Production mechanisms of aerosol chemical species, in terms of primary and secondary processes, were studied using vertical concentration gradient measurements at the coastal research station in Mace Head, Ireland. Total gravimetric PM1.0 mass, sea salt and water insoluble organic carbon (WIOC) concentration profiles showed a net production at the surface (i.e. primary production), while nssSO4 and water soluble organic carbon (WSOC) concentration profiles showed a net removal at the surface. These observations indicate that WSOC was predominantly of secondary origin and that WIOC was predominantly of primary origin. Derived PM1 mass fluxes compared reasonably well with those previously obtained from an eddy covariance (EC) technique following a power law relationship with the wind speed (FPM1 = 0.000096*U4.23). For cases with clear primary organic mass fluxes in the flux footprint WIOM mass fluxes ranged between 0.16 and 1.02 ng m−2 s−1 and WIOM/sea salt mass ratio was 0.34–3.6, in good agreement with previous measurements at Mace Head.
 Marine aerosols, despite their low concentrations, contribute significantly to the global aerosol budget considering 70% Earth's coverage by oceans. Significant impact of marine aerosols on global climate was first demonstrated by Charlson et al. . Sulphate aerosols produced from plankton sulphur emissions were considered a dominant species in marine atmosphere. Later O'Dowd and Smith  argued that significant amount of submicron sea salt particles can be found in marine atmosphere and reported the first wind-speed relationship for accumulation mode sea salt. Also O'Dowd et al.  have shown that sea salt aerosol may significantly affect the properties of marine CNN. More recently, O'Dowd et al.  have suggested that CCN carbon concentration in marine aerosols can be linked to primary production at the ocean surface. The detailed study by Yoon et al.  revealed strong seasonal cycle of physico-chemical properties of marine aerosol and linked it to the observed seasonal cycle of chlorophyll-a concentration in ocean surface waters. However, hitherto, definitive proof, linking the organic matter enrichment in marine aerosol to primary production processes, has been lacking.
 Previous flux measurements in North Atlantic marine air using an eddy covariance (EC) technique [Geever et al., 2005] have found a strong dependency of primary sea-spray flux on the wind speed (Flux > 106 m−2 s−1 at wind speeds of > 10 ms−1) and that approximately 50% of the number flux was attributed to the Aitken mode (10–100 nm) and 50% to the accumulation mode (100 nm–500 nm). These measurements represented the first pseudo-size-segregated aerosol flux measurements in marine environment indicating a significant contribution of primary marine particles to the submicron Aitken and accumulation modes. However, the eddy covariance technique, including its modifications (relaxed eddy accumulation or disjunct eddy covariance) with near real-time measurement techniques, do not produce unambiguous results for aerosol chemical fluxes.
 Given that off-line chemical analysis requires relatively long sampling times (of several days) to obtain sufficient particulate matter in clean marine air, alternative approaches to micrometeorological flux techniques are required. In this study we adapt an experimental set-up used by Valiulis et al.  to determine near-surface chemical gradients and resultant fluxes of aerosols in NE Atlantic marine air.
2. Experimental Methods
 Vertical concentration profiles of sub-micron aerosol chemistry were taken in clean marine air at the Mace Head NE Atlantic research station. The sampling was conducted on a 22m tower approximately 80 m from the shoreline at 3, 10 an 30 m heights (the highest sampling point was installed on an extended pole) relative to the ground which was approximately 5 m above MSL. The tower is located upwind of the station buildings relative to the ocean. The tower location relative to the sampling sector ensured minimal if any contamination from building induced turbulence [de Leeuw et al., 2002].
