Oceanic samples were collected at 150 m depth using a PPS3/3-24 S time-series sediment trap with a sampling area of 0.125 m² and a sampling frequency of 12 hours. Half the samples were taken in diurnal periods, from 6 a.m. to 6 p.m. (UTC, solar time), and the other half in nocturnal periods, from 6 p.m. to 6 a.m. The sediment trap array was deployed twice, the first time from 2 to 6 April 2011, and then from 7 to 13 April 2011. No samples were collected on 6 April 2011 because the day was used for the collection procedure and the new deployment. The sediment trap was installed as a free-drifting system, initially located at a distance of nearly 50 km north of Gran Canaria Island (Fig. 1). The samples were filtered just after collection, using a glass fibre filter (GF/C) with a pore size of 1.4 mm to avoid possible silica dissolution (Bauerfeind and Bodungen 2006). The filters were split in two, one for lithogenic and biogenic particles and the other for POC analysis. The filters used for lithogenic and biogenic quantification were treated with hydrogen peroxide (3%) previously heated to 50°C in order to eliminate organic matter. Gelatinous matter agglutinating particles persisted after treatment. This gelatinous matter had a similar visual aspect to other transparent masses observed by other authors (Martinez-Ruiz et al. 2019) and defined as TEPs (Alldredge et al. 1993, Passow 2002, Mari et al. 2017). Mineralogical determination of lithogenic particles was implemented using scanning electron microscopy/energy dispersive x-ray spectroscopy (SEM-EDX) at the Microscopy Service of the University of Huelva, Spain. The filter fractions used for POC analysis were wrapped in pre-combusted aluminium foil and frozen at –20°C until laboratory determination using a CHN analyser (Carlo Erba EA 1108 Elemental), according to procedures described by Alonso-González et al. (2010a). Swimmers (zooplankton entering the trap, Michaels et al. 1988) larger than approximately 1 mm were manually removed under a stereoscopic microscope. Dust events were identified using the dust load charts provided by the DREAM forecasting model at 0.0 h (Pérez et al. 2011, Haustein et al. 2012, Fig. 2). The DREAM prediction model of the Saharan plume dust also showed real time information at 0.0 h jointly with meteorological data; the daily operational model evaluation included satellites (MODIS and MSG) and Aerosol Robotic Network (AERONET) sun photometers (Basart et al. 2012). This real time information (expressed in dust concentration and flux) was used in this study.

The recorded Saharan dust period was from 30 March to 12 April (Fig. 2). Two Saharan dust plumes were observed during the sediment trap sampling period (Figs 2 and 3). The first and most intense was from 3 March to 3 April, with a peak of 15 mg m² day^–1, and a second dust plume was recorded from the second half of 6 April to the first half of 8 April, showing a peak of 1.5 mg m² day^–1. At the peak of lithogenic flux (2 April), there was no Saharan dust in the atmosphere. The dust input to the ocean was in dry deposition excepting a small amount of rain, 0.2 mm, on 2 April (source: automatic station C659M-Plaza de la Feria, City of Las Palmas de Gran Canaria, from the Spanish meteorology state agency, AEMET).

Images from the filters were obtained with a Leika© MZ6 stereoscopic microscope equipped with a photographic camera. They were taken at a resolution of 5.0 megapixels, illuminated by multiple non-polarizing natural light foci in order to avoid shadows. An 8x zoom and a grid of 4 mm were used to scan the whole filter, taking photographs in each grid area. A set of 396 images were obtained in total, at about 22-27 images for each filter. The image analysis was designed to guarantee a standard process for each filter. Firstly, the threshold for each picture was adjusted using the Corel PaintShop program. Brightness and contrast levels were established using a lens in order to minimize the background noise. Scaling of the images was performed using ImageJ software and the ratio between pixel size and the actual dimensions of the grid was measured. Each particle, agglomerate and biogenic formation was identified as irregular polygons, thus allowing their morphometric characteristics to be easily obtained. For a better discrimination between the background and particles, the images were binarized from RGB to black and white. The error uncertainty of the image treatment measurements is the pixel size (2 mm). No particles below this size were measured. This uncertainty is similar to the filter procedure, in which particles smaller than the filter voids would be lost during sample filtering. We observed no particles smaller than 2 mm in the SEM analyses performed on selected samples.

