Interactive comment on “ Rapid formation of large aggregates during the spring bloom of Kerguelen Island : observations and model comparisons ”

Jouandet et al present an interesting study of the evolution of particle and biogeochemical properties over the course of a 1 month study during the Spring bloom near Kerguelen Island. In general, the study is well presented and provides a few new insights into these processes. In particular, the use of models and observational data side by side should enable the critical evaluation of each, yielding new insights and understandings. However, the lessons derived from the comparisons done here are somewhat weak in hat they emphasize similarities but do not clearly discuss how these similarities lead to improved understanding of the system. Differences between the model and observations also were not fully described. These differences could be used to strengthen the interpretation of both the observational data and also the realism of the model. The issues I raise below have the potential to affect the interpretations of the study results and models, and therefore, I recommend that the details as well as the implications of these issues should be thoroughly addressed. The study assumes a one dimensional (depth) view of the temporal evolution of particles and plankton. How might advective considerations influence the results, observations, and interpretations? Can advective processes be ruled out as one of the possibilities for explaining the temporal changes?


Introduction
Biological particle production and sedimentation out of the euphotic layer to underlying waters is the major mechanisms for atmospheric CO 2 removal and the redistribution of carbon and associated nutrients in the ocean. The fate of this exported particulate carbon is a function of the plankton community producing it in the upper layer and particle transformation by microbes and zooplankton during their descent to the deep sea. Physical aggregation of particles is one key process in this transformation and transport and can explain the rapid formation and export of large particles during bloom conditions.
The Southern Ocean is the largest high-nutrient, lowchlorophyll (HNLC) region of the global ocean. However, several areas in this biological desert display strong seasonal Published by Copernicus Publications on behalf of the European Geosciences Union. phytoplankton blooms. Since the HNLC regions result from low supplies of the crucial nutrient iron, the hypothesis is that these blooms are supported by natural sources of iron, most likely supplied from local islands and shallow sediment (Moore and Abbott, 2002;Tyrrell et al., 2005;Blain et al., 2007;Pollard et al., 2007).
The impact of iron on the biological carbon pump has been investigated in these natural bloom regions (Blain et al., 2007;Pollard et al., 2007) and in patches formed by adding iron to localized HNLC regions (Boyd et al., 2000(Boyd et al., , 2004Gervais et al., 2002;Buesseler et al., 2004Buesseler et al., , 2005de Baar et al., 2005;Hoffmann et al., 2006;Smetacek et al., 2012;Martin et al., 2013). The observations made during the naturaliron fertilization programs KEOPS1 (Kerguelen Ocean and Plateau Compared Study) and CROZEX (CROZet natural iron bloom and EXport experiment) documented a twofold greater carbon export flux downward from the mixed layer (ML) in the naturally iron-fertilized bloom relative to that in unfertilized surrounding waters (Jouandet et al., 2008(Jouandet et al., , 2011Savoye et al., 2008;Pollard et al., 2009). An increase in Particulate Organic Carbon (POC) flux after artificial fertilization experiments was detected only during SOFeX (Southern Ocean Fe Experiment; Buesseler et al., 2005) and EIFEX (European Iron Fertilization Experiment; Smetacek et al., 2012).
Optical examination of particles trapped in polyacrylamide gels during KEOPS1 found that export at 100-430 m was dominated by fecal pellets and fecal aggregates (Ebersbach and Trull, 2008), which can be considered as a form of indirect export. (Note that we consider direct export to be the flux of phytoplankton cells, either alone or in aggregates.) By contrast, the CROZEX experiment observed direct export of surface production by a diverse range of diatoms (Salter et al., 2007), consistent with phytoplankton aggregation enhancing particulate flux. The lack of phytoplankton aggregation due to insufficient biomass has been invoked as the reason for which carbon export flux in SOIREE (Southern Ocean Iron Release Experiment) was not enhanced (Waite and Nodder, 2001;. The different results for these systems reflect differences in physical forcing factors, experimental duration, and seasonal evolution of the biological community. Because of the complexity of the export system, there are still extensive unknowns about the effect of iron fertilization on carbon export from the surface to the bottom layer. The aim of our study is to investigate processes responsible for the formation of large particles (> 52 µm) on a short timescale during bloom development in the surface ML.
We combine multiple vertical profiles of large-particle size spectra collected over a relatively short period during KEOPS2 with a one-dimensional particle dynamics model that incorporates phytoplankton growth as a function of light and nitrate concentration and coagulation as function of aggregate size. We measured particle distributions using the Underwater Vision Profiler (UVP) deployed at a bloom sta-tion above the Kerguelen plateau under pre-bloom conditions and at an early bloom stage during a period of rapid change. The coagulation model used here is an extension of a zerodimensional model that simulates abundances of phytoplankton cells in the surface mixed layer as well as the size distributions of settling particles (e.g., Jackson and Kiørboe, 2008). Here it has been extended into a one-dimensional model to describe the vertical distribution of phytoplankton in the mixed layer and the formation, distribution, and flux of aggregates. The comparison between observed and modeled particle size distribution provides a unique opportunity to test the usefulness of the coagulation theory to explain rapid formation of large aggregates during the early stage of a phytoplankton bloom.