 PM1.0 size selective inlets (PMX inlet, Sven Leckel Ingenieurburo GmbH, Germany) were used to collect aerosols on quartz filters. An automated clean sector sampling system ensured sampling only of air (1) reaching the site from a controlled sector between 190°–300° and (2) with a total particle concentration (d50% = 14 nm) below 700 cm−3. The system monitored both parameters at a time and activated sampling when both conditions were met. The sampling criteria ensured that air mass was of clean marine origin. Post analysis also confirmed air mass cleanliness with black carbon concentrations of less than 50 ng m−3 and a transit time of backward trajectories of 4–5 days over the ocean [O'Dowd et al., 1993; Cooke et al., 1997; Cavalli et al., 2004]. Eighteen 7-day samples were obtained, with the actual accumulated sampling times ranging from 20 to 98 hours (10 to 58% of the time) depending on the above conditions. Quality control resulted in 50% of the sample sets being discarded due to too short sampling periods and 9 profile sets retained for this analysis.
 Samples were collected from April to October 2005, which covered the period of high biological activity over the Northeast Atlantic. Samples were analysed for gravimetric mass, Na+, SO42−, water soluble (WSOC) and insoluble carbon (WIOC). Analytical details can be found in Cavalli et al. .
3. Results and Discussion
3.1. Source Region and Flux Footprint
 Previous flux measurements in marine air at Mace Head [Geever et al., 2005] demonstrated that the flux measured at the 22 m height had a flux footprint peak at approximately 600–1000 m from the coastline with significant contributions at distances of up to 10 km from the coastline. In this study three levels were used for the measurements, therefore, a slightly different approach to estimate the flux footprint was necessitated. The extent of the source region (flux footprint area) contributing to the concentration gradient and the potential influence of surf-zone emissions to the difference in concentration between 3 to 10 and 10 to 30 meters is estimated using the methodological approach used by Valiulis et al. . In this analysis, we use the micrometeorological data set from the Geever et al.  study as that study corresponds to a range of similar meteorological conditions at the same location. The analysis of the flux footprint and source regions influencing the different measurement heights found that surf zone emissions at a range of 80–180 m from the tower could have contributed up to 20% of the difference in concentration between the 3 m and 10 m levels and <5% between 10 m and 30 m level. The difference in concentration (or gradient) between 3 and 10 m reaches 90% of it's value at 1170 m from the coast line, while the difference between 10 and 30 meters reaches 90% of its value at 4840 m (Figure 1). Emissions from greater distances have minimal contribution to the flux footprint and gradient profile in terms of upward fluxes, but clearly, such emissions can influence the absolute concentration which in turn influences the deposition flux magnitude. In other words, the gradient flux footprint was limited to within 5 km region of open water adjacent to the coast line for detecting upward fluxes while the concentration footprint is typically 10–100 times greater distance upwind.
3.2. Aerosol Chemical Gradients
 The average percentage contribution of aerosol chemical components to PM1 mass at 10 m level was as follows: sea-salt - 24%, nssSO4 - 32%, WSOM - 25% and WIOM - 19%. The WIOM/sea salt mass ratio varied from 0.34 to 3.6, in good agreement with previous measurements at Mace Head [Cavalli et al., 2004; Yoon et al., 2007]. In order to deal with the wide range of concentrations encountered in each profile, concentrations were normalised to the sum of the total gradient concentration of individual chemical species mass. Such normalisation gives equal weight to each individual profile. After normalisation, the profiles of each mass category were averaged, resulting in statistically-meaningful variances around the mean value. Figure 2a shows the averaged normalised concentration profiles of different species for the whole sampling period. Differences between concentrations at all three heights were generally statistically significant. PM1 mass, sea salt and WIOC exhibited a decreasing concentration profile with height, while nssSO42− and WSOC showed the opposite pattern. The decreasing profile represents an upward mass flux and, therefore, respective chemical species are produced at the sea surface, while species exhibiting an increasing vertical profile (with height) point to a source either produced homogeneously in the boundary layer or from aloft (for example cloud processing or entrainment from the free troposphere). Thus, it can be concluded that sea salt and WIOC are produced at the ocean surface via primary or bubble-mediated production mechanism while nssSO42− and WSOC must be produced via secondary (or gas-to-particle) aerosol formation processes. It should be noted that the removal of certain species at the surface is caused by dry deposition only, as wet scavenging below the cloud evenly removes aerosol particles.
