Chemical fate and settling of mineral dust in surface seawater after atmospheric deposition observed from dust seeding experiments in large mesocosms

. We report here the elemental composition of sinking particles in sediment traps and in the water column following four artiﬁcial dust seeding experiments (each representing a ﬂux of 10 g m − 2 ). Dry or wet dust deposition were simulated during two large mesocosms ﬁeld campaigns that took place in the coastal water of Corsica (NW Mediterranean Sea) representative of oligotrophic conditions. The dust additions were carried out with fresh


Introduction
Dust transported in the atmosphere from the desert areas is known to be a major contributor to oceanic sedimentation in certain regions, notably in the Mediterranean (Loÿe-Pilot et al., 1986;Bergametti et al., 1989). In this region, the dust inputs are usually related to strong deposition pulses of mineral dust from the Sahara (Guerzoni et al., 1999). Thus, the fluctuations of past atmospheric dust fluxes can be used to reconstruct the atmospheric response to climate oscillations throughout the Mediterranean region (e.g., Moreno et al., 2002;Frigola et al., 2007). Moreover, atmospheric dust deposition constitutes the major source of nutrients (N, P, Si, Fe and trace metals) in the Mediterranean surface water (Krom et al., 2004;Bonnet and Guieu, 2006;Pulido-Villena et al., 2010). Dust deposition can be also an efficient mechanism to remove dissolved nutrients from ocean surface waters, notably by adsorption onto sinking particles (e.g., . Thus, dust deposition plays an important role on biogeochemical elemental cycling by acting as both a source and a sink for dissolved nutrients in the Mediterranean surface seawater. Finally, dust can also affect the carbon export in marine environment via a ballast effect on POC export (Ternon et al., 2010), by increasing the sinking velocity of organic particles (Ploug et al., 2008). In consequence, a quantification of dust deposition is essential for assessing the past and present role of dust on the Mediterranean Sea.
Atmospheric dust inputs to the Mediterranean Sea are indirectly assessed from accumulation rates in sediments, and from sediment traps in the water column. Sediment traps in the Mediterranean are also used to quantify and characterize the atmospheric flux of elements from the surface to the deep sea (Goutx et al., 2000;Heimbürger et al., 2014). These marine-based methods are used to validate modeled atmospheric dust fluxes, assuming a conservative dust transfer through the water column. However, the atmospheric and oceanic fluxes estimated from simultaneous measurements of dust fluxes by marine sediment traps and by atmospheric deposition method are not linear (Bory et al., 2002;Neuer et al., 2004). The export of dust to the bottom of the sediment traps is linked to the ballast effect of organic matter produced by biological activity, and hence an efficient downward export of the dust particles to the sediment traps demands a biological activity (Bory et al., 2002;Fischer et al., 2009;Ternon et al., 2010). The high dust deposition triggers a large increase in particles concentrations, enabling aggregation processes and hence inducing a differential settling rate of dust (Lee et al., 2009). Moreover, dust particles can be horizontally advected or redistributed in the water column, before reaching the sediment trap. A number of physical and biological mechanism control oceanic dust fluxes that, so far, are difficult to discriminate and parameterize.
Practically, the estimation of dust export is made by the determination of lithogenic material in sediment traps or records. Generally, the lithogenic component is estimated from Al content in sediments. This method consists of considering that all Al recovered in sediment traps is associated with lithogenic material and that lithogenic material is mainly from dust. By using the average Al content in dust, the dust inputs corresponding to the lithogenic fraction found in the trap can be calculated (e.g., Bory et al., 2002). The interelemental ratios with Al, like Si / Al, Ti / Al or Zr / Al ratios in the Mediterranean region have been also extensively used as proxies for dust input (e.g., Moreno et al., 2002;Frigola et al., 2007). On the basis that Ba / Al ratio in dust is known and likely constant, Ba / Al is considered as a proxy for productivity and used to estimate POC export (Paytan and Kastner, 1996;Mahiques et al., 2009). However, this proxy implies that the major source of elemental Ba to sediments is marine barite, and that there is no significant contribution of Ba from other sources, or that components of Ba in excess other than barite are related to C export in a predictable way. In a similar manner, the ratio Ba / Ti is also used as a proxy for productivity (Averyt and Paytan, 2004). Such approaches assume that the Al, Ba, Ti, Si or Zr dust content is constant during the settling of dust particles. However, the composition of sinking particles could deviate from the initial ratio during sinking since more labile elements are released more quickly than refractory ones. For example, Ba in dust is known to be more soluble than Al (Desboeufs et al., 2001).
