. Retrieval of the trap at station SS1 was precluded by unfavorable sea conditions. Consequently, we utilized sediment trap data acquired during a 53 h deployment in July 2019. Our choice of alternative data collection is unlikely to engender bias in our analysis, as evidenced by the limited interannual variability in 15N signals of sinking particles obtained from sediment traps (see details in Sect. 3.5). At each depth of stations SEATS and SS1, 12 cylindrical acrylic tubes (with a height of 50 cm and diameter of 10 cm) were assembled for different biogeochemical measurements. Before deployment, the tubes were filled with prefiltered surface seawater, and NaCl was added to supersaturation. After recovery, the tubes were stored at temperatures under 4 ∘C until the particles settled to the bottom. After removing the overlying supernatant, the particles were prefiltered through Nitex filters (120 µm pore size) to remove the visible zooplankton and were then collected on QMA filters (1.0 µm pore size) for elemental and isotopic analyses.

The small-volume (4 L) MnO2 co-precipitation method was used for the total 234Th analysis (Benitez-Nelson et al., 2001; Cai et al., 2006). The efficiency of thorium precipitation was monitored by 230Th. In detail, the seawater samples were acidified after collection and spiked with 200 µL of 230Th (17.38 dpm mL-1). After an 8 h period to allow equilibration between samples and tracers, the pH of seawater was raised to 8.05–8.20 using NH3 ⚫ H2O before 0.375 mL KMnO4 (3.0 g L-1) and 0.20 mL MnCl2 (8.0 g L-1) were added. The MnO2 precipitates were collected for a total 234Th analysis. The particles filtered for particulate 234Th from the seawater samples on a QMA filter (25 mm, 1.0 µm) were dried in the oven overnight under 45 ∘C. The filters were then packed with Teflon rings and disks (diameter of 23.5 cm, produced by RISØ National Laboratory, Denmark) covered by aluminum foil (density: 6.45 mg m-2) and Mylar film. A gas-flow-proportional low-level RISØ beta counter (model GM-25-5) was used for 234Th counting. The first count was carried out immediately after the samples were set up, and the second count was carried out after >6 months for the background measurement. All 234Th samples were counted for 1000 min each time. The 230Th recoveries were determined by inductively coupled plasma mass spectrometry (ICP-MS) (Agilent 7700x). The average of all the recoveries was 88 ± 12 % (mean ± 1σ, n=97, range 73 %–98 %). All 234Th data were recovery- and decay-corrected to the sampling time. The uncertainties of 234Th data were propagated from the counting error, uncertainty from recovery and detection efficiency. The 238U activity was estimated by the following equation assuming conservative behavior with respect to salinity (Owens et al., 2011): 1238U=0.0786×S-0.314.

Upon 234Th counting, the particulate samples were carefully removed from the disks and placed in glass dishes. Subsequently, the filters were dried at 50 ∘C for 24 h after adding 0.4 mL of HCl (1.0 µmol L-1) to remove inorganic carbon. POC and particulate nitrogen (PN) concentrations were determined by an elemental analyzer-isotope ratio mass spectrometer (EA-IRMS) system (EA: vario PYRO cube; IRMS: Isoprime 100). At station SS1, we conducted 10 replicate POC samplings at 5, 100 and 200 m water depth to investigate the precision of bottle-collected POC. Our results show that the standard deviations of our analyses were better than 13 %, which agrees well with the result from the JGOFS cookbook (Knap et al., 1996). The errors were included in the subsequent calculation of POC export fluxes. The particles from the sediment traps were treated the same as the suspended particles. The C and N contents and the isotopes of sinking particles were also analyzed by EA-IRMS.

The euphotic zone depth (Zeu or the Ez base, in m) is defined optically, based on Wu et al. (2021), as the depth where the usable solar radiation (USR) equals 0.9 % of the surface USR, which is close to the depth where the photosynthetically available radiation (PAR) equals 0.5 % of the PAR value at the sea surface. In situ Zeu during the cruise was obtained from profiling PAR data recorded by the optical sensor (Biospherical QCP2300-HP) on the CTD.

Nutrients were analyzed on board using a Four-channel Continuous Flow Technicon AA3 Auto-Analyzer (Bran-Lube GmbH). The detection limits for both N+N (nitrate plus nitrite – termed as dissolved inorganic nitrogen, DIN) and dissolved inorganic phosphorus (DIP) were 0.03 µmol L-1. The top of the nutricline in this study was defined as the depth at which the DIN concentration reached 0.1 µmol L-1 (Dore and Karl, 1996; Winn et al., 1995). The layers above and below to the base of Ez were defined as the as NDL and NRL, respectively.

