Atmospheric Deposition Over the Caribbean Region: Sea Salt and Saharan Dust Are Sources of Essential Elements on the Island of Guadeloupe

Dust emitted from North Africa is transported over long distances and has a strong impact on large areas over the North Tropical Atlantic Ocean. Sea salt emitted by the sea surface is the second source of essential elements transported in the atmosphere and plays a major role in the cycles of alkaline‐earth metals in the ecosystems of tropical North Atlantic Islands. The total atmospheric deposition fluxes were continuously sampled on a weekly basis in Guadeloupe, Lesser Antilles, from March 2015 to August 2018 (41 months). Elemental deposition fluxes, including Al, Ca, K, Mg, Fe, Na, P, S, and Zn, were measured for all samples to provide the first long time series of atmospheric elemental deposition fluxes over the Lesser Antilles region. It is shown that: (a) the three sources of atmospheric deposits in Guadeloupe for the presented elements are sea salt (for K, Ca, Mg, Na, and S), long‐range transported Saharan dust (for Al, Ca, K, and Fe), and biogenic particles (for P and Zn); (b) the average deposition mass fluxes of sea salt and Saharan dust are 17.4 and 11.2 g.m−2.year−1, respectively, without noticeable inter‐annual variations; (c) a pronounced seasonality is found for the Saharan dust deposition, for which maximum flux values are observed between June and July each year and 85% of the annual deposition flux occurs between April and September; (d) the deposition flux of sea salt is strongly correlated to local wind speed, without seasonality.

Most atmospheric deposition studies in the Caribbean have involved short-term direct deposition measurements, or indirect estimations from aerosol measurements or satellite observations for Saharan dust. Mahowald et al. (2005) pointed out the uncertainties of such estimations for iron fluxes due to spatial and temporal variability and the limited number of observations. These uncertainties reach a factor up to 10 for some fluxes estimated from satellite observations (e.g., Husar et al., 1997), and more than 50 for aerosol deduced fluxes (Heimburger et al., 2012).
Long-term studies of direct measurements of deposition fluxes are scarce. The longest study published today was conducted in Puerto Rico, where an analysis of total rainfall chemistry data spanning 20 years has been recorded (Heartsill-Scalley et al., 2007;McClintock et al., 2019). The elemental composition of atmospheric deposits was measured monthly in a highly polluted region of Cuba, from 2014 to 2016 (Morera-Gómez et al., 2019). The most documented study on trans-Atlantic dust deposition is a 3-year (from 1994 to 1996) terrestrial dust deposition record in Florida provided by Prospero et al. (2010), where both dry and wet deposition fluxes of dust were measured in nine different stations. Increasing the number and quality of the direct deposition measurements add more constraints to models and greatly improve their accuracy. These measurements are most useful when accumulated over long time periods to validate and to improve these estimations (Huneeus et al., 2011). All these studies show that it is necessary to acquire continuous atmospheric data in the Caribbean basin, over the long-term, to: (a) better investigate the contribution of dust in ecosystem nutrient cycling, in a tropical context, (b) investigate the seasonality and periodicity of this atmospheric input, and (c) improve the global aerosol models and the understanding of this environmental process.
It is therefore essential for model constraints to directly measure atmospheric deposition fluxes over long periods of time to better understand the past, present, and future of atmospheric evolution and its impact 3 of 19 on the earth and the bio-geochemical system (Albani et al., 2015;Goudie & Middleton, 1992;Mahowald et al., 2010Mahowald et al., , 2011Mahowald et al., , 2014. This study presents the temporal dynamic of atmospheric deposition for major and essential elements, including Al, Fe, Na, P, Ca, Mg, K, S, and Zn, during a continuous time series from March 2015 to August 2018.