The project DUNE (a DUst experiment in a low Nutrient low chlorophyll Ecosystem) aimed at better understanding the effect of dust deposition on the surface waters biogeochemistry of the Mediterranean Sea (Guieu et al., , 2014a. The approach applied in this project was to perform dust addition experiments onto large trace metals clean mesocosms. The original design of these mesocosms represented a unique opportunity to study the dust fate after deposition at the surface down to sediment traps. In particular, the use of mesocosms limits the problem of hydrodynamical artifact, i.e., lateral advection or losses of particles by currents, as observed for in situ sediment traps. In consequence, the DUNE experiments allowed assessing, in controlled conditions, the impact of dissolution, adsorption and chemical or biological process associated with the settling of particles in the surface water column in the case of high dust deposition events in an oligotrophic environment. The set of DUNE experiments simulated either wet (DUNE-P), dry (DUNE-Q) or a succession of two wet deposition fluxes (DUNE-R) of 10 g m −2 of Saharan dust. Here, we present the total mass and the elemental composition (POC, Al, Ba, Ca, Co, Cu, Fe, K, Li, Mg, Mn, Mo, N, Nd, P, S, Sr and Ti) of material collected in the traps and in the water column during the four dust addition experiments. The suite of elements was chosen to include nutrients (N, P, Si, Fe and trace metals: Mn, Cu, Co, Mo), elements used as proxies of marine productivity (POC, Ca, Ba), and elements used as proxies of dust input (Al, Ca, Ti, Nd). We examine the chemical composition of the added dust during its sinking and the  (1) to study the relevance of various proxies of terrigeneous or productivity fluxes and (2) to investigate the link between dust and POC fluxes as a function of mode of deposition. Dust and POC fluxes during DUNE-R presented here were also discussed in a companion paper (Bressac et al., 2014) in which the deconvolution of the different processes involved in POC export was proposed. Coupling metabolic rates in the water column and export fluxes, POC flux directly linked to new production by autotrophs stimulated by the dust deposition was found to represent 50 % of the flux, while the other 50 % were attributed to the "lithogenic carbon pump", a process due to the aggregation between organic material and dust. Their conclusions are discussed in this paper by comparing the dust fluxes between the four DUNE experiments.

Dust seeding and sediment traps sampling
The two series of mesocosm seeding experiments were undertaken at the beginning of summers 2008 and 2010 (Table 1). For these experiments, six (DUNE-P and Q) or seven (DUNE-R) mesocosms were deployed in the bay of Elbo (Scandola Marine preservation area −8.554 • E, 42.374 • N) during typical oligotrophic conditions 2014a). Three or four mesocosms (D1, D2, D3 and Dopt hereafter referred as "Dust-Meso") were seeded with 41.5 g of dust -corresponding to a deposition flux of 10 g m −2 using a trace metal clean spray. The time of the dust addition was the start of the experiment (t 0 ). For the experiments DUNE-P and DUNE-R, the seeding simulated a wet deposition event by spraying diluted cloud processed dust (see Sect. 2.2) in 4 L of ultrapure water. In the case of the experiment R, two successive seedings in the same mesocosms were carried out at time t 0 then at 7 days, i.e., 164 h after the first seeding (first and second seedings, hereafter referred as Dune-R1 and DUNE-R2, respectively). For the experiment DUNE-Q, the seeding mimicked a dry deposition event by spraying fresh dust dispersed in local seawater. In each experiment, three other mesocosms (C1, C2 and C3, hereafter referred as "Control-Meso") were kept unseeded for reference. Mesocosms were covered in order to avoid possible additional inputs from natural dust events.The sediment traps screwed to the base of mesocosms at 15 m depth were recovered and replaced by divers every 48 h for DUNE-P and Q and every 24 h for DUNE-R.

Dust characteristics
The fine fraction < 20 µm in diameter of a dry sieved alluvial soil sample collected in a dust source area in southern Tunisia (33.452 • N, 9.335 • E) has been used to seed the mesocosms . In order to obtain enough quantity of the same material, we used for seeding the fine fraction of soil as analog to Saharan aerosol particles (Desboeufs et al., 1999). Two campaigns of soil sampling were made in March 2007 and March 2009 corresponding to sieved soil Dust07 and Dust09, respectively. The comparison of the physico-chemical properties of both soil samples (chemical and mineralogical composition and size distribution) indicate a good consistency (e.g., for chemical composition in Table 2), both samples being characterized by a large proportion of quartz (40 %) and calcite (30 %), and different clay minerals (25 %) such as illite, kaolinite or palygorskite.
