Solubility of iron and other trace elements in rainwater collected on the Kerguelen Islands ( South Indian Ocean )

The soluble fraction of aerosols that is deposited on the open ocean is vital for phytoplankton growth. It is believed that a large proportion of this dissolved fraction is bioavailable for marine biota and thus plays an important role in primary production, especially in HNLC oceanic areas where this production is limited by micronutrient supply. There is still much uncertainty surrounding the solubility of atmospheric particles in global biogeochemical cycles and it is not well understood. In this study, we present the solubilities of seven elements (Al, Ce, Fe, La, Mn, Nd, Ti) in rainwater on the Kerguelen Islands, in the middle of the Southern Indian Ocean. The solubilities of elements exhibit high values, generally greater than 70 %, and Ti remains the least soluble element. Because the Southern Indian Ocean is remote from its dust sources, only a fraction of smaller aerosols reaches the Kerguelen Islands after undergoing several cloud and chemical processes during their transport, resulting in a drastic increase in solubility. Finally, we deduced an average soluble iron deposition flux of 27 ± 6 μg m−2 d−1 (∼ 0.5 μmol m−2 d−1) for the studied oceanic area, taking into account a median iron solubility of 82 % ± 18 %.


Introduction
The Southern Ocean is known to be the largest High-Nitrate Low-Chlorophyll (HNLC) oceanic area (de Baar et al., 1995).Such zones are characterized by a lack of micronutrients and trace metals in surface waters limiting phytoplankton growth (Martin 1990;Boyd et al., 2000Boyd et al., , 2007;;Blain et al., 2007).In HNLC area, primary production is especially limited by iron supply (Boyd et al., 2007) and could be co-limited by other transition metals, such as manganese (Middag et al., 2011), copper (Annett et al., 2008), cobalt (Saito et al., 2002), zinc (Morel et al., 1991) and nickel (Price and Morel, 1991).Atmospheric deposition is recognized to play an essential role in biogeochemical cycles in remote ocean areas (Duce and Tindale, 1991;Fung et al., 2000;Jickells et al., 2005), even at extremely low levels (Morel and Price, 2003): it brings new external trace metals into surface waters and thus vital bioavailable nutrients for marine biota.It is often assumed that the dissolved forms of trace metals in atmospheric deposition are directly available for phytoplankton because bioavailability is difficult to measure (e. g.Shi et al. [2012]).Indeed, bioavailability depends on several factors, which have to be taken into account to determine it, such as the presence of others nutrients in euphotic surface waters, the residence time of deposited atmospheric particles in surface waters, the soluble fraction and the physicochemical speciation of trace metals in seawater (Boyd, 2002;Boyd et al., 2010).Even if phytoplankton only uses a fraction of atmospheric soluble trace metals in its metabolism (Visser et al., 2003), the best proxy so far is taking the soluble fraction of metals as the bioavailable part of these metals for marine biota (Shi et al., 2012).This dissolved fraction expressed as percentage is referred to as "solubility", for which definition depends on the considered science field (e. g. oceanographic and atmospheric sciences) and the usage context.In this paper, we will define solubility in section 3.1.Numerous studies have been carried out on iron solubility and its controlling factors.Soluble iron in soil represents 0.5% of the total iron (Hand et al., 2004) while it ranges from 0.1% to 90% in aerosols, rains and snows, sampled at different places and times (e.g., Losno 1989;Colin et al., 1990;Zhuang et al., 1992;Guieu et al., 1997;Edwards and Sedwick, 2001;Kieber et al., 2003;Chen and Siefert, 2004;Baker et al., 2006;Buck et al., 2010b;Theodosi et al., 2010;Witt et al., 2010).Most of the solubility values for atmospheric samples are summarized in Mahowald et al. (2005) and Fan et al. (2006).
Variability of iron solubility in the atmosphere is controlled by interactions such as photochemical reactions, cloud processes and organic complexation (e.g., Losno 1989;Zhuang et al., 1992;Kieber et al., 2003;Hand et al., 2004;Chen and Siefert, 2004;Desboeufs et al., 2001Desboeufs et al., , 2005;;Paris et al., 2011), as well as mineralogy of dust sources (Journet et al., 2008) and the element's enrichment factor relative to its natural crustal abundance.Baker and Jickells (2006) also suggested that dust iron solubility may instead be controlled by particle size but this hypothesis was contradicted in Buck et al. (2010a) and Paris et al. (2010).All of these factors combined together can explain the wide range of iron solubility values found in the literature.But, it has to be noted here that part of this range is also due to different experimental protocols used by different researchers for investigating the solubility, which hinder our understanding of the factors controlling solubility (e. g. Baker and Croot, 2010;Witt et al., 2010;Shi et al., 2012 ;Buck and Paytan, 2012 ;Morton et al., 2013).Other studies have observed that the soluble part of other trace elements is highly variable and heterogeneous too.For example, reported solubility ranges from 0.1% to 90% for aluminium and from 10% to 100% for manganese (e.g., Jickells et al., 1992;Colin et al., 1990;Losno et al., 1993;Lim et al., 1994;Guieu et al., 1997;Desboeufs et al., 2005;Baker et al., 2006;Buck et al., 2010b;Hsu et al., 2010;Theodosi et al., 2010;Witt et al., 2010).
Compared to the North Hemisphere, atmospheric supply of micronutrients is believed to be small over the Southern Ocean (Fung et al., 2000;Prospero et al., 2002;Jickells et al., 2005;Mahowald et al., 2005) due to its remote distance from dust sources.In a previous paper, Heimburger et al. (2012a) demonstrated that atmospheric inputs have to be re-evaluated in the Indian part of the Southern Ocean: the authors found that direct measured dust flux is 20 times higher than the previous estimation calculated by Wagener et al. (2008).Therefore, it is highly probable that variation of atmospheric deposition in such an area may strongly influence marine biology and thus carbon sequestration since the Southern Ocean is depicted as the largest potential sink of anthropogenic CO 2 in the global ocean (Sarmiento et al., 1998;Caldeira and Duffy, 2000;Schlitzer, 2000).In this paper, we present measurements of soluble and insoluble composition for crustal elements, including iron, in rainwater samples collected on Kerguelen Islands in the Southern Indian Ocean.To our knowledge up to now such measurements have never been taken over this oceanic region.

