Seasonal evolution of net and regenerated silica production around a natural Fe-fertilized area in the Southern Ocean estimated with Si isotopic approaches

. A massive diatom bloom is observed each year in the surface waters of the naturally Fe-fertilized Kergue-len Plateau (Southern Ocean). We measured biogenic silica production and dissolution ﬂuxes ( ρ Si and ρ Diss, respectively) in the mixed layer in the vicinity of the Kerguelen Plateau during austral spring 2011 (KEOPS-2 cruise). We compare results from a high-nutrient low-chlorophyll reference station and stations with different degrees of iron enrichment and bloom conditions. Above the plateau biogenic ρ Si are among the highest reported so far in the Southern Ocean (up to 47.9 mmol m − 2 d − 1 ). Although signiﬁ-cant (10.2 mmol m − 2 d − 1 on average), ρ Diss were generally much lower than production rates. Uptake ratios ( ρ Si : ρ C and ρ Si : ρ N) conﬁrm that diatoms strongly dominate primary production in this area. At the bloom onset, decreasing dissolution-to-production ratios (D : P) indicate that the remineralization of silica could sustain most of the low silicon uptake and that the system progressively shifts toward a silica production regime which must be mainly supported by new source of silicic acid. Moreover, by comparing results from the two KEOPS expeditions (spring 2011 and summer 2005), we suggest that there is a seasonal evolution of the processes decoupling Si and N cycles in the area. Indeed, the consumption of H 4 SiO 4 standing stocks occurs only during the growing stage of the bloom when strong net silica production is observed, contributing to higher H 4 SiO 4 depletion relative to NO − 3 . Then, the decoupling of H 4 SiO 4 and NO − 3 is mainly controlled by the more efﬁcient nitrogen recycling relative to Si. Gross Si : N uptake ratios were higher in the Fe-rich regions compared to the high-nutrient low-chlorophyll (HNLC) area, likely due to different diatom communities. This suggests that the diatom responses to natural Fe fertilization are more complex than previously thought, and that natural iron fertilization over long timescales does not necessarily decrease Si : N uptake ratios as suggested by the silicic acid leakage hypothesis. Finally, we propose the ﬁrst seasonal estimate of the Si biogeochemical budget above the Kerguelen Plateau based on direct measurements. This study points out that naturally iron-fertilized areas of the


Introduction
Covering 20 % of the World Ocean, the Southern Ocean is considered a crucial component of the climate system since it represents a net sink for atmospheric CO 2 (Takahashi et al., 2009).It also plays a key role in the global silicon (Si) biogeochemical cycle because diatoms, a siliceous phytoplankton group, are one of the major primary producers in Published by Copernicus Publications on behalf of the European Geosciences Union.
this area (Buesseler et al., 2001;Quéguiner and Brzezinski, 2002;Tréguer and De la Rocha, 2013).As their cell wall is composed of biogenic silica (opal, amorphous SiO 2 .,nH 2 O, hereafter referred to as BSi), diatoms take up dissolved silicon (hereafter referred to as DSi) in the form of silicic acid (H 4 SiO 4 ) to produce their siliceous frustules.At the global scale, 56 % of this gross production is estimated to be directly recycled in the upper 100 m (Tréguer and De La Rocha, 2013) due to the combined effects of both physicochemical and biological processes (Kamatani, 1982;Bidle and Azam, 1999;Ragueneau et al., 2000).Only the material escaping dissolution is exported toward the deep ocean and eventually buried in sediment.Consequently, the marine Si biogeochemical cycle is dominated by biogenic silica production and dissolution in the surface mixed layer, and one atom of Si will undergo a cycle of biological uptake by diatoms and subsequent dissolution about 25 times before being removed to the seabed (Tréguer and De La Rocha, 2013).Thus, it is essential to estimate the balance between silica production and dissolution in the euphotic zone, which is best illustrated by the ratio of integrated dissolution to production rate (∫ D : ∫ P ; Brzezinski et al., 2003) or by the integrated net production rate (∫ ρSi Net = production minus dissolution).Globally, ∫ D : ∫ P values show an annual mean of 0.56 (Tréguer and De La Rocha, 2013) and range from < 0.1 to > 1, with low D : P values associated with diatom bloom events and D : P values exceeding 0.5 occurring during non-bloom periods (Brzezinski et al., 2001).However, the number of D : P estimates, due to a small number of Si uptake measurements and an even lower number of Si dissolution measurements, is insufficient compared to the high variability observed regionally and seasonally in the ocean, implying high uncertainty of the global D : P estimate and of the marine silicon budget overall.
Diatoms are ecologically widespread and dominate primary production in the Antarctic Circumpolar Current (ACC), especially south of the polar front (PF), where their productivity accounts for one-third of the global marine silica production (Pondaven et al., 2000;Buesseler et al., 2001).Consequently, this part of the Southern Ocean represents a key study area for improving our understanding of the global biogeochemical cycles of both carbon and silicon.Biological processes occurring in the Southern Ocean have indeed a significant impact on global biogeochemistry.For example, the large H 4 SiO 4 utilization by diatoms in the ACC, combined with the global overturning circulation, could determine the functioning of the biological pump of low-latitude areas by inducing strong silicic acid limitation (Sarmiento et al., 2004).In the Southern Ocean, a much larger depletion of silicic acid than nitrate in surface waters occurs (Trull et al., 2001), which results from the action of a silicon pump, i.e., the preferential export of BSi compared to particulate organic nitrogen (PON; Dugdale et al., 1995).This area is also the largest high-nutrient low-chlorophyll (HNLC) zone of the global ocean where dissolved iron limitation plays a funda-mental role in regulating primary production and the carbon cycle (De Baar et al., 2005;Boyd et al., 2007;Tagliabue et al., 2012).Indeed, phytoplankton community structure and nutrient cycling could be largely controlled by Fe availability, with the highest growth rates located close to iron sources such as continental margins, island systems and frontal regions (Blain et al., 2007;Tagliabue et al., 2012).
In this context, the KErguelen Ocean and Plateau compared Study (KEOPS) program, consisting of two expeditions (late summer 2005 and early spring 2011), was conducted to investigate a naturally iron-fertilized area located in the Indian sector of the Southern Ocean, where iron availability could potentially favor carbon and silicon biological pumps (Fig. 1; Blain et al., 2007).Two massive and complex blooms, which are clearly constrained by the local bathymetry, are observed annually over the Kerguelen Plateau and contrast with the HNLC character of the surrounding waters (Pollard et al., 2002;Mongin et al., 2008).The first KEOPS expedition (KEOPS-1, January and February 2005) highlighted the impact of natural iron fertilization on primary production and nutrient cycling, as well as the advantages of studying natural laboratories in the context of such ocean fertilization (Blain et al., 2008).The general purpose of KEOPS-2 (October and November 2011) was to improve our knowledge about the processes responsible for this iron fertilization and its impact on the seasonal variations of the mechanisms controlling primary production and carbon export.While the KEOPS-1 cruise was mainly directed towards the study of the bloom in the southeastern area of the plateau, KEOPS-2 focused mainly on the bloom located northeast of the Kerguelen Islands above the Kerguelen abyssal plain.
In this paper, we investigate the spatial and seasonal variability of silica production and dissolution in the surface waters of the Kerguelen area.The specific objectives are the following: -compare the Si cycle dynamics in contrasting productive environments such as the southeastern Kerguelen Plateau bloom, the northeastern Kerguelen bloom in a stationary meander southward of the PF and the warmer waters located north of this front, relative to the upstream HNLC area southwest of the Kerguelen Islands, and identify controlling processes -determine and quantify the seasonal evolution of processes which drive the Si biogeochemical budget in the upper layer of the Kerguelen Plateau, using the 30 Si stable isotope method (Nelson and Goering, 1977a;Fripiat et al., 2009) applied during KEOPS-2 and other techniques of mass and isotopic balance used during KEOPS-1 ( 32 Si radiogenic tracer incubations, Mosseri et al., 2008; and natural silicon isotopic composition, δ 30 Si, Fripiat et al., 2011a), in order to fully characterize the silicon cycle above the Kerguelen Plateau  -discuss the potential role of Fe in silica production/dissolution and in Si : N uptake ratios in the context of the silicic acid leakage hypothesis (Brzezinski et al., 2002;Sarmiento et al., 2004) -finally, compare our data set with previous results in other productive regions of the global ocean and discuss the different types of diatom-dominated regimes 2 Material and methods