 While it is well known that sea salt is a primary aerosol and nssSO42− is a secondary aerosol, the distinct new finding in this study is that WSOC and WIOC species exhibited contrasting profiles and thus different formation mechanisms. The WSOC showed consistently increasing concentration profiles, while the individual WIOC profiles showed variable gradients which warrant further exploration. In fact, the WIOC gradients were found to fit into three distinct classifications as shown in Figure 2b. The first classification exhibited a clear and strong negative gradient corresponding to a significant primary surface source (net production) within the flux footprint; the second exhibits net removal through a positive gradient between 3–10 m and almost zero gradient between 10–30 m; and the third classification exhibits a deposition flux gradient (positive) between 3–10 m and a source flux gradient (negative) between 10–30 m. Given that enrichment of organic component in primary aerosol is related to enrichment of organic matter at the ocean surface, this range of behaviour can be interpreted in terms of the location of biologically active region relative to the flux footprint. If this region is within the flux footprint (5 km from the coastline), a strong negative gradient as in classification 1 will result. However, if the oceanic organic matter or primary organic aerosol production region is significantly and exclusively upwind of the flux footprint (say 20 km), the aerosol will have become well mixed throughout the boundary layer and a clear deposition (positive) gradient will be observed as in classification 2. Classification 3 can result from a combination of both scenarios. The classifications above are consistent with satellite analysis of the spatial and temporal development of chlorophyll-a distributions [Yoon et al., 2007].
 These results are the first experimental evidence of different production mechanisms of the water soluble and water insoluble organic carbon in marine aerosol, indicating that WSOC is predominantly of secondary origin while WIOC is predominantly of primary origin.
3.3. Mass Flux of Chemical Species and Relationship to Wind Speed
 The averaged profiles presented in Figure 2a effectively show the mean concentration gradient and, hence, the direction of the flux over a period of a few months. While the direction of the flux was the main motivation for the study, an attempt was made to calculate the mass flux from individual gradients over each of the sampling periods using detailed meteorological measurements performed by Geever et al.  in similar marine conditions. First-order closure turbulent flux parameterisation, often known as a gradient transport theory or K-theory, can be expressed according to Stull  as following:
where F is the flux, Kz is the turbulent-transfer coefficient; dc/dz is the concentration gradient. Therefore, having Kz and the measured concentration gradient, it is possible to calculate mass flux of any chemical species. The approach, however, does not allow distinguishing between upward and downward fluxes, but rather enables estimation of the net fluxes. The uncertainty of the estimated flux would largely depend on the gradient function fitted to the measured profile and the variability of the turbulent-transfer coefficient at any given wind speed.
 The turbulent-transfer coefficient was derived from its relationship with the horizontal wind speed using the Geever et al.  data set obtained under similar marine conditions and same ensemble range. The relationship followed a power law (Kz = 0.00117 U2.067, r2 = 0.62, P ≪ 0.01). The turbulent-transfer coefficient representing each individual sampling period of this study was derived from 30 min averages of the horizontal wind speed measured during that period. Then the Kz values were averaged to obtain the resultant turbulent-transfer coefficient corresponding to each sampling period of this study. We admit that such an approach introduced a certain degree of variability, but long sampling time required for off-line chemical analysis necessitated averaging of environmental conditions. We believe that the magnitude of the flux error would largely depend on the variability of Kz value at any given wind speed.
 The mean flux (and range of fluxes encountered) of nine 7-day samples for individual chemical species at the 3 m level were: PM1 = 18 ng m−2 s−1 [4–106]; sea salt = 48 ng m−2 s−1 [11–400]; nssSO4 = −14 ng m−2 s−1 [−51–(−1.2)]; WSOM = −11 ng m−2 s−1[−104−0]; WIOM = 4.9 ng m−2 s−1 [1.3–35]. The mean fluxes were calculated using equation 1 and the best fit to the data points.
