Water samples were collected as part of the MarParCloud project Cabo Verde campaign in the nearshore water in São Vicente, on the research cruise HE491 in the North Sea/Norwegian Sea and fjords, and from the research cruise EMB184 in the Baltic Sea (Fig. 1). The sampling areas represent uniquely different regions; São Vicente has oligotrophic tropical water with large influences from Saharan dust deposition, the Norwegian fjords and Baltic Sea are both temperate climates, but the inner and outer Norwegian fjord systems have a large interaction with North Atlantic water, while the Baltic Sea is semi-enclosed with larger anthropological interaction and little interaction with North Atlantic water.

North Sea/Norwegian Sea and fjord samples were collected between 8 and 25 July 2017 aboard the R/V Heinke. Samples were collected once per day, weather permitting, from each station, with a total of 13 stations spanning inner fjord, outer fjord and open-ocean areas. Samples were collected from both the North Sea and the Norwegian Sea, but for the purpose of clarity, that campaign will be termed the “Norwegian Sea” campaign. Baltic Sea samples were collected between 30 May and 10 June 2018 aboard the R/V Elisabeth Mann Borgese, with a total of eight stations used. SML and ULW samples were collected using the radio-controlled sea surface scanner (S3), as described in Ribas-Ribas et al. (2017), which has six rotating glass discs partially immersed in the water to sample the SML by its surface tension. ULW (1 m depth) and SML water were pumped through two separate flow-through systems with onboard sensors at a rate of 1.2 L min-1 using peristaltic pumps. SML and ULW water are collected in 1 L bottles by the pilot's command and in addition collected into large volume carboys. Large sample volumes (∼20 L) were collected for multiple analyses by all groups involved in the campaigns. The S3 also records multiple meteorological parameters: photosynthetically active radiation (PAR), solar radiation, wind speed and humidity. Salinity was measured on SML and ULW using a multi-parameter meter (MU 6100 H, VWR) before the collection of the sample into a container; high-precision in situ temperature was constantly measured for the SML and ULW using a reference thermometer (P795, Dostmann Electronics GmbH). Specifications for instrument precision and accuracy can be found in Ribas-Ribas et al. (2017). All in situ data were averaged for the 2 h surrounding the sampling of discrete water samples. A new device termed the High-volume Sampler for the Vertical (HSV) was deployed to collect water from five depths between the SML and 2 m. The HSV is made of a vertical polypropylene pipe with five polypropylene tubes set at five distinct depths in the pipe and a float attached to the top which has been ballasted to ensure accuracy in depth. Peristaltic pumps, similar to that on the S3, pump water into collection containers. The HSV was deployed during the collection time of discrete SML and ULW samples by the S3 and close enough to the S3 so that it would sample the same body of water but would not interfere with the glass plate sampling.

Samples were taken once a day, weather permitting, between 18 September and 6 October 2016 within the same nearshore water (∼1 km), with a total of 12 stations sampled. SML and ULW samples were collected from fisher boats in the nearshore waters. SML samples were collected using the glass plate technique (Harvey and Burzell, 1972; Cunliffe and Wurl, 2014) and ULW was collected from 1 m depth using a large syringe. Wind speed was recorded using an anemometer placed at the nearby Cape Verde Atmospheric Observatory (CVOA) station. A handheld Global Position System (Garmin eTrex) was used to track fisher boat movement during sampling and for coordinates of each sampling station.

Samples for particulate organic carbon (POC), nitrogen (PON) and phosphorous (POP) were filtered onto acid-washed and precombusted glass-fibre filters (Whatman GF/C). Filters for POC and PON were dried at 60 ∘C for 3 d (Norwegian cruise) or 130 ∘C for 2 h (Baltic cruise and Cabo Verde), put in tin capsules and measured using an elemental analyser (Thermo, Flash EA 1112 and Elementar Analysensysteme, precision of 0.01±0.2 ‰). POP was measured by molybdate reaction after digestion with potassium peroxydisulfate (K2S2O8) solution (Wetzel and Likens, 2000). The filtered water was collected and analysed for dissolved nutrients (PO4, NO3) by a continuous-flow analyser according to Grasshoff et al. (1999).

During the Cabo Verde and Baltic (EMB184) campaigns, chlorophyll a (Chl a) was measured by filtering 500–1000 mL of seawater onto precombusted (4 h, 450 ∘C) GF/F filters (Whatman). The filters were stored frozen (-18 ∘C) until processed. Chl a was then analysed according to the method described by Wasmund et al. (2006) using a fluorometer (Jenway 6285, precision of 0.01± < 1 ng mL-1). During the Norwegian cruise (HE491), in vivo Chl a was measured with a hand fluorometer (Turner Designs, AquaFluorTM, precision of 0.001 absorption) and related to µg of Chl a using a calibration factor between filtered Chl a (Chl a standard in EtOH as reference) and in vivo absorbance.

