The CFE is comprised of an optical sedimentation recorder (OSR) interfaced with a conductivity–temperature–depth (CTD) equipped Sounding Oceanographic Lagrangian Observer (SOLO) float (Davis et al., 2001). The CFE dives repeatedly (daily in this study) below the surface to obtain OSR observations at three target depths. The CFE's optical sedimentation recorder (OSR; Fig. A1, Sect. A1 in the Appendix) awakes when the CFE reaches a target depth. Particles settle through a hexagonal celled baffle (1 cm opening, 5 cm length) into a high-aspect titanium funnel and settling column before depositing on a 2.54 cm diameter glass sample stage; the funnel diameter is 15.4 cm. Particles are imaged in three lighting modes: transmitted, transmitted–cross-polarized, and dark field. On first wake-up of a given dive, the sample stage is flushed with water and images of the particle-free stage are obtained. At timed intervals (∼ 20 min in data described here) the OSR repeats image sets, which register the sequential buildup of particles. After the predetermined number of image sets over ∼ 1.8 h, cleaning occurs and a new reference image set is obtained. After ∼ 5 h at a target depth, the OSR performs a final image and cleaning cycle and reference image set, and the SOLO surfaces to report GPS position, CTD profile data and OSR engineering data, and dives to its next target depth (Table 1). Every three dive cycles, the depth order is repeated; in the current study, target depths were 150, 300, and 500 m. A ballasting error in May 2012 led to the CFE operating 300 to 400 m deeper than planned. In the CFE described here, all images were stored aboard the OSR for post recovery analysis.

We also deployed an OSR tethered at 237 m below a drifting buoy array in January 2013. This system is referred to as BUOY-OSR in the text. Positions of CFE surfacing and of the surface-tethered BUOY-OSR are summarized in Table 1.

Images were taken under transmitted, transmitted–cross-polarized, and dark field illumination and were processed to yield three parameters that characterize particle abundance, according to the scheme depicted in Fig. 2. The parameters are particle attenuance, cross-polarized photon yield, and dark field counts. Attenuance is the best measure of particle loading as light is reduced exponentially as it propagates through the sample. Cross-polarized photon yield takes advantage of the birefringent property of calcium carbonate and is a measure of particulate inorganic carbon. Dark field counts reveal the color of particles. Details of the treatment of transmitted light and transmitted cross-polarized light images are covered in Sects. 2.2.1 and 2.2.2, respectively. Dark field images are normalized by illumination intensity and then transformed to a consistently scaled reflectance in counts above background. A full-resolution post-processing image of an aggregate particle from January 2013 is shown for all modes of illumination in Fig. 3.

Calibration samples from the BUOY-OSR system were not useful due to a 20-fold lower particle collection rate compared to that of the CFE (described below in Sect. 3.3). We thus estimate the amount of POC in our images in order to translate attenuance flux to POC flux.

Bishop et al. (1978) report information on the dry-weight density and geometric properties of particles of similar morphology as sampled here. Particularly, we use a particle dry-weight density of 0.087 g cm-3, and the Bishop et al. (1978) equation, which estimates aggregate thickness (h, in cm) as a function of equivalent circular diameter (d, in cm): h=0.052d+0.0045. We analyzed a series of five images collected on 20 January 2013 from 11:11:47 to 12:39:47 UTC (just before cleaning). The last image of this set had 65 % of the sample stage covered with particles. These images are provided in the Supplement for readers to independently analyze. The software package ImageJ (National Institutes of Health, USA – http://rsb.info.nih.gov/ij) was used to measure the equivalent circular diameter of aggregates. As aggregates overlapped in successive images due to the unexpectedly high sedimentation observed, successive images were stepwise subtracted in pairs prior to size analysis. Aggregate volume was calculated as the product of cross sectional area and thickness, h.