Field measurements
Station A3 (50 • 380 S, 72 • 050 E), located above the Kerguelen plateau, is characterized by a weak current (speed < 3 cm s −1 ; Park et al., 2008b), which results in a water mass residence time of several months. This long residence time allows the bloom to develop and persist over an entire season in response to natural-iron fertilization (Blain et al., 2007). During KEOPS2, Station A3 was sampled first during pre-bloom conditions on 21 October 2011 (A3-1) and was revisited during the early bloom from 15 to 17 November (A3-2), 2 weeks after the bloom had started. High sampling frequency started during the second visit at midnight on 15 November (Table 1).
The Underwater Vision Profiler 5 (UVP 5 Sn002) used in the present study was a component of the rosette profiler system. The UVP5 detects and measures particles larger than 52 µm on images acquired at high frequency (Picheral et al., 2010). Images were taken and data recorded at a frequency of 6 Hz, corresponding to a distance of 20 cm between images at the 1 m s −1 lowering speed of the conductivity-temperaturedepth (CTD) system. The observed volume per image is 0.48 dm 3 ; the total volume sampled for the 500 m depth profiles at Station A3 was 1.2 m 3 . The instrument takes a digital picture of a calibrated volume lit from the side. The image is scanned for particles and particle dimensions are measured. The pixel area (S p ) for each object is converted to cross-sectional area (S m ) using the calibration equation S m = 0.00018S 1.452 p . An equivalent spherical diameter d is calculated for that cross-sectional area. Hydrographic and biogeochemical properties, including density, fluorescence, and turbidity (as determined by a transmissometer using a wavelength of 660 nm and a 25 cm path length), were measured simultaneously with a conductivity-temperature-depth system (Seabird SBE-911 + CTD) linked to a Seapoint Chelsea Aquatracka III (6000 m) chlorophyll fluorometer and a WET Labs C-Star (6000 m) Transmissometer. We also present selected results of chlorophyll a (Chl a) and nitrate concentrations, as well as relative biomass of different phytoplankton size classes. Chl a and pigment concentrations were measured using high-performance liquid chromatography (HPLC) following the method described in Lasbleiz et al. (2014); the fraction of a phytoplankton group relative to the total biomass was calculated using the model of Uitz et al. (2006). Nitrate was analyzed with a Technicon Au-toAnalyzer as described in Tréguer and LeCorre (1975).

Data processing
The particles in each 5 m depth interval, with depth determined from the associated CTD measurements, were sorted into 27 diameter intervals (from 0.052 to 27 mm, spaced geometrically), and concentrations were calculated for each diameter and depth interval. We further analyzed size spectra having a minimum of five particles per size bin and depth interval; this criterion eliminated bins with d > 1.6 mm. The depth distributions of particles are summarized in terms of their total number N T (# L −1 ) and volume V T (mm 3 L −1 = ppm) concentrations. Particle number distributions (n) were calculated by dividing the number of particles ( N) within a given bin by the width of the ESD (Equivalent Spherical Diameter) bin ( d) and the sample volume. The resulting units are # cm −4 . The distributions are usually plotted in a log-log plot because of the large ranges in n and d. To compensate for these ranges, the results are often displayed as nVd spectra, where n is multiplied by the median diameter (d) and the spherical volume V = π/6d 3 for the particle size range. This form of the particle size distribution has the advantage that the area under the curve is proportional to the total particle volume concentration when nVd is plotted against log(d). The carbon export flux F POC (mg C m −2 d −1 ) can be estimated from the size spectra using the following empirical relationship: where d is the diameter in millimeters, A = 12.5, and b = 3.81 (Guidi et al., 2008). Guidi et al. (2008) developed the relationship by comparing particle size spectra to sediment trap collection rates at locations around the world. The value of b is less than the value of 5 expected for spherical particles of constant density (for which mass increases as d 3 and sinking speed as d 2 ). It is consistent with marine aggregates having increasing porosity with increasing size (e.g., Alldredge and Gotschalk, 1988).