The mass balance for 234Th in seawater can be described using Eq. (1) (Buesseler et al., 1992): 2∂AThtotal∂t=λAU-AThtotal-FTh+V, where FTh is the 234Th scavenging flux at the export horizon. AU and AThtotal are the 238U and total 234Th activities, and λ is the 234Th decay constant (0.02876 d-1); V, which is discussed below, is the term for physical effects, including advection and diffusion.

For particle export from the Ez, the deficit of total 234Th relative to 238U is integrated with depth to evaluate 234Th fluxes. Under the assumption of a steady state (∂AThtotal∂t=0) and no physical transport (V=0), the 234Th export flux from the Ez (FThEz) is integrated by Eq. (3) (as shown in Fig. 2): 3FThEz=∫0EzAU-ATh×λdz. Similarly, 234Th export flux from the base of the NDL, FThNDL, is estimated as follows: 4FThNDL=∫0NDLAU-ATh×λdz. However, the assumption of no physical transport needs to be verified before 234Th flux is calculated. In this study, the physical transport is estimated as follows: 5V=-u×∂ATh∂x-v×∂ATh∂y-w×∂ATh∂z+Kx∂2ATh∂x2+Ky∂2ATh∂y2+Kz∂2ATh∂z2, where u, v and w are the zonal, meridional and upwelling velocities, respectively; ∂ATh∂x, ∂ATh∂y and ∂ATh∂z are 234Th activity gradients from west to east, south to north and upward. Kx, Ky and Kz are diffusivities from west to east, south to north and upward, respectively, and ∂2ATh∂x2 and ∂2ATh∂z2 are the second derivatives of 234Th activity distributions (Benitez-Nelson et al., 2001; Cai et al., 2008; Buesseler et al., 2020b).

To better constrain the 234Th flux in the SCS basin, we estimated the horizontal and vertical transports of 234Th at the station SS1. The climatological w and Kz from modeling results (Gan et al., 2016) were applied to the equation to evaluate the impacts of vertical advection and diffusion on the 234Th flux. The vertical transport fluxes were -2.0 ± 0.4 and -11.4 ± 0.1 dpm m-2 d-1 at the bases of the NDL and Ez, respectively, which can be considered to be negligible (less than 10 %) compared to the vertical scavenging flux at the station SS1. This was in agreement with Cai et al. (2008), who also showed that the vertical term could be neglected for 234Th flux estimation in the SCS basin.

The apparent diffusivity around station SS1 is estimated as ∼4×105 cm2 s-1 (Okubo, 1971) from empirically derived oceanic-diffusion diagrams, and we simplified the horizontal diffusive term in Eq. (5) based on Benitez-Nelson et al. (2000) as follows: 6Vdiffusion=KxATh-H11-2×ATh-SS1+ATh-H012Δx2+KyATh-H08-2×ATh-SS1+ATh-H062Δy2‾. The Δx and Δy are the distance between the stations to evaluate the influences of physical terms (i.e., Δx is the distance between stations H01 and H11; Δy is the distance between stations H06 and H08). Δx and Δy were equal to 18 km in this study. Thus, the 234Th flux derived from horizontal diffusion was considerably low (approximately 0.1 dpm m-2 d-1).

The in situ horizontal current velocities at station SS1 from the acoustic Doppler current profiler (ADCP) exhibited a wide range from 0.01 to 0.3 m s-1 in the upper 200 m. Since these current velocities were measured instantaneously and since their timescales did not match those of 234Th (∼ 20 d), we applied model-derived time-integrated data (3-month average) to the equation instead. The model-derived u and v ranged from 0.007 to 0.2 m s-1 in the upper 100 m. Based on those velocities, the 234Th flux from horizontal transport was about 15 % of the 234Th flux estimated using the steady-state model in the upper 100 m, which is consistent with previous studies in oligotrophic ecosystems (e.g., Cai et al., 2008; Buesseler et al., 2020b). Thus, a 1D model assumption is applicable in this study for the subsequent 234Th flux estimation.