KEOPS-2 sampling campaign
The KEOPS-2 cruise was conducted in the Indian sector of the Southern Ocean during the austral spring 2011 (from 10 October to 20 November) on board the R/V Marion Dufresne (TAAF/IPEV) and was focused on the ironfertilized blooms observed around the Kerguelen Plateau region.This plateau is a large area of relatively shallow seafloor that acts as a barrier to the circumpolar flow of the ACC, forcing a large part of the current to pass north of the plateau.The remaining flow passes south of the Kerguelen Islands and forms the jet of the PF, which exhibits strong meandering and eddy activity (Park et al., 1998(Park et al., , 2008;;Roquet et al., 2009).As a consequence, the shallow region located south of the Kerguelen Islands represents a zone of weak northeastward circulation (Park et al., 2008;Roquet et al., 2009) and has potential for high chlorophyll a and BSi accumulation during phytoplankton blooms (Blain et al., 2001;Mosseri et al., 2008;Mongin et al., 2008) enhanced vertical mixing associated with internal waves interact with the local bathymetry (Park et al., 2008) and supply iron and macronutrients from depth to surface waters, enabling to fuel phytoplankton bloom (Blain et al., 2007;Fripiat et al., 2011a).
The cruise included eight long-term stations devoted to process studies with incubation experiments (Fig. 1).Except for one station (E4E), where we were not able to measure silica dissolution, Si fluxes were investigated at all of these process stations, whose characteristics are presented in Table 1: -an HNLC reference station (R-2) located in the deep waters southwest of the Kerguelen Islands -the Kerguelen Plateau bloom reference station of KEOPS-1 (A3-2) -a productive open ocean station (F-L) influenced by warmer Subantarctic Surface Water, located north of the polar front -a productive station (E-4W) located in the plume of chlorophyll observed downstream of the plateau and close to the jet induced by the PF -four stations (E-1, E-3, E-4E and E-5) constituting a pseudo-Lagrangian survey located in a complex recirculation zone in a stationary meander of the polar front characterized by strong mesoscale activity (Zhou et al., 2014)

Sample collection, spike and incubation conditions
The isotopic dilution technique adapted by Fripiat et al. (2009) from Corvaisier et al. (2005) is used to simultaneously determine the rates of Si uptake (i.e., silica production) and of biogenic silica dissolution in the same seawater sample.After spiking with a solution enriched in 30 Si followed by incubation of the samples, the production rate is estimated from the change in isotopic composition of the particulate phase (increase in 30 Si).Similarly, the isotopic dilution (increase in 28 Si) in the 30 Si enriched seawater, due to the dissolution of naturally 28 Si enriched BSi initially present, is used to estimate the dissolution rate.Production and dissolution rates were determined at seven and five depths, respectively, corresponding to different levels of photosynthetically active radiation (PAR), from 75 to 0.1 % of surface irradiance.Seawater was collected at defined depths in the euphotic layer using Niskin bottles mounted on a CTD (sonde) rosette.For each depth, 5 L of seawater were sampled.One liter was subsampled to obtain a natural silicon isotopic standard (i.e., not spiked with 30 Si) to be processed along with the samples to correct for the matrix effect and mass bias during isotopic analysis (Fripiat et al., 2009).These unspiked samples were immediately filtered on 0.8 µm Nuclepore polycarbonate membranes to separate biogenic silica from silicic acid.The membrane was dried at 50 • C overnight and the filtrate was directly preconcentrated (see Sect. 2.3) and stored at room temperature in the dark.
The remaining seawater volume was subsampled in 2 L aliquots spiked with H 30 4 SiO 4 -enriched solution (99 % 30 Si).Aliquots devoted to production measurements were spiked with a spike contribution representing usually less than 10 % of natural concentrations to minimize the perturbation of the natural DSi contents (Nelson and Goering, 1977a).In order to improve the method's dissolution detection limit, a second 2 L aliquot was spiked by adding the same amount of 30 Si as natural DSi (i.e., DSi spike addition at 100 % of the initial DSi).This provided sufficient sensitivity for the isotopic measurements of dissolution (see Sect. 3.1).
Immediately after spike addition and gentle mixing, 1 L was filtered following the same procedure as for the unspiked standard to determine the initial conditions (t 0 ).The second half of the sample was poured into polycarbonate incubation bottles and incubated under light conditions simulating those prevailing in situ for 24 h (10 % spiked samples) and for 48 h (100 % spiked samples).Deck incubators were fitted with blue plastic optical filters to simulate the light attenuation of the corresponding sampling depths, and temperature was regulated by circulating surface seawater.At the end of the incubation period, samples were filtered and treated as described above to characterize the final conditions of the incubation (t 24 or t 48 ).

Sample preparation and isotopic measurements
Preconcentration of H 4 SiO 4 in the seawater samples (for both production and dissolution measurements) was applied on board to increase the Si : salinity ratio because the maximum salinity of the solution that can be introduced into the mass spectrometer is about 2 ‰ (Fripiat et al., 2009).This step was achieved using a protocol adapted from the MAGIC method (Karl and Tien, 1992;Reynolds et al., 2006).The H 4 SiO 4 in seawater was scavenged by the brucite precipitate (Mg(OH) 2 ) obtained by adding 1 mL of 14 N NaOH to the 1 L of seawater sample and stirring strongly.The precipitate was recovered by decantation and centrifugation, and was then dissolved in 3 mL of 3 N HCl.
In the shore-based laboratory, polycarbonate membranes (t 0 , t 24 and t 48 for both production and dissolution measurements) were digested in one step using a protocol adapted from Ragueneau et al. (2005) with 4 mL of 0.2 N NaOH for 40 min at 100 • C to hydrolyze BSi.Samples were then neutralized with 1 mL of 1 N HCl to stop the reaction.
An aliquot of the solutions obtained after preconcentration and digestion was used to determine colorimetrically the DSi and BSi concentrations, following the method of Strickland and Parsons (1972).The remaining sample was diluted to 100 ppb Si in a 2 % HNO 3 solution to determine the initial and final Si isotopic composition of the dissolved and particulate phases using an Element 2 Biogeosciences, 11, 5827-5846, 2014 www.biogeosciences.net/11/5827/2014/high-resolution sector field inductively coupled plasma mass spectrometer (HR-SF-ICP-MS; Thermo Fisher) with the same configuration used by Fripiat et al. (2009).The sequence of the analysis is as follows: (1) blank, (2) natural standard, (3) spiked sample 1, (4) natural standard, (5) spiked sample 2, (6) natural standard, (7) spiked sample 3, (8) natural standard, and (9) blank.The average of the two blanks was subtracted from each standard and sample.To test whether our dissolution measurements were biased by a 30 Si contamination linked to a possible memory effect in the HR-SF-ICP-MS, we compared the average composition of the first natural standards (i.e., without contamination from memory effect, n = 55) with the composition of natural standards analyzed after a spiked sample (n = 102).There was no significant difference between natural standards before and after a 100 % DSi spiked sample (t test, p value < 0.001).We can thus exclude significant memory effect when applying the analytical sequence described above.