Model equations and parameterization
The biological model describes the growth rate of phytoplankton in the water column as a function of light and nutrient (nitrate) concentration. The model uses a maximum phytoplankton specific growth rate G max = 0.45 d −1 (Timmermans et al., 2004;Assmy et al., 2007). Phytoplankton cells are transformed into aggregates by differential settling and shear using the standard coagulation model of Jackson (1995). Aggregates are also fragmented into two similar parts using size-dependent disaggregation rates (Jackson, 1995). The primary phytoplankton cells are chosen to match the size of Fragilariopsis kerguelensis which was the dominant species under pre-bloom conditions (L. Armand, personal communication, 2014). A single phytoplankton cell has d 1 = 20 µm, a density of 1.0637 g cm −1 , and a resulting settling speed of v 1 = 1.05 m d −1 . The probability that two particles colliding stick together, α, is assumed to be 1. The average turbulent shear rate is γ = 1 s −1 . The initial abundance of phytoplankton is 10 cells cm −3 . These and other parameter values are shown in Table 2. The onedimensional model simulates the distribution of particles of different sizes, including solitary phytoplankton cells, and nitrate concentrations at 2 m depth intervals within the 0-150 m layer. This depth range corresponds to the average surface ML thickness during the survey (Table 1). Neither zooplankton grazing nor particle transformation by bacterial processes is included in these simulations. The model is described in greater detail in Supplement S1. While the concept of spherical diameter is simple for a solid sphere, it is not for irregular marine aggregates, with different shapes, assembled from multiple sources, having water in the interstices between their components and yielding different sizes for different measurement techniques (e.g., Jackson, 1995). The simplest diameter is the conserved diameter d c , i.e. the diameter if all the solid matter were to be compressed into a solid sphere. It has the advantage that when two particles collide and form a new particle, the conserved volume of the new particle is the sum of the conserved volumes of the two colliding particles. The particle diameter d, determined by the UVP, is larger than d c because aggregate size is determined from the outer shape of the aggregate and thus contains pore water between source particles. The relationship between the two measures of particle diameter is described using the fractal dimension (see Supplement S1). The model calculations use d c . However, all model results shown here use the apparent diameter d a , which is used to approximate the diameter reported by the UVP. The value of Average algal growth rate d −1 d a is calculated from d c using the fractal relationship and a fractal dimension of 2 (Supplement S1). Note that reported values of the fractal dimension vary widely, from 1.3 to 2.3 (Burd and Jackson, 2009). The value of 2 used here is in this range and yields peaks in the nVd distributions similar to those determined from UVP measurements, unlike values of 2.1 and 1.9 (not shown).

Biogeochemical and physical context
The water column was characterized by a deep mixed layer (∼ 150 m) during the pre-bloom and early bloom surveys, with a range of 120 to 171 m ( Figs. 1 and 2). Isopycnal dis-placements of up to 50 m can be seen in the density profiles. Such vertical movements are known to result from semidiurnal internal tides in this region (Park et al., 2008a). The fluorescence and Chl a concentrations show a fourfold increase from A3-1 (21 October) to A3-2 (15-17 November), with Chl a concentrations at the surface increasing from 0.5 to ∼ 2 µg L −1 (Figs. 1 and 2). The Chl a profile determined using bottle samples for Station A3-2 was characterized by a subsurface maximum at 170 m, at the bottom of the mixed layer (Fig. 3). fluorescence ( Figs. 1 and 2). The two were, in fact, highly correlated in the surface mixed layer (r = 0.95), which was not always the case in deeper layers. Nitrate concentrations at A3-1 were 28 to 30 µM in the mixed layer and then decreased by 4 µM at A3-2 (Fig. 3a). Pigment analysis (Fig. 3) and cell counts of phytoplankton captured in nets (L. Armand, personal communication, 2014) showed that the phytoplankton community was dominated by diatoms -Fragilariopsis at A3-1 and an assemblage of Fragilariopsis, Chaetoceros , and Pseudo-nitzschia at A3-2. The zooplankton community was dominated by copepods with a mixture of adult (50.5 %) and copepodites stage (49.5 %) at A3-2 . Zooplankton biomass increased from 1.4 g C m −2 at A3-1 to 4.1 g C m −2 at A3-2 over the 0-250 m layer, and was thus more than twofold lower than the mean biomass of 10.6 g C m −2 measured at A3 in summer during KEOPS1 (Carlotti et al., 2008).