The total cell numbers (TCNs) of prokaryotic and small autotrophic cells were determined by flow cytometry following a modified protocol from Marie et al. (2000). For determination of bacterial cell numbers, water samples were fixed with glutaraldehyde (1 % final concentration), incubated at room temperature for 1 h and stored at -18∘C until further analysis. Prokaryotic cells were stained with SYBR Green I (2.5 mM final concentration, Molecular Probes, Schwerte, Germany) for 30 min in the dark. Samples were measured on a flow cytometer (C6 FlowCytometer, BD Bioscience, fluorescence accuracy of FITC (fluorescein isothiocyanate) < 75; PE (phycoerythrin) < 50), and cells were counted according to side-scattered light and emitted green fluorescence. We used 1.0 µm beads (Fluoresbrite Multifluorescent, Polysciences) as an internal reference to monitor the performance of the device. Their cell counts include heterotrophic and photoautotrophic prokaryotes. Pico- and nano-autotrophic cells were counted after the addition of red fluorescent latex beads (Polysciences, Eppelheim, Germany) and were detected by their signature in a plot of red (FL3) vs. orange (FL2) fluorescence, and red fluorescence vs. side scatter (SSC). We did not further differentiate between different groups of prokaryotic and eukaryotic autotrophs.

TEPs were measured by filtering seawater, in triplicates, onto 0.2 µm polycarbonate filters under low vacuum (< 100 mm Hg) and staining with alcian blue solution (0.02 g alcian blue in 100 mL of acetic acid solution of pH 2.5) for 5 s. The 0.2 µm filters collect both large TEP aggregates and smaller colloidal TEP material. Filters were stored at -18 ∘C until processed. Alcian blue stain was extracted for 2 h in 80 % sulfuric acid, with gentle agitation applied to reduce bubble formation, and analysed using a spectrophotometer (VWR UV-1600PC, precision of 1±0.2 % T) and the spectrophotometric method (Passow and Alldredge, 1995). The stock solution of alcian blue was calibrated using the xanthan gum (Carl Roth) standard according to Passow and Alldredge (1995). TEP concentrations are shown in relation to xanthan gum equivalence. Recent calibration issues with xanthan gum were not observed in our studies, and thus the new method by Bittar et al. (2018) was not required.

To estimate local primary production, we used an adjusted version of the Vertically Generalized Production Model (VGPM) (Behrenfeld and Falkowski, 1997), as described by Wurl et al. (2011b). Estimation is based on concentration of Chl a, depth of euphotic zone estimated from the Secchi depth, photoperiod and photosynthetic active radiation (PAR).

Statistical analyses of the data set were performed using Graphpad PRISM version 8. Differences, null hypothesis testing, and correlation were considered significant when p < 0.05. The data were log transformed, if required, for parametric and analysis of variance (ANOVA) tests; further post-hoc Tukey analysis was run for comparison of means when the difference was significant in ANOVA. Unless otherwise indicated, results are presented as means ± standard deviations. Enrichment factors (EFs) were calculated as the ratio of concentrations in the SML sample to that of corresponding ULW taken at 1 m depth. For vertical sample profiles, the variance of each depth measurement from the average was used to determine homogeneity. Variance is the squared deviation from the mean of all depths and is thus given in units squared (e.g. µg2 L-2).

General characteristics of parameters for all three campaigns are shown in Figs. 2 and 3. We observed low (< 2 m s-1), moderate (2–5 m s-1) and high (> 5 m s-1) wind regimes (Wurl et al., 2011b). Average wind speed was 3.8±0.3, 4.2±2 and 5.6±1.8 m s-1 for the Baltic Sea, Norwegian Sea and Cabo Verde, respectively. PAR averages were 1172±145 and 739±251 µmol m-2 s-1 for the Baltic Sea and Norwegian Sea, respectively, and sea surface temperature (SST), measured from the SML, was 14.8±1.9 and 14.9±1.4 ∘C. Stations for the Baltic cruise were sampled within the same area (∼1 km) and thus had similar salinity (8.92±0.2) relative to those from the Norwegian cruise. The stations for the Norwegian cruise covered inner and outer fjord and open-ocean areas and thus had larger differences of salinity. SML salinity was (32.4±2; 23.5±0.4; 6.7±3.5) for outer fjord/open ocean, Trondheim fjord and Sognefjord, respectively. ULW salinity was 32.7±2, 23.9±0.2 and 6.6±3.5 for outer fjord/open ocean, Trondheim Fjord and Sognefjord, respectively. PAR, SST and salinity data were not collected for the Cabo Verde campaign due to logistical constraints. Primary production (PP) ranged from 426 to 734 mg m-2 d-1 during the Baltic cruise but had a higher range during the Norwegian cruise with 318–1194 mg m-2 d-1. Again, this is likely due to the differing water masses sampled during the Norwegian cruise.