Accuracy of the model, detection limit and standard deviation
To estimate the production and dissolution of biogenic silica (ρSi and ρDiss, respectively), two different models are available: the linear one-compartmental model described by Nelson and Goering (1977a, b) and the nonlinear twocompartmental model described in de Brauwere et al. (2005) and Elskens et al. (2007).In the latter, both isotopic composition and concentration changes occurring during the incubation time are taken into account to estimate production and dissolution rates simultaneously.Lack of consideration of these changes could induce significant biases in the estimated fluxes (Elskens et al., 2007).In this model the fluxes are calculated by resolving a system of four equations given by where [BSi] and [DSi] are the dissolved silicon and biogenic silica concentrations (in µmol L −1 ); aeBSi and aeDSi are the abundance in excess of 30 Si (measured minus natural abundances) in the particulate and dissolved phase, respectively; the subscripts t 0 and t refer to the initial and final incubation values.
The best solution is found numerically by optimizing parameter values (ρSi and ρDiss) and minimizing the cost function (weighted sum of squared differences between calculated and measured variables, [BSi], [H 4 SiO 4 ], aeBSi and aeDSi for the four equations simultaneously; de Brauwere et al., 2005;Elskens et al., 2007).
The relevance of the two models against a given data set has already been discussed by Elskens et al. (2007) and Fripiat et al. (2011b).Taking into account these considerations, and after testing the accuracy and the sensitivity of each model, we prefer using the nonlinear two-compartmental model to estimate the biogenic silica production and dissolution during KEOPS-2.This model was tested according to four criteria and the residual of the cost function was checked to follow a χ 2 distribution as detailed in Elskens et al. (2007).Due to unexpected sampling problems on board, we were not able to measure [DSi] t .Thus, in addition to the biogenic silica production and dissolution rates, this variable was also estimated by the model .Under these conditions, one degree of freedom is lost but the system remains identifiable with three unknowns and four equations.
KEOPS-2 took place during the onset of the blooms; the biogenic silica production rates were quite high and far above the detection limit, except for three depths of the HNLC reference station (R-2) and for the deepest value at each station (0.01 % PAR attenuation depth, eight samples).However, since biogenic silica dissolution rates were expected to be low in early spring, it is essential to determine the limit of detection for the 30 Si isotopic dilution.
In most cases, final aeDSi were significantly different from initial aeDSi (paired t test, p value < 0.001).The detection limit for isotopic dilution was then estimated as being the lowest difference between initial and final 30 Si isotopic abundances ( aeDSi) measurable by the instrument.Every aeDSi solution has been analyzed in duplicates with a pooled standard deviation of 0.32 % (n = 35).In addition, we analyzed the same in-house standard several times during every analytical session.This solution was a 10 % spiked seawater from Southern Ocean, having been analyzed for several years, with a aeDSi at 11.83 ± 0.43 % (n = 40).The relative standard deviation (RSD) on aeDSi of this standard solution is 0.43 % (n = 40) and represents the long-term reproducibility of HR-SF-ICP-MS measurements.Therefore, each KEOPS-2 incubation with a aeDSi between t 0 and t 48 higher than this RSD was considered to be significantly different from zero, and hence above the detection limit.This was the case for almost the entire KEOPS-2 data set (see, e.g., Fig. 2), except for seven values showing a change in 30 Si abundance below the detection limit.This included four samples from the HNLC reference station R-2 where biological activity was extremely low.
Due to time and sampled water volume constraints, the sampling strategy adopted for KEOPS-2 gave priority to a  higher vertical resolution instead of replicate incubations.Since only the analytical reproducibility was taken into account in the model, the standard deviations of Si uptake and dissolution rates were likely to be underestimated.Therefore, we will use a theoretical relative precision of 10 % for all incubation experiments, as estimated for Si uptake rates by Fripiat et al. (2009).

Physical, chemical and biological parameters
The vertical structure of upper layer waters in the area was characteristic of the Antarctic Surface Water in the vicinity of the polar front (Park et al., 1998(Park et al., , 2008)).The winter water (WW), identified by the minimum of temperature centered around 200 m, was capped by a homogeneous mixed layer (ML) induced by seasonal stratification.The boundary between the surface ML and the WW is usually marked by a strong seasonal pycnocline.However, at some stations, the stratification of the surface layer was relatively complex and showed two successive discontinuities evidenced by two different density gradients as indicated in Fig. 3.During KEOPS-2, the surface ML depth, defined by the density difference of 0.02 hg m −3 from the surface (Park et al., 2014), showed a large variability between stations (Fig. 3).A strong and shallow stratification was measured north of the polar front, while wind events induced weak stratification and deep ML in the stations above the plateau and in the HNLC area.Stations in the recirculation zone (E-1 to E-5) supported a complex stratification due to their highly spatial and temporal dynamics and were characterized by two distinct density discontinuities.
All the stations located south of the polar front had quite homogeneous Chl a, BSi and DSi stocks from the surface to the deepest density discontinuity (below the so-called ML; Fig. 3).North of the polar front, these stocks were higher at the surface and decreased with depth (Fig. 3c).Stations A3-2 and E-4W show similar BSi and DSi surface concentrations (Fig. 3e, g).At these two stations, DSi concentrations increase gradually while Chl a and BSi concentrations decrease drastically below the deepest density discontinuity.Station R-2 was distinguished from the latter stations by its low BSi, low Chl a content and relatively high DSi concentrations, confirming its HNLC character (Fig. 3a).During the Lagrangian survey (stations E-1, E-3, E-4E and E-5), we observed a DSi depletion from ≈ 15 to ≈ 10 µmol L −1 in surface waters, an increase of Chl a concentrations from < 1 to > 1 mg m −3 and a doubling of the BSi content from ≈ 1.5 to > 3 µmol L −1 (Fig. 3b, d, f, h).Such temporal variations were mainly driven by diatom production as described below.Table 2. Biogenic silica concentration ( [BSi]), Si-uptake ( ρSi), biogenic silica dissolution ( ρDiss), silica net production ( ρNet) integrated over the euphotic layer (1 % of surface Photosynthetically Active Radiation), and integrated specific rates of Si-uptake and silica dissolution (calculated as VSi = ρSi / [BSi] and Diss = ρDiss / [BSi], respectively).Meander 159.5 ± 0.4 27.5 ± 0.2 6.97 ± 0.03 20.5 ± 0.2 0.17 0.04 * Since no dissolution rates were measured at E4E, these values do not correspond to calculations from direct measurements but only to estimations.Dissolution was calculated as the average of all KEOPS-2 integrated dissolution rates.

Biogenic silica production and dissolution rates
Silica production rates were quite homogeneously distributed in the euphotic layer with the exception of the station F-L located north of the polar front where it decreases progressively with depth (Fig. 4a).Surface ρSi varied from 0.036 ± 0.003 µmol L −1 d −1 (R-2 in the HNLC area) to 1.28 ± 0.12 µmol L −1 d −1 (A3-2, above the plateau).All over the study area, Si uptake rates reached very low values at the base of the euphotic layer.Note that the same decreasing trend was also observed in primary production experiments performed in parallel (see, e.g., Cavagna et al., 2014).
BSi dissolution rates were considerably lower than Si uptake rates except in the HNLC area (R-2), and at station E-3 where ρSi was in the lower range of the KEOPS-2 data set.Vertical profiles of ρDiss (Fig. 4b) were quite homogeneous from the surface to the base of the euphotic layer and did not increase at depth.This indicates that the physical and biogeochemical processes affecting BSi dissolution did not vary significantly over the water column.This is also consistent with the low accumulation of biogenic silica observed at depth in spring (Lasbleiz et al., 2014) which contrasts with the occurrence of deep BSi maxima at the end of summer (Mosseri et al., 2008)  As silica production was close to zero below the euphotic layer, all the vertically integrated values presented in Table 2 were calculated from 100 to 1 % of the surface PAR.The integrated Si uptake rates (∫ ρSi) varied from 3.09 ± 0.01 mmol m −2 d −1 (R-2, in the HNLC area) to 47.9 ± 0.4 mmol m −2 d −1 (A3-2, above the plateau), and were among the highest reported so far in the Southern Ocean (see review in Fripiat, 2010).Integrated BSi dissolution rates (∫ ρDiss) were generally much lower than integrated production rates with values ranging from 3.79 ± 0.03 mmol m −2 d −1 north of the polar front (F-L) to 9.99 ± 0.03 mmol m −2 d −1 at E-3.Because ρDiss did not vary over depth or between stations, integrated dissolution estimates were correlated with the depth of the euphotic layer (Ze), with higher values at stations with deeper Ze, e.g., E-1 and E-3 (R 2 = 0.83, not shown).Net production rate of BSi in the euphotic layer (∫ ρSi net ) represents the difference between gross silica production and dissolution rates (Fig. 5) and could be associated with an uptake of "new H 4 SiO 4 ", i.e., uptake that does not come from remineralization processes within the ML.As for net primary production, net silica production could be defined as the part of the BSi that accumulates in the surface layer during the productive period, which would then be potentially available later for export to the mesopelagic layer (Brzezinski et al., 2001;Quéguiner, 2013).During the pseudo-Lagrangian survey, net silica production was quite low during the first two visits (E-1 and E-3 with, respectively, 9.6 ± 0.1 and 0.5 ± 0.1 mmol m −2 d −1 ) and reached a maximal value at the last visit (E-5, 20.5 ± 0.2 mmol m −2 d −1 ).The highest net production rate was observed above the Kerguelen Plateau (A3-2, 43.4 ± 0.4 mmol m −2 d −1 ).In the HNLC area (Station R-2), silica dissolution was higher than silica production, leading to a negative ∫ ρSi net (−1.78 ± 0.02 mmol m −2 d −1 ).