An ageing of fine fraction of soils has been made by mimicking cloud processing with the same procedure for the Dust07 and Dust09 . The fine fraction of soil which underwent the protocol of cloud processing is noted EC-Dust for evapocondensed dust, and the fresh soil is noted NEC-Dust. The Table 1 presents the type of soils used for the three experiments. The effect of the simulated cloud processing on the formation of sulfate and nitrate at the surface of dust was checked by electronic microscope observations. This showed an enrichment of nitrogen and sulfur via the neoformation of evaporite mineral like gypsum ( Fig. 1) consistent with the observations on the ageing of dust during atmospheric transport (e.g., Buseck and Posfai, 1999). The enrichment in sulfur and nitrogen was also observed on elemental composition of EC-Dust07 and EC-Dust09 by X-ray www.biogeosciences.net/11/5581/2014/ Biogeosciences, 11, 5581-5594, 2014 spectrometry fluorescence analysis (Table 2). This enrichment is associated with a decrease in carbon content in the EC-Dust due to the reactions between calcite (CaCO 3 ) and inorganic acids to form the evaporite minerals, as gypsum (CaSO 4 ) or calcium nitrate (Ca(NO 3 ) 2 ), which release CO 2 .

Chemical characterization of sediment trap samples
After recovery, the sample bottles of sediment traps were poisoned at 5 % with a solution of buffered formaldehyde to prevent microbial degradation and grazing by swimmers and were stored at 4 • C in the dark. The samples collected in the sediment trap were treated following the standard protocol developed at the national service "Cellule Piège" of the French INSU-CNRS (http://www.obs-vlfr.fr/ LOV/Pieges). Swimmers were removed by hand-picking under a binocular microscope. The sample was rinsed three times with 50 mL of ultrapure (MilliQ) water in order to remove salt and was then freeze-dried. Mass fluxes were measured by weighing the freeze-dried samples. The accuracy of the weighing was 1 % over the whole data series. Total concentration of carbon and nitrogen were measured in duplicate with a Perkin Elmer 2400 series II elemental analyzer (CHN) on aliquots of the desiccated samples (3-4 mg). Acid digestion described by Ternon et al. (2010)   sured on digested samples after dilution (1/100) by Ametek ICP-AES for Al, Ca, Co, Cu, Fe, K, Li, Mg, Mn, Mo, Nd, P, S, Sr and Ti. The recovery for all the elements was higher than 96 % in CRM samples indicating a good reliability of the digestion method. The accuracy of ICP-AES analyses was checked using SLRS-4 and SLRS-5 (River water standard materials from NRC) as CRM and the detection limits determined (Heimbürger et al., 2013). Reagent blanks were included as control for possible contamination during the analytical process.

Particulate concentration in the water column
In order to follow both the settling of the added mineral particles through the mesocosms and the change of their chemical composition, particulate concentrations for Al, Fe, Ca, Co, Cu, Fe, K, Li, Mg, Mn, Mo, Nd, P, S, Sr and Ti were measured in the water column during DUNE-R and only for Al and Fe during DUNE-P and DUNE-Q. The vertical profiles of particulate Al and Fe for the experiment P are already described in Wagener et al. (2010). The protocol used for sampling and treatment of DUNE-P is described in Wagener et al. (2010). In brief, particulate samples were collected on cellulose acetate filters by filtering one litre of seawater. For DUNE-P and DUNE-Q, the samples were collected from 0, 5 and 10 m depths at 6, 24, 46 and 70 h after seeding. For DUNE-R, the samplings were collected from six depths in the mesocosms (0, 2.5, 5, 7.5, 10 and 12.5 m) up to 164 h after seeding. After filtration, filters were dried under a laminar flow bench and kept at room temperature until analysis. One half of the collected filters were HNO 3 / HF aciddigested then diluted in 10 mL of 0.1 M HNO 3 after complete evaporation. The obtained solutions were analyzed at LOV (Villefranche sur mer) for Al and Fe with a Jobin Yvon (JY 138 "Ultrace") ICP-AES for DUNE-P and DUNE-Q (see Wagener et al., 2010). For DUNE-R, the digestion solutions were analysed at LISA (Créteil) by Ametek ICP-AES with the protocol used for the chemical characterization of sediments. The blank level was under the detection limits for the majority of elements except for Al, Fe, P and Ti with typical blank levels around 1 µg L −1 for Al and Fe and 500 ng L −1 for P and Ti. For Co, Cu, Li, Mo and Nd, the measured concentrations were mainly under the detection limits of this method (around 100 ng L −1 ). As the filters were not rinsed to remove salts, the data on Ca, K, Mg, S and Sr were highly affected by salts contained in the seawater and are not discussed here.

Total and elemental mass in sediment traps
The average total and elemental mass in the sediment traps for the four seeding experiments are presented in Table 3 for C, N, Al, Ca, S for the Control-Meso and Dust-Meso and in the supplementary section for Ba, Co, Cu, Fe, K, Li, Mg, Mn, Mo, Nd, P, Sr and Ti. The total and elemental masses were always higher in the Dust-Meso than in the Control-Meso, except for N in DUNE-Q.