 It was mentioned above that the uncertainty of the calculated flux will depend on the gradient fitting procedure. The concentration profiles in Figure 2a suggest a power function. However, due to the potential contribution of surf-zone emissions to the magnitude of the concentration at 3 meter level and hence, on the shape of the gradient, a linear gradient profile was considered too, ignoring the lowest 3 m-level concentration. Three different fitting exercises are presented in Figure 3a: using linear and two different power functions: f = a + bx, f = a + bxc and f = axb. PM1 mass concentration profiles were fitted with three different functions seeking the best correlation with the measurements and the fluxes calculated at 22 m level using equation 1. It has been shown, that the measurements performed at 22 m level are free from surf-zone influence [Geever et al., 2005]. The “linear gradient” approach yielded constant gradient. The gradient fitting procedure can be considered as a sensitivity analysis of the calculated flux. Then the calculated flux data were fitted with the power function to reveal the flux relationship to the wind speed and presented in Figure 3b. The impact of the fitting procedure is small up to wind speed values of about 9 m s−1, however, the curves start to diverge at higher wind speeds. The area between the broken curves can be considered as the uncertainty range of the calculated flux. Calculated fluxes of WIOM for cases with clear primary production ranged from 0.16 to 1.02 ng m−2 s−1 at a wind speed of 8.4 and 7.9 m s−1, respectively. Note that at similar wind speeds WISOM fluxes can differ by six times, which indicates the importance of seasonal biological activity in the flux footprint area.
 The presented mass flux calculation method involved averaging up of individual samples and needs an independent evaluation by comparing it with the flux estimated using eddy covariance method by Geever et al. . In order to compare the mass flux with the number flux, a particle diameter and density had to be set. The number flux can be converted into mass flux assuming modal diameter and density of accumulation mode particles, because contribution of Aitken mode particles to the PM1 mass is negligible. The density of the particles was set at 1.56 g cm−3 and the diameter of the accumulation mode was set at 140 nm based on long term observations of aerosol physical and chemical properties at Mace Head [Cavalli et al., 2004; Yoon et al., 2007]. The mass flux estimated by eddy covariance method compared best with the gradient flux, using the linear gradient fitting procedure (FPM1 = 0.000096*U4.23). The more important result though was that regardless of the fitting procedure, the flux relationship to the wind speed always followed a power law relationship (F = a*Ub, r2 = 0.76 ÷ 0.90). It can be argued that the uncertainty of the flux was not caused by the fitting procedure, but rather due to the lack of more high wind cases, which would have certainly helped to constrain a more reliable relationship. A generally good agreement between the two methods demonstrates that the number flux measurements using eddy covariance method can be successfully combined with off-line chemical measurements in the development of a combined organic-inorganic sea spray source function [O'Dowd et al., 2008].
 Production mechanisms and gradient fluxes of aerosol mass and inorganic and organic components in marine air over the Northeast Atlantic were studied using vertical concentration profiles measured at Mace Head. Total gravimetric mass, sea salt and WIOC exhibited a net upward flux, pointing to the primary source at the sea surface. NssSO42− and WSOC showed an opposite profile indicating a net downward flux, pointing to the removal of the species, which also indicates that it is produced via secondary processes. The difference in WSOC and WIOC concentration gradients in clean marine air masses is the first experimental evidence of the different formation mechanisms of these species, namely, secondary and primary, respectively. Calculated fluxes depend significantly on the gradient fitting procedure, but consistently demonstrated power law relationship with the wind speed. The estimated gradient fluxes are the first experimental estimates of the primary production rates for WIOC in the marine atmosphere. The total submicron aerosol mass flux relationship as a function of wind speed, described by a power lay (FPM1 = 0.000096*U4.23), agreed well with previous eddy-covariance measurements performed under similar conditions.
 This work was supported by EPA Ireland grant 2003-FS-CD-LS-12-M1, Lithuanian State Science & Study Foundation grant C-23/2005 and EU Framework 6 project MAP. The present research study was also supported by the strategic FISR Programme “Sustainable Development and Climate Changes” sponsored by the Italian Ministry of the University and Scientific Research (MIUR) and developed in the frame of the cooperative project between CNR and MIUR “Study of the direct and indirect effects of aerosols and clouds on climate (AEROCLOUDS)”. The authors also wish to thank anonymous reviewers for their constructive and helpful comments.