Site Description and Sample Collection
Basse-Terre Island is part of the Guadeloupe archipelago (Lesser Antilles, in the Caribbean Ocean), and has a surface area of 950 km 2 . The island is characterized by a wet tropical climate, with a mean annual temperature close to 25°C and 75%-95% humidity (Chaperon et al., 1985). The Guadeloupe climate is characterized by two seasons: the dry season from approximately January to June and the cyclonic wet season from July to December. During the wet season, hurricanes and tropical depressions produce intense individual rainfall events, most of which occur in August, September, and October in Guadeloupe (Zahibo et al., 2007). The total atmospheric deposition was sampled continuously on a weekly or bimonthly basis for 41 months from March 2, 2015, to August 3, 2018, at the OVSG site (Observatoire Volcanologique et Sismologique de Guadeloupe, 15°58′50″ N, 61°42′13″ W, altitude: 411 m, Figure 1). The site is located at the top of Houëlmont, 9 km south-west and outside the atmospheric influence of the La Soufrière volcano, 6 km west of the nearest up-wind beach, 5 km south-west of the nearest up-wind town (Trois-Rivières). The location of the sampling site prevents any local anthropogenic contamination, especially from the urban and industrial areas of Pointe-à-Pitre (Cécé et al., 2016).
The sampling system was the same as the one described in Heimburger et al. (2012) and used hydrochloric acid instead of nitric acid as the preservative solution. Briefly, a 12 cm diameter funnel made of Teflon™ was attached to a clean 1 L polypropylene bottle pre-loaded with 50 ml of ultra-pure 3% v:v HCl aqueous solution and left open to total atmospheric deposition. The sampling system was fixed 2 m above the roof of a building and approximately 12 m above ground level (Figure 1). Each time a sample was renewed, 50 ml of a 3% v:v HCl water solution was poured into the funnel to collect the last deposits and to rinse it, and the bottle was replaced by a new one. Each sample was then weighed and completely evaporated under ultra-filtered airflow at 85°C according to the system presented in Figure S1 in Supporting Information S1. Seventeen field blanks, including an evaporation step of hundred milliliters of rinsing solution, were made to check for any possible contamination. The dominant wind direction is from left to right in the figure (from east to west). The monthly rainfall data for 2015 was measured by a rain gauge installed by Météo France at the OVSG site and the rainfall data from 2016 to the end of the sampling period is measured by a rain gauge installed by IPGP at the same place.

Sample Digestion
All analyses were performed in clean rooms using Milli-Q™ ultra-pure water and sub-boiled ultra-pure acids. In the laboratory, 5 ml of a mixture of HCl:HF acids (3:1 v:v) was added in each evaporated bottle which was then sealed and heated on a heater plate at 70°C for at least 48 hr, producing a reflux of the acid mixture on the inner walls of the bottle. This mixture moved down the entire walls of the bottle and dissolved all of the mineral deposits. At the end of this first step, the bottle was left to cool down, its content was poured into a 60 ml Teflon™ vial and the bottle was rinsed with approximately 10 ml of water. The bottle was rinsed a second time with 3 ml of 65% HNO 3 which detached some of the remaining visible traces, likely organics, and the remainder of the contents in the bottle were recovered by rinsing the bottle twice using approximately 10 ml of water each. All of the rinsing solutions were added in the same vial as those used to collect the sample. Open vials were heated at 99°C for about 10 hr and then at 115°C until complete evaporation. During this sequence, water, hydrochloric acid, hydrofluoric acid, and silica (as SiF 4 ) were removed from the sample and at the end of the evaporation process, the remaining concentrated nitric acid destroyed the remaining organic matter and resulted in the full digestion of the sample. The temperature of the heater plate was then adjusted to 80°C, and 3 ml of diluted (1:1 v:v) nitric acid solution in water was added into each, now sealed, vial. After 2 hr, the full content of each vessel was completely transferred into a 125 ml polypropylene bottle (thoroughly ultra-cleaned with detergent/acid) and completed up to 100 ml with Milli-Q™ ultra-pure water. In addition, the whole Teflon vessel was also thoroughly cleaned previously, including with a blank digestion procedure.
To test the efficiency of our extraction and digestion procedures, a second extraction was performed starting from the reflux step for 13 samples (Table S1 in Supporting Information S1). The first extraction yield was calculated as the ratio between the amount extracted with the first extraction divided by the sum of the two extraction amounts. The median yield value obtained was larger than 96% for all the elements and validated a single extraction protocol.