A total of 127 aggregates from 500 µm to 8.3 mm in size were enumerated in all five images. A size cutoff of 800 µm was used to minimize contamination of the analysis due to the slight expansion of aggregate dimension over time as they rested on the stage. Although size distributions were initially quantified to a lower limit of 20 µm, it is important to note that 97 % of the volume attenuance of the sample was accounted for by particles > 800 µm in this sample. For the remaining 63 particles larger than 800 µm, we calculated an aggregate volume of 0.113 cm3. To estimate the POC flux, we assume an aggregate dry-weight density of 0.087 g cm3 (e.g., from Fig. 22 in Bishop et al., 1978) and thus calculate the weight of aggregate matter as 0.0098 g. Organic matter at depths between 100 and 500 m is typically 60 % (range 50–70 %) of dry weight (Fig. 11 in Bishop et al., 1978). Thus, organic-matter (OM) weight is 0.0059 g. We estimate POC (as carbon) by dividing this mass by the OM : C conversion factor, 1.88 from Hedges et al. (2002). We divide by 12.011, the atomic weight of carbon, to yield a POC loading of 0.26 × 10-3 moles. The opening of the OSR funnel has a diameter of 15.4 cm, which yields an assumed trap collection area of 186.3 cm2. The time interval for collection of this sample was 1.84 h, or 0.0766 days. Combining this information yields a carbon flux of 183 mmol m-2 d-1. The average POCATN flux for all particles > 13 µm in size during the same time interval was 66.2 mATN cm2 cm-2 d-1. As mentioned above, over 97 % of cumulative sample attenuance was in the > 800 µm size fraction. The conversion factor for POCATN flux to POC flux is the ratio of 183 to 66.2 × 0.97 = 2.8. This conversion factor allows the estimation of POC fluxes observed for a comparison with other data from nearby waters.

Alldredge (1998) collected marine snow using scuba at euphotic depths of 10 to 20 m in the nearby Santa Barbara Basin. They imaged particles in the laboratory, derived maximum and minimum dimensions, and assumed an elliptical volume. From this they calculated the equivalent spherical volume (ESV in mm3). Samples were analyzed for POC and a regression formula was derived (POC (µg) = 0.99 × ESV0.52). If we use the ImageJ 2-D analyses of minimum and maximum aggregate dimensions in our images to compute ESV following Alldredge's methods, we get a summed ESV value that is 8.5 times higher than that obtained using our approach. If we use the Alldredge (1998) conversion of ESV to POC for each aggregate and sum over all particles > 1 mm in size, then the amount of POC is 17 times lower than our estimate. The Alldredge method yields a POC density for aggregates equal to 0.00020 g C cm-3, while our method yields ∼ 0.028 g C cm-3. The carbon density of aggregate particles derived from their equations differs from ours by a factor of 140. We are not sure if the marine-snow particles sampled by Alldredge (1998) in shallow waters were mostly empty of material, or if they deflate as they sink below the euphotic zone. Bishop et al. (1978) describe aggregates sampled from 100–400 m that appeared similarly loaded to those described here; for this reason believe that our method is internally most consistent.

We stress that the factor 2.8 used for the conversion of POCATN flux to POC flux is approximate. Equation (1) for aggregate thickness is the only applicable published study that we are aware of. The dry-weight density factor (0.087 g cm-3) was taken for aggregates that were similarly optically dense as observed in this sample. Bishop et al. (1978) also reported dry density values as high as 0.24 g cm-3. The organic-matter percentage for large particulates used in the calculation above is typical of values obtained by our group using in situ filtration. We note that Collins et al. (2011) report upper water column POC flux and mass flux data from surface-tethered sediment traps in the San Pedro Basin (Fig. 1), closer to the shore, which yield OM values closer to 30 %, with the rest being contributed by inorganic terrigenous and biogenic phases. This would require our estimate of dry-weight particle density to be proportionately adjusted upwards to account for the greater fraction of inorganic ballasting material and thus compensate for the reduced organic percentage. A factor of 2 adjustment of the POC : POCATN ratio either way will not change our conclusions regarding the high magnitude of the sedimentation rate observed in January 2013 by the CFE, nor will it change the relative difference observed between CFE and BUOY-OSR presented below.

Retrievals of chlorophyll and POC from observations made by the Moderate Resolution Imaging Spectroradiometer (MODIS) on NASA's Aqua satellite are provided on the NASA Distributed Active Archive Center (DAAC). Level 2 Local Area Coverage (L2-LAC) data at 1 km resolution for the period of the three expeditions were downloaded on April 4 2013 and processed using SeaDAS software. Chlorophyll and POC were retrieved for pixels within a 2 km search radius of 33.72∘ N, 119.5∘ W and were selected using the standard ocean color processing quality flags. Means and standard deviations were calculated and data were further selected based on the requirement of five valid pixels. Further analysis of spatial and temporal trends of chlorophyll within our study area is presented in Appendix Sect. A3.

