Atmospheric deposition fluxes of 26 elements over the Southern Indian Ocean: Time series on Kerguelen and Crozet Islands


  • Alexie Heimburger,

    Corresponding author
    1. Laboratoire Inter-universitaire des Systèmes Atmosphériques, UMR CNRS 7583, Université Paris Est-Créteil, Université Paris Diderot, Créteil, France
    Search for more papers by this author
  • Rémi Losno,

    1. Laboratoire Inter-universitaire des Systèmes Atmosphériques, UMR CNRS 7583, Université Paris Est-Créteil, Université Paris Diderot, Créteil, France
    Search for more papers by this author
  • Sylvain Triquet,

    1. Laboratoire Inter-universitaire des Systèmes Atmosphériques, UMR CNRS 7583, Université Paris Est-Créteil, Université Paris Diderot, Créteil, France
    Search for more papers by this author
  • Elisabeth Bon Nguyen

    1. Laboratoire Inter-universitaire des Systèmes Atmosphériques, UMR CNRS 7583, Université Paris Est-Créteil, Université Paris Diderot, Créteil, France
    Search for more papers by this author


[1] Atmospheric deposition of dust is suspected to have a significant impact on biogeochemical processes in high-nutrient-low-chlorophyll waters of the open ocean. In this study, we report time series of atmospheric deposition samples collected over two years at three different sites on Kerguelen and Crozet Islands in the Southern Indian Ocean. Total atmospheric deposition fluxes were measured for a large suite of elements (Al, As, Ba, Ca, Ce, Co, Cr, Cu, Fe, K, La, Li, Mg, Mn, Na, Nd, Ni, Pb, Rb, S, Si, Sr, Ti, U, V, and Zn). Most of them are identified as coming from sea-salt or crustal sources, but enrichment factor variabilities of Pb, As, Cr, Cu, and V highlight an anthropogenic contribution during the austral winter for these five elements. For Al, Fe, Mn, and Si, deposition fluxes are similar for both Kerguelen and Crozet Islands. Fluxes for the other non-sea-salt elements exhibit differences below a factor of five with a decreasing gradient from Crozet to Kerguelen.

1 Introduction

[2] Dust is recognized to be a major factor affecting biogeochemical cycles through atmospheric deposition [Duce and Tindale, 1991; Jickells et al., 2005]. It has come to be viewed as the main new external source of micronutrients necessary for oceanic productivity in the remote ocean [Fung et al., 2000] even at extremely low concentrations [Morel and Price, 2003]. Iron supply especially limits phytoplankton growth in high-nutrient-low-chlorophyll (HNLC) oceanic surface waters [Martin, 1990; Martin et al., 1994; Coale et al., 1996; Boyd et al., 2000; Blain et al., 2007], such as those of the Southern Ocean [de Baar et al., 1995]. In addition to iron limitation, other nutrients are suspected to colimit phytoplankton growth because of their lower concentration levels in oceanic surface waters, such as cobalt [Saito et al., 2002], zinc [Morel et al., 1991], nickel [Price and Morel, 1991], and manganese in the Southern Ocean [Middag et al., 2011]. Annett et al. [2008] have also shown that some diatoms significantly increase their copper demand in response to iron limitation. Therefore, variabilities in atmospheric input to surface waters could change marine biology and thus carbon sequestration, particularly for the Southern Ocean, which is considered to be the largest potential sink for anthropogenic carbon dioxide in the global ocean [Sarmiento et al., 1998; Caldeira and Duffy, 2000; Schlitzer, 2000].

[3] Atmospheric inputs are 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 remoteness from dust sources. Very little observational data on aerosol concentrations and dust depositions exists for this ocean [Planquette et al., 2007; Wagener et al., 2008, Heimburger et al., 2012a] for comparison with predictions given by global dust deposition models [Jickells et al., 2005, Mahowald, 2007]. In an earlier paper focusing on dust deposition at one site on Kerguelen Islands in 2005 and 2009–2010 [Heimburger et al., 2012a], we found dust flux values of one order of magnitude higher than previously estimated by indirect measurements [Wagener et al., 2008], suggesting that atmospheric contribution to nutrient supplies is significant in this part of the Southern Ocean. In this study, we present total atmospheric deposition fluxes obtained from time series recorded at two locations (Kerguelen and Crozet Islands in the Southern Indian Ocean) for a large set of elements, for which we identified crustal, sea-salt, or anthropogenic origins. Flux values are compared in order to assess or limit possible extrapolations over the Southern Indian Ocean.