In this study, the 234Th-derived POC export flux was calculated using the following equations: 7FPOCEz=FThEz×POCEzPThEz, where FThEz, POCEzPThEz and FPOCEz are the 234Th flux, particulate POC/234Th ratio and POC flux at the Ez base, respectively; 8FPOCNDL=FThNDL×POCNDLPThNDL, where FThNDL, POCNDLPThNDL and FPOCNDL are the 234Th flux, particulate POC/234Th ratio and POC export flux at the NDL base, respectively.

Sediment-trap-derived POC export fluxes were calculated as follows: 9FPOC-Trap=POCMeasuredΔt×ATrapTube, where POC is the concentration of organic carbon on the particles collected by the traps, Δt is the duration of trap deployments, and ATrapTube is the area of the trap tube.

The profiles of temperature and salinity reveal distinctive hydrological features between stations in the SCS basin (Fig. 3). The surface mixed-layer depth (MLD, defined as the depth where the potential density σθ increased by 0.03 kg m-3 compared to the value at the sea surface; Cornec et al., 2021) at stations SEATS, A1, A2 and C1 was shallower (20–39 m) than at other stations (MDL > 40 m; Table 2 and Fig. 3). The shallower MLD and isoclines (i.e., thermocline and halocline) might indicate upward displacement of waters at those stations. Du et al. (2021) attributed such vertical shifts in isoclines to mesoscale processes or basin-scale circulation. Indeed, most of these stations (SEATS, A1 and C1) were under the influence of eddies during the sampling periods, as revealed by the sea level anomaly (SLA) map (Fig. S2 in the Supplement); modeling results indicate stations C1 and A2 were impacted by cold water sourced from the southwestern SCS basin and derived from upwelling off the coast of Vietnam (Fig. S3 in the Supplement) (Gan et al., 2016).

As shown in Fig. 3c, nutrients were depleted in surface waters until the top of the nutriclines, where concentrations started to rapidly increase. The depth profiles also show a clear relationship between nutriclines and subsurface nutrient concentrations. The four stations (i.e., stations SEATS, C1, A1 and A2) with shallower nutriclines correspond well with higher subsurface nutrient concentrations. For example, DIN concentrations at 125 m in these four sites ranged from 13.1 to 17.0 µmol N L-1, averaged at 14.0 ± 2.1 µmol N L-1; in contrast, DIN concentrations at the same depth in other sites with deeper nutriclines ranged from 6.5 to 12.1 µmol N L-1, averaged at 8.9 ± 2.4 µmol N L-1 (Fig. 3c).

The partitioned nutrient inventories (i.e., nutrient inventories within the NDL and Ez) also showed such a trend (Fig. 4a, b). The average Ez inventory of DIN was 196 ± 30 mmol N m-2 at stations with shallow nutriclines compared to 29 ± 19 mmol N m-2 at other stations, and the average inventory of PO43- (DIP) in the Ez at stations with shallow nutriclines was 13 ± 1 mmol P m-2 compared to an average of 3 ± 2 mmol P m-2 at other stations. The pattern of nutrient inventories across stations possibly results from vertical water displacement induced by horizontal divergence at the mesoscale and/or basin scale (Du et al., 2021).

Chl a concentrations at the four stations with shallower nutriclines were consistently enhanced in response to elevated nutrient levels, resulting in shallower depths of subsurface Chl a maxima (SCM; Fig. 3d) relative to other stations (55–80 m vs. 85–108 m). Chl a inventories at these stations with high nutrient inventories (23.6–52.2 mg m-2, average 29.6 ± 4.8 mg m-2) were significantly higher (p<0.05) than at others stations (8.0–22.8 mg m-2, average 14.0 ± 4.6 mg m-2; Fig. 4c).

Variations of total 234Th and Chl a versus depth are shown in Fig. 5. The activities of total 234Th ranged from 1.70 ± 0.05 to 2.73 ± 0.05 dpm L-1, with an average of 2.30 ± 0.31 dpm L-1 (n=97; Fig. 5), and all stations displayed similar patterns. Generally, 234Th was deficient relative to 238U in the upper Ez and was in equilibrium or excess at the base of and/or below the Ez. The 234Th deficit peaked within the NDL and largely diminished in the NRL, implying that a large amount of particle removal occurred in the NDL but that low export or high remineralization occurred in the NRL. The 234Th activity minimum (1.70 ± 0.17 dpm L-1) appeared at a depth of 25 m at station A1 (one of stations characterized by a shallow MLD and nutricline). 234Th activity at the stations surrounding station SS1 showed little spatial variability: the differences in 234Th activity were less than 0.1 dpm L-1 at the same depth.

