Investigating Particle Size-Flux Relationships and the Biological Pump Across a Range of Plankton Ecosystem States From Coastal to Oligotrophic

Sinking particles transport organic carbon produced in the surface ocean to the ocean interior, leading to net storage of atmospheric CO2 in the deep ocean. The rapid growth of in situ imaging technology has the potential to revolutionize our understanding of particle flux attenuation in the ocean; however, estimating particle flux from particle size and abundance (measured directly by in situ cameras) is challenging. Sinking rates are dependent on several factors, including particle excess density and porosity, which vary based on particle origin and type. Additionally, particle characteristics are transformed while sinking. We compare optically-measured particle size spectra profiles (Underwater Vision Profiler 5, UVP) with contemporaneous measurements of particle flux made using sediment traps and 234Th:238U disequilibrium on six process cruises from the California Current Ecosystem (CCE) LTER Program. These measurements allow us to assess the efficacy of using size-flux relationships to estimate fluxes from optical particle size measurements. We find that previously published parameterizations that estimate carbon flux from UVP profiles are a poor fit to direct flux measurements in the CCE. This discrepancy is found to result primarily from the important role of fecal pellets in particle flux. These pellets are primarily in a size range (i.e., 100 – 400 µm) that is not well-resolved as images by the UVP due to the resolution of the sensor. We develop a new, CCE-optimized algorithm for estimating carbon flux from UVP data in the southern California Current (Flux = ∑_(i=1)^x ▒ 〖 n_i A Abstract 31 Sinking particles transport organic carbon produced in the surface ocean to the ocean interior, leading to 32 net storage of atmospheric CO 2 in the deep ocean. The rapid growth of in situ imaging technology has the 33 potential to revolutionize our understanding of particle flux attenuation in the ocean; however, estimating particle flux from particle size and abundance (measured directly by in situ cameras) is challenging. 35 Sinking rates are dependent on several factors, including particle excess density and porosity, which vary 36 based on particle origin and type. Additionally, particle characteristics are transformed while sinking. We 37 compare optically-measured particle size spectra profiles (Underwater Vision Profiler 5, UVP) with 38 contemporaneous measurements of particle flux made using sediment traps and 234 Th: 238 U disequilibrium 39 on six process cruises from the California Current Ecosystem (CCE) LTER Program. These 40 measurements allow us to assess the efficacy of size-flux relationships for estimating fluxes from optical 41 particle size measurements. We find that previously published parameterizations that estimate carbon flux 42 from UVP profiles are a poor fit to direct flux measurements in the CCE. This discrepancy is found to 43 result primarily from the important role of fecal pellets in particle flux. These pellets are primarily in a 44 size range (i.e., 100 – 400 µm) that is not well-resolved as images by the UVP due to the resolution of the 45 sensor. We develop new, CCE-optimized parameters for use in an algorithm estimating carbon flux from 46 UVP data in the southern California Current (Flux = ∑ 𝑛 𝑖 𝐴𝑑 𝑖𝐵 ∆𝑑 𝑖 𝑥𝑖=1 ), with A = 15.4, B = 1.05, d = 47 particle diameter (mm) and Flux in units of mg C m -2 d -1 . We caution, however, that increased accuracy in 48 flux estimates derived from optical instruments will require devices with greater resolution, the ability to 49 differentiate fecal pellets from low porosity marine snow aggregates, and improved sampling of rapidly 50 sinking fecal pellets. We also find that the particle size-flux relationships may be different within the 51 euphotic zone than in the shallow twilight zone and hypothesize that the changing nature of sinking 52 particles

Sinking particles transport organic carbon produced in the surface ocean to the ocean interior, leading to 32 net storage of atmospheric CO2 in the deep ocean. The rapid growth of in situ imaging technology has the 33 potential to revolutionize our understanding of particle flux attenuation in the ocean; however, estimating 34 particle flux from particle size and abundance (measured directly by in situ cameras) is challenging. 35 Sinking rates are dependent on several factors, including particle excess density and porosity, which vary 36 based on particle origin and type. Additionally, particle characteristics are transformed while sinking. We 37 compare optically-measured particle size spectra profiles (Underwater Vision Profiler 5, UVP) with 38 contemporaneous measurements of particle flux made using sediment traps and 234 Th: 238 U disequilibrium 39 on six process cruises from the California Current Ecosystem (CCE) LTER Program. These 40 measurements allow us to assess the efficacy of size-flux relationships for estimating fluxes from optical 41 particle size measurements. We find that previously published parameterizations that estimate carbon flux 42 from UVP profiles are a poor fit to direct flux measurements in the CCE. This discrepancy is found to 43 result primarily from the important role of fecal pellets in particle flux. These pellets are primarily in a 44 size range (i.e., 100 -400 µm) that is not well-resolved as images by the UVP due to the resolution of the 45 sensor. We develop new, CCE-optimized parameters for use in an algorithm estimating carbon flux from 46 UVP data in the southern California Current (Flux = ∑ ∆ =1 ), with A = 15.4, B = 1.05, d = 47 particle diameter (mm) and Flux in units of mg C m -2 d -1 . We caution, however, that increased accuracy in 48 flux estimates derived from optical instruments will require devices with greater resolution, the ability to 49 differentiate fecal pellets from low porosity marine snow aggregates, and improved sampling of rapidly 50 sinking fecal pellets. We also find that the particle size-flux relationships may be different within the 51 euphotic zone than in the shallow twilight zone and hypothesize that the changing nature of sinking 52 particles with depth must be considered when investigating the remineralization length scale of sinking 53 particles in the ocean. 54 55