2 Materials and Methods

2.1 Sampling Sites

[4] To measure a global dust deposition flux over the Southern Indian Ocean, three different continuous time series were recorded under the program FLATOCOA (FLux ATmosphérique d'Origine Continentale de l'Océan Austral), supported by IPEV (Institut polaire française Paul Emile Victor). One site was set up on Crozet Island (46°S, 51°E) and the two others on the Kerguelen archipelago (49°S, 69°E) (Figure 1a). Crozet and Kerguelen Islands are approximately 1300 km distant from each other, 2000 km north of the Antarctic coast, and 2800 and 3800 km southeast of South Africa, respectively. On Crozet Island, atmospheric deposition sampling was performed monthly from 12 January 2010 to 11 November 2010 on the site “Pointe Basse” (PB, 46°21'47.0" S, 51°42'35.3" E, altitude: 70 m; Figure 1b) located 14 km northwest of the base “Alfred Faure” (AF), the only permanently occupied location on the island. There is occasional human occupation 150 m southeast of the sampling station, i.e., downwind. The two sampling sites on Kerguelen were set up in two different places: one on the top of the “Guillou” Island (G) (49°28'42''S, 69°48'41.7''E, altitude: 90 m), the other in the foothills of “Monts du château” (named “Jacky” (J), 49°18'42.3''S, 70°07'47.6''E, altitude: 270 m) (Figure 1c). These two sites were 30 km apart; they were located 38 km southwest and 8 km northwest, respectively, of the archipelago's only permanently occupied base, known as “Port-aux-Français” (PAF). G was run between 21 November 2008 and 22 December 2009 on a monthly basis. J was run semi-monthly from 24 November 2008 to 7 September 2010 [Heimburger et al., 2012a].

Figure 1.

(a) Locations of Kerguelen and Crozet Islands in the Southern Ocean; sampling sites on (b) Crozet Island and on (c) Kerguelen archipelago.

2.2 Sampling

[5] The setting up of sampling sites and the sampling protocol were identical for J, G, and PB. Two devices (A and B) were installed as duplicates on the top of a PVC pipe stand, measuring 100 mm in diameter and 2 m in height, erected vertically with Kevlar® shrouds. Each device consisted of a 1 L polypropylene Nalgene® bottle containing 100 mL of 1% v/v Romil-UpATM nitric acid, screwed into a Teflon® PTFE funnel (for all of the details, see Heimburger et al. [2012a]). A and B duplicates were 5 m apart from each other at G, 20 m at J, and 8 m at PB. At the end of each sample collection, the funnel was rinsed with 60 mL of 1% v/v Romil-UpATM ultrapure nitric acid in ultrapure water, before the 1 L bottle containing the sample was replaced by a new one. Field blanks were carried out from time to time in order to estimate possible contamination of the samples occurring during the collection procedure due to low dust concentrations found over the area. All of the protocols used ultraclean procedures and thorough washing as described by Heimburger et al. [2012a]. Overall, 36 duplicates (A and B) and 11 single (A) samples were collected at J, 15 duplicate samples at G, and 13 at PB. Operation of the B duplicate device at J stopped on 31 May 2009 and resumed on 4 December 2009.

2.3 Sample Analysis

[6] Back in the laboratory, a large set of elements was analyzed in our samples using high resolution-inductively coupled plasma-mass spectrometry (HR-ICP-MS, ThermoFisher Scientific Element 2) and inductively coupled plasma-atomic emission spectrometry (ICP-AES, Spectro ARCOS) coupled with CETAC ultrasonic nebulization, both installed in a clean room and calibrated by diluted acidified multi-element external standards. Analytical blanks (n = 28) were performed using 1% v/v Romil-UpATM HNO3 in order to determine the analytical detection limits (DL) of both methods (Table 1). The results for 26 elements (Al, As, Ba, Ca, Ce, Co, Cr, Cu, Fe, K, La, Li, Mg, Mn, Na, Nd, Ni, Pb, Rb, S, Si, Sr, Ti, U, V, and Zn) were validated taking into account accuracy and reproducibility computed from the results of both analytical techniques. Details on analysis procedures are available in the work of Heimburger et al. [2012a].