Particulate 234Th ranged from 0.13 ± 0.01 dpm L-1 to 0.47 ± 0.01 dpm L-1 (with an average of 0.25 ± 0.11 dpm L-1, n=83; Fig. 6). At most stations, the profiles of particulate 234Th shared similar depth patterns with Chl a, with the maximum values appearing in the subsurface water, while at stations H01 and H06, particulate 234Th generally increased with depth in the upper 100 m and showed little station-to-station variability. The maximum of particulate 234Th appearing at both the surface and subsurface at station B2 suggested complicated biogeochemistry of 234Th on particles.

POC concentrations ranged from 0.83 to 2.5 µmol L-1 (with an average of 1.2 ± 0.44 µmol L-1, n=83; Fig. 6). At most stations, the POC concentration was low (with an average of 1.1 ± 0.2 µmol L-1) in surface water and generally increased with depth until it reached its maximum at the SCM layer, and then it decreased again with depth. However, at some stations (i.e., C1, B2), there were POC peaks that appeared in both the surface water and the SCM layer.

Calculated 234Th fluxes at different export horizons (i.e., the NDL and Ez base) are shown in both Table 2 and Fig. 4d. 234Th fluxes at the Ez base mostly ranged from 267 ± 69 to 839 ± 59 dpm m-2 d-1. 234Th fluxes at the NDL base ranged from 204 ± 57 to 624 ± 52 dpm m-2 d-1, which accounts for 69 %–100 % of 234Th fluxes at the Ez base. We found that the 234Th fluxes remained rather low, mostly <800 dpm m-2 d-1, during our study, which was close to the threshold for the validity of the steady-state assumption, as shown in many prior studies (e.g., Savoye et al., 2006; Resplandy et al., 2012). The sea surface Chl a also indicated that no bloom was observed during the survey in June 2017 (Fig. S4 in the Supplement), suggesting that the study area retained its biogeochemistry under the steady-state condition.

234Th fluxes at the Ez base were within the range of 62–1365 dpm m-2 d-1, similar to those found in prior studies in the SCS basin (e.g., Cai et al., 2008, 2015; Zhou et al., 2013, 2020a). Given that our high-vertical-resolution sampling mode was only applied to stations SEATS and SS1, we estimated 234Th fluxes at the Ez base by reducing the vertical resolution to a 25 m interval so as to be consistent with other stations. This exercise resulted in values of 490 ± 60 and 655 ± 71 dpm m-2 d-1, respectively, for stations SEATS and SS1 compared to 522 ± 43 and 631 ± 48 dpm m-2 d-1 under the high-resolution sampling mode. The low-resolution sampling thus might induce an uncertainty of less than 6 % for the 234Th flux. However, high-resolution sampling is essential in order to examine the partitioning of carbon export in the upper water column, especially for the oligotrophic ocean characteristic of low export fluxes.

Based on the high-resolution total 234Th pattern at stations SEATS and SS1, we first determined the 234Th deficit in the NDL, showing substantial particle scavenging and POC export at the NDL base at both stations, and we subsequently found similar patterns at the rest of stations where we estimated the partitioning in POC export fluxes.

Besides the 234Th–238U disequilibrium method, sediment-trap-derived 234Th fluxes at SEATS were 589 ± 2 dpm m-2 d-1 at the NDL base (50 m), representing over 50 % of the 234Th flux, and 830 ± 2 dpm m-2 d-1 near the Ez base (100 m). The trap-derived 234Th fluxes were higher, but within a factor of 2, than the fluxes derived from bottle-sampled 234Th (362 ± 34 dpm m-2 d-1 at 50 m and 471 ± 46 dpm m-2 d-1 at 100 m) at both export horizons. Although 234Th fluxes at the base of the NDL had rarely been quantified in prior studies, we estimated the particle-scavenging rate at the corresponding export horizon with the historical data of 234Th. Our recalculation using these literature data showed that 234Th fluxes at the NDL base averaged 349 ± 296 dpm m-2 d-1 (n=36) with limited spatial and temporal variations in the oligotrophic SCS. This is also consistent with the fact that these prior measurements of 234Th shared similarities in terms of activities in the NDL (Cai et al., 2015; Zhou et al., 2020a). Nevertheless, the partitioning in particle fluxes between the NDL and Ez, based on both techniques employed in this study, is similar, which further supports that our 234Th–238U-disequilibrium-based fluxes are representative.