A one-way Tukey analysis of variance was used to compare the concentration and enrichment of the main three parameters between all regions: TEPs, Chl a and POC. Figure 2 shows that TEP concentrations were significantly higher in the Baltic Sea compared to Cabo Verde and the Norwegian Sea, and significantly lower in Cabo Verde compared to the Baltic Sea and Norwegian Sea (SML: p < 0.0005, n=11; ULW: p < 0.0009, n=11). TEP enrichment was significantly higher in Cabo Verde compared to the other regions and significantly lower in the Baltic Sea compared to the other regions (p < 0.0418, n=8). Samples from the Norwegian Sea cruise fell between the other two cruises in significance for all parameters. Chl a concentrations matched TEPs with significantly higher concentrations in the Baltic Sea and lower in Cabo Verde (SML and ULW: p < 0.0001, n=8). However, Chl a enrichment had no significant difference found between the regions, likely due to overall low enrichment values. POC enrichment was significantly lower in the Norwegian Sea than in Cabo Verde (p < 0.0062, n=5) but not compared to the Baltic Sea. Statistical analysis for POC could not be run using data from the Baltic Sea due to a low number of samples (n=4). However, both SML and ULW POC concentrations were not significantly different between the Norwegian Sea and Cabo Verde but POC enrichment was (p < 0.01, n=6).

TEP concentrations at various depths (in metres) are shown in Table 3. Variance between concentrations is used to express the relative homogeneity of the parameter within the upper 2 m and results are given in units squared. During the Baltic cruise, there was a distinct change in TEP distribution between the first and second halves of the cruise (Fig. 4). TEP concentrations were lower and homogenous (average variance of 8.63×103 µg XG eq2 L-2) for stations 3–8 but became higher in concentration and heterogeneous (average variance of 6.47×105 µg XG eq2 L-2) for stations 9–12. Variance of Chl a was highest at stations 9, 10 and 12 and lowest at stations 3–5, showing a positive linear correlation between average vertical concentration and homogeneity (R2=0.95, p < 0.0001, n=8): no such correlation was observed for TEPs (R2=0.29, p=0.16, n=8). The vertical profiles for microbial counts were also taken in the Baltic Sea to investigate if there was any correlation to TEP depth profiles due to the importance of the microbial loop in TEP production and consumption (Yamada et al., 2013; Busch et al., 2017). However, no correlation or direct connection could be found between TEP profiles and microbial profiles, given as TCNs and small autotroph profiles (Fig. S1 in the Supplement).

During the Norwegian cruise, the vertical distribution of TEPs varied greatly between stations with the highest variance at station 3 (open-ocean station) and the lowest variance at stations 5, 7 and 11 (fjord/nearshore). There was no relation between TEP variance and geographical location, e.g. nearshore vs. fjord systems, vs. open ocean. Additionally, no correlation was found between TEPs and turbulent kinetic energy (TKE), measured with an acoustic Doppler velocimeter (data not shown). TKE data during the Norwegian cruise are presented in Banko-Kubis et al. (2019). TEP profiles shown in Fig. 5 were chosen based on minimum, median and maximum variance, and presented as such, since no correlation could be found to any other parameter. It is important to note that station 3 had a variance nearly 24 times larger than the second highest variance (station 9). Chl a and POC showed a moderate correlation between concentration and variance (Chl a: R2=0.67, p < 0.0006, n=13; POC: R2=0.63, p < 0.0013, n=13). However, TEPs showed no similar correlation when the putative outlier variance from station 3 was excluded (R2=0.02, p=0.66, n=12). During both cruises, TEPs were found to be enriched even when POC was not, but POC was never enriched without TEPs also being enriched.

Comparing the enrichment of the SML between each region showed a higher variability of enrichment within each region than between the regions (Table 2), supporting the notion that SML enrichments are a global phenomenon (Wurl et al., 2011b). Interestingly, while there was a significant but weak correlation between Chl a and TEP concentration in both the ULW and SML (ULW: R2=0.32, p < 0.0007, n=30; SML: R2=0.36, p < 0.0005, n=30), there was no significant correlation between the enrichment of TEPs and enrichment of Chl a (R2=0.045, p=0.27, n=30). This suggests that, while phytoplankton are the main source for TEP production, the transport mechanisms for TEPs and phytoplankton differ. This is in large part due to the motility of phytoplankton species, which are known to have vertical migration patterns (Bollens et al., 2010; Schuech and Menden-Deuer, 2014) and can have motility responses to physical changes like turbulence (Sengupta et al., 2017).