Table 1. Detection Limits of ICP-AES and HR-ICP-MSa
ElementMethodm/z (res.) or λ (nm)DL (ng L–1)
  1. a

    MS: elements analyzed by HR-ICP-MS; AES: elements analyzed by ICP-AES; m/z = mass of the considered isotope; res. = resolution; h = high resolution (>10,000), m = medium resolution (≈4000); l = low resolution (≈300); DL = detection limits; λ = wavelength.

AsMS75 (h)0.73
BaMS137 (l)20
CeMS140 (l)0.016
CoMS59 (m)0.23
CrMS52 (m)0.25
CuMS63 (m)0.77
LaMS139 (l)0.021
MnMS55 (m)2.9
NdMS146 (l)0.041
NiMS60 (m)2.51
PbMS208 (l)0.12
RbMS85 (l)0.23
SrMS88 (l)2.09
UMS238 (l)0.013
VMS51 (m)0.22
ZnMS66 (m)20

3 Results

3.1 Detection Limits, Accuracy, and Reproducibility

[7] Detection limits are almost one order of magnitude below concentrations in samples for most of the elements except for Ba, Li, Ni, and Zn (Table 1). Accuracy (expressed as recovery rate: RR% = mean of measured standard concentrations/certified or published values) and reproducibility (expressed as relative standard deviation: RSD% = SD/mean; SD: standard deviation) of measurements were checked using the two certified reference materials (CRMs) SLRS-4 [Yeghicheyan et al., 2001] and SLRS-5 [Heimburger et al., 2012b], which usually control trace metals analysis (Table 2). Both CRMs were diluted 10 times using 1% v/v Romil-UpATM ultrapure nitric acid in ultrapure water in order to obtain the same level of concentrations in the CRMs and the samples (Figure 2), allowing significant recovery rate and reproducibility to be calculated [Feinberg, 2009]. RR% is generally between 90% and 110%, except for Ca (88% for SLRS-4), K (119% for SLRS-5), Pb (83% for SLRS-4 but 100% for SLRS-5), Ti (89% and 87% for SLRS-4 and SLRS-5, respectively), and Zn (138% for SLRS-5 but 102% for SLRS-4). RSD% is less than 10% for most of the elements, except for As, K, Na, and Zn for both CRMs and Ni and Pb for SLRS-5 only (Table 2). Uncertainties related to measured concentrations (ΔCX) from both ICPs are computed using the mathematical approach of exact differential [Feinberg, 2009]:

display math(1)

where CX is the concentration measured by ICP-AES or HR-ICP-MS for an element X and (1 – RR%) is the accuracy error. The relative concentration uncertainty ΔCX / CX is between 5% and 60% depending on the element, sample, and analytical batch.

Table 2. Accuracy and Reproducibility of Measurements
ElementCertified or Published Values ± SD (µg L-1)Measured Values ± SD (µg L-1)Recovery Rates (%)RSD%
  • a

    No value available in the literature.

  • b

    Uncertainties not provided.

Al54 ± 453 ± 1992
As0.68 ± 0.060.73 ± 0.1210716
Ba12.2 ± 0.612.3 ± 0.51014
Ca6200 ± 2005477 ± 135882
Ce0.360 ± 0.0120.357 ± 0.012993
Co0.033 ± 0.0060.036 ± 0.0031089
Cr0.33 ± 0.020.31 ± 0.02936
Cu1.81 ± 0.081.77 ± 0.11986
Fe103 ± 596 ± 2942
K680 ± 20732 ± 8110811
La0.287 ± 0.0080.269 ± 0.006942
Li0.54 ± 0.070.53 ± 0.04988
Mg1600 ± 1001492 ± 33932
Mn3.37 ± 0.183.29 ± 0.27978
Na2400 ± 2002257 ± 2759412
Nd0.269 ± 0.0140.247 ± 0.009924
Ni0.67 ± 0.080.72 ± 0.051077
Pb0.086 ± 0.0070.072 ± 0.003834
Rb1.53 ± 0.051.54 ± 0.061004
Sa2363 ± 105 4
Si1864 ± 481776 ± 48953
Sr26.3 ± 3.228.0 ± 0.91063
Ti1.46 ± 0.081.30 ± 0.10897
U0.050 ± 0.0030.049 ± 0.001973
V0.32 ± 0.030.34 ± 0.021077
Zn0.93 ± 0.100.95 ± 0.2310224
Al49.5 ± 5.048.6 ± 1.7983
As0.41 ± 0.040.41 ± 0.0710017
Ba14.0 ± 0.514.0 ± 0.61004
Ca10,500 ± 4009801 ± 366934
Ce0.236 ± 0.0080.248 ± 0.0131055
Co0.050b0.051 ± 0.0051029
Cr0.21 ± 0.020.23 ± 0.021089
Cu17.4 ± 1.317.5 ± 1.21017
Fe91.2 ± 5.887.8 ± 2.8963
K839 ± 36999 ± 11011911
La0.196 ± 0.0060.194 ± 0.009995
Li0.50 ± 0.130.49 ± 0.03986
Mg2540 ± 1602395 ± 99944
Mn4.33 ± 0.184.11 ± 0.37959
Na5380 ± 1005544 ± 86510316
Nd0.185 ± 0.0100.181 ± 0.008985
Ni0.476 ± 0.0060.524 ± 0.06311012
Pb0.081 ± 0.0060.081 ± 0.01410017
Rb1.23 ± 0.041.23 ± 0.071006
Sa2330 ± 146 6
Si1881 ± 991779 ± 52943
Sr53.6 ± 1.352.7 ± 2.1984
Ti2.28 ± 0.051.99 ± 0.12876
U0.100b0.093 ± 0.003934
V0.32 ± 0.030.35 ± 0.0310910
Zn0.85 ± 0.101.16 ± 0.3813832
Figure 2.