For all experiments in the Dust-Meso sediment traps, the cumulative mass increase was linear with time and significant in the first 90 h, then the cumulative mass is constant until the end of the experiments (Fig. 2a). The maximum mass collected was reached between 24 h and 72 h for the experiment P and R1, whereas for the experiment R2 this maximum was obtained for the first trap samples, i.e., in the first 24 h (Table 3). The cumulative mass at the end of experiment Q was 1 order of magnitude lower compared to the other experiments. In terms of chemical composition, C was the preponderant element in the sediment traps of Control-Meso, whereas Ca, C then Al were predominant in the Dust-Meso sediment traps. The dominance of dust in the sediment traps for Dust-Meso was supported by the visible presence of dust in the samples after collection (Fig. 2b). The highest elemental concentrations for C, N, Al, Ca and S were observed for the experiments DUNE-P and DUNE-R1 and the lowest for DUNE-Q, according to the variability of the cumulative mass (Table 3).
For most elements, it appears that the elemental mass concentrations were linearly correlated with the total mass with a correlation coefficient higher than 0.98, except for N which presented a much larger dispersion (Al and N are shown in Fig. 3). The linearity of the relationship implies that the sediment composition was quasi-constant from 24 h after seeding up to the end of the experiment (168 or 172 h after seeding), showing that the chemical composition of sinking particles collected in sediment traps did not evolve after the first 24 h after dust seeding. Linear regression between total mass and elemental mass enables one to estimate the % of a given element in the collected sediments. For instance, Al was 4.82 ± 0.12 % of the total mass during DUNE-P (Fig. 3), a value significantly higher than the initial Al contribution to total mass in the seeded dust (4.12 ± 0.39 %, Table 1). Such higher elemental concentration in the sediment traps compared to initial concentration in the dust proxy was also observed for all the other studied elements except Ca, S and N. For these three elements, the mass fractions were significantly lower in the sediment traps than in the added dust.
In order to compare the composition of added mineral dust with the particles collected in the sediment traps, we normalized the elemental (X) concentration ratio of X / Al in sediment traps to the X / Al ratio in dust (Table 4). We used Al since dissolved Al measurements in the water column during DUNE-R were shown to be negligible with fractional solubility ranging from 0.74 to 0.84 % (Wuttig et al., 2013). Doing this, we identified enrichment or depletion of elements X independently of total mass variations. No significant change of Ba, Fe, Ti, Nd, Mo and Li contents was observed between added dust and collected particles in the Dust-Meso sediment traps. In contrast, for C, Co, Cu and K, a systematic enrichment in all experiments was found. Inversely, an important depletion of Ca and S was observed. The behavior of N and P was contrasted depending on the experiments: during DUNE-Q, N and P were highly enriched in the sediment traps compared to the added dust; N was depleted in the sediment traps compared to the added dust during DUNE-P and DUNE-R; no significant enrichment or depletion in P was observed during DUNE-P and DUNE-R.
Regarding the depletion of Ca, N, S, they are the major constituents of evaporite minerals as gypsum (CaSO 4 ) or cal-   cium nitrate (Ca(NO 3 ) 2 ), which have been formed by cloud processing into EC-Dust. These minerals are known to be water soluble (Sullivan et al., 2007). Moreover, dissolution experiments performed in laboratory on EC-Dust07 showed that 100 % of N, associated with the neoformation of calcium nitrate was dissolved as nitrate in seawater (Ridame et al., 2013). Therefore, the depletion of Ca, S and N was due to the dissolution of sulfate and nitrate containing particles into seawater after seeding. The depletion was observed for all the samples, whatever the time, after seeding (not shown), showing that this dissolution took place during the first 24 h.

Time series of elemental particulate concentrations in the water column
The profiles of particulate aluminum (pAl) in the water column are presented for DUNE-P, -Q, -R1 and -R2 in Fig. 4. Analogous profiles for Ba, Fe, Mn, P, and Ti are given in supplementary information. The particulate concentrations in Control-Meso were always lower than the ones found in Dust-Meso at the same depth. It is obvious by comparing profiles in Dust-Meso and Control-Meso that lithogenic particles correspond mainly to added dust: particulate Al was thus used as a tracer of this dust. The highest pAl concentrations were observed in the first 5 m of mesocosms in the first 24 h for all the experiments (Fig. 4). For DUNE-R, higher pAl concentrations were found below 10 m after 48 h. A large part of pAl stock in DUNE-R2 remained at the surface until 72 h, whereas this stock was homogeneously distributed over the whole mesocosm during DUNE-R1. This is consistent with the difference of masses collected in sediment traps between DUNE-R1 and DUNE-R2; 164 h after the seeding, pAl concentrations were always higher in Dust-Meso compared to Control-Meso (not shown). No measurement of pAl was made at −2.5, −47.5 and −10 m for DUNE-P and DUNE-Q, limiting the conclusions on the location of added dust for those experiments. The particulate concentrations of Ba, Fe, Mn, P and Ti followed the patterns of Al probably indicating their lithogenic origin.

