12 Carbon System Imbalances in the Sargasso Sea

Anthony F. Michaels

Bermuda Biological Station for Research

Ferry Reach GE-01, Bermuda

(edited for this report by H. Ducklow)

The export of organic matter from the illuminated upper ocean to the deep sea is a fundamental part of the ocean carbon cycle. This export can occur via three processes: gravitational settling of particles, transport by mixing, and the vertical migration of zooplankton. Of these processes, the settling of large particles has been considered dominant (McCave ref) and sediment traps are regularly used to measure this flux. However, there are no direct methods for determining the accuracy of sediment traps; their acceptance has come through the consistency of the early trap data with existing paradigms (JGOFS report). At the U.S. Joint Global Ocean Flux (U.S. JGOFS) time-series station near Bermuda, we compare the sediment trap fluxes with independent estimates of the carbon cycle. The particle flux estimated by the sediment traps is nearly 7-fold lower than required by the seasonal changes in the other carbon species. We conclude that inaccuracy in the sediment trap flux accounts for a large part of this discrepancy, while advection may also play a role. We suggest that the accuracy of sediment traps, both in general and in each specific application, must be more carefully documented to justify their continued use and that a more complete understanding of the affects of horizontal advection is a prerequisite to closure on the carbon cycle.

The U.S. JGOFS Bermuda Atlantic Time-series Study (BATS) includes water column hydrography profiles collected on a biweekly to monthly basis, primary production measurements using trace-metal clean in situ techniques and freely-drifting sediment trap array deployments of 3 days duration every month (see Lohrenz et al., 1994, Michaels et al., in press for methods details). The trapping uses the cylindrical MultiPITs (Knauer, 1979) and follows protocols that are nearly identical to those used in all JGOFS process studies, the VERTEX program and many other studies. The BATS program began in October 1988 and measurements are made near a site 45 nm southeast of Bermuda (31°50'N, 64°10'W). We use data from this study to constrain the carbon mass balance in the upper 150 m of the ocean near Bermuda and compare these mass balance considerations with the data from the sediment traps.

There is a strong seasonal cycle in total primary production with significant interannual variability (Figure 12.1). Spring blooms are a result of the nutrients supplied by the deep winter mixing and the strength of the bloom is related to the duration and depth of mixing. Sediment trap fluxes at 150 m are less variable. During the first two years, the production peaks were coincident with peaks in trap collection (Asper et al., 1993); however, in subsequent years this pattern has been much less consistent, calling into question some of the earlier conclusions about the tight linkage between production and export (Asper et al., 1993). The annual mean organic carbon flux is 0.71 moles C/m2/y. This export is only 6% of the mean annual primary production of 11.9 moles C/m2/y. These low export rates are typical for the paradigm of an oligotrophic environment (Eppley and Peterson).

We compare these trap estimates of carbon export rates with the mass balance for DIC in the upper 150 m. There is a strong seasonal cycle in upper ocean DIC; the 0-150 m integrated stocks are low in the fall and highest during the spring period of deep winter mixing when DIC rich water from below is brought back to surface. During the period between January and April, mixed layers usually extend below 150 m, making comparison of the hydrography and the 150 m traps more difficult. However, from April to the end of each year, mixed layers are less than 150 m and the vertical mixing is slowed by the stratification. Over this period, there is a very consistent decline in the integrated stock of DIC, an average drawdown of 20 umoles/kg or 2.6 moles C/m2 in each year (Figure 12.1). This DIC drawdown occurs in the absence of measurable nitrate and is an extreme example of the non-Redfield ratio DIC drawdowns observed in the North Atlantic (Sambrotto). Such non-Redfield ratio biogeochemical processes are one of the major conceptual uncertainties in our understanding of the oceanic carbon cycle. The magnitude of the seasonal change in DIC is in good agreement with other historical geochemical measurements of new production at this site (Sarmiento, Jenkins and Goldman, Spitzer and Jenkins). Recent seasonal measurements of DOC near Bermuda indicate that from April to December, the integrated 0-150 m stock of DOC also declines by of the order 0.35 moles/m2 (Carlson et al., submitted, Figure 12.2). Thus the total loss of suspended carbon is approximately 3.0 moles/m2. Over the same period, the sediment trap flux averages 0.45 moles C/m2. The carbon flux estimate from the sediment traps is nearly 7 times lower than the apparent loss of DIC and DOC.

There are five possible ways to resolve the difference between the DIC budget and the sediment trap flux. 1. The DIC is converted into organic carbon. 2. The carbon is exported by a different process than settling particles. 3. The traps collect the ambient field of sinking particles accurately, but the BATS sampling misses the major flux events. 4. The seasonal cycle at BATS is largely determined by the seasonally varying advection of different water masses. 5. The sediment traps at BATS are inaccurate.

The largest pool of carbon after DIC is the dissolved organic carbon pool which, as noted above, also declines over this period (Carlson, submitted). The stocks of particulate organic carbon are small (0.3 moles/m2) and relatively constant over the same period. Thus the 2.6 moles of DIC drawdown do not simply get transferred into an organic pool.

Gas exchange results in an annual flux of carbon dioxide from the atmosphere into the ocean of 0.5 to 1.1 moles C/m2/y depending on the year (Bates et al submitted). Much of this ingassing occurs in the winter and between April and December there would be a net transfer of DIC from the ocean to the atmosphere of 0.05-0.09 moles C/m2 (with a possible uncertainty of 2-4 fold depending on the choice of gas exchange coefficient).