Materials
Rains were sampled using a collector placed on top of a 100 mm diameter and 2 m high vertically erected PVC pipe (Fig. 2a).This collector is made from a 24 cm diameter low density polyethylene (PE) funnel attached to an on-line filtration device (Fig. 2b).The filtration device is composed of several parts: a machined high density PE cable fitting holds the bottom end of the funnel and supports a Teflon ® filter holder equipped with a clipped Nuclepore ® polycarbonate membrane (PC) filter (porosity : 0.2 µm, diameter : 47 mm) on a PC supporting grid.The filter holder is placed on the top of a 30 cm high closed section of tubing that is fitted to a 500 mL polypropylene (PP) bottle.
A small Teflon ® pipe lets filtered water flow freely into the bottle.The insoluble fraction of rainwater remains on the surface of the PC filter while the soluble fraction flows by gravity into the PP bottle (Nalgene ® ).The only pieces of equipment that touch the rainwater are the funnel, the Teflon ® filter holder, the PC filters, the PC filter supporting grid and the PP bottles (Fig. 2b).
All the sampling materials were thoroughly washed in the laboratory before the campaign.The 500 mL PP bottles and Teflon ® parts underwent the same washing protocol as described in Heimburger et al. (2012a) for total deposition devices.All of the other materials were: i) washed using ordinary dish detergent in an ISO 8 controlled laboratory room, ii) soaked from two days to one week in a bath of 2% Decon ® detergent diluted with reverse-osmosed water (purified water) and iii) soaked from two to three weeks in 2% v/v Normapur ® analytic grade hydrochloric acid.
Extensive rinsing was performed between each step with reverse-osmosed water.Materials were then transferred to an ISO 5 clean room and: iv) rinsed in Elga TM Purelab ultra ® pure water and v) soaked in a high purity hydrochloric acid solution (2% Merk TM Suprapur ® ), except for the funnels, which were too large for our soaking baths.In an ISO 1 laminar flow bench, these materials were finally: vi) rinsed once (three times for the funnels) with 2% high purity hydrochloric acid solution, vii) five times with ultra pure water and viii) left until dry (two to four hours).Once all the materials had been washed and dried, the funnels were mounted on their high density PE cable fittings under the ISO 1 laminar flow bench and the last three steps of the washing protocol were repeated.They were then individually placed in bags that had been washed in the same way as the materials, and were stored until being used only once in the field.The Nuclepore ® PC filters (0.2 μm porosity, diameter : 47 mm) were i) washed in a bath of 2 % v/v Romil-UpA TM HCl for almost 2 h under the ISO 1 laminar flow bench, then ii) rinsed with ultra pure water, iii) clipped with special rings (FilClip ® ), previously washed by the protocol for materials described above, and iv) stored individually in washed polystyrene Petri dishes until use.