Mass budget in the sediment traps
A mass budget of dust integrated from the surface to the sediment traps was calculated from the mass budget in the sediment traps and pAl concentrations in the water column. In order to estimate the fraction of dust in the total mass in the sediment traps, the dust mass which is lost from the dissolution of evaporite minerals needs to be quantified. First, we used the Al content in the initially added dust to estimate a theoretical dust mass in the sediment traps (Table 5) which corresponds to the total dust mass without dissolution. This estimated mass corresponds with the total dust mass without dissolution process. Then, for DUNE-P and DUNE-R, we assessed the mass of seeded dust lost after dissolution of CaSO 4 and Ca(NO 3 ) 2 from the depleted part of Ca, N and S in the sediment traps (Table 5). For DUNE-Q, we considered only a potential dissolution of CaCO 3 , the major material www.biogeosciences.net/11/5581/2014/ Biogeosciences, 11, 5581-5594, 2014  containing Ca in NEC-Dust07. So, for DUNE-Q, we used the Ca content in sediment traps material to estimate the dissolved part, considering that all the depletion of Ca is associated with the dissolution of calcium carbonate (Table 5). These estimations also show that in DUNE-Q, only 70 % of the total mass collected in the traps was dust, whereas for the other experiments it accounted for more than 93 % ( Table 5). The dissolution of dust constituted a mass loss of around 7 g (i.e., around 17 % of initial dust mass) in each experiment (Table 6). That implies that only about 34 g from the 41.5 g actually seeded remained under particulate form in the mesocosms. A large part of the introduced dust was not recovered in the sediments traps even after taking account of the dissolution (Fig. 5). We further used the estimated mass percentage of Al in the sediments traps from the Fig. 3 (top) in order to also consider the dissolution of dust particles during their settling. Doing this, 7 days after seeding, only 52, 11, 57 and 41 % by mass of the lithogenic particles initially added were recovered in the sediment traps in DUNE-P, -Q, -R1 and -R2, respectively (Fig. 5). For the DUNE-P and DUNE-R, the temporal change of dust settling was very homogeneous up to 72 h. After 72 h, the settling of dust particles in DUNE-R2 was significantly lower in comparison to the DUNE-P and DUNE-R1. The low recovery of dust mass in the sediment traps even 7 days after seeding suggests that more than 45 % of dust particles (Fig. 5) had sinking velocities below 2.1 m d −1 , whereas the recovery after one day indicates that less than 15 % of dust presented sinking velocities higher than 14.7 m d −1 for DUNE-P and DUNE-R. This is consistent with the results from Bressac et al. (2012) showing that the higher settling velocity of Saharan dust particles could reach 24 to 87 m d −1 during DUNE-R. For DUNE-Q, 89 % of particles had sinking velocities below 2.1 m d −1 , and 1 % of particles had sinking velocities higher than 14.7 m d −1 (Fig. 5). This means that the large majority of deposited dust remained in the water surface layer even 7 days after the seeding. , 11, 5581-5594, 2014 www.biogeosciences.net/11/5581/2014/ Using pAl concentrations in the water column, the remaining mass of dust still in suspension in the mesocosm were estimated to be 12.8, 1.1, 10.8 and 12.9 g, respectively, for DUNE-P after 5 days, DUNE-Q after 3 days, DUNE-R1 and DUNE-R2 after 6 days (these times corresponding to the last sampling in the water column) ( Table 6). When correcting these numbers by the mass fraction that dissolved from dust, the recoveries were 96, 25, 99 and 82 % of the initial dust mass (Table 6). This mass budget shows that a half of dust was found in the sediment traps for DUNE-Q and DUNE-R2 and two-thirds for DUNE-P and DUNE-R1. A critical point of uncertainties in this calculation is the integration of pAl within the water column to estimate the mass of dust in suspension. As previously noted, no measurement of pAl was available below 10 m for DUNE-P and DUNE-Q and a potential high concentration of pAl could have been missed, underestimating the final estimated mass of dust. Thus, it is probable that this low depth resolution was insufficient for the case of the Q experiment, which presented the largest mass fraction in suspension at the end of the experiment. In consequence, this could explain, at least in part, the low rate of recovery for this experiment. Although the low depth resolution could increase the uncertainty on the estimated mass, it is important to note here that the lowest recovery was obtained for DUNE-Q experiment mimicking a dry deposition (see discussion Sect. 4.2).