Specific rates of production and dissolution
The specific Si uptake rate (VSi, d −1 ) and dissolution rate (VDiss, d −1 ) give the fraction of the BSi pool produced or dissolved in 1 day as follows: (5) and VDiss = ρDiss VSi is mainly impacted by nutrient and/or light limitation (Frank et al., 2000;Claquin et al., 2002) and by the diatom community composition (Leynaert et al., 2004).During KEOPS-2, VSi values (profiles not shown) presented the same decreasing trends with depth as Si uptake, which is consistent with an impact of light limitation on silica production.Globally, relatively high "integrated specific Si uptake rates" (∫VSi, calculated by the averaged integrated ρ divided by the integrated BSi) prevailed for KEOPS-2 (≈ 0.1 to ≈ 0.3 d −1 ; Table 2).Such values are not different from those of nutrient-replete diatoms growing in the open zone of the Southern Ocean (Brzezinski et al., 2001).By contrast, the HNLC area showed a ∫VSi value below 0.1 d −1 , suggesting non-optimal conditions for the growth of diatoms and/or artifacts of siliceous detritus, which is important in other HNLC regions (e.g., Krause et al., 2010;Fripiat et al., 2011b).

Seasonality of the balance between silica production and dissolution
The D : P ratios integrated between the surface and the 1 % PAR attenuation depth (∫ D : ∫ P ; also summarized in Table 3) are presented in Fig. 5.At the HNLC reference station R-2, a ∫ D : ∫ P value > 1 indicates that the integrated dissolution rate exceeds the measured integrated production rate (note that both fluxes were very low at R-2).This situation leads to a net loss of biogenic silica by dissolution in the euphotic zone and suggests that a short development of diatoms could have occurred before our sampling.This observation  is in accordance with the high barium excess measured between 200 and 400 m at R-2 (Jacquet et al., 2014), indicating a high carbon mineralization activity in the mesopelagic zone, which could be likely associated with a surface production event prior sampling.High ∫ D : ∫ P values have already been measured occasionally in the Southern Ocean during the summer bloom (review in Tréguer and De La Rocha, 2013).
In the Kerguelen bloom area, ∫ D : ∫ P ratios ranged from 0.09 (station A3-2) to 0.95 (station E-3) and depended on the stage of the blooms.The ∫ D : ∫ P ratios were relatively high at stations visited in the beginning of the cruise, indicating that a significant fraction of silica were recycled in the surface waters in early spring, and then decreased as the bloom took place.The highest ∫ D : ∫ P ratio occurred at station E-3.This station was characterized by a low BSi stock (83.6 mmol Si m −2 ), a low integrated BSi production rate (10.5 mmol Si m −2 d −1 integrated over the euphotic layer; Table 2), a dissolution rate close to the mean for all stations and a high specific dissolution rate.This may evidence a higher relative proportion of detrital silica free of organic matter at this station which could be due to stronger bacterial and/or grazing activities inducing a top-down control on diatom growth.Without considering E-3, ∫ D : ∫ P ratios decreased progressively from E-1 to E-5 and showed low values at the most productive stations E-4W, A3-2 and F-L.Here, ∫ D : ∫ P ratios were similar to those measured in nutrient-replete conditions such as productive upwelling regions (Brzezinski et al., 2003).
High ∫ D : ∫ P ratios in winter and in early spring indicate that silica dissolution is sufficient to sustain a large fraction of the low Si uptake rates observed during non-bloom conditions and during the bloom onset, i.e., when primary production is still low.Indeed there is a temporal decoupling of silica production and dissolution since the dissolution kinetics are slow.It is only after diatom death and removal of their protecting organic coating by microorganisms that the silica frustules can dissolve (Kamatani, 1982;Bidle and Azam, 1999;Bidle et al., 2003).By contrast, the progressive decrease of the ∫ D : ∫ P values implies that the majority of gross silica production is sustained by the silicic acid pool supplied from below (winter water) as the bloom develops.This pool can be regarded as the "new" Si reservoir, similar to nitrate for N. Thus, we observe a seasonal shift from Si uptake behaving mainly as a regenerated production before the bloom onset, when silica production is still very low, to then behaving more like a new production during the bloom, when we observe higher Si uptake rates.
An opposite shift at the end of the productive period was suggested by Brzezinski et al. (2001) in the upwelling system of Monterey Bay with ∫ D : ∫ P ratios increasing following the bloom development.In this case, higher ∫ D : ∫ P values were associated with an increase of the relative proportion of detrital BSi in the water column.Similarly, the occurrence of an accumulation of dissolving BSi in subsurface waters following productive periods inducing a net loss of BSi in late summer (∫ D : ∫ P = 1.7) was already identified in the Australian sector of the Southern Ocean (Fripiat et al., 2011b).Since KEOPS-2 took place at the start of the bloom and since there was no silica dissolution rate measured from KEOPS-1, such an increase of ∫ D : ∫ P ratio in the Kerguelen area at the end of the blooming season was not observed but will be discussed in Sect.4.3.
Because silica dissolution profiles were not significantly different from each other between all the KEOPS-2 bloom stations (Fig. 4b), silica dissolution can be ruled out as a process explaining the variability in ∫ D : ∫ P ratios.The observed decreasing trend of ∫ D : ∫ P ratios was actually mainly driven by the increase of BSi production rates (from 3.09 ± 0.01 to 47.9 ± 0.4 mmol Si m −2 d −1 ) and by the accumulation of living diatoms with high specific Si uptake rates in the euphotic layer (Table 2).
The fraction of silica production supported by new silicic acid is estimated by 1 − D : P .During KEOPS-2 it ranged from −0.58 at the HNLC station, where we observed a net loss of biogenic silica, to a maximum of 0. the plateau, where maximum production rates were recorded (A3-2, Table 3).When plotting 1 − D : P vs. gross silica production rate, Brzezinski et al. (2003) found that eight regional estimates of this parameter representing a large range of ocean environments fall along a hyperbolic curve, and thus it might be possible to predict the strength of the silicon pump in a system based on its mean silica production.To obtain a zero-intercept of the curve satisfying the assumption that when the fraction of silica production supported by new silicic acid approaches 0 the production must also be 0, we have plotted 1 − D : P as a function of the net silica production (instead of the gross production in Brzezinski et al., 2003).Since these two variables are not fully independent, the equation of the model matching all data follows a rectangular hyperbola (Fig. 6).This fitting has been obtained from KEOPS-2, Brzezinski et al. (2003) and Fripiat et al. (2011b) data.
Using Fig. 6, we can identify several parameters characterizing the distribution of both KEOPS-2 stations and other oceanic regions.The 1 − ∫ D : ∫ P max is centered around 1 as it is not possible to have more than 100 % of silica production supported by new H 4 SiO 4 .K ρnet (5.89 ± 2.24 mmol m −2 d −1 ) represents the value of net silica production at which the system shifts from a regenerated to a new biogenic silica production (i.e., 1− D : P = 0.5).
In KEOPS-2 stations showing net silica production below K ρnet , the development of diatoms is mainly controlled by recycled sources of silicon; above this value, the supply of new H 4 SiO 4 is the main source of nutrients for biogenic silica production.It is important to keep in mind that this model is mainly governed by the dependency of its two variables, and thus we cannot use it to make predictions about the silicon cycle functioning of stations where only one of the two parameters was estimated (e.g., net silica production estimated from the change in BSi concentrations).However, it allows us to sort KEOPS-2 stations into specific groups and to compare them with other oceanic regions.For example, Fig. 6 is remarkably helpful in differentiating stations with low net production rates (x axis) that have very variable fraction of new Si production (y axis).
KEOPS-2 stations follow the same trend as that of Brzezinski et al. (2003) and encompass almost the full range of variability observed in very contrasting oceanic regions (e.g., HNLC, oligotrophic, coastal upwelling, river plume).They can be sorted in three functional groups: -The "low activity stations" group includes the HNLC reference station R-2 and station E-3 that showed a net loss of BSi with negative values of ∫ ρSi net or close to 0 (Fig. 5, Table 2).In Fig. 6, the HNLC station falls in the negative part of the hyperbolic curve, close to  Fripiat et al., 2011b).Despite its low iron concentration, the high ∫ D : ∫ P ratio observed at R-2 suggests that a short development of diatoms could have occurred before our sampling, in agreement with Jacquet et al. (2014) and Dehairs et al. (2014).This kind of low diatom production in the HNLC area surrounding the Kerguelen Plateau at the end of summer has already been suggested by Mosseri et al. (2008).Since net silica dissolution is not sustainable, the values measured at R-2 should represent conditions that prevail only over a short period of time.Production and dissolution rates are indeed snapshot measurements of over 24 h or 48 h.Although H 4 SiO 4 concentrations were not limiting in surface waters, E-3 was characterized by very low silica production that could have been sustained exclusively by recycled silicic acid (1− D : P = 0.05) and seemed to have approached steady-state conditions as siliceous biomass cannot increase in a system supported solely by regenerated silicic acid (Brzezinski and Nelson, 1989).This situation could be the result of a previous attempt to bloom that would have failed due to the destabilization of the ML.
-The "starting bloom" group is represented by station E-1 which was visited in the beginning of the KEOPS-2 cruise (early November).Although carbon incubation experiments reveal that the bloom began to grow at this station (Cavagna et al., 2014), low Si uptake (Fig. 4a) and low net silica production (Fig. 5) were still observed.A moderate 1 − D : P ratio (0.58) indicates that BSi production at E-1 is controlled both by new and regenerated sources of H 4 SiO 4 .
-The "spring bloom" group includes stations in waters with a high capacity for BSi accumulation, i.e., with low dissolution rates and high net silica production rates.Figure 4a allows us to distinguish between stations from the Lagrangian study E-4E and E-5, with only moderate surface ρSi values (respectively, 0.62 ± 0.06 µmol L −1 d −1 and 0.57 ± 0.06 µmol L −1 d −1 ) and ∫ D : ∫ P ratio close to 0.3 (Table 3); and stations A3-2, F-L and E-4W, showing particularly high surface production rates (> 1 µmol L −1 d −1 ) and a ∫ D : ∫ P ratio close to 0.1 (Table 3).Blooms with such a low ∫ D : ∫ P ratio have the potential to accumulate a large fraction of BSi production and/or export a large amount of BSi into the deep ocean (Quéguiner, 2013;Tréguer and De La Rocha, 2013).Despite their location on both sides of the polar front and in different parts of the Kerguelen bloom, stations E-4W and F-L fall close to each other along the hyperbolic curve (Fig. 6).Consequently, they should operate in comparable ways in terms of silica production dynamics, which is quite similar to the average value of PFZ spring bloom conditions measured by Brzezinski et al. (2001).Therefore, even though complex physical settings (Park et al., 2014) are very different between E-4W (which is not part of PF with high surface DSi concentration of 17 µmol L −1 ) and F-L (with lower surface DSi concentration of 6 µmol L −1 ), diatom production regimes behave as at typical PFZ stations.This was also observed with carbon export and mesopelagic remineralization by Jacquet et al. (2014).
Compared to F-L and E-4W, the water surrounding A3-2 is highly active in terms of silica production and can be compared to the Amazon river plume and coastal upwelling systems such as Monterey Bay or Peru.This highlights once again the exceptional character of diatom-dominated ecosystems sustained by natural iron fertilization in the Southern Ocean.