Rain sampling
A clean hood (AirC2, ISO 2 quality), which provided an ultra-clean work zone, was installed inside a dedicated clean area (ISO 6-ISO 7 quality) in the PAF scientific building (see Heimburger et al. (2012a) for more details).It allowed us to prepare rain devices before sampling: i) a clipped filter was placed in the Teflon ® filter holder, ii) a 500 mL PP bottle without its cork was introduced into the 30 cm high closed tubing (the cork was stored in a clean box intended for this purpose) and iii) a funnel with its cable fitting + Teflon ® filter holder were screwed on to the top of the closed tubing.
The plastic bag protecting the funnel's aperture had to be kept in place; a crack was made at the level of the cable fitting.
The sampling started at the beginning of a rain event.A prepared rain device was placed on the top of the PVC pipe; the plastic bag protecting the funnel was removed and conserved.Once the rain event had finished, the funnel was covered by its plastic bag and the device was brought into the clean hood in the scientific building.A vacuum was applied to the section of tubing to help the last rain drops to pass through the filter.The funnel was then removed and no longer used (a new one was used for each sampling).The clipped filter was stored in a clean Petri dish and the 500 mL bottle was weighed.Finally, less than half on hour after the collection of the sample, part of the soluble fraction of rain was stored in a 60 mL Teflon ® bottle.Teflon ® bottles have undergone the same washing protocol as the 500 mL bottles.They contained enough Romil-UpA TM HNO 3 to give a 1% concentration of acid when filled with the collected rain ; the acid solution was used to prevent adsorption of trace metals into the Teflon ® bottle walls during the storage of samples (between six months and two years) before trace metal analyses back in the laboratory.During the 2008 campaign, the pH of samples was immediately measured after sampling: it is equal to 5.4 ± 0.2 (mean ± σ, σ = standard deviation) for all the samples.The Teflon ® filter holder was then rinsed once with 2% Merk TM Suprapur ® hydrochloric acid solution, five times with ultra pure water and allowed to dry in the clean hood before being used for the next sampling.Four laboratory blanks and eight field blanks were performed by simulating a rain event with Elga TM Purelab ultra ® pure water in an ISO 5 clean room and in the field respectively.