Estimation of fluxes associated with dust deposition
Settling particles consist of four major components: biogenic opal (opal), biogenic carbonate (bCaCO 3 ), lithogenic particles, and organic matter (POC). In the Dust-Meso, the lithogenic particles corresponded essentially to added dust. In consequence, the fraction of dust was calculated from the estimated mass % of Al in the sediments traps from the Fig. 3, as for the mass budget. The fraction corresponding to biogenic opal was determined from the measurement of biogenic Si, obtained from sequential leaching following Mosseri et al. (2005). In comparison to open ocean sediment traps studies, we have seen that a large part of Ca measured in the traps was from added dust and the total mass of Ca is the sum of Ca as Ca(NO 3 ) 2 , CaCO 3 and CaSO 4 present in dust, plus the bCaCO 3 . Sulfur concentrations in the traps indicate that all the produced gypsum by dust treatment was not completely dissolved. The undissolved mass of CaSO 4 was estimated from the mass of particulate S. The total carbonate mass was assessed from the total mass of Ca minus the mass of Ca related to gypsum. The biogenic carbonate was finally estimated from the total carbonate mass minus the estimated CaCO 3 issued from dust (estimated from the Ca to total mass scatter plot as for Al in Fig. 3). The organic matter was estimated as 2.4 times the organic carbon (Klaas and Archer, 2002), which was issued from the total carbon mass less the total carbonate fraction of carbon. The masses in the Control-Meso were typically at least one order of magnitude lower than the masses obtained in the Dust-Meso. The exported material in Control-Meso was dominated by the POC fraction (30-50 %), and by the lithogenic fraction (20-30 %) regardless of the experiment. Inversely, the lithogenic fraction was the main component of the mass in the Dust-Meso, representing between 66 to 96 % of the total mass, the lowest percentages being measured 6 days after seeding (not shown). POC represented up to 14 % of the total mass. The highest POC contribution was obtained for DUNE-P and the lowest for DUNE-Q. The total mass, POC, dust, opal and bCaCO 3 fluxes have been estimated from the calculated mass fraction in the collected material www.biogeosciences.net/11/5581/2014/ Biogeosciences, 11, 5581-5594, 2014  (Table 7). All fluxes were significantly lower in the DUNE-Q compared to the other experiments, in agreement with the low dust recovery found in the sediment traps of this seeding.

Elemental composition of sediment records: a good proxy for atmospheric inputs?
Chemical elements (except N) and total mass are well correlated in the material collected in sediment traps. Such linearity means that the process likely involved in a modification of dust composition (such as dissolution, adsorption, precipitation, aggregation) occurred in the water column before t = 24 h after the seeding. This confirms that the study of these processes demands a high temporal resolution of dissolved concentrations monitoring as observed by Wuttig et al. (2013). This is in particular the case of evaporite minerals dissolution, releasing Ca, S and N. According to our data, the mass content of those elements is altered after dust deposition. Such dissolution implies also a decrease of dust mass during settling (Table 5), modifying the mass percentage of elements in the collected material in the sediment traps (e.g., Al in Fig. 3). In such very controlled conditions, we observed that the estimation of dust mass from Al is on average 30 % larger than the actual added mass of dust. As mentioned before, Al is often used for estimating the dust mass in sediment traps (e.g., Bory et al., 2002). Our results show that the dissolution of evaporite minerals formed during atmospheric dust transport due to cloud processing could generate an overestimation of the dust total mass estimated in this way. The typical comparison between atmospheric and marine dust fluxes estimated from Al contents shows that estimated atmospheric deposition fluxes are 2-3 times lower than oceanic sediment trap fluxes (Bory et al., 2002;Ternon et al., 2010). This suggests that the uncertainties on the estimation of lithogenic fluxes from Al content in sediment traps could be explained in part by this discrepancy, at least in areas where dust ageing is observed. Practically, the interelemental ratios found in sediments are usually used as proxies of terrigenous input (Ti / Al, Fe / Ca and Ti / Ca) (Mahiques et al., 2009;Govin et al., 2012), and of productivity (Ba / Al and Ba / Ti) Kastner, 1996 andMahiques et al., 2009). Recent studies have shown that the potential of elemental ratios including Ca as Fe / Ca or Ti / Ca are too sensitive to dilution effects by biological components to allow reliable imprints of terrigeneous inputs (Govin et al., 2012). Our data support this conclusion by showing that the high dissolution of Ca in dust triggers an increase in the Fe / Ca and Ti / Ca ratios in particles, making it difficult to use of these ratios to estimate accurate lithogenic fluxes. On the contrary, our results show the stability of content of Al, Fe and Ti in dust during their sinking in the water column, confirming the reliable use of their interelemental ratios as dust proxies. Ba in excess, i.e., the fraction of total Ba not associated with the lithogenic material, i.e., marine barite, is used to estimate the C export flux (e.g., Paytan and Kastner, 1996). During DUNE, the ratio Ba / Al was stable in material collected in the sediment traps, meaning that the Ba in the sediment traps corresponds mainly to the Ba from dust. Indeed, the biological origin of this element was masked by the high dust mass found in the sediment traps, preventing the calculation of Ba in excess. In consequence, the use of the Ba / Al or Ba / Ti ratio as productivity proxies is likely not recommended in case of large dust events such as the one simulated during DUNE. On the Biogeosciences, 11, 5581-5594, 2014 www.biogeosciences.net/11/5581/2014/ contrary, the systematic enrichment observed for Co whatever the experiment (Table 4), confirms a supplementary biological source of this element. For example, Co is known to substitute Zn in the enzyme carbonic anhydrase in some phytoplankton species (e.g Sunda and Huntsman, 1995). An increase of chlorophyll concentrations was observed after dust seeding for DUNE-P and DUNE-R, but not during DUNE-Q (although a strong increase of N 2 fixation by diazotrophs was observed, Ridame et al., 2014). Co / Al was consistent with these results since it was higher when the autotrophs response was higher and inversely. We propose that the ratio Co / Al could be a good productivity proxy even in the case of large dust inputs. However, it should be tested in "real" (deeper) sediment traps material records.