Downward mixing of POC or DOC cannot reconcile the difference between the declines in the DIC stocks and the trap collections. DIC increases with depth and the vertical gradients of DIC at 150 m are much larger than for DOC or POC, both with concentrations that decline with depth. Thus, any physical mixing that transfers POC or DOC to depth would mix even larger concentrations of DIC to the surface and thus would reduce the seasonal disappearance of DIC. Vertical migration of zooplankton and the subsurface excretion of food ingested at the surface has also been suggested to transfer carbon to depth (Longhurst and Harrison ref). Measurements from the Sargasso Sea near Bermuda yield fluxes at 150 m of 0.25-0.75 mmoles C/m2/d (Longhurst et al ref), equal to 0.06 to 0.2 moles C/m2 for the April to December period. The BATS sampling involves a single, 3 day trap deployment every month. From 1989-1993, there are 458 three-day intervals for the period between April 1 and December 31 with BATS trap deployments on 45 of these. With this sampling frequency, there is a 95% probability that one or more of the BATS deployments would have co-occurred with a high flux event if there had been 30 of these events over the 5 year period and a 99% probability if there were 45 events. To balance the observed discrepancy, the average daily flux for each 3-day, high flux event would be 1666 or 1111 mg C/m2/d (for 30 and 45 events respectively). These rates exceed the highest measurements of total primary production in the BATS data set and the sum of the daily rates over the three day deployment would exceed the total stock of POC in the water column. The highest measured flux at the BATS site is 70 mg C/m2/d. Thus, it is statistically highly improbable that the difference between the predicted and measured trap fluxes could be due to imperfect sampling of an episodic particle flux.

Although the mean flow for the Sargasso Sea is small, it could result in the net transport of water masses over distances of 300-800 km on seasonal time scales, probably from the northeast as part of the Gulf Stream recirculation (Worthington). North-south transects over these distances indicate small variations in DIC, inadequate to explain the discrepancy. Much of the DIC drawdown and measurement discrepancy happens over a few months in the summer, a shorter period where local effects should dominate over net advection. However, it is not possible to completely rule out advection, a conclusion which would have important consequences for the conduct of biogeochemical field studies. If the entire signal at BATS is caused by horizontal advection, it implies that we will be unable to understand the ocean carbon cycle, even in a weakly advected regime, without a full, three-dimensional realization of the flow fields. Given the data and arguments above, we also conclude that the BATS sediment traps likely are inaccurate for the intended task of estimating the total particle flux at 150 m at this site. Particles dominate the export of organic matter at this site as the other transport mechanisms involve relatively little carbon. However, the sediment trap underestimates the magnitude of the predicted particle flux between April and December (and likely over the entire year) by nearly 700%. The similarity between the patterns of sediment trap data at BATS and at most other oligotrophic and mesotrophic sites suggests that these other studies may have uncertainties of similar magnitude, however, the quantity or quality of ancillary measurements has been insufficient in other studies to unambiguously make this comparison.

Trap inaccuracies can take three forms: swimmer biases, solubilization of particulate matter in the trap, and hydrodynamical biases. Swimmer removal is difficult and imperfect removal will increase the apparent flux. As we have improved our swimmer-removal methods at BATS the magnitude of our minimum fluxes has declined by 25% (Figure 12.1). Since late 1994, each trap sample is examined by two different people, the second fresh person concentrating on removing the very small and cryptic swimmers (Michaels et al., 1990).

Much of the difference between the BATS trap collections and the required total particle flux may be due to physical collection biases. These traps experience an apparent flow at the trap mouth as they are pulled through the water by the drag on the mooring line, subsurface floats and the surface flotation. At BATS, these flow rates vary from 3-30 cm/s and usually are between 10 and 15 cm/s (measurements on 20 different deployments over two years). Although these rates are largely lower than the recommended threshold for hydrodynamic biases (JGOFS report), other research indicates that collection biases are a relatively continuous function of flow speed over the range of speeds from 3-30 cm/s (Gust flume, Gust DSR).

The most direct assessments of the accuracy of drifting traps have used 234Th. This, particle-reactive, radionuclide is produced from a conservative parent isotope, 238U and the equilibrium activity of 234Th is determined by the production and decay rates. Deviations from equilibrium must involve the transfer of 234Th from one watermass to another in particulate form, usually on the same kinds of large sinking particles that mediate the fluxes of carbon.

A time-series of 234Th scavenging profiles and 234Th collection by sediment traps was collected on the monthly BATS cruises in 1993. Three 234Th profiles were collected on each cruise to allow an assessment of day-to-day variability or non-steady-state behavior of the 234Th. The 234Th profiles between April and December show a larger predicted export than actually observed by the sediment traps, particularly in June, 1993 (Figure 12.2). There is a strong seasonal pattern to the predicted export, a pattern that is absent in the trap collections. The traps appear to collect as little as 25% of the predicted flux in mid summer. We do not know that the trap collection biases for the particles which mediate 234Th fluxes are the same as for the particles which cause the export of organic C, so it is difficult to use the 234Th calibration to directly correct the carbon flux. However, the 234Th bias shows a pattern in the April-December period that is consistent with the export-deficit patterns for carbon. Thus these data strongly corroborate our conclusion that the drifting traps are underestimating the sinking particle flux during this time.

In summary, there are large discrepancies between the predicted carbon export on particles and the carbon flux measured by sediment traps at BATS. We use the wealth of ancillary data to conclude that, on average, the sediment traps are undercollecting particles between April and December, and likely for the entire year. Horizontal advection could conceivably account for some of the seasonal DIC changes and the resolution of advective processes is at least as difficult as resolving the accuracy of sediment traps. Many published disagreements about the relationship between particle fluxes and production near Bermuda can be resolved if sediment traps are determined to undercollect at this site (Jenkins and Goldman, Lohrenz et al., Michaels et al., 1994). Ultimately, a full three-dimensional realization of the carbon cycle will have to be performed (a control volume experiment) to finally determine both the level of our understanding of carbon cycle processes and the accuracy of our tools.