Sample preparation and analyses
Back in the laboratory and just before analyses, the soluble fractions of rains (stored in 60 mL 1% HNO3 acidified Teflon ® bottles) were transferred into PP 15 mL sample vials that had been thoroughly washed (see Heimburger et al. (2012a) for details of the washing protocol).The contents of vials were analysed using High Resolution -Inductively Coupled Plasma -Mass Spectrometry (HR-ICP-MS, Thermo Fisher Scientific TM Element 2), which was installed in an ISO 5 clean room and calibrated by diluted acidified multi-element external standards.The sample introduction system was protected by an ISO 1 box.
The contours of the clipped filters, which contained the insoluble fractions of rains, were cut using a new clean stainless steel scalpel blade.The filters of rain samples, laboratory blanks and field blanks were then digested using 4 mL of a HNO 3 / H 2 O / HF solution (proportion: 3 / 1 / 0.5 of pure Romil-UpA TM HNO 3 / ultra pure water / Merk TM Ultrapur ® HF) during 14 h in an air oven at 130°C in closed Savillex TM PFA digestion vessels.Vessels had undergone the same washing protocol as described in Heimburger et al. (2012a) followed by a trial digestion.These vessels were then rinsed and filled with 2% Romil-UpA TM HCl until being used.At the end of digestion, the HF was completely evaporated on a heater plate.5 mL of 1% Romil-UpA TM HNO 3 plus 0.5 mL of Romil-UpA TM H 2 O 2 were then added and left on the plate for 30 minutes.Finally, the content of each vessel was transferred into a 60 mL PP bottle (same washing protocol as for the bottles containing rain samples) with the 1% Romil-UpA TM HNO 3 solution used to rinse the vessel walls.These samples were then analyzed by HR-ICP-MS as well.Seven blank Nuclepore ® PC filters underwent the digestion protocol in order to estimate possible contamination from the filters and the digestion experiments.6 mg of BE-N (Basalt from SARM laboratory, France) and 8.6 mg of SDC-1 (Mica Schist from USGS, USA) geostandards, crushed prior to use, also underwent this protocol in order to estimate the yield and accuracy of our digestion method.
Analytical blanks (n = 7) were carried out using 1% v/v Romil-UpA TM HNO 3 in order to determine the analytical detection limits (DL) of the HR-ICP-MS method.The accuracy (expressed as recovery rate: RR% = mean of measured standard concentrations / certified or published values) and reproducibility (expressed as relative standard deviation: RSD% = σ / mean) of measurements were checked using the certified reference material (CRM) SLRS-5 (Heimburger et al., 2012b) commonly used to control trace metals analysis.This CRM was diluted ten times using 1% v/v Romil-UpA TM ultra-pure nitric acid in ultra-pure water in order to find more similar concentrations between the SLRS-5 and the ones found in samples, allowing calculation of significant RR% and RSD% (Feinberg, 2009).Table 1 presents DL, RSD% and RR% for a set of analysed elements, for which results were validated (see section 3.1) and so discussed afterwards.All the measured concentrations including blanks were above DL: they are three times higher than DL in samples, except for Nd for the soluble fraction.Reproducibility of SLRS-5 measurements is under or equal to 10% for all the elements; accuracy is between 94% and 109%.Measured concentrations in BE-N and SDC-1 geostandards are fairly consistent with the certified ones: RR% are generally equal to 100% ± 30% .