Certified or published concentrations of the two SLRSs diluted 10 times as function of median concentrations for all of the measured elements in samples.

3.2 Total Atmospheric Deposition Fluxes

3.2.1 Measured Fluxes

[8] Total atmospheric deposition fluxes, i.e., dry and wet deposition fluxes, were computed using the following formula:

display math(2)

where Qtot.dep is the quantity in deposition samples corrected from blanks, Sfunnel is the aperture area of the funnel (equal to 0.0113 m2), and Texposure is the sample exposure time.

display math(3)

Vsample is the volume of samples. Qfield blank(=Cfield blankVfield blank) is the median quantity found in all the field blank samples, representative of possible contamination during sampling. For almost all of the elements, this median field blank quantity generally never exceeds 10% of the amount found in samples and is under or near DLs half of the time. We can nevertheless observe that this quantity can reach 20% for Pb. Exceptional cases must also be noted: (1) Zn, for which field blank quantities are often higher than 10% in all of the samples but never exceed 50%, and (2) Si, for which field blanks performed in 2009 at J represent up to 40% of the amount in samples, which is not observed at G in 2009 or at J and PB in 2010.

[9] Relative uncertainties related to Vsample, Sfunnel, and Texposure are estimated at a maximum of 0.2%, 0.6%, and 0.8%, respectively; these are much lower than relative concentration uncertainties. They are therefore considered to be negligible and are not expressed in equation (4). Deposition flux calculation uncertainties (ΔF) take into account those related to field blank quantities and are computed as follows:

display math(4)

[10] Correlation coefficients (Rc) between A and B duplicates were calculated for each time series. They are up to 0.9 for all of the elements analyzed on the three sites, except for Zn and some elements at the G site, for which correlation coefficients can be as low as 0.7 (Table 3). Since correlations between duplicates are fairly good, A and B fluxes are merged together when both are available (Faveraged; flux data in the supporting information; Al fluxes on the three sites shown on Figure 3). The analytical uncertainties of the obtained fluxes (ΔFanalytical) from duplicate averages are computed by propagation of flux uncertainties:

display math(5)

with ΔFA and ΔFB flux uncertainties from equation (2). In addition, field experiment uncertainties (ΔFexp) are taken into account and defined as the discrepancy between A and B duplicate fluxes. The total uncertainties of averaged daily flux values (ΔFaveraged) for each sample are given in the supporting information and calculated as follows:

display math(6)
Table 3. Correlation Coefficients (Rc) between A and B Duplicates for All of the Analyzed Elements on the Three Sites
ElementKerguelen (J) 2009–2010Kerguelen (G) 2009Crozet (PB) 2010
Figure 3.

Al deposition fluxes (µg m-2 d-1) at Jacky (black line) and Guillou (dotted line) (Kerguelen Islands) and at Pointe Basse (gray line) (Crozet).