Chemical Analysis
An ARCOS (Spectro-Ametek) ICP-AES equipped with a CETAC ultrasonic nebulizer or a concentric nebulizer with a cyclonic chamber was used to determine the elemental concentrations in the digested sample solutions. An external linear calibration was performed for all the analyzed elements with ICP-AES (Al, Ca, Fe, K, Mg, Na, P, S, and Zn) by measuring a set of multi-elementary solutions with a concentration between zero and 500 ppb for the calibration slope. The calibration intercept was calculated as the average of eight replicates of a blank sample (diluted ultra-pure nitric acid in Mili-Q water). Because of the well-known matrix effects in plasma with alkali metals, Na and K were also analyzed using a flame photometer (Cole Parmer 360) with an external calibration between 0 and 20 ppm. Flame photometry results were used for these two elements except for the less concentrated samples found close to or below the flame photometry detection limit, for which the ICP results were used. The wavelengths used for analysis were: 167.078 nm for Al except for the highest concentrated samples for which saturation was observed and the much less intense line at 176.641 nm was used; 317.933 nm for Ca; 238.204 nm for Fe; 285.213 nm for Mg; 177.495 nm for P; 180.731 nm for S; and 206.200 nm for Zn. For each element, the detection limit was calculated to be three times the standard deviation of the analytical blanks and converted in element masses and fluxes for each sample. The analytical relative standard deviation of the replicates was approximately 3% for values that were far above the detection limit.
Analyses of the geostandard-certified samples were performed to check the accuracy of the whole digestion and analytical methods: BE-N (SARM, Nancy, France), BHVO-1, SCO-1, MAG-1, STM-1, SDC-1 (USGS, Washington, USA). A mass between 15 and 45 mg of the previously crushed standard was placed into the same bottles which were used in the field to mimic a real deposition sample. These geostandard samples underwent the same procedures used for digestion and analysis as those for real deposition samples. The average recovery rate for each element for each geostandard is listed in Table S2 in Supporting Information S1. The recovery rates were in the range of 88%-114% (first and third quartiles) with a minimum of 70% (Ca in STM-1) and a maximum of 157% (P in BHVO-1).

Determination of the Saharan Dust and Sea Salt Deposition Flux
For each sample, the elemental deposition fluxes (F i ) for element i were calculated by dividing the mass of measured element i (Q i ), corrected from field blanks (Q 0,i ), by the aperture area of the funnel (S funnel = 0.0113 m 2 ), and by the exposition duration (T exposure ). Blank data are reported in Tables S3 and S4 in Supporting Information S1.
Because two samples are missing, the mean deposition flux value for a given period was calculated as the average flux measured with the existing samples: The measured fluxes were broken down into three sources: North Africa dust (F dust ), sea salt (F sea salt ), and excess flux (F excess ) probably from biological sourced elements such as living materials debris (insect, plants, etc.) or microorganisms.
The mass deposition fluxes of Saharan dust (F dust ) were estimated assuming a pure crustal origin of aluminum: where wt% Al is the abundance of Al in the crustal soil model proposed by Rudnick and Gao (2014), equal to 8.15%. The mass deposition fluxes of sea salt (F sea salt ) were estimated using Brewer's (1975) model assuming that the entire Na deposition flux came from dust and sea salt (Rahn, 1976).
Because the concentrations of most of the major elements in sea spray were similar to those of sea salt, we used the sea salt composition (Brewer, 1975) and the (X/Na) sea salt ratio to determine the sea salt deposition flux of the major elements F (X) sea salt .

of 19
The excess deposition flux after subtraction of the sea salt and dust contribution was calculated as: The enrichment factor of an element X with respect to a given source (crustal or sea salt) was calculated as: A good correlation is found between the collected water and rain gauge indications ( Figure S2 in Supporting Information S1), up to an accumulation of 85 mm (maximum accumulation) because the rainwater overflowed the 1-L sampling bottle and therefore part of the deposition sample was lost. Corrections could not be applied to the deposited flux in this case because of unknown variations in the deposition rate during the sampling period and the subsequent heterogeneity of the sample. If accumulated, the amount of lost rainfall represents 7% of the total amount of water collected during the whole sampling period and is therefore considered negligible.

Sampling Representativeness
To identify potential dust sources, the periods for which the deposition flux of aluminum is higher than 2.9 mg.m −2 .day −1 are selected because this represents the most loaded top third of the sample set. Over these periods, a backward-trajectory was calculated every hour over a period of 12 days ending 2,000 m above the sampling site using the Hysplit 4 rev. 761 model (Stein et al., 2015) based on the Global Data Assimilation System (GDAS). This backward-trajectory ensemble is plotted using the "openair" R package with the "trajLevel" function (Carslaw & Ropkins, 2012;R Core Team, 2021) showing the flyover frequencies ( Figure S4 in Supporting Information S1). It can be observed that during these higher aluminum deposition flux episodes, most of the air parcels originated from the Saharan regions.

Atmospheric Deposition Fluxes of the Major Elements in Guadeloupe
The deposition fluxes averaged on the whole period (F whole period ) of each element have been reported in Table 1 and compared to the fluxes previously measured in Puerto Rico by Heartsill-Scalley et al. (2007), Cuba by Morera-Gómez et al. (2019) and Guadeloupe by Dessert et al. (2015). The larger flux was measured for Na, followed by Al, S, Mg, K, Ca, Fe, P, and Zn. Reported for the whole sampling period, the mean dust deposition flux F mean is found to be 30.6 mg.m −2 .day −1 , and the mean deposition flux of sea salt is equal to 47.7 mg.m −2 . day −1 . Figure 2 illustrates the temporal chemical flux variation for each element. The seasonality and contribution of each major source (Saharan dust, sea salt, and unknown sources are marked in different colors). It can be observed that the aluminum, iron, and calcium fluxes present a strong seasonality, with the highest deposition occurring between April and September. The potassium flux also seems to present seasonal trends but not as strong as those observed for aluminum, iron, and calcium. The trends of sodium, sulfur, and magnesium are very similar, all marked by obvious single episodes and without seasonality. The trends of the phosphorus and zinc deposition fluxes are marked by intense episodes. To understand the temporal evolution of each species, the contribution of various sources to the deposition flux of each element is investigated in the following sections.