Surface photosynthetically active radiation (PAR) estimated from MODIS Aqua data for January, March, and May expeditions was 25, 45, and 60 einsteins m-2 d-1, respectively. Corresponding values for diffuse attenuation coefficient for PAR (kPAR) from beam attenuation coefficient profiles (Bishop and Wood, 2009) were 0.068, 0.20, and 0.18 m-1. Euphotic zone depths corresponding to the 1 % light level were 68, 23, and 26 m. The euphotic zone depths based on the 1 E m-2 d-1 isolume were 50, 20, and 25 m.

Figure 5 (above) shows the time series of POCATN flux and PICPOL flux. In contrast to surface-tethered sediment traps (Martin et al., 1987), we found that the influence by/presence of swimming organisms was rarely significant in our image time series (see Supplement animations of CFE imagery). The highest POCATN flux was measured in January 2013, intermediate flux was observed in March 2013, and lowest flux was observed in May 2012. Similar temporal–depth systematics were observed for PICPOL flux (Fig. 5b); however, the relative ratio of PIC / POC flux was highest in March 2013.

The hydrographic context for our observations was established by CTD and transmissometer casts made to full water depth three to four times per day during each expedition. We found that mixed layer depths were mostly shallower than 10 m during all expeditions, reflective of the California coastal upwelling regime and net divergence of surface waters in the Santa Cruz Basin. We calculate properties of the upper 20 m, as these are most closely comparable to remotely sensed surface properties. Upper 20 m temperatures for January, March, and May averaged 12.6, 13.5, and 15 ∘C, respectively, reflecting seasonal warming. Over each cruise, averaged 0–20 m POC concentrations derived from transmissometer profiles (±SD; Appendix Sect. A2; Fig. A2) were 4.1 ± 0.7 (January), 17.5 ± 3.4 (March), and 15.4 ± 5.7 µM (May). POC was most variable in May 2012, with concentrations rising from ∼ 10 to 25 µM over the period of observations reflecting the progression of a plankton bloom southward from Point Conception. Corresponding euphotic zone depths (EZDs) based on the 1 einstein m-2 d-1 isolume (EZD for 1 % light level in parentheses) were 50 (68), 20 (23), and 25 (26) m, respectively (Sect. 2.4). January and March 2013 showed little temporal POC trend, indicating a quasi steady-state balance of photosynthesis and loss processes (Appendix Sect. A2; Fig. A2).

Figure 6 shows MODIS Aqua chlorophyll and POC time series, averaged MODIS chlorophyll and POC values for each cruise period and the previous week, averaged upper 20 m POC from CTD casts, and depth profiles of POCATN and PICPOL flux. Satellite-retrieved POC values agreed well with transmissometer estimates for March and May and were lowest in January 2013, although the January values were approximately 2.5 times higher than transmissometer estimates. Imagery of the area in January the week prior to our cruise showed a patchy (rather than organized) structure in chlorophyll distribution. During the time of the expedition, distributions were nearly spatially invariant on the scale of the study area (Supplement, Appendix Sect. A3, Fig. A3). From all perspectives, January 2013 had the lowest phytoplankton biomass but by far the highest fluxes of POC and PIC.

Averaged POCATN and PICPOL flux profiles provide systematic quantification of the depth trends of sedimentation. In January 2013, POCATN fluxes at depths of 150, 300, and 500 m were 67, 85, and 35 mATN cm2 cm-2 d-1, respectively; in March 2013 at similar depths, flux was between 6 to 7 mATN cm2 cm-2 d-1; values in May 2012 for the 450–900 m interval fell between 2 and 3 mATN cm2 cm-2 d-1 with no depth trend. PICPOL flux trends roughly followed those for POCATN flux. PICPOL fluxes at 150, 300, and 500 m depths were 96, 122, and 63 ppm cm2 cm-2 d-1 (January 2013) and 33, 15, and 16 ppm cm2 cm-2 d-1 (March 2013); the apparent decrease in PIC flux with depth from 2.8 to 0.5 ppm cm2 cm-2 d-1 in May 2012 may be a consequence of the low flux condition.

For both POCATN and PICPOL flux, it is also notable that the 1.8 h averaged fluxes for each depth ranged over a factor of 10 relative to their average for the May and March periods but ranged by less than a factor of 2 in the high flux period of January 2013. This suggests that sedimentation is more uniform when large aggregates dominate.