3.2.2 Calculation of Averaged Deposition Fluxes Over the Entire Sampling Period

[11] Averaged total deposition fluxes (Fglobal) for each element and each site are computed over the entire sampling period for J, G, and PB, respectively. These fluxes and their associated uncertainties are given by the following formulas (Table 4):

display math(7)
display math(8)

with the coverage factor of k = 2 [Feinberg, 2009], which allows us to obtain an expanded uncertainty representing a confidence level of 95%.

Table 4. Daily Deposition Fluxes (µg m–2 d–1) Averaged Over the Entire Sampling Period, at J, G, and PB
ElementKerguelen (J) 2009–2010Kerguelen (G) 2009Crozet (PB) 2010
Al53.6 ± 2.355.3 ± 5.648.8 ± 3.8
As0.026 ± 0.0020.014 ± 0.0020.052 ± 0.004
Ba0.22 ± 0.030.13 ± 0.020.37 ± 0.04
Ca242 ± 12223 ± 17985 ± 94
Ce0.062 ± 0.0030.056 ± 0.0060.090 ± 0.008
Co0.017 ± 0.0010.027 ± 0.0030.024 ± 0.002
Cr0.012 ± 0.0010.025 ± 0.0040.049 ± 0.010
Cu0.056 ± 0.0030.053 ± 0.0070.121 ± 0.005
Fe29.1 ± 1.338.7 ± 4.134.6 ± 2.1
K305 ± 22342 ± 361407 ± 77
La0.025 ± 0.0010.023 ± 0.0020.042 ± 0.005
Li0.13 ± 0.020.13 ± 0.020.70 ± 0.05
Mg830 ± 32843 ± 513437 ± 208
Mn0.74 ± 0.030.90 ± 0.090.95 ± 0.13
Na7466 ± 5538784 ± 76533,420 ± 10,039
Nd0.025 ± 0.0010.020 ± 0.0020.034 ± 0.003
Ni0.027 ± 0.0030.101 ± 0.0150.085 ± 0.007
Pb0.017 ± 0.0020.008 ± 0.0010.051 ± 0.006
Rb0.086 ± 0.0030.077 ± 0.0040.322 ± 0.010
S579 ± 15436 ± 182434 ± 71
Si81.7 ± 28.392.2 ± 10.197.8 ± 9.9
Sr4.32 ± 0.133.75 ± 0.1917.31 ± 0.69
Ti2.14 ± 0.181.40 ± 0.212.96 ± 0.35
U0.0027 ± 0.00010.0025 ± 0.00020.0091 ± 0.0007
V0.058 ± 0.0030.034 ± 0.0030.135 ± 0.007
Zn0.16 ± 0.030.073 ± 0.0530.37 ± 0.08

3.2.3 Local Contamination From Crustal or Anthropogenic Sources

[12] On Kerguelen Islands, 32 potentially erodible soils were sampled over a 500 km2 windward area around the J and G sites to check for possible crustal contamination from local soil emissions [Heimburger et al., 2012a]. The Ti/Al ratio in soil samples equals 0.15 ± 0.05 (mean ± SD), whereas the Ti/Al ratio in deposition samples is 0.04 ± 0.01 at J and 0.025 ± 0.005 at G. On Crozet, Ti/Al ratios in soils are 0.20 ± 0.06 over the entire island [Segard et al., 2011], whereas the Ti/Al ratio in deposition samples is 0.06 ± 0.02. These results show that significant local crustal contamination of deposition samples from soils can be excluded because Ti/Al mean ratios are more than three times higher in soil than in deposition.

[13] As described in section 1, the G and J sites were placed 38 km southwest and 8 km northwest, respectively, of the only permanently occupied base PAF and could therefore receive contamination from this anthropogenic source (respectively situated northeast and southeast of the sampling sites). A meteorological station was installed at J and recorded wind direction from December 2008 to May 2010. During this period, the southeast sector, and thus wind direction coming from PAF toward J, had a 0.5% median monthly occurrence probability expressed as frequency (Figure 4) with an exceptional maximum probability of 5.43% in February 2010. When the G site was in operation, from December 2008 to December 2009, the northeast sector had a 0.3% median occurrence probability (frequency), with one outlier at 2.9% in January 2009. No particular anthropogenic signal on deposition was observed for either of these exceptional events. We thus deduced that PAF was not a source of anthropogenic contamination in our samples on Kerguelen. On Crozet, no wind directions were recorded at PB, so we used Global Data Assimilation System (GDAS) reanalyzed archives and the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model from the NOAA Air Resource Laboratory [Draxler and Rolph, 2012; Rolph, 2012] with reanalyzed archived meteorological data (FNL) to calculate air mass back trajectories arriving at the site and wind roses for the entire sampling period. This method was verified at J for wind directions obtained from GDAS archives and for those recorded by our meteorological station. Both give similar results for all of the sampling years and as observed in the particular case of February 2010 in Figure 5. On Crozet, wind from GDAS archives and HYSPLIT simulations indicate that most of the wind went from north-northwest to south-southwest sectors at PB in 2010. Thus, the same conclusion can be drawn as on Kerguelen, with regard to the influence on deposition samples of the AF base, situated southeast of PB.