Introduction 56
Each year, approximately 40 -50 Pg of carbon dioxide (CO2) is fixed into organic matter in the 57 ocean via photosynthesis (Le Quéré et al., 2018). The majority of this fixed carbon fuels the surface 58 ecosystem and is quickly respired back into CO2, which then equilibrates with the atmosphere. A small 59 fraction of the organic matter produced by primary productivity escapes the euphotic zone and is 60 transported to depth, primarily as sinking particles (Ducklow et al., 2001;Siegel et al., 2016). This 61 process, known as the biological carbon pump (BCP), isolates carbon from the atmosphere for decades to 62 centuries (Volk and Hoffert, 1985), and is estimated to transport between 5 and 13 Pg C from the euphotic 63 zone each year (Henson et al., 2011;Laws et al., 2011;Siegel et al., 2014). Since marine photosynthesis 64 accounts for about half of global photosynthesis (Field et al., 1998), the BCP is a key component in 65 determining global and regional carbon budgets, which in turn are important for understanding climate 66 change and for predicting environmental changes in future climate scenarios. Unfortunately, due to the 67 numerous and complex processes that contribute to and influence the BCP, predicting its responses to 68 climate change remains difficult (Passow and Carlson, 2012;Boyd, 2015;Burd et al., 2016). 69 The BCP is comprised of a suite of processes including active transport by vertically-migrating al., in press). Quantifying the responses of the BCP to predicted changes in temperature, stratification, and 77 surface wind stress thus requires sustained measurements of sinking particle flux across a wide range of 78 marine ecosystems. 79 Sinking particle flux has traditionally been measured using sediment traps (Martin et al., 1987;80 Buesseler, 1991;McDonnell et al., 2015). These instruments have known issues associated with 81 hydrodynamic biases, particle dissolution, and contamination by swimming zooplankton (Knauer et al.,82 1984; Baker et al., 1988;Lee et al., 1992;Buesseler et al., 2007). However, short-term deployments using 83 surface-tethered traps with a large aspect ratio (height:diameter) and free-floating, neutrally-buoyant 84 sediment traps seem to give accurate estimates of particle flux from the euphotic zone. Unfortunately, 85 these short-term deployments require a substantial ship-time investment, because they typically require 86 that a large research vessel remain in the vicinity of deployment for a period of days. This substantial cost 87 has limited such time-series to only a few oceanic regions (Church et  Ohman, 2015). By microscopically quantifying fecal pellet flux and analyzing sinking particles and 133 aggregates collected in polyacrylamide gels we further investigate the processes that alter the particle 134 size-flux relationship. We then develop an algorithm, or more specifically a set of parameters to be used 135 in an algorithm, optimized for the CCE that can potentially be used to estimate particle flux using both the 136 UVP and fully autonomous instruments such as the Zooglider . We present evidence 137 that as our dataset encompasses a decade's worth of data in the CCE, including various frontal regimes, 138 mesoscale, and submesoscale phenomena, these parameterizations are far more accurate to the region 139 than values previously published. We caution, however, that more accurate quantification of flux from 140 I n r e v i e w optical measurements likely requires additional information beyond particle size spectra, including 141 characteristics related to the composition and identification of the particle or aggregate.  productivity, mesozooplankton abundance, carbon export) varied substantially between the offshore side, 171 coastal side, and core of these fronts. On the P1706 cruise our goal was to elucidate the role of mesoscale 172 filaments in offshore transport of nutrients, carbon, and plankton communities. Cycles on this cruise were 173 conducted over a reasonably large geographical area, as a result of the large spatial extent of the filament. 174 However, these cycles can be considered a relatively coherent analysis of the temporal progression of a 175 water parcel that is upwelled near the coast and ages over a period of weeks as it is advected from shore. 176 177