Evolution of the total abundance and volume distributions in the mixed layer
In the pre-bloom profile, total particle abundance (N T ) and volume (V T ) distributions at Station A3 were characterized by a two-layer structure (Fig. 1b). The shallower layer had relatively uniform N T (V T ) values of 90 ± 5 particles L −1 (0.3 ± 0.1 mm 3 L −1 ) between 0 and 100 m; the second layer, from 100 m to the base of the ML (166 m), had subsurface N T and V T maxima of 142 particles L −1 and 0.45 mm 3 L −1 .
There was a twofold increase in N T at the first cast of the early bloom (A3-2/1), with values reaching 200 ± 7 # L −1 in the first 100 m and a subsurface maximum of 300 # L −1 (Fig. 4). V T also increased by one order of magnitude reach-  ing a value of 3 mm 3 L −1 at the depth of the subsurface maxima (Fig. 4). In subsequent casts, there was a 40 m thick surface layer with constant N T and V T and a subsurface maximum present at variable depths. Particularly striking was the rapid and continuous increase of both N T and V T from A3-2/1 to A3-2/5 over a roughly 24 h time period. This was more than a redistribution of aggregates, as N T and V T integrated over the ML increased from 282 to 743 # m −2 and from 101 × 10 3 to 1500 × 10 3 mm 3 m −2 . There was a further increase by the end of the survey in the maximum V T to 25 mm 3 L −1 , almost two orders of magnitude greater than for the pre-bloom situation.

Evolution of size distributions with depth and time during the early bloom phase
The particle size distributions (PSD) calculated from the UVP observations provide additional insight into the change in particle abundance during the 2 d spring observation period. In order to display the vertical structure of PSD, we compare the average over the nominal euphotic zone (0 to 40 m) to the average over the 40 m layer centered on the subsurface particle maximum. Particles larger than 129 µm were more abundant in the subsurface layer (Fig. 5a). Consistent with the analysis in the previous section, the smallest difference between the two layers occurred during the pre-bloom sampling (A3-1). The maximum increases were in the 0.128-0.162 mm and 0.204-0.257 mm size classes, with abundance increases of 66 # L −1 and 62 # L −1 for A3-2/3. The increase was also substantial in the 0.4-0.5 mm size range. The cumulative volume distribution in the 0-40 m euphotic zone shows that increased particle volumes resulted from formation of larger particles (Fig. 5b).
Within the vertical particle maxima, half of V T was in particles with d > 0.5 mm at the start of the survey, while these larger particles provided more than 80 % at the end. The largest change in size spectra was in the approximately 17.5 h period between the morning (A3-2/2) and the middle of the night (A3-2/5) of 16 November.
The nVd size distribution for profile A3-2/5 is shown in detail in Fig. 6. The area under the curve at a constant depth is proportional to the particle volume V T at that depth. Between the surface and 60 m most particle volume is made up of the smallest size class with particles d ranging between 200 and 500 µm. Massive changes occurred with depth with an increase of the volume and d. The volumes from 60 m to 150 m are supported by larger particles ranging between 0.65 mm and 1.1 mm, with a peak of 30 ppm for a d of 1 mm.

Particle distributions below the ML
In the first 50 m below the ML, V T values mirrored those in the overlying waters, increasing to 20 ppm by the end of the survey period (A3-2/7) (Fig. 7). V T decreased from the base of the ML to 200 m by about a factor of 20 for A3-2/6 and A3-2/7. Below 200 m, the depth limit for winter mixing, there was no change in V T during the 2 d survey. The average V T was 0.40 ± 0.10 and 0.38 ± 0.10 mm 3 L −1 at 250 and 350 m. There was an increase in V T at about 475 m caused by resuspension from the bottom, as documented during KEOPS1 (Chever et al., 2010;Jouandet et al., 2011). The particle number distribution (n) decreased from the base of the mixed layer to 350 m in all size classes, particularly for particles larger than 500 µm, which were no longer detectable (Fig. 7b).