Solubility uncertainties
The solubility in rainwater is expressed as follows: where S X % is the solubility of an element X, [X] soluble is the soluble concentration of X, [X] insoluble is the insoluble concentration of X and [X] total is the sum of [X] soluble and [X] insoluble .The soluble fraction is defined here as the amount of metals in rainwater which passes through the 0,2 µm PC membrane filter.The insoluble one is defined as the amount which stay on the PC filter.If we assume that rainwater is aerosol particles trapped in water drops, solubility is then defined as the fraction of metals that is dissolved in rainwater (i.e. the metal content in the filtrated rain divided by the total metal content in rain) (e.g.Lim et al., 1994;Buck et al., 2010b).This solubility is related to the "fractional solubility" defined by Baker and Croot (2010) for laboratory experiments on aerosol dissolution.Filtration of rainwater during the sampling provides a direct measurement of natural solubility.
To determine [X] soluble and [X] insoluble , we took into account the contamination observed in the different blanks performed (laboratory blanks, field blanks, blank filters; see Sect.2.) for both soluble and insoluble fractions respectively.This contamination is caused by elements remaining in sampling devices, including filters and the walls of equipment in contact with samples.For a given element X, we computed its quantities (Q i ) in each blank by multiplying measured blank concentrations by blank volumes.For the elements presented in this paper, these quantities are found to be similar for both laboratory and field blanks; the quantities in filter blanks are also equivalent to the ones in laboratory and field insoluble blanks.Therefore, all the blanks were pooled together for both fractions respectively in order to extract a global blank defined as the median quantity of all the blank quantities.Figure 3 represents ratios of this median quantity in blanks relative to the one in rainwater, for all the analysed elements in the soluble and insoluble fractions respectively.Expressed as a percentage, these ratios are under 10 % for Ce, La, Mn and Nd for both fractions, under 20% for Al and Fe for both fractions, and reach 35% for Ti for the insoluble fraction only.It has to be noted here that other elements (Co, Cr, Cu, Ni, V, Pb, Zn) were also analysed in rainwater but their ratio values (median quantity in blanks relative to the one in rainwater) were higher than 40 % for the both soluble and insoluble fractions, and even equal to 100 % for Ni and Cu.Thanks to all the blanks we performed, this contamination was identified as coming from PC filters.Although careful washing of these filters, filter blanks exhibit high quantities of Co, Cu, Cr, Ni, V, Pb and Zn compared to the median quantities found in rain samples for these elements after blank corrections.It leads to a contamination of the soluble fraction of laboratory and field blanks, for which no other significant contamination were observed.
For the validated elements (Al, Ce, Fe, La, Mn, Nd, Ti), the median quantity in blanks was subtracted from the ones found in rain samples for each element.[X] soluble and [X] insoluble are consequently given by the following formulas: where [X] analytical represents measured concentrations, V rain the volumes of collected rainwater, and V insoluble the dilution volumes of the digested insoluble fraction.Uncertainties associated with [X] analytical (σ([X] analytical )) are computed using standard deviations and the mathematical approach of exact differential (Feinberg, 2009).Because the quantities of all the blanks are not normally distributed, we used robust statistics for a better estimation of the blank distribution range (Feinberg 2009).
where (1 -RR%) is the accuracy error from SLRS-5 measurements.Standard deviations of [X] soluble and [X] insoluble are then computed as follows: with median absolute deviation MAD = median(|Q i -median(Q i )|) representing the dispersion of blank distribution.Finally, solubility uncertainties are given by the Eq. ( 7): with the coverage factor of k = 2 (Feinberg, 2009), which allows us to obtain an expanded uncertainty representing a confidence level of 95%, i.e. this expanded uncertainty includes 95% of possible solubility values.

Local contamination issues
Rain samples may be contaminated by local soil emission due to human activities on PAF occurring not far enough from the sampling site: soil portions are occasionally moved because of track maintenance generating exposed surfaces that produce local emission spots.Heimburger et al.
(2012a) demonstrated that Ti/Al ratio is a suitable tracer for such contamination: the authors reported that these ratios are equal to 0.15 ± 0.05 (mean ± σ) and 0.04 ± 0.01 in soil and atmospheric deposition samples respectively.Consequently, the [Ti] total /[Al] total ratio was computed for each rain sample (Fig. 4).Uncertainty on this ratio was computed by the following formulas: Rains from P6_09 to P5_08 on Fig. 4 present Ti/Al ratios consistent with the one found in Kerguelen's soil, which is not compatible with pure long range transported particles, and so they were not discussed afterwards.Rain P3_10 exhibits a Ti/Al ratio incompatible with local soil contamination and in the range found in deposition samples (Heimburger et al., 2012a).Four rains (P1_10, P3_08, P6_08, P3_09) have a Ti/Al ratio between the ones in soils and in deposition.If we take into account standard deviation calculated with the Eq. 8 and Eq. 9, a local soil contamination is less probable for P1_10 and P3_08 than for P6_08 and P3_09, for which a small recovery of ranges of both soils and samples is observed.Nevertheless no strong discriminating criterion was found for these four rains, they will be included with rain P3_10 in the following discussion.
To insure that no other local contamination from anthropogenic activities taking place on PAF, we used gdas re-analyzed archives (Draxler and Rolph, 2012;Rolph, 2012) to observe wind direction during the respective sampling times of the five kept rains.The base PAF is located East of the sampling site.For the five rains, winds came from opposite sectors of PAF, excluding wind transported contamination from the base (Table 2).