Bottle-derived POC/234Th profiles in the Ez are shown in Fig. 7. They ranged from 2.6 to 15.7 µmol dpm-1 (with an average of 5.6 ± 3.3 µmol dpm-1, N=83), peaked in the upper 25 m and generally decreased with depth at all stations. POC/234Th differences between most stations gradually diminished with depth and converged near 4.2 ± 1.6 µmol dpm-1 at the base of the Ez. The decreasing pattern of POC/234Th was not observed at stations SEATS and H11 (as noted in Fig. 7a).

POC/234Th ratios from sediment traps were only measured at station SEATS and were 4.7 and 3.2 µmol dpm-1 at 50 and 100 m, respectively (Fig. 7b). These values are comparable with the bottle-derived POC/234Th ratios from the same site as that sampled during the cruise.

The δ15NPN values for the trap samples varied between 2.6 ‰ to 6.7 ‰ in the upper 200 m at stations SEATS and SS1, showing an increasing trend with depth. Specifically, the lowest δ15NPN of 2.6 ‰ was observed at 50 m within the NDL, and the δ15NPN increased to 4.7 ‰ at the Ez base (about 100 m). Below the Ez, the δ15NPN value increased to 6.7 ‰ at 200 m at station SEATS. A similar pattern of δ15NPN was also found at station SS1, with the lowest δ15NPN of 4.1 ‰ at 50 m, an intermediate value of 5.8 ‰ at 100 m and the highest value of 6.0 ‰ at 200 m. The observed δ15NPN values at both stations were comparable to previous results (3.3 ‰–7.3 ‰) from sinking particles collected by sediment traps in the upper 500 m around station SEATS (Kao et al., 2012; Yang et al., 2017). Yang et al. (2017) found a δ15NPN value of 4.9 ‰ at 100 m at station SEATS, which was very consistent with our observation at the same depth at station SEATS. These results suggest that interannual variations in δ15NPN from the upper ocean in the SCS may be limited, and the δ15NPN value at station SS1 from the cruise in 2019 could be comparable to that in this campaign. We thus diagnose the nutrient sources of sinking particles at stations with different environmental settings without focusing on temporal variability.

The particle flux at the NDL base was comparable (88 ± 11 %) to that at the Ez base. This vertical structure indicates that the NDL base should be a hotspot for particle scavenging. The trap-derived 234Th fluxes (589 ± 2 dpm m-2 d-1 at 50 m and 830 ± 2 dpm m-2 d-1 at 100 m) were slightly higher compared to bottle-derived 234Th fluxes (362 ± 34 dpm m-2 d-1 at 50 m and 471 ± 46 dpm m-2 d-1 at 100 m). The higher trap-derived 234Th fluxes might possibly be related to incomplete removal of zooplankton (Buesseler et al., 2020b). In addition, the discrepancy in between could be due to the difference in their timescales (Umhau et al., 2019). Regardless of the differences in 234Th flux estimations from the separate methods, the similar vertical partitioning from both bottle- and trap-derived 234Th fluxes indicated substantial particle scavenging at the bases of both the NDL and Ez in the oligotrophic SCS.

It is also interesting to note that, at stations with higher nutrient inventories, 234Th fluxes (362 ± 34–624 ± 52 dpm m-2 d-1, average 547 ± 107 dpm m-2 d-1 at the NDL base, and 522 ± 45–839 ± 59 dpm m-2 d-1, average 637 ± 120 dpm m-2 d-1 at the Ez base) are significantly higher (by approximately 100–200 dpm m-2 d-1) than those at other stations (210 ± 38–520 ± 31 dpm m-2 d-1, average 359 ± 90 dpm m-2 d-1 at the NDL base, and 204 ± 57–613 ± 42 dpm m-2 d-1, average 427 ± 105 dpm m-2 d-1 at the Ez base; Fig. 4d). This regional pattern of 234Th fluxes might result from differences in nutrient distributions, as 234Th has thus far been an indispensable tool to trace biogenic particle scavenging (Ceballos-Romero et al., 2022, and references therein). Whether these high and low 234Th fluxes would respectively drive similar POC export fluxes at stations with high and low nutrient inventories remains to be determined.