Relationship between particle volume and fluorescence
As mentioned, there was no simple correlation between V T and fluorescence. However, separating the observations by depth layers (the mixed layer, the base of the ML to 200 m and deeper than 200 m) reveals a pattern (Fig. 8). In the shallowest layer, there was an increase from the pre-bloom values of low fluorescence and particle volume for A3-1 (21 October) to high fluorescence and low particle volume for A3-2/1 (15 November, 23:20 LT). This is consistent with an increase in phytoplankton biomass without aggregate production. For A3-2/2, there are hints of an increase in V T , which became pronounced in subsequent casts. The increased particle concentrations were accompanied by a slight decrease in fluorescence. For the seven casts performed during the early bloom stage, the correlations between fluorescence and V T were negative (−0.53), with a slope of −0.015 µg Chl mm −3 . In the second layer, immediately below the surface mixed layer, fluorescence and V T increased together, with a positive correlation coefficient (0.68) and a slope of 0.036 µg Chl mm −3 (Fig. 8). This is consistent with  no phytoplankton growth in this depth layer, but with phytoplankton and aggregates arriving together from above, presumably in aggregates. There was no correlation between fluorescence and V T below 200 m during this period.

POC flux
The flux at 200 m computed from the UVP particle size distributions increased from 1.8 mg m −2 d −1 during pre-bloom conditions to 23 mg C m −2 d −1 during the early bloom (last cast of the survey). This increase over time as estimated from UVP measurements was also evident at 400 m but with a smaller change, with F POC increasing from 1.04 to 3.5 mg C m −2 d −1 at 400 m (Table 3).
Our POC flux estimates at 200 m for the spring bloom are in the range of the POC flux estimated from the sediment trap PPS3/3 (27 ± 8 mg C m −2 d −1 ) and below the estimates made from the gel trap (F POC = 66 mg C m −2 d −1 ) and from the thorium deficit (F POC−Th = 32 mg C m −2 d −1 ) Planchon et al., 2014). F POC−Th at 100 m increased from pre-bloom to early bloom but stayed unchanged at 200 m. The F POC−Th at 200 m was estimated at A3-2/1, consistent with UVP observations that did not record any V T increase.

Development of the phytoplankton bloom
The phytoplankton in the model grew exponentially in the upper part of the water column for the first 8 d of the simulation, slowing down as light limitation became important (Fig 9a). The specific rate of integrated population growth (0 to 150 m) was ∼ 0.06 d −1 for this initial period. The peak phytoplankton biomass was 2 µg Chl L −1 at about 10 m depth on day 13. The phytoplankton biomass decreased slightly when coagulation became an important removal mechanism by day 20, with surface phytoplankton biomass of 1.7 µg Chl L −1 , a maximum concentration of 1.9 µg Chl L −1 at 15 m, and a minimum concentration of 0.2 µg Chl L −1 at 150 m. Surface nitrate concentrations decreased from the initial 30 to 25 µM by day 20 (Fig. 9b).

Development of the aggregate volume
Aggregates with d a > 100 µm appeared by day 14, when the total volume peaked at 1.3 ppm at 40 m (Fig. 9c). As the phytoplankton biomass increased, the maximum total volume also increased. The depth of the aggregate maximum deepened as the aggregates sank, becoming 17 ppm below 140 m on day 18. By day 20, the initial rapid coagulation phase ended, with the maximum phytoplankton biomass decreasing Table 3. Comparison of the POC fluxes (F POC in mg m −2 d −1 ) derived from particle size distributions from the UVP, particle distributions from gel-filled sediment traps and sediment trap PPS3/3 Technicap Inc, France  during KEOPS2 and KEOPS1 (Jouandet et al., 2011, Ebersbach and  slightly in the upper 50 m and the aggregates at the base of the mixed layer slowly decreasing. The vertical size distribution at day 20 provides further details on the system (Fig. 10).
The nVd a size distribution shows the distribution of particle volume, with the area under the curve being proportional to the particle volume when displayed with a logarithmic d a axis, as here (Fig. 10). Most particle volume at the surface is made up the smallest particles, the single phytoplankton cells. At 10 m depth, aggregates appear with a maximum nVd a value at d a = 200 µm. With increasing depth, the total volume and the diameter of the maximum nVd a both increase. The d a at the maximum became 0.9 mm at about 70 m depth, remaining constant with increasing depth, even though the total volume continued to increase with depth.