Temporal Evolution of the Saharan Dust Deposition Flux (F dust )
The In Figure 3, we report the dust deposition fluxes for each month during the whole sampling period and the associated standard deviations deduced from the inter-annual variations. These average values take into account the monthly inter-annual variability in the intensity of the dust deposition fluxes at the 4-year scale (March 2015 to August 2018). The average dust flux is minimal in December (5.0 ± 2.0 mg.m −2 .day −1 ) and maximal in June (75 ± 6 mg.m −2 .day −1 ). This seasonal dust profile is consistent with the results of Prospero et al. (2014) and Prospero and Mayol-Bracero (2013), who found a similar pattern for the PM 10 and mineral dust atmospheric concentrations in Barbados. We note that the Saharan dust deposition flux has large standard deviations in May and October (44 ± 33 and 22 ± 16 mg.m −2 .day −1 , respectively), showing a large inter-annual variability of the dust deposition for these 2 months. However, this variability does not have any impact on the inter-annual variability of a full-year deposit because it only shows a shift at the beginning and end of the dust season.
Direct monthly chemical measurements in Cuba between 2014 and November 2016 (Morera-Gómez et al., 2019) are compared to the results of this work. The mean deposition flux in Cuba is equal to 8.3 mg.m −2 .day −1 . This flux measured in Cuba is four times lower than those measured in this work. In addition, less seasonality of dust deposition is observed in Cuba. This shows that Saharan dust transport has less of an impact at the Cuba sampling site than in Guadeloupe. This regional variation may be explained by the fact that the island of Guadeloupe is much closer to the center of the trans-Atlantic Saharan dust plume than Cuba, as demonstrated by model studies (e.g., Mahowald et al., 2006) Figure 2. Deposition flux of the major elements. The source data are given in Table S5 in Supporting Information S1. The contributions of Saharan dust, sea salt, and a non-attributed source are shown in orange, light-blue, and gray. The figure is plotted using the "openair" R package with the "timePlot" function (Carslaw & Ropkins, 2012; R Core Team, 2021).

of 19
in Cuba are slightly higher than the highest model estimations but with the same relative proportion as the modeled values.
Other Our results highlight that the Saharan dust deposition fluxes on the island of Guadeloupe are higher than those measured in the Northern Caribbean region, and they show that Guadeloupe is in the central part of the trans-Atlantic Saharan dust plume. This observation also matches expectations based on concentrations compared between Guadeloupe and Barbados (Prospero et al., 2014). This experimentally confirms the substantial regional variation of dust deposition between the Southeast and Northwest Caribbean, already documented by satellite observations and model studies.