Rain event fluxes
Deposition fluxes generated by single rain events were computed by dividing the quantities found in each validated rain sample by the surface of the funnel aperture (0.045 m²).In Heimburger et al. (2013), the authors found that atmospheric total deposition fluxes for the oceanic area of Kerguelen and Crozet Islands, averaged over 2009-2010, are equal to 53 ± 2 µg m -2 d -1 and 33 ± 1 µg m -2 d -1 for Al and Fe respectively.Here, we found averaged rain fluxes (wet fluxes) equal to (mean ± σ) 24 ± 18 µg m -2 per rain events for Al and 14 ± 10 µg m -2 per rain events for Fe (Table 3).Because dust deposition is controlled by wet deposition on Kerguelen Islands (Heimburger et al., 2012a), we can neglect the dry deposition flux and thus we can assimilate total deposition flux to the wet deposition one (rainwater events).Taking into account meteorological data that we recorded 8 km from PAF, rain events occur from once a day to every two days, and so with a frequency of 0.5 to 1 per day.Applying this frequency on deposition flux values from Heimburger et al. (2013), the averaged deposition flux on Kerguelen Islands is 51 to 110 µg m -2 per rain event for Al and 32 to 68 µg m -2 per rain event for Fe.These flux values are higher than the ones found in rainwater but they have the same order of magnitude.We can then conclude that rain samples studied in this paper are not unusual events.

Solubility
Before this study, no observed solubility values in rainwater were available in the literature for the oceanic area of Kerguelen Islands.Our values can help to better quantify and model (chemistry and transport) the part of atmospheric iron, which can be bioavailable for phytoplankton in the Southern Indian Ocean.Solubilities in rains are reported in Table 4: they are higher than 70% for all the elements (Al, Ce, Fe, La, Mn, Nd, Ti) for the five considered rain, except for Ti (33% ± 44% and 46% ± 32%) and Fe (57% ± 17% and 51% ± 22%) in P1_10 and P3_09 respectively.The rare earth elements (La, Ce and Nd) also exhibit high solubility values ranging from 68% to 98%.In contrast, solubilities measured for the rejected rain samples show much lower values, for example with a median of 17% for Ti, 9% for Fe and 30% for Al.High solubilities were already observed for some of these elements in the literature.Siefert et al (1999) wrote that "labile Fe" solubility in the fine dust fraction is more than 80% in aerosols collected on-board, while Edwards and Sedwick (2001) reported a Fe solubility ranging from 9% to 89% in snow samples collected in Antarctica and Baker and Croot (2010) modelled a Fe solubility between 0.2% and 100% over the Southern Indian Ocean.Witt et al. (2010) found that Al solubility can reach 91% ± 66% when the soluble fraction of aerosols collected in the North Indian Ocean was extracted with a pH 1 solution.Mn solubility can reach more than 90% in oceanic areas (Baker et al., 2006) and is known to be highly variable (Losno, 1989;Desboeufs et al., 2005;Buck et al., 2010b).Nonetheless, Ti solubility generally exhibits a lower value (<15%) (Buck et al., 2010b;Hsu et al., 2010) than the ones found on Kerguelen Islands (median = 76% ± 13%) although Ti remains the least soluble element in our samples.We did not find any previously published solubility values for La, Ce or Nd.High solubility of Ti informs us that dissolution processes in the atmosphere are very efficient and probably destroy all the solid phases forming original aerosols, including the ones containing REE.
Several studies demonstrate that aerosol solubility increases during particle transport, especially due to cloud processes (Zhuang et al., 1992;Gieray et al., 1997;Desboeufs et al., 2001).It is believed that during their transport in the atmosphere aerosols typically undergo around 10 condensation/evaporation cloud cycles (Pruppacher and Jaenicke, 1995).In clouds, trace gases, such as HNO 3 , SO 2 and NH 3 , are present and modify the pH of cloud droplets, which can increase the soluble fraction of mineral particles.Organic molecules can also increase solubility, e.g.oxalate complexation promoting iron solubility (Paris et al., 2011), as well as photochemistry processes, as reviewed in Shi et al. (2012).Moreover, the average size of mineral aerosols decreases with distance from dust sources, as a result of higher deposition rates for larger particles (Duce et al., 1991).When mineral aerosol size becomes smaller, a greater proportion of their volume is exposed to surface processes (Baker and Jickells, 2006) and is therefore available for dissolution.Ito (2012) support the hypothesis that smaller dust particles yield increased iron solubility relative to larger particles as a result of acid mobilization in smaller particles.In consequence, the smaller the aerosols are and the further they are from their source area, the more soluble they are (Baker and Jickells, 2006).Taking into account both of these hypotheses, we can explain the high solubilities observed on Kerguelen Islands by long range transport from dust sources, which have been identified as South America, South Africa and/or Australia (Prospero et al., 2002;Mahowald et al., 2007;Bhattachan et al., 2012).Indeed, Wagener et al. (2008) and Heimburger et al. (2012a) noted that particles observed on Kerguelen Islands at sea or ground level exhibit 2 µm median diameters, suggesting that only the fine dust fraction, which is believed to be more soluble than the larger dust fraction, reaches Kerguelen Islands.In addition, air mass back trajectories computed from a Hybrid Single Particle Lagrangian Integrated trajectory from the NOAA Air Resource Laboratory (HYSPLIT) model (Draxler and Rolph, 2012;Rolph, 2012) with re-analysed archived meteorological data (gdas) show that air masses travelled for at least five days over the ocean before arriving at our sampling location during the five rain collection period.These air masses did not pass over continents and so did not gain new less soluble continental aerosols.In consequence, continental aerosols coming to Kerguelen Islands underwent several cloud processes during their long range transport in the atmosphere and over the ocean, which probably dramatically increased their solubilities.