Sediment Traps 178
We used VERTEX-style drifting sediment traps with 69.85-mm inner diameter and an 8:1 aspect 179 ratio (height:diameter) topped by a baffle constructed from smaller beveled tubes with a similar aspect 180 ratio (Knauer et al., 1979). During P0704, cross frames consisting of 8 or 12 trap tubes were deployed at 181 100 m depth. On P0810, P1106, and P1208 cruises, traps were deployed at 100 m depth and also just 182 below the base of the euphotic zone (if the euphotic zone was shallower than 75 m). On P1408, P1604, 183 and P1706 cruises traps were deployed at the base of the euphotic zone, 100 m, and 150 m. Trap tubes 184 were filled with a hypersaline, poisoned brine solution (0.4% formaldehyde final concentration). Upon 185 recovery, the overlying seawater was gently siphoned before the samples were gravity filtered through a 186 200-μm filter. This filter was then examined under a stereomicroscope to enable manual removal of 187 swimming mesozooplankton taxa. After swimmer removal the <200-μm and >200-μm size fractions were 188 recombined and samples were split for multiple analyses using a Folsom splitter. Samples for particulate 189 organic carbon (POC) and nitrogen were filtered through pre-combusted glass fiber filters and frozen 190 prior to fumigation with HCl and analysis with a CHN analyzer or isotope ratio mass spectrometer.

191
Samples for C: 234 Th ratios were filtered through pre-combusted quartz (QMA) filters and analyzed as 192 below. Chl-a and phaeopigments were measured by the acidification method (Strickland and Parsons, 193 1972). Samples for biogenic silica were filtered through a 0.6-µm polycarbonate filter and analyzed using On the P0704, P0810, and P1604 cruises, samples (typically ~1/2 tube) were stored in the trap 199 brine (kept in the dark) and analyzed under stereomicroscope on land to enumerate and size 200 mesozooplankton fecal pellets. Samples were placed in a settling chamber and fecal pellets were allowed 201 to sink to the bottom for >6 hours. Water was then gently decanted through a 60-μm filter to ensure that 202 none remained in the overlying water. Pellets were then transferred to a gridded petri dish and random 203 grids were selected for analysis. Fecal pellets were identified by eye and imaged with a dedicated digital 204 camera. From calibrated images, the length and width of each fecal pellet was recorded and a shape-205 specific volume and ESD were calculated. A shape-specific carbon to volume relationship (determined 206 from CCE fecal pellet samples, Stukel et al., 2013) was applied in order to calculate the carbon content of 207 each pellet. 208 On the P1706 cruise we also deployed sediment trap tubes containing a polyacrylamide gel 209 designed to enable imaging of intact aggregates (Ebersbach and Trull, 2008; McDonnell and Buesseler, 210 2010). These trap tubes were identical to tubes used for other analyses, except that they were unbaffled to 211 ensure that there was no bias against >1-cm aggregates. The gel was contained in a custom-made acrylic 212 chamber with an 82.55-mm inner diameter (i.e., slightly larger than the trap tube above it), because on 213 previous cruises we found that material often clustered near the edge of the gel. Immediately after the 214 cruise, a series of images covering the entire gel was taken under both bright field and dark field at 7.5X 215 magnification (effective resolution 11.6 μm/pixel) with a Zeiss Discovery V20 stereomicroscope. 7.5X 216 magnification bright field images were manually stitched together using Photoshop software. Particles 217 were then manually circled and length, width, and area were extracted using Image J processing software 218 (dark field images were simultaneously viewed to aid in particle and aggregate identification). A random 219 set of locations on the gel was also imaged at 20X and 40X magnification. These higher magnification 220 images were used only for qualitative analysis of smaller particles, because non-uniform loading of 221 particles across the gel traps limited our ability to compute abundances from these images. Because the 222 P1706 cruise had considerably higher particle flux than measured on previous cruises (beneath a coastal 223 filament), particle loading at times obscured portions of the gel. Hence we were not able to image all of 224 the particles on the gel and do not consider our analyses to give accurate estimates of the total flux of 225 particles in any given size bin. Nevertheless, manual inspection of the images suggests no bias towards or 226 against any particle size classes, so we believe that the slope of the particle size spectrum determined 227 from the 7X magnification images of the acrylamide gels is accurate, although the intercept will not be. and/or near the deep chlorophyll maximum depending on water column structure. Samples were 234 immediately acidified and spiked with a 230 Th tracer. After an equilibration period, NH4OH was added to 235 adjust to a pH of 8-9 and manganese chloride and potassium permanganate were added. >8 hours later, 236 manganese oxide precipitate was vacuum filtered onto a QMA filter, which was dried and mounted on a 237 RISO sample holder.