Decoupling of Si, C and N cycles in the Kerguelen area
In the Kerguelen area, the high NO − 3 concentrations in surface waters compared to H 4 SiO 4 depletion observed annually at the end of the bloom period suggest a strong decoupling of the seasonal consumption of these two nutrients (Mosseri et al., 2008).This situation could be partly induced by differential recycling processes between Si and N strengthening the silicon pump.Si is thus primarily exported to deeper water through the sinking of biogenic silica while PON is mostly recycled in the ML and used as nitrogen source for the development of new phytoplankton organisms including diatoms.Since organic matter is more quickly and efficiently remineralized compared to silica, this decoupling also occurs between Si and C.
The strength of the silicon pump could be investigated by comparing the Si : C and Si : N uptake ratios.In this study, we use only the gross uptake ratios (∫ ρSi : ∫ ρN and ∫ ρSi : ∫ ρC), calculated, respectively, as and which reflect only the stoichiometry of phytoplankton nutrient uptake.We will not consider net uptake ratios that could be calculated but would be biased by the significant rates of nitrification estimated at all KEOPS-2 stations (see Cavagna et al., 2014, andDehairs et al., 2014).Note that both ∫ ρSi : ∫ ρC and ∫ ρSi : ∫ ρN uptake ratios are underestimates of actual diatom uptake ratios because of the simultaneous C and N uptake by non-siliceous organisms.Diatoms growing in nutrient-replete conditions present Si : C and Si : N elemental ratios around 0.13 (from 0.09 to 0.15) and 1 (from 0.8 to 1.2), respectively, with the variability of these ratios depending on diatom species, size classes and growth rates (Brzezinski, 1985;Martin-Jézéquel et al., 2000).Si : C and Si : N uptake ratios are strongly impacted by colimitations which alter growth rates and in most cases increase silicification processes, and thus lead to higher uptake ratios (Claquin et al., 2002;Leynaert et al., 2004;Bucciarelli et al., 2010).During the KEOPS-2 study, ∫ ρSi : ∫ ρC and ∫ ρSi : ∫ ρN uptake ratios vary from 0.10 to 0.38 and from 0.32 to 1.51, respectively (Table 3).
In our study, these variations could not be explained by limitations from a lack of macronutrients (such as silicic acid), as already proposed by Nelson and Tréguer (1992) and Quéguiner (2001).Indeed, KEOPS-2 took place at the beginning of the growth period (October and November) and a bloom onset was observed above the plateau (Blain et al., 2014).It is thus not surprising that macronutrient concentrations in the surface layer did not limit diatom growth (cf.Cavagna et al., 2014, for N uptake).For silicic acid, kinetic experiments conducted during the cruise at all sites demonstrated the lack of response of phytoplankton to H 4 SiO 4 enrichment (data not shown).Indeed, at all stations, mixed layer silicic acid concentrations were high (from 6.2 to 18.5 µmol l −1 ; Fig. 3) preventing the limitation of biogenic silica production by H 4 SiO 4 availability as V Si values at ambient DSi were always similar to V max , the maximum uptake rate achievable when Si is not limiting (data not shown).
Results from a previous cruise in the same area (austral spring 1995) already highlighted the crucial role played by the light-mixing regime on the control of diatom growth in the nutrient-replete waters (Blain et al., 2001).Light limitation also takes part in the decoupling of the Si, N and C cycles by decreasing the growth rate and consequently increasing Si : N and Si : C uptake ratios (Claquin et al., 2002).At all stations the ρSi : ρC uptake ratios increase slightly with depth and reach a maximum at the bottom of the euphotic layer (data not shown), in agreement with the fact that C assimilation is light-dependent through photosynthesis while silicification processes mainly involve energy coming from respiration (Martin-Jézéquel et al., 2000;Claquin et al., 2002).However, because Si uptakes reach very low values at the base of the euphotic layer (Fig. 4a), our data suggest that BSi production rates were not fully independent of light levels and that there was a close coupling between C and Si assimilation processes (see Cavagna et al., 2014, for carbon uptake).This coupling of ρSi with light was also observed in other regions such as the equatorial Pacific and the North Pacific Subtropical gyre by Krause et al. (2011).Note, however, that this is not in contradiction with Claquin et al. (2002) given the different timescales of the two studies.
Limitation by trace metals (especially iron) also alters the stoichiometry of phytoplankton nutrient uptake and its elemental composition and, eventually, contributes to the decoupling of the Si, N and C cycles.From bottle enrichment experiments, it has been argued that diatoms have higher Si : N uptake ratios under Fe stress (Takeda, 1998;Hutchins and Bruland, 1998;Franck et al., 2000).Interestingly, relatively high ∫ ρSi : ∫ ρN ratios were measured for the Kerguelen spring bloom (Table 3), with the highest value above the plateau (1.5, station A3-2) although this area was naturally Fe-enriched (Sarthou et al., 2014;Queroue et al., 2014).In the KEOPS-2 productive stations, diatoms can take up more H 4 SiO 4 compared to nitrogen and carbon.Indeed, these organisms are known to store silicic acid in their vacuoles and are linked to other intracellular components (Martin-Jézéquel et al., 2000;Hildebrand, 2008), or could built more silicified frustules.By contrast, three stations showed low ∫ ρSi : ∫ ρN uptake ratios (0.44, 0.74 and 0.32 for R-2, E-3 and F-L, respectively).Lower Si : N and Si : C integrated uptake ratios in these areas might be partly due to changes in phytoplankton composition.By measuring phytoplankton pigment composition in the HNLC station, Lasbleiz et al. (2014) estimated a lower contribution of microphytoplankton due to a higher proportion of nanophytoplankton organisms such as nanoflagellates.Thus, since the phytoplankton community at R-2 may contain a significant fraction of non-siliceous organisms, the C and N uptake ratios were not solely prescribed by diatoms and thus this could explain the low ∫ ρSi : ∫ ρN uptake ratios observed.However, the higher concentrations in fucoxanthin over the other pigments at all other stations clearly evidence the dominance of large diatoms in the Kerguelen blooms (Lasbleiz et al., 2014).The very low ∫ ρSi : ∫ ρN uptake ratios estimated at F-L could not result from a dominance of non-siliceous phytoplanktonic organisms but likely from different diatom communities showing contrasting degrees of silicification and adapted to the specific hydrological and biogeochemical conditions occurring north of the polar front.Indeed, previous studies in the Southern Ocean have already shown that diatom community composition could better explain differences in silicification than physiological responses to environmental factors such as iron concentration (Baines et al., 2010;Assmy et al., 2013).
Thus, at a seasonal scale and above the plateau, the combination of this increased Si uptake and nitrogen regeneration processes including nitrification (see, e.g., in Cavagna et al., 2014;Dehairs et al., 2014) in the beginning of the bloom, and preferential recycling of organic matter at the bloom offset helps to explain the depletion of most of the Si from the surface layer observed between early spring (18.7 µmol L −1 , averaged in the upper 80 m at station A3-2, KEOPS-2) and the end of summer (1.9 µmol L −1 , averaged in the upper 80 m at station A3, KEOPS-1; Mosseri et al., 2008), while nitrate remains abundant (23 µmol L −1 , averaged in the upper 80 m at station A3, KEOPS-1; Mosseri et al., 2008).Fe-replete diatom assemblages, such as those found at A3-2 (Sarthou et al., 2014), will deplete silicic acid from the water column before nitrate.Such a silicon pump above Kerguelen Plateau would then not be driven solely by Fe-limitation, contrary to incubation experiments from coastal upwelling systems (Hutchins and Bruland, 1998) bottle experiments fertilize, over short timescales (days), a diatom community that is not adapted to higher Fe levels.
Here, by comparing ∫ ρSi : ∫ ρN uptake ratios of different natural communities adapted to their specific ambient Fe levels, our results suggest that natural Fe fertilization might favor diatoms with higher Si : N ratios.Above the Kerguelen Plateau, diatoms seem to maintain a relatively higher degree of silicification until the demise of the bloom, as Mosseri et al. (2008) observed the same range of ∫ ρSi : ∫ ρN at A3 (1.6 ± 0.5, n = 3).
If confirmed in other naturally fertilized regions, these observations may have great implications in our understanding of the past and future functioning of the Southern Ocean and its role in the regulation of climate.Indeed, the silicon pump occurring in the ACC results in an Antarctic Surface Wa-ter (AASW) relatively replete with NO − 3 but strongly depleted of H 4 SiO 4 as observed by Blain et al. (2007) and Mosseri et al. (2008) in the Kerguelen region.This property is then exported toward lower latitudes by the Antarctic Intermediate Water (AAIW) and Subantarctic Mode Water (SAMW) (Sarmiento et al., 2004), currently favoring nonsiliceous organisms production in these regions.Thus, any change in Si : N uptake ratios south of the region of AAIW and SAMW formation might in turn modify diatom productivity at low latitudes.Matsumoto et al. (2002) proposed that a silicic acid leakage hypothesis (SALH) could explain the drop of atmospheric pCO 2 during glacial times and would be mainly driven by changes in Si : N ratios of diatoms induced by an increase in iron supply to the Southern Ocean.This hypothesis is based on in vitro and artificial Fe-enrichment www.biogeosciences.net/11/5827/2014/Biogeosciences, 11, 5827-5846, 2014 experiments showing lower Si : N uptake ratios for diatoms in non-iron-stressed conditions (Takeda, 1998;Hutchins and Bruland, 1998), so that the increase of Fe-deposition in the Southern Ocean during glacial times would drive AASW towards NO − 3 depletion instead of the actual H 4 SiO 4 depletion (Brzezinski et al., 2002).
Our results could indicate that glacial Fe fertilization may not have necessarily resulted in a decrease in Si : N uptake ratios in AASW contrary to what has been proposed as an explanation for the SALH (Brzezinski et al. 2002).Other processes leading to SALH could be invoked, such as shifts in AAIW and SAMW formation rates (Crosta et al., 2007), variations in phytoplankton assemblages (including relative contribution of non-siliceous organisms) or changes in NO − 3 remineralization efficiency.Since this interpretation is based on several incubations for several samplings (during spring and summer) at a single at a single station characterized by Fe-fertilized waters (A3, n = 4) and only one station in the HNLC region (R-2), further investigation concerning the seasonality of net silica production and Si : N uptake ratios in both naturally Fe-fertilized and HNLC areas of the Southern Ocean is clearly needed to validate this observation.