The profile of PICPOL / POCATN ratio (Fig. 7) shows a depth-dependent increase from 1.4 to 2.4 in January, indicating preferential loss of POC with sinking. In March, PICPOL / POCATN ratios slightly decreased with depth from 4.6 to 3.7. In May 2012, the PIC / POC ratio decreased from 1.5 to 0.3 between 430 and 900 m. The fact that PIC / POC can decrease with depth suggests preferential carbonate dissolution can and does occur in shallow waters, a finding consistent with Bishop and Wood (2008). These observations confirm that the relative contribution of carbonates to aggregate ballasting varies with time.

In January 2013, an OSR system, identical to the OSR on the CFE, was deployed at 237 m tethered below a buoy system at the surface (Appendix Sect. A4, Fig. A4). This BUOY-OSR system was engineered to collect samples needed to calibrate the POCATN and PICPOL data; however, this approach did not succeed as intended. The POCATN time series from CFE and BUOY-OSR are compared in Fig. 8. The BUOY-OSR yielded POCATN and PICATN fluxes of 3.6 mATN cm2 cm-2 d-1 and 2.2 ppm cm2 cm-2 d-1, respectively, approximately 25-fold and 50-fold lower than CFE fluxes at 144 and 320 m depths. At the same time, BUOY-OSR and CFE images showed that the two systems collected aggregates of a completely different morphology and size distribution (Fig. 9). The CFE images showed abundant 5 mm to centimeter-sized aggregates. The BUOY-OSR collected only fragments of these aggregates and few particles larger than 2 mm.

Figure 10 shows cumulative volume attenuance size distributions corresponding to samples collected in January 2013 by the CFE and BUOY-OSR. We selected CFE images for depths of 144 m (* in Fig. 8a), 320, and 507 m. We also selected the BUOY-OSR image (* in Fig. 8b) and another taken at the exact same time as the first CFE image. Other BUOY-OSR images were analyzed for the rest of the time series. CFE images were either the first or second image after stage cleaning since subsequent images had multiple cases of particle overlap. The BUOY-OSR images analyzed were from just before stage cleaning since no particles were overlapping. Size distribution analysis was performed using ImageJ software, with an attenuance threshold > 0.02 and a size threshold of at least two pixels for a particle. Particles were classified according to their equivalent circular diameter (ECD), with the lower size limit of 20 µm. Volume attenuance for each particle was computed (the product of particle area and attenuance). Results were summed cumulatively from largest to smallest particle size and then normalized by total volume attenuance (Table 2; Fig. 10). In January 2013, > 96  % of normalized cumulative volume attenuance (NCVA) was found in the > 1000 µm size range in CFE data at all depths. In contrast, BUOY-OSR results typically show less than half of the cumulative volume attenuance in aggregates larger than 1000 µm; this is consistent with the particles being fragments of the larger aggregates and consistent with the factor of 20 difference in observed attenuance flux. If we were to eliminate all aggregates larger than 1500 µm from January CFE data, then there is close agreement with the NCVA size distributions derived from BUOY-OSR data. During the period when current flow across the BUOY-OSR fell below 2 cm s-1, analysis of the BUOY-OSR image shows that nearly 50 % of the aggregate volume attenuance was in the 2000–5000 µm, i.e., large, size range. In other words, the surface-tethered OSR began sampling larger aggregates.

Although BUOY-OSR and CFE were operating at different depths, and were separated by ∼ 9 km in our study area (Fig. 1, Table 1), there was no indication in CTD or optics profiles that there was any discontinuity or advective feature at 237 m that would lead to biasing the BUOY-OSR results low. Nor was there any indication in remote sensing imagery (Supplement) of nearby fronts or eddy structures. Furthermore, spatial chlorophyll variability was less than 20 % in our study area (Appendix Sect. A3).

Figure 11 shows CFE NCVA-PSDs from all three expeditions compared with January BUOY-OSR data and further demonstrates that the CFE captured a greater loading of large particles than the BUOY-OSR. All lines of evidence indicate a major bias against aggregates larger than millimeter in size by the surface-tethered BUOY-OSR.

In the absence of calibration sample data, calculations based on particle geometry, thickness, dry-weight density, and organic-matter percentages (Sect. 2.3 based on Bishop et al., 1978) yield a scaling factor of 2.8, which can be used to convert the optical POCATN flux (mATN cm2 cm-2 d-1) to POC flux (mmol m-2 d-1). POC flux was thus estimated to be 190, 240, and 100 mmol C m-2 d-1 in January 2013 (at 144, 320, and 506 m), ∼ 18 mmol C m-2 d-1 in March 2013 (from 170 to 508 m), and ∼ 6 mmol C m-2 d-1 in May 2012 (430 to 900 m). We similarly estimate ∼ 10 mmol C m-2 d-1 at 237 m during the January 2013 BUOY-OSR deployment.