Evolution of the Sea Salt Deposition Flux (F sea salt )
Marine year −1 (16 and 27 mg.m −2 .day −1 ) for Guadeloupe and Puerto Rico, which is also lower than our direct measurements and still much lower than data from Puerto Rico. These two models did not take into account the coastal surge in Guadeloupe and Puerto Rico even though coastal surf-zones are important sources of sea salt aerosol emissions which can produce as much as 30%-50% of the total sea salt aerosol in the air of coastal regions and islands (e.g., Im, 2013). This may explain the fact that both our direct measurements and those obtained in Puerto Rico are higher than the model predictions.
It has been shown that the surface wind speed is the major driver for the sea salt aerosol emission, as a higher wind speed results in higher emission fluxes for all particle sizes of sea salt (Lewis & Schwartz, 2004;Marks, 1990;Monahan et al., 1983). Figure 4 shows the sea salt deposition fluxes as a function of wind speed assimilated near the island of Guadeloupe using 6-hourly wind data at 10 m above the up-wind sea surface (ERA-Interim; Dee et al., 2011). The data are assimilated 100 km upwind from Guadeloupe so as to take the possible coastal surge of the sea and the wind condition over the Atlantic Ocean into consideration. It can be observed that the wind speed over the Atlantic Ocean upwind of Guadeloupe also has an impact on the sea salt deposition fluxes in Guadeloupe (Figure 4). In our case, the highest deposition fluxes of sea salt often correspond to a high wind speed or heavy rain (with an average daily rainfall over 10 mm.day −1 ) which are not necessarily cyclone events. The name of the five cyclone events occurring during the sampling period (except for Maria) have been reported in Figure 4. To investigate the influence of rainfall and wind speed on the deposition flux of sea salt, we used two Spearman correlations between the sea salt deposition fluxes and rainfall intensity, respectively ( Figure S3 in Supporting Information S1). Wind speed explains 59% of the sea salt deposition flux variance whereas rainfall explains 16% of the variance, showing that wind speed is the more influential parameter controlling sea salt deposition.
To conclude, in the Caribbean region, wind as well as coastal surges may have a large impact on sea salt emission and deposition processes. Rainfall and wind patterns vary substantially over different locations, which leads to huge variations in the sea salt deposition flux, on different islands. Table 2 summarizes the respective contributions of sea salt and Saharan dust to the total deposition flux of each element during the whole sampling period, using Equations 5, 9, and 10. Sodium and magnesium are 95% and 89% explained, respectively, by the sea salt source and therefore could be considered as elements that have been sourced from sea salt alone. Fries et al. (2019), using magnesium isotopic measurements on rainwater collected in Guadeloupe, have already suggested that sea salt contributes to almost 100% of the atmospheric magnesium deposition flux. The aluminum and iron fluxes are explained with Saharan dust inputs alone. The calcium and potassium fluxes are supplied by both Saharan dust and sea salt sources, with a residue of 16% for calcium and 24% for potassium. Most of the sulfur flux is attributed to sea salt input (62%), as the crust-dust influence is negligible (1% only). For one third of the sulfur fluxes and all of the phosphorus and zinc fluxes, the source of this flux is neither sea salt nor Saharan dust. Another source must be investigated for these three elements.

Contribution of Biogenic and Anthropogenic Sources
Small plant debris measuring close to 100 μm have been observed under a lens in our samples, highlighting the significant biogenic input. Overall, pachosomes (∼1 μm), fungal spores (∼1-5 μm), plant debris (∼100 μm), pollen (∼5 μm), fern spores (∼20 μm), and fungal hyphae (∼50 μm) are present in the atmosphere and have been observed in primary biological particles collected in the Amazon Basin. They may be a local major source of aerosols in tropical forests (e.g., China et al., 2018;Martin et al., 2010). These particles may be the sources of the phosphorus, zinc, potassium, and sulfur in the samples studied here, as these elements are normally enriched in biologic tissues (e.g., Adachi et al., 2020;Barkley et al., 2021;Bowen, 1966;Fraund et al., 2017).
The presence of African biomass burning smoke in the Marine Boundary Layer is also well documented on the island of Barbados in the Caribbean region (Kristensen et al., 2016;Quinn et al., 2022;Royer et al., 2022;Figure 4. Sea salt deposition fluxes as a function of the averaged wind speed showing the daily precipitation rate levels (mm.day −1 ), with the name of the five cyclone events occurring during the sampling period. The smooth correlation curve is obtained using the "ggplot2" R package with the "geom_ smooth" function, to see the trend of the nonlinear relationship between two variables (R Core Team, 2021;Wickham, 2016). As the average daily rainfall ranges from 0 to 67 mm.day −1 during the whole sampling period, and because only two samples correspond to a sampling period with average daily rainfall over 20 mm.day −1 , all the samples corresponding to daily rainfall levels over 20 mm.day −1 are shown in a dark blue color.