Conclusion
Out of a total of 14 single rain events collected on Kerguelen Islands, five samples considered as free of local contamination were validated and are representative of long range transported particles deposited by rain events.Uncertainties are computed by propagating standard deviations of Eq. 5 and Eq. 6. Absolute uncertainties (±) are computed using Eq.7 for each rain sample.

Figure captions:
Figure 1: a) Kerguelen Islands in the Southern Indian Ocean.b) Port-aux-Français on Kerguelen Islands plus picture of rainwater sampling device on PAF.

Figure 2 :
Figure 2: (a) Rainwater sampling device on the top of its PVC tube, (b) drawing of the sampling device, the sampling funnel is cut here.

Figure 3 :
Figure3: Ratio of the median quantities in blanks (all the blanks pooled together) relative to both median soluble (grey) and median insoluble (black) quantities in rainwater samples for all the measured elements.

Figure 4 :
Figure 4: Ti/Al ratios in rainwater samples (grey histogram), in soil samples (dotted black line + hatched rectangle for uncertainties; Heimburger et al., 2012a) and in deposition samples (black line; Heimburger et al., 2012a).Ti/Al in P3_10, P1_10, P3_08, P6_08 and P3_09 exhibit values not compatible with the range of Ti/Al found in soil collected on Kerguelen Islands; these five rains were then considered as not significantly influenced by local soil contamination and so representative of long range transport particles.
Figure 1: a) Kerguelen Islands in the Southern Indian Ocean.

Figure 3 :
Figure 3:Ratio of the median quantities in blanks (all the blanks pooled together) relative to both median soluble (grey) and median insoluble (black) quantities in rainwater samples for all the measured elements.

Table 2
Sampling conditions for the discussed rain events.The funnel collecting surface is 0.045 m².