Link between dust deposition state and POC fluxes
The total and lithogenic fluxes obtained for DUNE-P and DUNE-R (Table 6) Bressac et al. (2014) showed a high degree of covariance between POC and lithogenic fluxes in the Dust-Meso for DUNE-R. They explained this link through the ballast effect of added dust on the organic matter present in the mesocosms. This conclusion was supported by the optical measurements during DUNE-R showing that the high sinking velocities of Saharan dust pool (24 to 87 m d −1 ) correspond to the formation of organic-mineral aggregates within the upper few meters of the water column after seeding (Bressac et al., 2012). We observed that the positive covariance between lithogenic and POC fluxes existed also for DUNE-P and DUNE-Q (not shown), indicating a link between dust and POC export in all the experiments. However, our results show that the pattern of sinking of particles was not equivalent for all the experiments (see Sect. 3.3 and Fig. 5). A much slower settling was observed for DUNE-Q, simulating dry deposition of dust. The highest settling was observed for DUNE-P and DUNE-R1 simulating wet deposition of dust.
In order to explain the difference of dust settling in relation with POC, we estimated the mass ratios of lithogenic matter (i.e., dust in the Dust-Meso) to organic carbon in the sediments traps (Litho / POC in Table 7), i.e., dust fluxes normalized to the POC fluxes in the collected material. For all the experiments simulating wet deposition, the mean ratios obtained were very consistent and around 30. This value is in the range of values found in the case of "real" wet dust deposition events onto surface seawater with high organic matter concentrations (Ternon et al., 2010). The lowest ratio (13) is obtained for DUNE-Q corresponding to a dry deposition.
This value is consistent with the ratio observed by Ternon et al. (2010) in Mediterranean summer conditions with a strong stratification of the water column and low Chl a concentrations. To explain these different ratios, it is important to consider the difference of seeding protocols simulating wet or dry deposition (Table 1), but also the physical and biogeochemical conditions in the four experiments (Guieu et al., , 2014a. During experiment P, stratification of the column water inside the mesocosms was not marked whereas stratification was observed during the whole DUNE-Q experiment and toward the ends of both R-seeding periods. However, the dust settling pattern was quasi-similar for DUNE-P and DUNE-R1, meaning that the stratification effect is probably low. Moreover, initial biogeochemical conditions were typical of oligotrophic conditions for all the experiments with very low Chl a concentrations in the range 0.07-0.11 µg L −1 . The chlorophyll concentration was at least doubled for DUNE-P and DUNE-R, proving a fertilizing effect of dust on phytoplankton . Inversely, no Chl a increase was observed after seeding in the Q experiment. Our results show that the highest POC and lithogenic fluxes were observed when an increase of chlorophyll concentrations was observed. On the contrary, the observations for DUNE-Q simulating dry dust deposition showed a slower dust settling and a low POC export related to an ineffective fertilizing effect for autotroph community. In the case of two successive wet deposition simulations (DUNE-R1 and R2), dust export was less efficient in the second seeding even if the chlorophyll increase was equivalent. However, the initial Chl a concentrations were higher for the second seeding, meaning that fresh organic matter produced after the first seeding had not totally disappeared. This observed primary-productivity dependence of lithogenic fluxes in our controlled oligotrophic conditions shows that a high dust export after a dust deposition event needs both a fertilizing effect to produce new organic matter and mineral ballast. This conclusion supports the work from Ternon et al. (2010) that suggested that the high lithogenic fluxes associated with dust deposition likely occur only when there is simultaneous presence of organic matter and lithogenic material (Ternon et al., 2010). This organic matter could be freshly produced due to a fertilizing effect of deposited dust or older organic matter. The high covariance observed between lithogenic and POC fluxes is similar for all the experiments simulating wet deposition, suggesting that the measured ratio Lithogenic / POC fluxes around 30 (Table 7) could be used as reference to estimate the POC export triggered by a wet dust deposition event.