Element
Al Note. The proportion of the deposition flux derived from sea salt is calculated using Equation 9; the proportion derived from the crust is calculated using Equation 5; and the proportion of the residues is calculated using Equation 10.  Savoie et al., 2002). Potassium is enriched in biomass burning smoke and can be used as indicator of smoke together with black carbon (BC) and carbon monoxide (Andreae, 1983;Artaxo et al., 1994;Hand et al., 2010;Hudson et al., 2004;Li et al., 2003;Murphy et al., 2006;Pósfai et al., 2003;Reid et al., 2005). Also, African biomass burning smoke particles have been shown to contributing to sulfur deposition in South America (Ansmann et al., 2009;Barkley et al., 2019). To investigate the potential contribution of biomass burning smoke to the deposition flux of major elements in Guadeloupe island, the time series of F(S) excess , F(K) excess , F(Ca) excess , F(Zn) excess, F(P) excess , and F(BC) are plotted on Figure S5 in Supporting Information S1 together with the monthly data on atmospheric concentration of carbon monoxide, BC and the concentration of SO 2 . The carbon monoxide concentration presents a marked seasonality with higher values detected between March and May. The BC concentration presents high values in August and October of each year. None of these seasonalities are clearly observed on any excess proportion of a major element of this study. We conclude that biomass burning smoke is not the major source of the excess of these elements.
The estimated mean value of the total deposition flux of BC from smoke F(BC) smoke is 0.134 mg.m −2 .day −1 (with a min. value of 26 and a max. value of 0.67 mg.m −2 .day −1 , also plotted in Figure S5 in Supporting Information S1). Using the tropical forest fire smoke composition data measured by Yamasoe et al. (2000) and the African Savana biomass burning composition summarized by Reid et al. (2005), we were able to estimate the elemental ratio in the smoke. This leads to elemental ratios ranging from 0.06 to 0.3 for (S/BC) smoke , ranging from 0.11 to 0.80 for (K/BC) smoke , and equal to 0.008 for (P/BC) smoke and 0.001 for (Zn/BC) smoke. We then roughly estimated the deposition flux derived from biomass burning smoke for these elements using the elemental ratio and deposition flux of BC: 0.008-0.041 mg.m −2 .day −1 for F(S) smoke , 0.014-0.109 mg.m −2 .day −1 for F(K) smoke , 0.001 mg.m −2 .day −1 for F(P) smoke and 0.0001 mg.m −2 .day −1 for F(Zn) smoke . For S, P and Zn, these values are much smaller than those measured in this work (Table 3) and therefore this source could be neglected for these elements.

Potassium
On our sampling site, the excess potassium flux is 0.374 mg.m −2 .day −1 accounting for one third of the total potassium deposition flux. The highest estimation of potassium coming from African biomass burning is 0.109 mg.m −2 .day −1 which represents one third of the excess potassium or one tenth of the total potassium atmospheric deposition flux. Taking the uncertainties attached to these estimations into account, it is difficult to come to more quantitative conclusions on the contribution of African biomass burning.

Sulfur
The source of excess sulfur on the island of Guadeloupe can be traced to dimethyl sulfide (DMS), H 2 S, or CS 2 emitted from marine or local biogenic sources (e.g., plants, microorganisms, and marine plankton) and long-range transported sources (e.g., biomass burning and anthropogenic emissions from Africa and America) (Martin et al., 2010;Vet et al., 2014). In previous studies conducted in the Caribbean region between 1988 and 1990, around 50% of nss-sulfate in aerosols has been estimated to be brought by anthropogenic emissions transported from Europe, North America, and Africa to Barbados (Savoie et al., 2002). However, it has been observed that due to the North American and European SO 2 -S emission reductions in response to emission reduction programs, the wet deposition of nss-S between 2005 and 2007 is only around 50%-70% of the wet deposition of nss-S between 2000 and 2002 in the Caribbean region (Vet et al., 2014). Since then, the anthropogenic sulfur emissions in Africa have been increasing and the impact of anthropogenic sulfur on the Caribbean region is not clear. Biogenic gases containing sulfur as DMS and H 2 S and the product of oxidation of these gases in the form of sulfate also exist in the marine boundary layer (Charlson et al., 1987;Huang et al., 2018;Shon et al., 2005). The excess sulfur (37%) in Guadeloupe represents a deposition flux F(S) excess of 0.702 mg.m −2 . day −1 . The Great Atlantic Sargassum belt form every year and produces massive influxes of Sargassum in the Caribbean region with substantial progression since 2015 (Wang et al., 2019). The decomposition of sargassum produces huge amounts of hydrogen sulfide (H 2 S) in the atmosphere of the Caribbean coastal region which causes heavy pollution at the local scale (Cabanillas-Terán et al., 2019;Chávez et al., 2020;Rodríguez-Martínez et al., 2019). The deposition fluxes of sulfur derived from these biogenic sources are still not clearly quantified. Global model predictions neglect the impacts of sargassum decomposition at the regional scale and only take into Note. F(X) excess is calculated using Equation 10.

Table 3 Excess Deposition Flux of Elements With a Possible Biogenic Source
account the DMS produced by phytoplankton activities and, to a lesser extent, the dissolution of global anthropogenic sulfur dioxide (SO 2 , e.g., Vet et al., 2014). The model predictions of Vet et al. (2014) give a range of 0.137-0.274 mg.m −2 .day −1 for the wet deposition of nss-S and 0.027-0.137 mg.m −2 .day −1 for the dry deposition of nss-S, leading to a total of 0.164-0.411 mg.m −2 .day −1 for the deposition of nss-S. The highest value predicted by this model represents 60% of our excess sulfur measurement (0.702 mg.m −2 .day −1 of excess S calculated with Equation 10), resulting in the correct order of magnitude to explain the observed amount of excess sulfur coming from large-scale DMS marine emissions. The remaining 40% of excess sulfur observed may be attributed to uncertainties and variability related to the evaluation of the DMS emission fluxes or to the still unknown contribution of the Great Atlantic Sargassum belt.