The high January 2013 POC flux values of 190 and 240 mmol C m-2 d-1 to depths of 144 and 320 m are remarkable from several perspectives. Firstly, cases where the imaging stage was almost completely covered with material in 1.8 h had been never before encountered in our previous deployments of CFEs for missions totalling over 100 days (mostly in summer time conditions). Secondly, these high particle accumulation rates were observed in the wintertime at a time when both remotely sensed and in-water optical measures of particle stocks were low. Thirdly, the magnitude of the January 2013 POC flux, estimated from CFE data, was approximately 8 times higher than the highest sedimentation rates measured using surface-tethered (at 100 and 200 m) and moored sediment traps (near 500 m) in the nearby Santa Barbara (Thunnel, 1998; August 1993 to September 1996) and San Pedro basins (Collins et al., 2011; January 2004 through December 2007). In both these studies, POC fluxes in moored trap samples collected over 4 years at a depth of 500 m were typically 3 mmol C m-2 d-1, with rare values as high as 15 mmol m-2 d-1. The much sparser surface-tethered trap observations (Collins et al., 2011) were no higher than 28 mmol C m-2 d-1. The finding of a high sedimentation rate during wintertime low-biomass conditions in California coastal waters has not been reported in the multiyear moored sediment trap observations.

The May 2012 fluxes of 6 mmol m-2 d-1 observed by CFE were the lowest of the study, at a time when surface waters had high particle concentrations, thus implying high recycling and shallow remineralization. The finding of low export beneath plankton blooms in coastal waters reinforces the findings in the open ocean that high surface biomass does not imply high export to depth (Bishop and Wood, 2009; Lam and Bishop 2007; Ebersbach et al., 2011).

The low surface biomass observed in January 2013 could indicate either a state of nutrient limitation of photosynthesis or a state of efficient transfer of primary produced organic matter to grazers and efficient particle export through the upper 500 m. We rule out nutrient limitation in our case, as mixed layers consistently shallower than 10 m and surface temperatures of 12.5 ∘C indicated that upwelling conditions were prevalent. Furthermore, we observed intense activity of marine mammals and sea birds feeding on abundant macro-invertebrates and fish during this period – which suggests that consumers were the source of the sinking aggregated material in the water column at this time. Many of the large aggregates seen in both January and March 2013 closely resemble abandoned appendicularian houses described by Alldredge and Silver (1988). Animations of CFE imagery are provided in the Supplement for further exploration.

Physically driven aggregative processes were not likely in January and March 2013, as winds were below 5 m s-1 and averaged 2.5 m s-1. In May 2012, winds were below 10 m s-1 and averaged 3 m s-1. Surface waves and swell were almost completely absent in January and March. Acoustic Doppler current profiler (ADCP) records showed that currents were dominated by tidal fluctuations and not by eddy structures. The low biomass condition observed in January is therefore most consistent with biologically mediated export.

We note that the 5-day running mean of satellite chlorophyll and POC levels appears to be nearly halved in a week (Fig. 6a, b), just before the study period in January. The decrease suggests that active grazing and export were in progress. Satellite imagery from January 2013 (Supplement) shows a patchy POC and chlorophyll distribution without obvious eddy structures or fronts nearby.

Our observations show that POCATN flux profiles (Fig. 6) attenuate with depth, albeit not following the classic Martin et al. (1987) curve fit (Φz=ΦZref/(Z/Zref)b, Z is depth, Zref = 100 m, Φ is flux, and b= 0.858). Using light-based euphotic depths of 50, 20, and 25 m for January, March, and May as the Zref values for particle remineralization (Buesseler and Boyd, 2009), we derive Martin's “b” factors of 0.36, 0.22, and 0.17, respectively, far lower than the accepted b value (Table 3). There are multiple limitations to this exercise; these include likely temporal variation of flux at a particular depth with time of day (see, e.g., Fig. 15 in Bishop et al., 1987), depths sampled being 100 m deeper than the base of the euphotic zone, and the limited time of the study. That said, it is clear that in January 2013, the export may have been high enough to overwhelm the capacity of detritus feeders to reduce the flux. In May and March, all evidence is consistent with high recycling of the sinking POC in waters shallower than our 400 and 125 m observation points, respectively.