Recently, Bressac and Guieu (2013) defined the "lithogenic carbon pump" to describe the relation between the lithogenic ballasting and POC export, independently of the biological contribution to POC export stimulated by the dust deposition. They suggested that the age and quantity of organic matter could be also essential to estimate the efficiency of the "lithogenic carbon pump". From this concept, Bressac et al. (2014) calculated that this lithogenic carbon pump represented 50 ± 8 and 42 ± 3 % of the total POC fluxes during DUNE-R1 and DUNE-R2, respectively. They propose that the relative decrease in the lithogenic ballasting after the second seeding was due to the scavenging of large quantity of organic matter from the water column following the first seeding. Comparing these conclusions with our observations on the POC fluxes during DUNE-P and DUNE-Q suggests that the "lithogenic carbon pump" was inefficient for DUNE-Q since the POC fluxes in the Dust-Meso were similar with the ones in the Control-Meso. This implies that the initial organic matter probably presented an insufficient concentration or an inappropriate quality (e.g., thickness) to induce lithogenic ballasting in this experiment. However, it is difficult to estimate what was the effect of strong stratification during this experiment on the low POC fluxes. On the contrary, the induced new production during DUNE-P and DUNE-R provided sufficient fresh organic matter to activate the lithogenic carbon pump, while the water column was not strongly stratified. Our results suggest that the "lithogenic carbon pump" mechanism after a dust deposition is more efficient when new production (and thus production of fresh DOM) is induced by the deposition and the stratification is not too marked.

Conclusions
Elemental particulate composition in the water column and sediment traps constitutes useful data for assessing the fate of mineral dust particles deposited at the ocean surface. From controlled artificial seeding experiments in large mesocosms, we have shown that the dust predominated the particulate phase exported at the base of mesocosms (15 m depth) and that dust particles were still in suspension in the enclosed seawater body (52 m 3 ) 164 h after the seeding. Lithogenic and POC fluxes were consistent with direct measured fluxes in sediment traps at 200 m depth in the water column following a strong desert dust deposition event (NW Mediterranean Sea; Ternon et al., 2010). This confirms that data obtained from our experimental mesocosm approach captured the mechanisms of export following a natural dust deposition event.
About 15 % of the initial dust mass introduced was dissolved in the water column in the first 24 h after seeding. This loss was due to the rapid dissolution of calcite for DUNE-Q and from the new minerals, as gypsum or calcium nitrate formed by artificial cloud processing of seeded dust in DUNE-P and DUNE-R. In spite of these dissolutions, the interelemental ratio Ti / Al of seeded dust remained constant during the dust settling, confirming that this ratio is a good proxy for marine lithogenic fluxes. We showed that relatively high Ba content in dust prevents the use of Ba / Al as productivity proxy in the case of high dust deposition such as the ones mimicked during DUNE. Instead, we identified that the ratio Co / Al was linked to the marine productivity and could be a good candidate as a productivity proxy.
The mass budget in the sediment traps and in the mesocosm revealed differences in the dust settling between the different seeding experiments. The higher mass recoveries were measured in DUNE-P and DUNE-R (Fig. 5) and were associated with the highest POC fluxes. This corresponded to the seeding experiments carried out with EC-Dust, i.e., "aged" dust and simulating wet deposition, when a significant Chl a increase after seeding was observed and the stratification was not marked. Inversely, the experiment Q, simulating a dry deposition event of NEC-Dust, i.e., "fresh" dust, presented the lowest recovery of dust mass in the sediment traps, with around 89 % of dust remaining in the water column after 6 days (Fig. 5). This low dust recovery in the sediment trap was concomitant with (1) a low Chl a increase during this experiment, (2) a low POC export and (3) and a strong stratification. We hypothesize that because dry deposition of fresh dust was inefficient to strongly fertilize autotrophs communities in the mesocosms, the dry deposition of fresh dust in our oligotrophic and stratified conditions was inefficient to induce the necessary dissolved organic matter to trigger high POC fluxes by a ballast effect. On the contrary, wet deposition of aged dust was very efficient to trigger high POC fluxes following new production induced by the new nutrients from the dust. The lithogenic fluxes in this case were typically 30-fold higher than the POC fluxes. The different dust deposition simulated during DUNE highlighted a series of processes that modulate the export of lithogenic material and POC after a dust deposition. These processes include the fertilizing effect of dust on the autotroph community, the ballast effect between lithogenic particles and dissolved organic matter and the intensity of the stratification. Further studies should focus on the link between the intensity of the POC export and the type of deposition (dry or wet) since our data do not enable us to conclude if this is a critical parameter.
The Supplement related to this article is available online at doi:10.5194/bg-11-5581-2014-supplement.