Figure 4.

Wind rose representing wind direction at the J site (Kerguelen Islands) from December 2008 to May 2010. Most of the wind went from west sectors, excluding significant anthropogenic contamination from PAF placed east of sampling sites.

Figure 5.

Three hours wind directions at J for February 2010. Wind directions are obtained from (1) our meteorological station records (" station," full black line) and (2) from GDAS archives (" gdas," dotted gray line). Wind direction (Y axis) is given in degrees with 0° = north, 90° = east, –180° and 180° = south, and –90° = west. We can observe that both methods display similar wind directions for this month, which is also the case for the rest of the sampling period.

[14] Occasional human activities involving two to six people occurred 150 m southeast of the PB station, i.e., generally downwind of the station. They included the use of a domestic waste incinerator at the end of each occupation period. Even if wind went predominately from west sectors at PB during the whole sampling period, we should consider the fact that occasional east winds brought some emissions from the incinerator into our deposition samples. To assess this possible local contamination, K/Na ratios were looked at as an indicator of contamination from refuse incineration in marine air [Ooki et al., 2002]. The characteristic K/Na value is 0.036 in seawater [Thurman, 1994] and can reach 0.045 in sea salt generated in a laboratory [McInnes et al., 1994], while it is 1.2 in fly ash from the incinerator [Mamane, 1988]. K/Na ratios in our deposition samples from Crozet range from 0.0402 to 0.0456 (one sample excluded), with a median value of 0.0428 ± 0.004 (median ± SD) and so are consistent with a sea-salt origin. The excluded sample collected from 18 to 26 October 2010 exhibits a higher K/Na ratio (0.0574), suggesting K contamination. For this sample, both HYSPLIT back trajectories and computed wind roses show possible transport from the local domestic waste incinerator to the sampling device, when the incinerator was potentially used, i.e., at the end of a PB occupation period. This “bad” wind sector is also observed for two other samples, collected from 12 January to 11 February 2010 and from 3 to 11 September 2010, which do not exhibit anomalous K/Na ratios or high fluxes of anthropogenic trace metals. In all other cases when the incinerator was suspected to be used, wind directions from wind roses exclude direct transport from this anthropogenic local source, and wind trajectories generated by HYSPLIT come directly from the ocean.

4 Discussion

4.1 Flux Variabilities and Element Sources

[15] In order to determine different groups of elements depending on the variability of their fluxes, principal component analyses (PCA) were carried out using the free R software (, R Development core team [2011]) and its FactoMineR and MASS packages; correlation coefficients (Rc) were also computed. We observe for each site separately that:

  1. [16] At J (Kerguelen), PCA shows two distinct groups of elements well separated from each other (Figure 6a): (1) Ca, K, Li, Na, Mg, Rb, S, Sr, and U, and (2) Al, As, Ba, Ce, Co, Cu, Cr, Fe, La, Mn, Nd, Ni, Pb, Si, Ti, V, and Zn. Assuming that Al is an exclusive indicator of continental dust transport in atmospheric samples [e.g., Mahowald et al., 2005] and of dust deposition to the ocean [Measures and Vink, 2000; Han et al., 2008], the second group (2) can be identified as having a crustal origin. In this group, we sorted the elements depending on their Rc with Al. We observe that Pb and As are not well correlated with Al (Rc = 0.27 and 0.42, respectively), whereas the Rc of the other crustal elements is between 0.7 (Cr, Ni, and Cu) and 0.99 (Ti and Fe). We also computed the contribution of crustal Na following equation (7) in the work of Losno et al. [1991]. This crustal contribution represents on average 0.34% of the total Na collected at both Kerguelen and Crozet Islands and so implies that Na can be considered as a sea-salt indicator only. Elements in group (1) therefore have a sea-salt origin.