Role of coagulation in the rapid changes observed
There are several striking correspondences between the observations at A3 during KEOPS2 and the one-dimensional coagulation model used here. First, the formation of large aggregates observed over the short timescale (< 2 d) was mimicked by the model. The simulation results highlight the ability of coagulation to change the system state on short timescales that require a frequent sampling regime for them to be observed. The shapes of the nVd spectra at the base of the mixed layer, centered at 0.9 mm for the model and 1 mm for the observations, with half widths of 1 mm for the model and 0.6 mm for A3-2/5 (Figs. 6 and 10), were very similar. The transition to rapid coagulation took place when relatively little of the initial nitrate had been consumed in the model (4 µM), consistent with the 3.6 µM decrease observed from A3-1 to A3-2 (Figs. 3 and 9). Coagulation theory has been used to predict the maximum phytoplankton biomass in the ocean (e.g., Jackson and Kiørboe, 2008). Coagulation of phytoplankton cells is a nonlinear process. Rates increase dramatically as phytoplankton biomass increases, eventually removing cells as fast as they divide. The volume concentration at which this occurs is the critical volume concentration : (2) For this calculation, we assumed an average specific growth rate for the population increase rate µ = 0.1 d −1 , in agreement with measurements made by Closset et al. (2014), α = 1, and γ = 1 s −1 . Note that the average increase rate is not the same as the peak rate G max . For a POC : volume ratio of 0.17 g C cm −3 (Jackson and Kiørboe, 2008) and a carbon to chlorophyll ratio of 50 g C : g Chl, this is equivalent to 1.5 µg Chl a L −1 . This value for V cr is remarkably close to the maximum concentrations of 2-2.2 µg Chl a L −1 observed during the particle formation at A3-2. The rapid production of aggregates at Station A3 observed in this study provides an impressive example of the importance of coagulation in controlling PSD and vertical export of primary production.
The nature of the exported material collected in a freedrifting sediment gel trap at 210 m supports also the importance of algal coagulation in forming the exported material . Their analysis shows that the particle flux number and volume were dominated by phytoaggregates over the 0.071-0.6 mm size range.

Limitations of the model
There are, not unexpectedly, differences between model results and observations. To start, fluorescence profiles are relatively constant through the surface mixed layer in the observations, but have a pronounced shallow subsurface chlorophyll maximum in the model because of the higher light levels near the surface. Increased mixing in the model could smooth the chlorophyll profiles, as well as the distribution of particle volume. Simulations made using a much larger Figure 10. Distribution of apparent particle volume, nVd a , as a function of depth and d a as calculated by the model at 20 d. Because the value of d a is plotted on a logarithmic scale, the area under the curve for each depth is proportional to total particle volume V Ta . mixing coefficient (1000 m 2 d −1 ) yield a smaller difference in chlorophyll between the surface and 150 m, but there is still a difference of 0.8 µg Chl L −1 over the depth range (results not shown). The vertical mixing rate estimated for the iron fertilization experiment EIFEX, 29 m 2 d −1 , was actually smaller than that used in these simulations, namely 100 m 2 d −1 (Smetacek et al., 2012). A previous model of phytoplankton growth in the Kerguelen region discussed large-scale horizontal patterns but unfortunately did not display vertical distribution (Mongin et al., 2008). Whatever the reason for the relatively uniform fluorescence profile, it is not simply a faster diffusive mixing rate. Those differences illustrate the difficulty of building a realistic phytoplankton growth model in the region to drive the coagulation model. The shallower phytoplankton distribution does affect the distribution of aggregates as well.
In a model such as the one used in the present study, there are many parameters and modeled processes that influence the final results. These include parameters such as the fractal dimension, the size of the phytoplankton cells, or processes to describe diatom chain growth, disaggregation rates, and grazing. While the parameters could be tuned systematically to give an improved fit, what is striking is the similarity between observations and the model without such a systematic fitting procedure. One important parameter that was varied during model development to adjust the results was the fractal dimension. Decreasing it decreased the diameter of the peak value of nVd. The value that was chosen, D fr = 2, was similar to some of the estimates of fractal dimension noted above and did provide the correct nVd distribution when coagulation occurred.
Other processes are known to affect particle concentrations and fluxes, most notably physical process such as advection and biological processes such as zooplankton grazing and fecal pellet production (e.g., Lampitt et al., 1993;Stemmann et al., 2000;Turner et al., 2002). The importance of advection could be inferred from time series measurements of LADCP (Lowered Acoustic Doppler Current Profiler). The results indicated a current below 0.1 m s −1 , with negligible changes over the survey in the 0-200 m depth layer (Y. H. P. Park, personal communication, 2014). The abundance and volume of zooplankton larger than 0.7 mm, as well as fecal sticks/pellets and aggregates, were estimated from the identification of organisms in the vignettes recorded by the UVP using the Zooprocess imaging software (see Picheral et al., 2010). The volume of copepods did not increase through the early bloom survey, suggesting that they were not responsible for the observed rapid increase in particles. Ingestion rates were also estimated from zooplankton biomass using the relationship detailed in Carlotti et al. (2008) using the biomass results integrated over the 0-250 m layer. The ingestion rate was 1.36 mg C d −1 during the early bloom cast and lower than during the KEOPS1 summer cruise. In addition, fecal pellet production should have a diurnal signal , which was not observed in the V T profiles. Lastly, fast-sinking fecal pellets are much smaller than the aggregates observed here. For example, fecal pellets falling at 100 m d −1 are typically 2-5 × 10 6 µm 3 , equivalent to d = 200 µm (Small et al., 1979), compared to the millimeter-sized aggregates dominating at A3. Thus, changes in zooplankton populations can be ruled out as an explanation of the observed V T increase at this time, although not through the entire season. Modeling the dynamics of the entire season would require integrating zooplankton activity.