Seasonality and budget of the silicon cycle above the Kerguelen Plateau
In order to investigate the seasonal evolution of the Si biogeochemical cycle in the Kerguelen iron-fertilized bloom (station A3), in Fig. 7 we combined the KEOPS-1 and KEOPS-2 silicon fluxes measured using different isotopic approaches (stable and radiogenic isotope tracer incubations: 30 Si and 32 Si, respectively, and, natural silicon isotopic composition, δ 30 Si, of both diatoms and seawater).Assuming that the whole water mass above the Kerguelen Plateau may have been significantly ventilated with surface waters from the HNLC area annually, Fripiat et al. (2011a) used the HNLC winter water (WW) characteristics to detail the initial conditions of the winter H 4 SiO 4 stock in the fertilized area surface layer (0-100 m).Because the biological activity (Si and C assimilation) took place only in the upper 80 m during KEOPS-2, Si fluxes, stocks and values estimated by Fripiat et al. (2011a) discussed in this section were recalculated from the surface to 80 m (instead of 100 m in Fripiat et al., 2011a).The winter supply of DSi estimated from the seasonal H 4 SiO 4 depletion during KEOPS-1 from the Upper Deep Circumpolar Water (UDCW) to the WW is around 2 mol m −2 , which is not significantly different from the 1.9 mol m −2 DSi stock we measured in mid-October during KEOPS-2 (A3-1).This confirms that, as suggested by Fripiat et al. (2011a), the HNLC winter water is representative of the Si source with initial Si pool conditions prevailing before the bloom onset in the fertilized area.At this time, silica production (which was not measured during this first visit) should be very low since only little BSi accumulation was observed in the ML (79.1 mmol m −2 ), and only 2−1.9 = 0.1 mol m −2 of H 4 SiO 4 was consumed in the surface water before our sampling compared to the estimated WW initial stock.The system was likely exclusively driven by regenerated silica production inducing a potentially high ∫ D : ∫ P ratio (close to 1).Irradiance and mixed layer regimes should be the more likely dominant factors controlling the bloom development for eukaryotes in winter and in early spring (Boyd et al., 2001;Blain et al., 2013).Models have previously reported that the interannual variability of mixed layer depth significantly affects both the date of the bloom onset and the maximum chlorophyll concentration in the region (Pondaven et al., 1998).Then, because H 4 SiO 4 and Fe were not at limiting concentrations in the surface layer (Sarthou et al., 2014), the light-mixing regime that occurred above the plateau in mid-October should have been still unfavorable to diatom growth.
As the surface irradiance becomes more favorable with time, biogenic silica production progressively increases and reaches the highest net production value (46.8 mmol m −2 d −1 ) measured during our second visit to A3 (mid-November).Between these two samplings (i.e., for 28 days), the H 4 SiO 4 depletion in the ML (1.9−1.5 = 0.4 mol m −2 ) yields an average ∫ ρNet of 14.3 mmol m −2 d −1 .Although it was located in different blooms of the Kerguelen region with different diatom communities, this value is in good agreement with the net silica production measured at station E-1, which we characterized as a starting bloom dynamic.Consequently, we can predict that simultaneously with the rise of silica production above the plateau, the ∫ D : ∫ P ratio should decrease toward values around 0.5 as measured at E-1, and that the silica production could be controlled by both new and regenerated sources of H 4 SiO 4 .In this situation, almost all the net BSi production is accumulated in the surface water: 348.5 mmol m −2 (BSi stock integrated over 80 m, this study) and very low export of biogenic silica is possible from the ML to the bottom layer: 4.7 mmol m −2 d −1 (estimated from the difference between net production calculated with DSi standing stocks and that calculated using BSi standing stocks, 14.3 and 9.6 mmol m −2 d −1 , respectively).These observations are in good agreement with the low carbon export (4 % of the surface primary production) estimated during the first visit at A3 by Jacquet et al. (2014).Indeed, such an early spring bloom generally starts by the development of lightly silicified diatoms with potentially low sedimentation rates (Quéguiner, 2013).
The H 4 SiO 4 stock measured during the second visit at A3 (1.5 mol m −2 ) would have been able to sustain the strong net silica production rates measured there for 32 days.Using a natural Si isotopic approach, Fripiat et al. (2011a) suggested that diatoms could additionally receive at least 1.2 mol m −2 of DSi coming from the WW during the productive period in the upper 80 m of the fertilized area, allowing 26 supplementary days of growth for diatoms with the same high Si uptake rate.Taking into account the sum of such winter and summer vertical Si supply, the high productive period, as  De Brauwere et al. (2012) suggested that the bloom could also persist over the same duration without considering this summer Si supply from deep waters.This appears unlikely because the high net silica production we measured at A3-2 could not have been sustained for more than 32 days.Even if such high net silica production could not have been representative of the Si uptake by diatoms over the plateau during the rest of bloom period, an additional source of Si would be needed.The very good accordance of our results with the bloom duration from Mongin et al. (2008) suggests that this summer H 4 SiO 4 input in the ML is realistic and could sustain a significant part of the phytoplankton growth above the Kerguelen Plateau.Indeed, high internal wave activity above the plateau is assumed to be a major process for vertical dissolved iron (and other nutrients) supply on the upper waters (Park et al., 2008;Blain et al., 2008).These vertical fluxes allow a 2-month period of highly active blooming, with a system controlled exclusively by new sources of H 4 SiO 4 (∫ D : ∫ P = 0.09, A3-2, Table 3), inducing the strong DSi depletion observed in January (DSi stocks at 0.2 mol m −2 , KEOPS-1; Mosseri et al., 2008) and the high BSi accumulation in the ML (348.5 mmol m −2 , KEOPS-2; this study) which could be exported at the end of summer when the water column stratification becomes weaker.The gross ∫ ρSi measured here is in the upper range of published values in the Southern Ocean and the specific Si uptake rate was relatively high (respectively, 47.9 mmol m −2 d −1 and 0.28 d −1 in the euphotic layer; Table 2).This could indicate that diatoms have already reached their maximum BSi production rate, and that our second visit to A3 could represent the maximum of the bloom dynamics above the plateau.This is quite consistent with the date of the bloom peak estimated in early December both by modeling and satellite approaches (Mongin et al., 2008, andde Brauwere et al., 2012).The H 4 SiO 4 standing stock in the ML would then be depleted by mid-January (0.2 mol m −2 , KEOPS-1; Mosseri et al., 2008).Then the bloom would shift toward a steady-state dynamic, almost entirely controlled by regenerated Si with decreasing gross silica production down to 10.7 mmol m −2 d −1 , KEOPS-1 (Mosseri et al., 2008) and where H 4 SiO 4 concentration in the ML becomes limiting for diatom growth.
This progression of Si limitation could be associated with a change in the phytoplankton community as observed between the different visits at A3 during KEOPS-1 (Mosseri et al., 2008;Armand et al., 2008), and could be related to the selection of species with higher affinnities with silicic acid, resulting in a better ability to grow at low H 4 SiO 4 concentrations.Such a change in the community structure in response to physical and biological forcing was also proposed in a conceptual scheme by Quéguiner (2013).In spring, diatoms with high growth rates and low degree of silicification dominate the bloom whose development is mainly controlled by new nutrients sources.This diatom assemblage is soon affected by the availability of both silicic acid and iron, and will change for a population showing lower growth rates exclusively sustained by regenerated sources of H 4 SiO 4 .This second assemblage, which begins to dominate in January, could thus persist at steady state until May, when PAR decreases below the threshold of 1 mol photon m −2 d −1 (Blain et al., 2013) in waters that are quite depleted in H 4 SiO 4 , as it is composed by small diatoms with high affinities for silicic acid (Mosseri et al., 2008) and a deep silica maximum (DSM) characterized by strongly silicified large-sized diatoms growing at the base of the ML in the nutrient gradient (Quéguiner, 2013).Under such conditions, a very low net silica production and a ∫ D : ∫ P ratio close to 1 are expected.Unfortunately, no silica dissolution measurements are available from KEOPS-1.However, we can try to estimate the net production from the difference between the KEOPS-1 average gross silica production (10.7 mmol m −2 d −1 ; Mosseri et al., 2008) and average silica dissolution measured at all stations south of the polar front during KEOPS-2 (10.2 mmol m −2 d −1 ).The choice of a constant silica dissolution rate throughout the productive season may seem surprising at first sight.However, there are several reasons to support this hypothesis: (i) Brzezinski et al. (2001) show that the seasonal variability of silica dissolution rates in the Southern Ocean is very low (6.7 mmol m −2 d −1 in October and November to 6.6 mmol m −2 d −1 in February and March); (ii) ∫ ρDiss measured close to the polar front by Fripiat et al. (2011b) in late summer gave values close to our spring dissolution values (4.9 to 6.6 mmol m −2 d −1 ); and (iii) as best seen in Fig. 4b, all KEOPS-2 stations except the HNLC one (R-2) have similar silica dissolution profiles.Thus, it is reasonable to assume that the net silica production estimated above the plateau in late summer should be around 0.5 mmol m −2 d −1 .
The BSi standing stock observed in the upper layer in late summer (275.3 mmol m −2 , KEOPS-1; Mosseri et al., 2008) is lower than that measured in spring (348.5 mmol m −2 ).This could indicate that a part of the BSi that was produced before reaching a silica production in steady state (i.e., sustained almost entirely by regenerating production) was not accumulated in the surface ML at the end of the productive period.In fact, part of the summer production can be stocked in the form of a DSM as measured in late summer above the plateau (550.1 mmol m −2 between 100 and 200 m; Mosseri et al., 2008) which was not yet present at KEOPS-2.This subsurface biogenic silica accumulation mainly results from the combination of sedimentation of living but inactive cells and the occurrence of phytoplankton populations living at depth (Uitz et al., 2009;Fripiat et al., 2011a).The part of the summer net silica production which is not accumulated below the ML should thus be exported to deeper waters as the seasonal stratification breaks down with the intensification of vertical mixing.In  Jacquet et al., 2008).As proposed by Quéguiner (2013), a massive export of biogenic silica and organic matter (58 days × 46.8 mmol m −2 d −1 − (275.3 + 550 mmol m −2 ) = 2.2 mol Si m −2 ) should occur and could then represent the major annual event of the silicon and biological carbon pumps.
Although our budget of the silicon biogeochemical cycle above the Kerguelen Plateau is based on different silicon isotopic approaches and is sustained by silicon stocks and fluxes coming from different years, it matches very well with all previous individual findings.The seasonal variations of ∫ D : ∫ P ratio are in accordance with those observed by Brzezinski et al. (2001) across the polar front zone and are quite well represented by the recent model of Coffineau et al. (2013), who estimated a ∫ D : ∫ P ratio ranging from 0.64 in winter to 0.19 during the spring bloom.The good accordance of our approach with outcomes from different studies also highlights that to fully characterize the silicon cycle in a region of interest we need to measure both silica production and dissolution rates.Indeed, taking only into account the gross silica production in such a synthesis exercise (i.e., without measuring silica dissolution as it is the case in most studies) could lead to misinterpretations of the silicon pump functioning.For instance, we would not be able to identify shifts between "new" and "regenerated" silica production or accurately calculate the real bloom duration without considering silica recycling.Without taking into account dissolution, silica production would have been overestimated by 21 % in the mixed layer (58.2 mmol m −2 d −1 compared to the 47.9 mmol m −2 d −1 of real net silica production), and the bloom duration computation would have yielded 74 days, which is not consistent with the 85 days of KEOPS-1 bloom duration observed by Mongin et al. (2008).