  2. [17] At G (Kerguelen), there is no clearly distinct group on the PCA graph (Figure 6b): most of the elemental fluxes are grouped together, except for Pb, As, Li, Zn, and Cu. Even if PCA results do not show any distinction between flux behaviors, Rc values with Al exhibit the same pattern for Pb and As as at J (Rc = 0.08 and 0.32, respectively). The Rc with Al for sea-salt elements defined previously is from 0.5 to 0.8, whereas it is at least 0.8 for crustal elements.

  3. [18] At PB (Crozet), Al, Ba, Ce, Co, Fe, La, Mn, Nd, Ni, Si, Ti, and V flux variabilities are correlated together (crustal origin); Ca, K, Li, Na, Mg, S, and Sr are also correlated together (sea-salt origin); Cu, Cr, Zn, As, Pb, Rb, and U are separated from the two previous groups (Figure 6c). Again, Pb and As show poor correlation coefficients with Al (Rc = 0.21 and 0.08, respectively).

Figure 6.

Principal component analyses of deposition flux variabilities for all of the elements, highlighting different behaviors and thus different groups of elements for (a) J fluxes, (b) G fluxes, and (c) PB fluxes.

[19] To summarize these results, most of the elements belong to the same group at the three sites. However, Cr and Cu, which are clearly included in the crustal group at J, are detached at PB, and Cu is outside the crustal group at G. Pb and As are never well correlated with elements of crustal or sea-salt groups but well correlated together. Zn never displays a particular correlation with the other elements; this is certainly due to analytical and field blank issues, as mentioned above.

[20] In the light of PCA behavior at all of the sites, three sources must be considered: (1) a sea-salt origin; (2) a crustal origin from South America, South Africa, and/or Australia [Prospero et al., 2002; Mahowald et al., 2007; Bhattachan et al., 2012]; and (3) a possible anthropogenic contribution. To track this third source, we observed Pb behavior. Pb enrichment factors (EF(Pb) = [Pb]/[Al] [Al]Ref/[Pb]Ref) were computed using (1) Al as the crustal reference element and (2) the reference crustal soil composition from Bowen [1966]. As shown in Figure 7, temporal variabilities of EF(Pb) are similar for the three sites and for both sampled years: they clearly increase during the austral winter (June, July, August, and September). During this period, EF(Pb) values are up to 8 (±3) at G, 9 (±2) at J, and 25 (±5) at PB compared to the rest of the year, when their median ± SD values are equal to 0.7 ± 0.2, 1.4 ± 0.7, and 3 ± 2, respectively. Other known anthropogenic elements are As, Cu, and Cr; their EF varies similarly to EF(Pb). Although V is associated with the crustal group only from PCA results and exhibits high correlation coefficients with Al, its EF shows a seasonal pattern similar to EF(Pb) too. In comparison, median EF(Fe) is 1.31 ± 0.19 at G, 1.01 ± 0.11 at J, and 1.39 ± 0.22 at PB for all sampling periods (Figure 7). EF(Fe) is stable over time without noticeable temporal variabilities; the same is true of the other crustal elements, i. e., Ba, Ce, Co, La, Mn, Nd, Ni, Si, and Ti. We can thus suppose that an excess of Pb, As, Cu, Cr, and V comes from an anthropogenic source to the Southern Indian Ocean during the austral winter, which impacts Crozet more than Kerguelen Islands.

Figure 7.

Enrichment factors of Pb (solid line) and Fe (dotted line) at J (black), G (light gray), and PB (dark gray) (left Y axis: J and G results; right Y axis: PB results). Al is used as the crustal reference element and the work of Bowen [1966] as the crustal soil composition reference. During the austral winter and for the both sampling years, we observed Pb enrichment, which is three times higher on Crozet than on Kerguelen Islands.

4.2 Geographical Deposition Trends and Extrapolations

[21] Deposition flux values averaged over the entire sampling period for each site are shown in Table 4. Compared to the literature, the total Fe deposition flux is in good agreement with the one proposed by Planquette et al. [2007] on Crozet (50 µg Fe m–2 d–1), but not with Wagener et al. [2008] on Kerguelen Islands (0.8–2.6 µg Fe m–2 d–1), who underestimated atmospheric fluxes in this oceanic area as demonstrated by Heimburger et al. [2012a].