The fact that the PIC / POC ratio can decrease with depth suggests that carbonate dissolution may occur in shallow waters of the Pacific, a finding consistent with Bishop and Wood (2008). Also, we note that living foraminifera and empty foraminifera shells were predominantly associated with large aggregates in January. While the occurrence of empty shells in aggregates would not be exceptional, the finding of living foraminifera suggests an active association with these particles.

Our comparison of fluxes recorded by the CFE and a surface-tethered BUOY-OSR in January 2013 found a factor of 20 difference in collection rate (Figs. 8 and 9), consistent with the loss of all aggregates > 1500 µm in size. CFE collection rates, ignoring the > 1500 µm fraction in March 2013 and May 2012, would be approximately 4 times lower and 3 times lower, respectively (Fig. 11). This was surprising in light of reports that there are less than “factor of 2” differences in sedimentation measured using surface-tethered particle interceptor traps (Martin et al., 1987; PITS) and neutrally buoyant sediment traps (NBST) deployed in oligotrophic waters near Bermuda (Buesseler et al., 2000; Stanley et al., 2004; Owens et al., 2013). In our study, the particles mostly absent from the BUOY-OSR were large aggregates in the millimeter to centimeter size class (Figs. 9, 10, 11). The only time when large aggregates appeared in the BUOY-OSR imagery was during periods when the flow of water across the mouth of the BUOY-OSR was below 2 cm s-1 (Fig. 8). Given that marine-snow aggregates typically settle at rates of 0.1–0.2 cm s-1 (Alldredge and Silver, 1988), the trajectory of the aggregates across the BUOY-OSR funnel would be almost horizontal, even in currents as weak as 2 cm s.-1 The large aggregates are thus likely discriminated against entering the centimeter-sized trap baffle openings, which are of the same size as the particles and we hypothesize that the aggregates simply bounce back into the flow after shedding a few fragments, rather than entering the trap.

Further support for a higher coastal sedimentation than is presently recognized is provided by Stukel et al. (2011), who worked just offshore of our study area in the California Current. They compared the new production (Eppley and Peterson, 1979) “f ratio” (the fraction of primary production supported by upwelled nitrate) to the 234Th disequilibrium-based “e ratio” (the fraction of primary production exported as particles), and the two should be equivalent in a food web in equilibrium. While methods agreed 300 km offshore, the f ratio exceeded the e ratio by factors of 5–7 at locations within 50 and 100 km from the coast. Simply stated, there is insufficient time for the 234Th (half life 24 days) deficit relative to its parent 238U to reach steady state in coastal waters. Subsequent surface-tethered particle interceptor trap deployments (Stukel et al., 2013) in the same area showed consistently low exports in this region where f ratios are typically high (Eppley and Peterson, 1979). Our location is closer to shore. Even taking into account factors such as food web and lateral transport processes, multiple lines of evidence strongly indicate that large > 1 mm sized aggregates are severely undersampled by surface-tethered baffled traps.

It is important to note that sediment traps have been the sole method to date for the assessment of carbon export in coastal waters. The single profile 234Th / 238U method is not applicable in the coastal environment since it relies on an assumption of a steady-state deficit and requires a timescale of multiple half-lives (24.1 d) to reach this condition. Although the 234Th / 238U method may be used to infer export from time series sampling of a water parcel, such a study has not been undertaken in nearshore California waters to our knowledge. Also, e-ratio estimates are calibrated with large particles sampled either using traps or pumps and the assumption is that the Th : C ratio of all large particle phases is the same. Both methods have been shown to have large biases; in situ pumps are discussed by Bishop et al. (2012), and results presented here show undersampling of large aggregates by surface-tethered baffled traps. As there is scant literature on the size dependence of Th–C in different large particle size classes in the coastal environment, it is thus impossible to know the accuracy of e-ratio export estimates.

Coordinated at-sea studies are required to intercompare the various methods of measurement of carbon export.

Data for plots, representative images used in this paper, and animations of image time series (at 1/4 resolution) are available as Supplement and are available from the author on request. Shipboard data sets, including meteorological, surface hydrography, and CTD cast data, were contributed to the Rolling Deck to Repository (R2R) program (http://www.rvdata.us) archives or, in the case of ADCP data, are available through the University of Hawaii.