KEOPS1
The comparison of our results with the size spectra obtained from UVP measurements at Station A3 during the early bloom (KEOPS2) and the late stage of the bloom (KEOPS1) allows us to investigate the seasonal variability of particle production in the 0-200 m layer and the POC export flux (Fig. 11, Table 3). During summer (KEOPS1), the phytoplankton community was also dominated by Chaetoceros but shifted to Eucampia antarctia by the end of the bloom (Armand et al., 2008). Zooplankton abundance was tenfold higher than during the early bloom and the community was dominated by copepods at copepodite stage (Carlotti et al., 2008). The mixed layer decreased from 150 m during early bloom to 70 m during summer. During KEOPS2, V T increased more than twentyfold from pre-bloom conditions, probably as a result of the higher diatom biomass (L. Armand, personal communication, 2014), and coagulation as described in Sect. 4.1. The value of V T achieved by the time of the bloom decline in February (KEOPS1) was quite similar to that measured during early bloom for KEOPS2 but the vertical structure was different, with two subsurface maxima during KEOPS1, the first one present at the base of the ML (70 m). The larger V T measured in January was associated with an increase in the fraction of large particles (Fig. 11c). Below 200 m depth, V T was still 10 times higher during the peak bloom as compared to early bloom. This resulted in forty-(at 200 m) and tenfold (at 400 m) higher carbon export fluxes during the peak bloom than the early bloom (Table 3). During the decline of the bloom, V T and POC flux were still higher than during early bloom. This is consistent with the general scheme of low-production-high-export at the end of the bloom put forward by Wassmann (1998). Our results provide insights into particle production and size distributions at different stages of the seasonal bloom. The early bloom occurs before zooplankton grazing dominates. This leads to a large increase in diatom abundance resulting in rapid aggregate formation and export from the surface ML. Later in the season, export becomes controlled by zooplankton grazing and fecal pellet production, as found through the gel trap analysis (Ebersbach and Trull, 2008). Despite the importance of zooplankton grazing in the late season, the presence of V T maxima at the base of the ML indicates that coagulation still occurred during summer. An increase of aggregate formation through coagulation as a result of high cell numbers in the ML and their disappearance due to grazing between the base of the mixed layer and 200 m traps could also ex-plain the dominance of fecal aggregates in the gel traps during the summer deployments. Combining KEOPS cruises to describe temporal scales of particle production and export (transient versus seasonal) is useful as a first step, but our limited observations highlight the need for high-frequency data collection over long periods.