UVP Measurements and Data Processing 281
The UVP was attached downward facing on the CTD Niskin-rosette, allowing for 282 contemporaneous particle imaging and water column characterization, typically on 10 casts per cycle. For 283 P0810, P1208, P1408, and P1604 the UVP5 standard definition model was used while for P1106 and 284 P1706 a zooming version and HD model were deployed respectively. Each instrument was cross-285 calibrated in the lab prior to deployment. The UVP images a volume of ~1 L with each image and takes 6 286 images per second. It was typically deployed to a depth of 500 m. Post-cruise, data were processed using 287 published Matlab scripts and uploaded to the Ecotaxa website (https://ecotaxa.obs-vlfr.fr/). 288 Particle data were obtained from the Particle Module of the Ecotaxa website. While the UVP 289 reliably images organisms with a size >600-μm by converting the pixel border length to metric 290 dimensions, it can also be used to quantify particle size and abundance down to the calibrated pixel size 291 of the sensor, though for such small particles this is not as straight-forward a task. As a result of light-292 scattering, the relationship between the pixel area of particles and their corresponding metric area is 293 nonlinear and follows a power law function (Picheral et al., 2010). This equation may be defined as = 294 ⋅ ( ) where Sp is the surface of the particle in pixels, Sm is particle area in squared millimeters, and α 295 and β are calibration constants for the UVP that were determined every year prior to deployment. 296 Equivalent spherical diameter (ESD) was then calculated for each particle using this relationship. The 297 unfiltered size range of particles measured by the UVP ranged from 42 µm to 81 mm, but to be consistent 298 with previous literature we kept the upper size limit for flux calculations of 1.5 mm used by Guidi et al. 299 (2008). As for the lower size limit, in addition to the limits imposed by camera resolution many of the 300 exceptionally small signals were likely the result of noise or otherwise could not be reliably distinguished 301 as actual particles, therefore raw particles were aggregated based on sampling year and used to plot 302 particle size spectra (Supp. Figure 1). From these spectra the point at which the particle abundance 303 relationship no longer followed a power law function was determined. However, because the camera and 304 I n r e v i e w calibrations varied between cruises, there was no consistent size range at which the slope of the power 305 law function changed, yet it was necessary to apply a uniform cutoff in order to determine a single 306 parameter set representative of the entire dataset. 200 μm was chosen as the minimum particle size as it 307 represented a reasonable compromise when comparing the particle spectra and considering the 308 importance of ~200 um fecal pellets in sediment trap derived export (discussed further in Section 3.2). 309 However, we note that, particularly for the 2008 cruise, the UVP is likely underestimating the abundance B values), we extended our particle size lower limit to 100 μm, when computing a CCE-optimized 344 algorithm, because we found that small particles are particularly important to flux in the CCE. 345 346

Algorithm-data comparisons 347
In assessing the accuracy with which the UVP-dependent algorithms (i.e., models) approximate 348 flux in the CCE, we treat sediment trap and sediment trap-234 Th blended in situ measurements as "truth", 349 but fully acknowledge that error also exists in these data (Lynch et  The AE was -21 mg C m -2 d -1 and the ANAE was 0.19. The non-parametric Pearson'scorrelation 376 coefficient (r) between sediment trap and 238 U-234 Th deficiency-derived flux was 0.64 (p < 0.001). The 377 weakest correlations between measurements occurred during the P1106 and P1208 cruises. This was not 378 surprising, because on these cruises we focused on measurements within and immediately adjacent to 379 strong mesoscale fronts. Residence time of water in these fronts (days) was substantially shorter than the 380 temporal integration time-scale of 234 Th measurements (~1 month), but similar to the time-scale of 381 sediment trap measurements (2 -4 days). Cruise-average measurements determined by sediment traps 382 and 234 Th were however, in close agreement for these two cruises ( Figure 2B). When these cruise 383 averages were used for P1106 and P1208 instead of the individual cycle averages, r increased to 0.76 (p = 384 5.2×10 -8 ). This overall agreement gives us confidence that the sediment traps were accurately collecting 385 sinking particles and that the mismatch between sediment traps and 238 U-234 Th deficiency were primarily 386 caused by differences in time scales and by occasional invalidation of the no-upwelling assumption in the 387 238 U-234 Th deficiency calculations. However, we cannot rule out the possibility of a sampling bias for any 388 specific deployment. 389 390

Comparison of Sediment Trap and UVP-Flux Estimates 391
In contrast to the sediment trap-thorium comparisons, we expect no mismatch in time scales with AE = 754.3 mg C m -2 d -1 . Both the mean (916 mg C m -2 d -1 ) and the median (600 mg C m -2 d -1 ) flux 407 estimates from the Iversen algorithm were higher than the carbon flux quantified during any of our 408 sediment trap deployments. 409 410