[22] Because sampling periods are not the same for the three sites (J sampled over two years, G in 2009, and PB in 2010) and thus the given fluxes on Table 4 cannot be compared directly with each other, we computed averaged fluxes corresponding to common sampling times for (1) J and G in 2009 and (2) J and PB in 2010. Using this approach, G/J and PB/J flux ratios for non-sea-salt elements were computed and are shown in Figure 8. For Al, Fe, Mn, and Si, ratios are very close to unity: their respective deposition fluxes are quite similar at the three sites. Weighted averaged fluxes can therefore be considered for the studied area: they are equal to 53 ± 2 µg m-2 d-1 for Al, 33 ± 1 µg m-2 d-1 for Fe, 0.83 ± 0.04 µg m-2 d-1 for Mn, and 88 ± 14 µg m-2 d-1 for Si. These averaged fluxes for Al and Fe are similar to the ones found in the work of Heimburger et al. [2012a]; dust flux derived from Al measurements is then the same as in this previous study. Flux seasonality for Al and Fe reported by Heimburger et al. [2012a] can be applied for Si and Mn since these four elements are strongly correlated together, as shown by the PCA results. Since our time series recorded deposition during 18 months only, we presently have no clear idea of the interannual variability of those deposition fluxes over the Southern Indian Ocean. Longer-term measurements will be necessary to address this question fully.

Figure 8.

G/J (square) and PB/J (diamond) flux ratios for non-sea-salt elements. The black line represents J/J ratios, which are equal to 1. Ratios were computed from averaged flux values obtained for the same sampling periods (1) at J and G, i.e., 2009, and (2) at J and PB, i.e., 2010.

[23] For the other elements, we can observe in Figure 8 flux variabilities generally larger (1) between PB and J than between G and J, and (2) for elements for which an anthropogenic contribution is suggested. The largest flux differences between J and G are observed for Ni and Zn. No evidence was found to explain Ni behavior, especially since sample analyses for this element were validated (section 3.1, Table 2) and their duplicates correlated (Table 3), which is not the case for Zn. Zn and Ni fluxes are nevertheless of the same order of magnitude on the three sites, respectively, and so can be taken as indicative values. However, if Ni and Zn are excluded, differences between sites never exceed a factor of two relative to J, except for V, Pb, and Cr (2.3, 2.2, and 3.6 between J and PB, respectively): deposition flux extrapolations could then simply be applied over the studied oceanic area with a maximum uncertainty given by the maximum discrepancy observed for each element (Table 4).

5 Conclusion

[24] We measured total atmospheric deposition fluxes for an extensive set of elements (Al, As, Ba, Ca, Ce, Co, Cr, Cu, Fe, K, La, Li, Mg, Mn, Na, Nd, Ni, Pb, Rb, S, Si, Sr, Ti, U, V, and Zn) continuously over two years at three sites on Kerguelen and Crozet Islands. Using principal component analyses on the temporal variability of elemental fluxes and enrichment factors, we deduced a possible anthropogenic contribution for some elements (Pb, As, Cu, Cr, and V) over the Southern Indian Ocean during the austral winter, highlighting that both locations are influenced by different atmospheric processes over the year, as transport and/or sources. We also show that Al, Fe, Mn, and Si fluxes are similar at both Kerguelen sites and between Kerguelen and Crozet, which are situated 1300 km apart. For the other non-sea-salt elements, we observed larger differences between flux values from a factor of two to a factor of five with a decreasing gradient from Crozet to Kerguelen Islands. To gain a better understanding of the biogeochemical processes taking place in the Southern Ocean, new field experiments, including long time series (at least two years), still need to be performed at several sites, such as the ones proposed by Schulz et al. [2012], which include Kerguelen, Crozet, and also Falkland Islands. New field experiments also must take into account the recommendations of Heimburger et al. [2012a], who advise to directly sample atmospheric deposition using duplicate devices with a maximum care of contamination issues. One of such issues is caused by local anthropogenic activities; more remote sampling sites from human occupation should be chosen, implying the use of automatic sampling stations.


[25] We would like to thank the French polar institute " Institut polaire français Paul Emile Victor" (IPEV), which supported our program " FLux ATmosphérique d'Origine Continentale sur l'Ocean Austral" (FLATOCOA). We also thank the " Terres Australes et Antarctiques Françaises" (TAAF) team for their help. The authors gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for the provision of the HYSPLIT transport and dispersion model and/or READY website ( used in this publication.