Conclusions
Our study addressed the seasonal evolution of the efficiency of the silicon pump and of the biogenic silica fluxes in the mixed layer under different naturally iron-fertilized bloom conditions around the Kerguelen region.Integrated Si uptake rates were among the highest reported so far in the Southern Ocean.They varied from 3.09 ± 0.01 mmol m −2 d −1 in the HNLC area (R-2) to 47.9 ± 0.4 mmol m −2 d −1 above the plateau (A3-2) and seemed to be strongly coupled with C uptake over depth.Indeed, C and Si assimilation were very low below the euphotic layer, indicating the occurrence of a subsurface accumulation of living but inactive diatoms.Although significant, silica dissolution rates were generally much lower than production rates and did not vary between bloom stations or over depth.
We observed a shift from a BSi production regime based on the regeneration of H 4 SiO 4 during the early stages of bloom onset (with an averaged ∫ D : ∫ P ratio of 0.98) to a regime based on a new production during the bloom development (with an averaged ∫ D : ∫ P ratio of 0.18).This change switched on an active silicon pump which led to the decoupling of the Si and N cycles as well as a strong H 4 SiO 4 depletion of surface water by late summer, with significant implications for global biogeochemical properties.Indeed, the system progressively shifted toward a stronger silicon pump as Si uptake rates increased and nitrogen became more efficiently remineralized when the bloom was well established.This led ultimately to a strong Si limitation and drove the system toward a regenerated silica production regime which allowed the persistence of the bloom in a steady state despite the low concentrations of silicic acid concentrations.Our results confirm and help complete the concept of a seasonal transition from a diatom new production to a diatom regenerated production already proposed in the Antarctic zone by Brzezinski et al. (2003).
Moreover, in contrast to previous artificial Fe-enrichment bottle experiments outcomes, Si : N and Si : C uptake ratios during KEOPS-2 were not higher in the HNLC area compared to the fertilized region.This observation could have great implications on our understanding of processes involved in setting atmospheric pCO 2 during glacialinterglacial transitions.Our results suggest that the increase of low-latitude diatom production observed during glacial periods is probably not controlled primarily by a shift in the nutrient uptake stoichiometry of Antarctic diatoms induced by an enhanced iron supply, as proposed in the silicic acid leakage hypothesis (Matsumoto et al., 2002;Brzezinski et al., 2002), but further investigation is clearly needed to validate this idea.
The combination of the results from the two KEOPS cruises (early spring and late summer) and of different isotopic approaches allowed the first seasonal estimate of a closed silicon biogeochemical budget in the iron-fertilized area above the Kerguelen Plateau based on direct measurements.Our estimates emphasize the benefit of combining different tracers and methods with different sensitivities to physical and biological processes to better constrain and quantify all the processes simultaneously.The major outcome of this seasonal budget is that the winter and summer silicon sources of the mixed layer (3.2 mol m −2 yr −1 ) seem to be well balanced by the combination of biogenic silica accumulation (both in the upper layer and in the winter waters) and late summer BSi export (0.3 + 0.6 + 2.2 = 3.1 mol m −2 yr −1 , respectively).This confirms the occurrence of a significant summer Si supply from winter water as suggested by Fripiat et al. (2011a) sustaining the diatom bloom over the Kerguelen Plateau.
Finally, a striking feature of this study is that naturally iron-fertilized areas of the Southern Ocean, like the Kerguelen Plateau, could sustain a biogenic silica production regime similar to those observed in coastal upwelling systems or in large river plumes.This highlights the exceptional character of diatom-dominated ecosystems associated with natural iron fertilization in the Southern Ocean and their significant role in the global Si biogeochemical cycle.Even if the outcomes of this budget are consistent with previous measurements, large uncertainties remain about the seasonal evolution of dissolution rates at the end of the productive period.Indeed, in order to fully characterize the Si biogeochemical cycle in a region of interest, it is recommended to measure both BSi production and dissolution rates.In combination, the natural silicon isotopic composition (δ 30 Si) of diatoms and seawater represents a powerful tool for identifying silicon sources and silica production over larger temporal and spatial scales (Fripiat et al., 2011a;Tréguer and De la Rocha, 2013) and will be also examined during KEOPS-2.The combination of several isotopic approaches as well as modeling exercises (such as in Pondaven et al., 1998;De Brauwere et al., 2012;Coffineau et al., 2013), by allowing better constraint and quantification of different physical and biogeochemical processes simultaneously, would strongly improve our understanding of the regional Si biogeochemical cycle and its implications for global ocean biogeochemistry.
The Supplement related to this article is available online at doi:10.5194/bg-11-5827-2014-supplement.