Sinking Particle Size Spectrum 411
To investigate particle size variability and its relationship to flux, we computed the spectral slope 412 of the modelled particle size spectrum for the mixed layer (to elucidate the relationship between mixed 413 layer dynamics and sediment trap-derived flux) and for the specific depths of sediment trap deployments 414 (to compare particle size at trap depth to sediment trap-derived flux directly). We found a slight though 415 statistically-insignificant positive trend in the mixed layer (0-20 meters) between mean spectral slope and 416 mean particle abundance ( = 0.335, p = 0.081; Figure 4A), potentially indicating that mean particle size 417 increases when particle abundance increases. This is likely the result of the dominance of large 418 phytoplankton, particularly diatoms, during bloom conditions (Goericke, 2011b; a), which promote the 419 presence of large zooplankton, large fecal pellets, and large aggregates that can all be detected by the 420 I n r e v i e w UVP. Particle flux at 100 m depth was generally elevated when particle abundance and size spectral slope 421 in the mixed layer were high. A weak potential trend was also seen when examining mean spectral slope 422 as a function of mean particle abundance specifically for the depths at which sediment traps were 423 deployed ( = 0.242, p = 0.075; Figure 4B). Interestingly, no relationship between mean spectral slope 424 and sediment trap flux at corresponding depths was observed. However, particle flux was correlated with 425 particle abundance at the same depth. 426 To compare the particle flux size spectrum estimated by UVP to the sinking particles collected 427 directly by sediment trap, we quantified the abundance and size of recognizable fecal pellets collected in 428 sediment traps on the P0704, P0810, and P1604 cruises (Stukel et  well as pellets that were smaller than ~60-µm. However, we also note that it is possible that some of these 432 pellets were transported to depth within larger aggregates that would register as large marine snow 433 particles in the UVP. Across all cruises and cycles, 50% of fecal pellet carbon flux was mediated by fecal 434 pellets <272 µm in ESD. This is notable for two reasons. algorithm, that algorithm assigns very little flux to particles in the size range of even large fecal pellets. 449 There is, however, still a substantial misfit for cycles with low total flux. Despite the fact that 450 recognizable fecal pellets were relatively few in these samples, the Guidi et al. (2008) algorithm  451 substantially underestimated flux. These samples typically had a more negative size spectral slope ( Figure  452 4). Hence it is possible that <60 µm fecal pellets were an important contributor to flux in these cycles or 453 that smaller aggregates are more important than predicted by the Guidi et al. (2008) algorithm. 454 To further investigate the size-flux relationship, on the P1706 cruise we utilized polyacrylamide 455 gels located in the bottom of sediment trap tubes to collect particles and aggregates without destroying the 456 structure of marine snow during sediment trap recovery ( Figure 6). Materials seen in the gels were 457 primarily of fecal pellet origin and had primarily sunk as independent pellets, rather than within marine 458 snow particles. Total volume flux was typically dominated by particles in the <600 µm size range ( Figure  459 7). In one sample (Cycle P1706-1, 150 m depth), >50% of volume flux was dominated by greater than 1 460 mm particles ( Figure 7A, B); however, these large particles were not marine snow. They were identified 461 as fish (likely anchovy) fecal pellets. In three samples (P1706-2, 50 m; P1706-3, 60 m; and P1706-4, 50 462 m) marine snow aggregates were also noted in the samples. However, they appeared to be less dense than 463 the acrylamide gel and hence were not in the same focal plane as the rest of the particles. Therefore they 464 were not included in the data assembled in Figure 6, but their impact will be further considered in the 465 Discussion section. For P1706-2, three marine snow aggregates (equating to flux of 218 aggregates m -2 d -466 1 ) were noted with an average ESD of 2.3 mm. For P1706-3, 24 marine snow aggregates (1958 m -2 d -1 ) 467 were noted with ESD ranging from 1.6 -7.1 mm. For P1706-4, an estimated 57 marine snow aggregates 468 (6550 m -2 d -1 ) with typical ESD of 2.5 mm were noted. However, in this sample, the aggregates were 469 clumped together making it difficult to get an accurate count. In all samples, these marine snow 470 aggregates appear to be approximately spherical and highly porous ( Figure 6G, H). Some contained 471 phytoplankters and a few contained small fecal pellets. However, their contribution to total fecal pellet 472 flux into the samples was negligible (likely <1%), and we cannot rule out the possibility that the fecal 473 pellets only became attached to the aggregates after sinking into the traps. Regardless, the high porosity of 474 the aggregates, relative dearth of phytoplankton and fecal debris associated with them, and the high 475 abundance of fecal pellets elsewhere in the samples, suggested that these marine snow particles were only 476 minor contributors to the total carbon flux in all samples. No such marine snow aggregates were noticed 477 in the acrylamide gel traps deployed at 150 m. 478 Although there is a notable difference between the size relationships of volume-weighted flux 479 estimates determined using acrylamide gels (mid-point of volume flux at ~500 µm) and the carbon-480 weighted flux estimates based on analyses of fecal pellets only (mid-point of mass flux at ~270 µm) we 481 suspect that this discrepancy is derived from the different carbon:volume ratios of different classes of 482 fecal pellets (large krill fecal pellets had lower carbon density than small copepod and appendicularian 483 pellets), rather than any other difference resulting from the different methodologies. We caution, 484 however, that all cycles of the P1706 cruise were associated with either the early, middle, or decline 485 phases of a coastal diatom bloom and had above average carbon flux. Hence, these results do not inform 486 the question of what may be the source or size of sinking particles during low flux periods. relative to their standing stock as a result of increased mass, increased settling velocities, or a combination 496 of the two. Since particle volume increases with ESD 3 , a value of B < 3 suggests that either the density or 497 sinking speed of particles is inversely correlated with particle ESD. 498 The AE of our parameterization compared to sediment trap derived flux was 43.5 mg C m -2 d -1 , 499 with a median misfit of -.142 mg C m -2 d -1 , suggesting that the algorithm had relatively low bias. The