Potential impact of coagulation after iron fertilization
Our results can be compared to those from other iron fertilization experiments to understand the relative roles of coagulation and zooplankton grazing on particle export during different parts of the bloom cycle. However, it must be remembered that fertilization experiments differ in important aspects, including location, physical and chemical regimes, and observational techniques applied to determine stocks and fluxes. In addition, conclusions about carbon export from the surface often depend on sediment traps that are usually located well below the euphotic zone or surface ML, sampling events that have been filtered by intervening processes and offset by transit times. With this in mind, we compare our results to those from other iron fertilization studies by classifying them into those with phytodetritus export driven by diatoms on the one hand and the rest, including those with a zooplankton-mediated export, on the other hand. The artificial iron fertilization experiment SOIREE (February 1999) found an increase in phytoplankton biomass (Chl a = 2 mg m −3 ) as a result of the iron addition, but no rapid removal of phytoplankton production. The export flux was low and driven by phyto-detrital aggregates (Waite and Nodder, 2001).  argued that the final abundance of phytoplankton cells was too low for rapid coagulation and sinking. There was a change in diatom settling rate associated with a change in iron status. The persistence of the bloom after iron was depleted implies that zooplankton grazing was not removing the particulate material.
The EIFEX (February-March 2004) environment was remarkably similar to that of KEOPS2 (Smetacek et al., 2012). The mixed layer was slightly shallower during EIFEX (100 m) than during KEOPS2 (150 m), but still relatively deep; the phytoplankton accumulation rates were also similar (0.03 to 0.11 d −1 ). Iron fertilization stimulated a large diatom bloom that reached concentrations of about 2 mg Chl a m −3 4 weeks after the fertilization started. There was little effect on vertical export during the first 4 weeks, but export then increased rapidly to 110-140 mmol C m −2 d −1 . This change was associated with mass mortality of several diatom species that formed rapidly sinking, mucilaginous aggregates of entangled cells and chains (Smetacek et al., 2012). This pattern of rapid formation of algal cells late in the bloom is similar to what we observed.
CROZEX investigated the impact of high biomass (Chl a = 2 mg m −3 ) associated with the bloom decline in carbon export during 2 legs (November 2004 and January www.biogeosciences.net/11/4393/2014/ Biogeosciences, 11, 4393-4406, 2014 2005) . Carbon export fluxes estimated from a sediment trap in the highly productive, naturally ironfertilized region of the subantarctic waters were two to three times larger than the carbon fluxes from adjacent HNLC waters (Pollard et al., 2009). Vertical flux was dominated by a diverse range of diatoms, which suggests an important role for direct export, such as by coagulation.
In contrast, the particulate flux project in a region of elevated biomass (Chl a = 1.9 mg m −3 ) in the Subantarctic Zone east of Tasmania fueled by enhanced iron was dominated by fecal aggregates (Ebersbach et al., 2011).
The LOHAFEX (loha is the Hindi word for iron) iron fertilization experiment was one of the few to use a particle measuring system for the water column, also the UVP (Martin et al., 2013). A cyclonic eddy low in silica on the Antarctic Polar Frontal Zone was fertilized with iron. In response, phytoplankton biomass almost doubled to 1-1.5 mg Chl a m −3 , but 90 % of it was in flagellates of less than 10 µm instead of diatoms. There was no observable change in concentrations of particles larger than 100 µm or in vertical particle flux. There were several reasons proposed for the low export, including the lack of diatoms in the lowsilicate water and intense particle consumption, particularly at the base of the mixed layer (66 m).

Conclusions
It is clear that particle flux in the ocean is the result of many interacting processes, and none of these has been identified as dominant across systems. In the present study, we were able to observe rapid aggregate formation and sedimentation of high concentrations of diatoms from the euphotic zone. Our observations are consistent with results from a onedimensional model that includes only phytoplankton growth and coagulation. Our results demonstrate the utility of coagulation theory in understanding vertical flux and its importance in initiating the formation of large particles in the mixed layer and their subsequent transfer to depth during a bloom. Nevertheless, efforts are still required to measure large aggregates distribution at a high frequency to fill the temporal window between these short time events taking place during the early bloom and the possibly slower dynamics of summer. In addition, more effort is required to understand vertical variations better at a fine scale for all times and particularly to estimate the transformative roles of microbes and zooplankton in decreasing the total particle volume exported from the euphotic zone.
The Supplement related to this article is available online at doi:10.5194/bg-11-4393-2014-supplement.