Phosphorus
Little is known about the phosphorus deposition flux all around the world because of the relatively small amount of observational data on atmospheric phosphorus and uncertainties with regard to atmospheric phosphorus sources and deposition rates Vet et al., 2014). Phosphorus is not routinely measured during aerosol monitoring programs, these measurements are normally rare and very few long-term measurements are available Vet et al., 2014). Model studies have concluded that the most important external source of phosphorus for tropical forests is long-range transported dust (82%), followed by long-range transported biogenic particles (12%) and biomass burning smoke (5%) (e.g., Mahowald et al., 2008;. The simulation of the atmospheric transport of phosphorus associated with mineral dust, combustion products, primary biogenic particles, marine (sea salt) aerosols, and volcanic emissions suggests a total deposition of 0.014-0.027 mg.m −2 .day −1 in the North Caribbean regions. Among all the external sources of phosphorus for the Caribbean region, Saharan dust is considered as the largest source and contributes more than 90% of the total external phosphorus input . Field measurements without the influence of local emission conducted in Barbados and Miami also show that the Saharan dust is the predominant source of total phosphorus in aerosols and that contributions from sea spray, pollution, and biomass burning are minor (Zamora et al., 2013).
The deposition flux of dust-derived phosphorus roughly estimated using Zamora et al.'s (2013) measurement taken in Barbados is 0.015-0.093 mg.m −2 .day −1 , which supports Mahowald et al.'s (2008) model. The measured average deposition flux of phosphorus in Guadeloupe during this study F whole period is 0.186 mg.m −2 . day −1 and the deposition flux derived from Saharan dust is 0.02 mg.m −2 .day −1 (calculated from Equation 5). The phosphorus deposition flux derived from Saharan dust is within the same range as the model prediction largely based on dust from Mahowald et al. (2008) and is consistent with the rough estimation made using Zamora et al.'s (2013) measurement. The deposition flux of phosphorus roughly estimated from the BC concentration F(P) smoke is negligible (0.001 mg.m −2 .day −1 ). The Saharan dust-derived phosphorus F(P) dust represents only 10.8% (9.2%-12.4%) of the measured flux. This amounts to 0.160 mg.m −2 .day −1 of excess phosphorus (calculated using Equation 10) from probably local biogenic and long-range transported biogenic sources.
Almost all biogenic particles are normally highly enriched in potassium and phosphorus (Bowen, 1966). Pollen and other biogenic sources may dominate atmospheric phosphorus in and near forests or other densely vegetated areas, for example, where our sampling site is located. The importance of biogenic aerosols can be characterized as local and regional but not global (Doskey & Ugoagwu, 1989). Particles originating from plants contain substantial phosphorus levels and are considered to be the main contributor of the observed phosphorus deposition in tropical regions (Graham et al., 2003). As far as we are aware, only one publication has taken such pollen deposition and phosphorus flux measurements, around 0.083 mg.m −2 .day −1 from pine pollen at a site in northern Wisconsin (Doskey & Ugoagwu, 1989), which is within the same order of magnitude as our excess phosphorus deposition flux (0.160 mg.m −2 .day −1 ). Other recent studies conducted in the Amazon and Tropical Atlantic Ocean region with very low Saharan dust input suggest that the sources of phosphorus are more complex with substantial contributions from long-range transported African biomass burning in these regions where the Saharan dust input is minor (Ansmann et al., 2009;Barkley et al., 2019;Giglio et al., 2013;Kristensen et al., 2016;Quinn et al., 2022). However, the contribution of biomass burning smoke is estimated to be minor in regions close to Guadeloupe island in the Caribbean region (e.g., Mahowald et al., 2008;Zamora et al., 2013).