Algorithm-data misfits 508
After the UVP carbon export estimates were obtained using the fitted parameters, the misfit 509 (defined as UVP-ST flux) was examined against a variety of environmental parameters including primary 510 productivity, depth, the slope of the size spectra, and the ratios of biogenic silicon, chlorophyll, 511 phaeopigments, and nitrogen to organic carbon sinking material in order to evaluate covariance ( Figure  512 8). In most cases these relationships were not significant. However, misfits were typically lower (i.e. UVP 513 algorithm underestimated flux) when Si:C ratios were low ( = 0.46, p = 0.004; Figure 8B). This is the 514 opposite of the effect we would expect if discrepancies were driven by increased silica-ballasting causing 515 an increase in sinking speeds of similarly sized particles. This misfit also tended to be negative during 516 cycles with the lowest primary production ( = 0.312, p = 0.026; Figure 8D). 517 518

Particle flux vertical profiles 519
Comparison of our CCE-optimized UVP-derived flux estimates to flux profiles determined using 520 a blended sediment trap-234 Th approach (from 50 -200 m depth) showed relatively good agreement 521 ( Figure 9A). There was a strong correlation between the two measurements (Spearman's ρ = 0.59, p < 522 0.001). The AE was 28.6 mg C m -2 d -1 , showing no substantial bias in the data overall, while the ANAE 523 was 0.68. However, over the 50 -200 m depth range we did find a slight difference in algorithm accuracy 524 with depth ( Figure 9B). The AE was 112.5 mg C m -2 d -1 at 50 m, 39.9 mg C m -2 d -1 at 100 m, 12.5 mg C 525 m -2 d -1 at 150 m, and 108.9 mg C m -2 d -1 at 200 m. This suggests that (at least over the limited depth range 526 of the upper twilight zone) changing relationships between particle size and flux do not lead to major 527 biases in profiles of vertical flux. However, manual inspection of profiles showed that for most cycles 528 there was a depth near the base of the euphotic zone at which UVP-derived particle flux estimates 529 substantially overestimated measured flux ( Figure 10). Since UVP flux estimates were based on particle 530 size and abundance, this suggests that near the base of the euphotic zone there is a large abundance of 531 particles that are sinking more slowly than similarly-sized particles deeper in the euphotic zone. 532 suggest that 238 U-234 Th deficiency measurements provide a more robust approach for estimating particle 549 flux than UVP-based estimates. Despite not being calibrated to the CCE region, the thorium approach 550 showed no significant bias with respect to estimation of particulate 234 Th fluxes (although we note that in 551 situ pump values of the C: 234 Th ratio, as would normally be used to convert to carbon fluxes on a survey 552 cruise, underestimated sinking particle C: 234 Th ratios in the CCE by an average of 44% (Stukel et al.,553 2019)). The greater accuracy of the 238 U-234 Th deficiency approach is not surprising, since it estimates 554 carbon flux from a direct flux measurement (i.e. 234 Th flux), while the UVP-based approach infers flux 555 from measurements of particle standing stock. Nevertheless, the increased cost, required expertise, and 556 added water budget requirements of 234 Th measurements suggest that they are not likely to be as widely 557 used as optical particle measurements. Furthermore the suitability of optical sensor deployment for 558 autonomous deployments highlights the utility of algorithms that can quantify flux from such 559 measurements, even if these results require careful (and potentially sustained) tuning and validation. 560 Most optical approaches rely on a strong covariance between particle size (or a proxy for particle 561 size) and the property of interest, which for carbon export is mass flux. Previous work has demonstrated 562 trends between particle size and sinking velocity (Smayda, 1971; Alldredge and Gotschalk, 1988; 563 Stamieszkin et al., 2015), as well as particle size and mass (Alldredge, 1998). Therefore a reasonable 564 approach was undertaken in Guidi et al. (2008) to couple these properties into mass flux for the UVP 565 platform. However, the use of previously published parameterizations for estimating flux from UVP data 566 led to substantial misfits with our measurement data. With the Guidi et al. (2008) global algorithm, the 567 UVP flux estimate was biased low, often underestimating actual flux by ~90%. The misfit was 568 particularly stark during low-biomass, low-flux periods when the UVP often predicted flux <7 mg C m -2 569 d -1 , even though sediment trap measured flux was never <26 mg C m -2 d -1 . In contrast, the Iversen et al.