While the highest abundances of TEPs were found in the Baltic Sea and the lowest abundances in Cabo Verde (Fig. 2d), the highest enrichment factors were found in Cabo Verde (Fig. 2c). As the nearshore Cabo Verde waters are oligotrophic, this complements previous studies by Wurl et al. (2011a, b), which found the highest enrichment of surfactants to be in oligotrophic waters compared to mesotrophic and eutrophic. While manual sampling techniques were employed in Cabo Verde in comparison to rotating glass disc samples in the other campaigns, earlier comparative studies by Shinki et al. (2012) found both methods to collect similar SML thickness and associated biochemical parameters. Since our catamaran was modelled after Shinki et al. (2012), we are able to compare the results from both versions of the glass plate method.

We observed enrichment of all parameters irrespective of either instantaneous wind speeds (2 h average) or wind speed history (24 h average), including higher wind speeds > 7 m s-1. This supports previous studies which found enrichment of material even at wind speeds > 8 m s-1 (Reinthaler et al., 2008; Kuznetsova et al., 2004), including the enrichment of TEPs in the SML at moderate wind speeds (Wurl et al., 2009). Breaking waves from moderate wind regimes can create bubble plumes in the near-surface water (Deane and Stokes, 2002; Blanchard and Woodcock, 1957), and this bubbling has proven to be an effective transport mechanism for TEPs and DOM (Robinson et al., 2019a; Zhou et al., 1998) to the SML. Thus, bubbling and turbulence at moderate wind speeds can induce more complex enrichment processes and subdue any direct correlation with wind speed. We never observed wind speeds greater than 8 m s-1, which has been found to be the threshold speed for the breakup of TEPs during experiments in a wind-wave tunnel by Sun et al. (2018). Thus, the moderate wind speeds we observed likely had an indirect positive effect on enrichment.

While TEP concentrations mimic Chl a or PP due to the large contribution phytoplankton play in TEP creation (Passow, 2002a; Ortega-Retuerta et al., 2017), the enrichment of TEPs is driven by many other processes. Wurl et al. (2011a) found TEP enrichment to be irrespective of PP or negatively related with highest enrichment in oligotrophic waters with the lowest PP. Considering the relationship of PP and TEP enrichment within each region, we found this to be true for the Baltic Sea but not for the Norwegian Sea. In the Baltic Sea, as PP increased, enrichment of TEPs decreased, due to a larger increase of TEP concentration in the ULW caused from a post-bloom state (Fig. 3). However, in the Norwegian Sea, TEP enrichment matched PP, most likely due to the changing water bodies in that study, whereas the same body of water was sampled over time for the Baltic Sea and offshore Cabo Verde. While we do not have PP data for Cabo Verde, considering the positive relationship between PP and Chl a concentration in the SML for the Baltic and Norway data, Chl a concentration in the SML can be used here as proxy for PP in Cabo Verde. Under this premise, Chl a in Cabo Verde SML was similar to the Baltic in that, while Chl a and TEP SML concentrations were correlated (R2=0.68, p < 0.012, n=8), enrichment was not (R2=0.19, p=0.24, n=8).

The oligotrophic waters of Cabo Verde present an interesting scenario for TEP production. We found the highest enrichment of TEPs in the SML here, while simultaneously observing the lowest concentrations of both Chl a and TEPs. We suggest that this is due to the abundance of precursor material as well as the increased formation of TEPs via the abiotic pathway. A tank experiment with the same setup as used by Robinson et al. (2019a), with different techniques to bubble the water, was also employed in Cabo Verde, using water from the same nearshore area as field samples. The tank was made of 10 mm polyvinyl chloride (PVC) plates in a size of 120 cm length ×110 cm width ×100 cm height. The tank had a volume of 1400 L with a 500 L aerosol chamber on top. Materials in contact with seawater were made from teflon, including liners for the wall using teflon bags. The unfiltered seawater was bubbled using the waterfall technique (Cipriano and Blanchard, 1981; Haines and Johnson, 1995) and via bubbling, large abundances of TEPs were created (Fig. S2). This suggests that, while TEP concentration in the nearshore water was low, the colloidal and precursor material for TEPs was present and only required sufficient formation mechanisms to form aggregates in the size range to be identified as TEPs. Such precursor material may have been deposited from the atmosphere; Cabo Verde is known for its Saharan dust deposition events and indeed dust events were observed during our campaign (supporting data will be shown in this special issue). This dust deposition has been shown to increase the abiotic formation of TEPs (Louis et al., 2017) and is potentially a large contributing factor to higher enrichment of TEPs in Cabo Verde. If true, this presents interesting implications for the residence time of dust which ends up floating in the SML with sufficient time for photochemical processing.