Figure 1 .
Figure 1.Map of the KEOPS-2 cruise area (Indian sector of the Southern Ocean) showing the location of stations discussed in this study.Dotted line represents the position of the polar front from Park et al. (2014).

Figure 2 .
Figure 2. Comparison of changes in 30 Si abundance of seawater for each incubation (symbols) with detection limit of the 30 Si isotopic dilution method (solid line) estimated from the reproductibility of an internal standard (0.43 %, n = 40).The dotted line represents the detection limit obtained from the average reproductibility of all dissolution duplicates (0.32 %, n = 35).

Figure 3 .
Figure 3. Vertical distribution of chlorophyll a (solid black line; estimated from CTD fluorescence), biogenic silica concentration ([BSi], light dots) and H 4 SiO 4 concentration ([DSi], dark dots).Dotted lines show the bottom of the euphotic layer (1 % of photosynthetically active radiation, Ze) for each station.Dark dashed lines represent the mixed layer depth (MLD; estimated by Park et al., 2014) and grey dashed lines correspond to a second density gradient identify from the density CTD profile.

Figure 4 .
Figure 4. Vertical distribution of Si uptake (ρSi, a) and biogenic silica dissolution (ρDiss, b) at the KEOPS-2 stations.Open symbols represent the depth at the bottom of the euphotic layer (1 % of photosynthetically active radiation) for each station.

Figure 5 .
Figure5.Si uptake (black), biogenic silica dissolution (white) and net silica production (grey) integrated over the euphotic layer (1 % of photosynthetically active radiation).Italic values correspond to the integrated dissolution-to-production ratio (∫ D : ∫ P ) * Since no dissolution rates were measured at E-4E, dissolution and net production do not correspond to calculations from direct measurements but only to estimations.Dissolution was calculated as the average of all KEOPS-2 integrated dissolution rates.

Figure 6 .
Figure 6.Fraction of biogenic silica production supported by new silicic acid (1 − D : P ) as a function of the integrated net silica production rate (∫ ρ net ) during KEOPS-2 (solid black diamonds) compared with different regions of the global ocean (open circles, Brzezinski et al., 2003, and references therein; and open diamonds, Fripiat et al., 2011b).Triangles and squares show the mean values for different growth seasons for SAZ and PFZ, respectively (Fripiat et al., 2011b).The solid line is a rectangular hyperbola made to fit all data points (grey symbols were excluded from the model since they either represent negative 1 − D : P values which are not allowed by the model or represent averages while all other symbols refer to single stations).The equation of the curve is 1 − D : P = 1.01 × ∫ ρ net / (5.89 + ∫ ρ net ).Dotted lines correspond to ±1 SD of the 1 − ∫ D : ∫ P max .

Figure 7 .
Figure 7. Schematic view of the seasonal silicon cycle in the mixed layer above the Kerguelen Plateau as estimated from natural and enriched Si isotopic measurements.Blue silicon fluxes correspond to estimated values while dark fluxes correspond to direct measurements.The three main water masses are represented in dark grey for Upper Circumpolar Deep Water (UCDW), medium grey for the winter water (WW) and light grey for the mixed layer (ML).Variation of biogenic silica and H 4 SiO 4 standing stocks integrated over 80 m (respectively, [BSi] and [DSi]) are shown in the upper panel.Vertical continuous arrows represent DSi supplies from deep water to the ML, and dotted arrows correspond to particulate silica fluxes.Integrated silica production rates are calculated from the surface to 80 m.Horizontal white arrows represent the state of the bloom (indicated by the D : P ratio) through time.Winter consumption has been estimated from the difference of winter mixing supply(Fripiat et al., 2011a) and the standing stock measured at the first A3 visit (KEOPS-2).The difference in standing stocks between the two visits at A3 (28 days, KEOPS-2) yields the net silica production of 14.3 mmol m −2 d −1 .The DSi standing stock at the second visit of A3 would have been able to sustain the net silica production measured for 32 days (this study) while the summer DSi supply estimated byFripiat et al. (2011a) would have been able to sustain the same net production by an extra 26 days.The fall production measurements and standing stocks are fromMosseri et al. (2008, KEOPS-1).Since no silica dissolution is available from KEOPS-1, fall net production has been estimated from the difference between average gross silica production during KEOPS-1(Mosseri et al., 2008) and average silica dissolution measured at all stations south of the polar front (KEOPS-2, this study).Deep BSi accumulation is calculated by integrating data fromMosseri et al. (2008) over the 100-200 m depth layer (see text for further details).

Table 3 .
Cavagna et al., 2014)tion ratio ( D : P ), fraction of the silica production supported by new silicic acid (1 − D : P ) and silicon-to-carbon (C) and silicon-to-nitrogen (N) uptake ratios (C and N assimilation were measured byCavagna et al., 2014).ρN represents both nitrate and ammonium uptake.All these values are integrated over the euphotic layer (1 % of surface photosynthetically active radiation).Since no dissolution rates were measured at E4E, these values do not correspond to calculations from direct measurements but only to estimations.Dissolution was calculated as the average of all KEOPS-2 integrated dissolution rates.