570
(2010) parameterization led to substantial overestimates of flux often by a 10:1 ratio, with misfits that 571 were greatest when particle loading was highest. This parameterization suggested that carbon flux should 572 exceed 1000 mg C m -2 d -1 for 21 (out of a total of 69) paired sediment trap-UVP measurements. Measured 573 sediment trap flux never exceeded 560 mg C m -2 d -1 . These results echo the conclusions of Iversen et al.

574
(2010) that UVP-flux estimates need to be tuned to specific regions. This suggests that extreme caution 575 should be utilized when attempting to quantify flux from UVP-profiles using a global algorithm. 576 Although Iversen et al. (2010) suggested that the Guidi et al. (2008) algorithm may work better in 577 offshore regions than coastal environments, we actually find the greatest mismatch to occur during 578 oligotrophic conditions. 579 When optimizing Eq. 4 to the CCE domain, we found a much lower exponent (B = 1.05) than that 580 obtained by either Iversen et al. (2010) or Guidi et al. (2008). This B value is similar to the theoretical B 581 value calculated by Iversen et al. (2010) from relationships between aggregate size-mass and size-settling 582 velocity relationships published in Alldredge and Gotschalk (1988). Nevertheless, this exponent is 583 surprisingly low for a system in which fecal pellets are a dominant contributor to flux and for which 584 limited evidence suggests that carbon density decreases only slightly with volume (Stukel et al., 2013). 585 Indeed, given reasonable exponents for the scaling relationship between fecal pellet ESD and mass (>2), a 586 value of B = 1.05, would imply that larger pellets sink more slowly than smaller pellets. This result would 587 I n r e v i e w be in stark contrast with multiple theoretical and experimental studies suggesting that larger fecal pellets 588 sink faster (Small et al., 1979;Turner, 2002;Giesecke et al., 2010). Instead, we suspect this small 589 exponent may reflect the fact that flux was dominated by particles with sizes near the lower size detection 590 limit of the UVP (i.e., 42 µm -125 µm depending on calibration). If 200 µm to 300 µm particles are 591 dominating flux, the UVP will struggle to accurately differentiate their sizes. This difficulty is 592 exacerbated by the non-spherical shape of fecal pellets, which had typical length:width ratios ranging 593 from 2:1 to 10:1 . Thus the abundance of such particles (rather than slight differences 594 in their sizes that may not be accurately recorded by the UVP) becomes a dominant predictor of flux. The 595 fecal pellet data presented here illustrate an important caveat for imaging systems. Imaging systems are 596 well suited to observing objects over a range of approximately 3 orders of magnitude (i.e. 10 pixels to 597 ~1,000 pixels in length), yet particles within the water column often range over 5 orders of magnitude 598 from <1 µm to >10 cm. The pixel size on the UVP5 thus results in coarse granularity of particles in 599 regions where fecal pellets are abundant and important to mass flux. One potential solution to this that is 600 increasingly being implemented in the field is the use of multiple instruments with varying resolutions to 601 adequately cover the full size range of marine particles (e.g., Jackson et al., 1997). Such increased 602 resolution would allow accurate particle shape (and potentially color) information to be used in 603 classifying particles and more accurately determining volume and carbon content. 604 Our results thus suggest that, in a region dominated by fecal pellet flux, more accurate estimates 605 of flux will require optical profiling instruments with finer resolution at small particle sizes. Furthermore, 606 in regions where the relative contribution of different particle classes (e.g. phytodetritus, fecal pellets, 607 mucous feeding webs, or marine snow of mixed origin) can vary substantially, optical approaches may 608 need to incorporate information other than size (shape, color, porosity) into flux calculations. Alternately, 609 the use of autonomous 'optical sediment traps' may provide a feasible approach for estimating flux driven 610 by heterogeneous particle classes, because the ability of such approaches to actually quantify particle flux 611 onto a glass plate obviates the need for an assumed particle size-settling velocity relationship (Bishop et