In contrast, in mesotrophic and eutrophic water, there is more biological activity present in the ULW which can increase the complexity of the system by which TEPs and their precursor material is recycled or altered before it can reach the SML. Such increase in biologically derived complexity can be seen in the depth profile data from the Baltic Sea, which showed increased heterogeneous mixing of TEPs in the water when PP was higher during the second half of the cruise. Indeed, the HSV data for all parameters show that the vertical flux processes are not straightforward to interpret.

Previous studies have found chemical characteristics to be heterogeneous in the upper 1–2 m (Goering and Wallen, 1967; Manzi et al., 1977; Momzikoff et al., 2004), substantiating the notion that vertical flux processes are too complex and strong to assume homogenous mixing. Additionally, phytoplankton communities and abundance have also been found to be heterogeneous in the water column (Cheriton et al., 2009; Dekshenieks et al., 2001; Mitchell et al., 2008). Considering the importance of ULW concentrations in estimating enrichment, the presumption of ULW as homogenous becomes problematic. When considering a parameter like TEPs which bridges the boundary of biological and chemical parameters and is so fundamentally affected by both, these studies become even more crucial indicators for the likelihood of TEP distributions to be heterogeneous, at least in surface water.

Vertical profiles were sampled for the Baltic and Norwegian seas, and in both regions, the vertical distribution of TEPs, Chl a, POC, PON were found to change from station to station. In the Baltic Sea, the vertical variance of TEPs appears to be linked to PP and the creation of TEPs from the biotic pathway. Higher deviation was seen in the second half when TEP and Chl a abundances were higher. One possible biological cause for this heterogeneous mixing could come from its link to phytoplankton. For example, Nielsen et al. (1990), Bjørnsen and Nielsen (1991) and Carpenter et al. (1995) found vertical phytoplankton patches within the water column in the open North Sea and Baltic Sea. Additionally, Cheriton et al. (2009) found that vertical oscillations cause stratification of phytoplankton into thin vertical patches. Thus, if these same processes were to occur in the near-surface environment, stratification of plankton could result in TEP precursor material being released in patches and, with rapid aggregation, create heterogeneous distribution of TEPs. This holds true especially with the majority of TEPs produced by diatoms, which are non-mobile phytoplankton that would be more susceptible to grouping and patchiness by physical forcing. This is hinted at by the heterogeneous mixing of Chl a observed during both cruises (Figs. 4 and 5), which always showed heterogeneous mixing. However, this cannot be the only mechanism, as the peak TEP concentration was not always seen at the same depth as Chl a. Further studies on vertical phytoplankton distribution in the near-surface (> 2 m) environment are needed in order to substantiate the role their patching might have on DOM and TEPs. Furthermore, there was a significant correlation between high variance of Chl a and high variance of TEPs in the Baltic Sea (R2=0.65, p < 0.028, n=7; station 4 excluded) but no correlation in the Norwegian Sea (R2=0.00, p=0.92, n=13). This suggests that with sufficient phytoplankton abundances, and reduced influence from the open ocean, the biological influences on TEP heterogeneity can dominate.

While biological sources are likely to determine the chemical characteristic of TEPs, they are not the only influence. TEPs are operationally defined polymers of acidic polysaccharides and naturally positively buoyant with highly surface active properties (Zhou et al., 1998). When unballasted by detritus or other organic matter, TEPs have a positive buoyancy and can rise to the surface at rates of 0.1–1 m d-1 (Azetsu-Scott and Passow, 2004). However, TEPs are never unattached from some type of OM, and it is this OM which helps to determine the sinking velocity of TEPs (Passow, 2002a). Additionally, the density of TEPs is dependent on the formation of its precursor material; e.g. the resulting density, and therefore sinking or rising velocity of TEPs produced from diatoms vs. bacteria will differ as well as TEPs produced from nutrient depletion vs. temperature stress (Mari et al., 2017). In near-surface water, where the ambient density of water is stratified, this could result in the immobilization of TEPs into layers of water with equal density. Thus, heterogeneous mixing of TEPs may be caused by a sort of vertical filtering via TEP density and surrounding water density.