Zinc
With regards to both sea salt and crustal sources, 97% of the zinc is found in excess amounts. Among all samples, the crustal enrichment factor of zinc varies between 5 and 2,400 with an average value of 30; the marine enrichment factor of zinc varies between 1,200 and 70,400 with an average value of 9200. In aerosol collected near non-polluted coastal regions, a high enrichment factor of zinc up to 250,000 relative to the sea salt composition has been observed all over the world (Richardson et al., 2001). Large trace metal enrichments in the marine surface micro-layers have been reported by Duce et al. (1972). A systematic enrichment in zinc has been attributed to the marine micro-organisms and materials associated with these marine micro-organisms. It is well known that diatoms and other phytoplankton, bacteria, viruses, exopolymers, and colloidal gels are present and enriched in the sea surface micro-layer, where sea water is aerosolized (e.g., Aller et al., 2005;Bigg & Leck, 2008;Blanchard, 1975;Blanchard Duncan & Lawrence, 1970;Kuznetsova et al., 2005;Wurl & Holmes, 2008). The presence of these materials in sea spray is also well documented (e.g., Aller et al., 2017;Alpert et al., 2011;Knopf et al., 2011;Ladino et al., 2016;Wilson et al., 2015). Enrichment factors equal to 26 for phytoplankton (Sutcliffe et al., 1963) and 10-1,000 for marine bacteria (Blanchard Duncan & Lawrence, 1970) compared to the bulk seawater have been found in the literature. During phytoplankton blooms, bubble bursting generated by phytoplankton activities can produce sub-micrometer particles in the air, thereby increasing these phenomena (Hallquist et al., 2009;Kanakidou et al., 2005;Meskhidze & Nenes, 2006). However, the observed zinc enrichment factors are much higher than those found in micro-organisms and a systematic enrichment of deposit samples in zinc relative to living organisms can be observed in Figure 5. Another enrichment mechanism should therefore be investigated. Zinc in sea water may be associated with very small organically-bound aggregates of colloidal detritus present in the micro-layer. They then form relatively stable attachments at the air-sea interface by means of inter-facial forces concentrating at levels that are substantially higher than their typical crustal and sea salt abundance (Hardy, 1982). This could explain the very high enrichment factors observed.

Conclusion
Very different sampling techniques for atmospheric deposits are applied with variable approaches during previous studies carried out in the Caribbean region. Each of these studies examined somewhat different time periods and different sets of elements, producing large uncertainties. Before this work, long-term direct measurements of the dust deposition flux were rare. Therefore, it has been difficult to produce a consistent view of the natural atmospheric deposition flux in the Caribbean region based on current field measurement data. This study is the first continuous long-term time series (from March 2015 to August 2018) with direct measurements of elemental atmospheric deposition fluxes on the island of Guadeloupe with a resolution of days and weeks without anthropogenic influence.
The mean value of the annual deposition flux of Saharan dust was determined to be 11.2 g.m −2 .year −1 over the 4-year sampling experiment. This value is slightly higher than recent model predictions, but within the correct order of magnitude. No inter-annual deposition flux variations were observed during this period. Saharan dust is the only source for aluminum and iron, and constitutes the largest source for calcium and potassium. A strong seasonality in the Saharan dust deposition is observed with deposition peaks between April and September, including a maximum deposition flux in July. The mean annual deposition flux value for sea salt is 17.4 g.m −2 .
year −1 , with poor inter-annual variations and no seasonality. Sea salt constitutes the only source of sodium and magnesium and is a strong secondary source for calcium and potassium. The major meteorological factor influencing the sea salt deposition flux is the local wind speed.
Phosphorus deposition is strongly dominated by local biogenic sources. Biogenic particles probably originated from local insects and vegetable debris, dead or living organisms and their tissues found in sea spray. Because the measured phosphorus flux is dominated by local biogenic sources, the measured atmospheric deposition of this element cannot be related to a net atmospheric contribution but rather to an indication of local recycling. This recycling local effect must be carefully taken into account for the interpretation of phosphorus atmospheric measurements in green environments. At this stage, the phosphorus atmospheric input must be estimated from the Saharan dust input composition rather than from direct deposition measurements. Zinc is also found to have a large biogenic origin, but its enrichment factor cannot be explained by insect or vegetable debris. It should be assumed that there is a large contribution of the marine micro-layer heavily enriched with zinc.

Data Availability Statement
Data used in the creation of the manuscript are available at French Generic Research Data Center at https://doi. org/10.57745/WEUNPS (Losno et al., 2022) and in the Supporting Information (Table S5 in Supporting Information S1). Figures 2-4, Figures S4 and S5 in Supporting Information S1 are made with R version 4.1.1 (R Core Team, 2021) using different packages. Figure 2, Figures S4 and S5 in Supporting Information S1 are made with "openair" package version 2.9.1 (Carslaw & Ropkins, 2012). Figure 4 is made with "ggplot2" package version 3.3.5 (Wickham, 2016). Figure 5 is made with "ggtern" package version 3.3.5 (Hamilton & Ferry, 2018). Backward trajectories displayed in Figure S4 in Supporting Information S1 were calculated using the Hysplit 4 rev. 761 model (Stein et al., 2015) based on the Global Data Assimilation System (GDAS).