Vertical particle flux and attenuation 616
Our results shed light on the processes driving patterns in vertical POC flux in the CCE. Fecal 617 pellets in the ~100 µm -400 µm size range appear to be the dominant contributors to export flux in the 618 shallow twilight zone (Figures 5-7). Flux of these pellets shows substantial spatial variability, with greater 619 fecal pellet (and total carbon) flux beneath coastal, bloom water parcels than beneath the oligotrophic 620 waters typically found offshore. The prevalence of small particles in flux leads to a low exponent (B = 621 1.05) for the relationship between particle size and mass flux. This stands in marked contrast to exponents 622 determined in other regions, such as the Mauritanian Upwelling region, where B values >3 (and in situ 623 estimates of the particle size-spectrum) suggest a greater importance for mm-sized marine snow 624 aggregates (Iversen et al., 2010). Notably, the Mauritanian Upwelling region is subject to substantial 625 Saharan dust deposition, which may contribute to mineral ballasting and higher settling velocities (Iversen 626 et al., 2010; van der Jagt et al., 2018). 627 Such large aggregates were in fact found in acrylamide gels placed in the bottom of sediment 628 traps on our P1706 cruise. They would have contributed a substantial amount of the volume of sinking 629 material in the samples if they had penetrated the gels sufficiently to enable their quantitative analysis. 630 However, their density appeared to be less than that of the acrylamide gel, resulting in them essentially 631 floating atop the gel. Despite their volume, however, visual inspection of this marine snow suggested that 632 they had only a marginal contribution to carbon flux, as a result of high porosity and a paucity of visible 633 material (fecal pellets and phytodetritus) within the aggregates. These marine snow particles were only 634 seen in acrylamide gel traps deployed at a depth of 50 m -60 m, however, and were absent from traps at 635 150 m. Taken together this suggests that: 1) these aggregates likely have only moderate sinking rates as a 636 result of high porosity and correspondingly low excess density, 2) these aggregates are likely a substantial 637 contributor to volume flux near the base of the euphotic zone, but only minor contributors to carbon flux 638 at this depth, and 3) these marine snow aggregates experience substantial flux attenuation in the shallow 639 euphotic zone. 640 The results above provide an interesting interpretation of contrasting results determined using 641 alternate approaches to quantify export in the CCE region. Jackson & Checkley (2011) used an 642 autonomous laser optical plankton counter to quantify particle abundance, volume, and flux (assuming 643 Stokes' Law relationships and constant carbon:volume ratios). They concluded that flux near the base of 644 the euphotic zone was quite high (with maximum values near 800 mg C m -2 d -1 ), but that flux attenuation 645 was exceedingly rapid in the shallow twilight zone, and decreased to ~10% (sometimes substantially less) 646 by a depth of 100 m. From these results, the authors concluded that the base of the euphotic zone was a 647 region where zooplankton play a key role as gatekeepers of flux into the mesopelagic. Our sediment trap 648 and 234 Th measurements, however, show no such zone of high carbon flux attenuation ( Figure 10). 649 Rather, our combination of direct flux measurements and in situ particle imaging paint a more nuanced 650 picture of particle transformations in this dynamic depth domain. We find that these depths are precisely 651 the depths where our CCE-optimized algorithm consistently overestimates flux relative to sediment traps 652 and 234 Th. This result strongly suggests that the relationship between size and sinking speed shifts near the 653 base of the euphotic zone, with rapidly-sinking particles (within any given size range) contributing more Our paired measurements of particle size and volume using UVP optical measurements compared with 673 carbon flux determined with sediment traps and 238 U-234 Th disequilibrium clearly show that previously 674 published algorithms for estimating particle flux from UVP data do not perform well in the CCE. This 675 adds to a growing body of literature suggesting that particle size-flux relationships are highly variable in 676 the ocean and that such variability must be taken into account when applying optically derived estimates 677 of flux. A CCE-optimized parameterization leads to a relatively low exponent (B = 1.05) for the size-flux 678 relationship. This low exponent likely results from the important role of 100-400 µm fecal pellets in 679 vertical carbon flux in the CCE, a carbon source that is poorly resolved in optical profiles. Further 680 improvements in optical estimates of carbon flux will require imaging systems with a higher resolution at 681 small particle size, the ability to discriminate between different classes of particles (e.g., fecal pellets and 682 marine snow), and accurate representation of rapidly sinking fecal pellets and other sources of organic 683 matter. Until such advancements are made it is advisable that future UVP derived estimates be paired 684 with other direct methods of flux estimation as has previously been suggested. 685 686

Acknowledgments 687
We would like to thank the captains and crews of the R.V.s Melville, Thompson