The geometrical properties (area, perimeter, major and minor axis) and morphometry features were computed. The lithogenic particles were characterized by grain size and morphometric descriptors: circularity index (CI) and roundness index (RI, Wadell 1934), which will be defined in the next section. The volume of the particles was calculated from their area (Okada et al. 2001). Particle volume was adjusted to an ellipsoid because it has been observed that an ellipsoidal volume is closer to the 3D shape of the mineral particles (Jaramillo et al. 2011). Ellipse axes were then calculated using their roundness and particle area, following the equations:

Minor axis = Major axis × Roundness

The software calculated the CI and the RI. These parameters describe the geometrical complexity of the particles. The CI approximates the particle sphericity and was defined by Blott and Pye (2008) as

The RI approximates the particle irregularity and was determined as the ratio between the minor and the major dimensions of the particle. Again, an ellipse was chosen as the best representation, and major and minor dimensions correspond to the major axis and minor axis, which was equivalent to the particle area. The expression is given below, following Blott and Pye (2008):

According to Jaramillo et al. (2016), the mean density of the lithogenic particles collected in the sediment trap, considering its mineralogical composition, was 3.4 mg mm^–3. Particle mass was therefore estimated from their volume and density. The lithogenic flux (in mg m^–2 12 h^–1) was determined considering the trap area (0.125 m²) and the sampling time (12 h). The intermediate axis of the collected particles was used for the accumulated grain size distribution.

Biogenic particles were estimated using a similar procedure to those used in lithogenics. Due to their irregularity (chitin), as well as to the fact that many of them have hollow shells (foraminifers and gastropods), their volume was calculated from the surface area multiplied by an estimated mean height of 0.012 mm for foraminifera and gastropods (carbonates) and 0.005 mm for crustaceans (chitin). Their density values were 1370 kg m^–3 for chitin (https://www.guidechem.com/dictionary/en/1398-61-4.html) and 2710 kg m^–3 for carbonates (as calcite, http://webmineral.com/data/Calcite.shtml#.XCZ1Wlz7SyI). POC analysis was performed according to the procedures described by Ariza et al. (2015). Dissolved organic carbon adsorption onto the filters ranged from 0.3 to 1.6 μmol C per 25 mm diameter GF/F filter (less than 3.5% of the POC signal), similar to the blanks reported by Alonso-González et al. (2010a).

No opal individuals or fragments (radiolarian or diatoms) were identified by binocular study in these images, due to the scale (pixel definition was 2 mm). A binocular study was used for particle geochemistry identification through SEM images with only four filters (2d, 2n, 4s, 4n) and only two diatoms or radiolarians were found. The abundance of silica biogenic particles has been determined in other studies (using the automated wet leaching method; Müller and Schneider 1993) at this same area and depth, and also resulted in low values (0.21 mg m^–2 day^–1, in Neuer et al. 2004).

The theoretical settling velocity (V) was calculated by the Stokes’ law as follows (Allen 1992):

where ρ_s is the quartz spheres density (ρ_s=2650 kg m^–3), ρ_f is the seawater density (ρ_f=1025 kg m^–3), µ is the seawater viscosity (9.05×10^–4 N s m^–2 at 10°C), g is the standard acceleration due to gravity (g=9.80665 m s^–2), and d is the particle diameter, in mm.

A typical Saharan dust event has an average grain size distribution of 30±19 mm (dashed line [pink] distribution in Figure 4) (Menéndez et al. 2009), but the size distribution observed at 150 m depth in this study after a dust event was significantly coarser (average grain size of 57±16 mm). This mismatch could be due to differences in the sinking rate between large and small particles, as was also observed in mesocosm experiments for particles larger and smaller than 5 mm (Bressac et al. 2014). Smaller particles could be retained in the water column because of their low weight and density, while larger ones could sink at higher velocities. Biological activity could also retain small particles in the upper layers as an effect of ingestion by micro- and mesozooplankton (see Jaramillo et al. 2016). The aggregation effect of the organic matter (Jackson et al. 2015) and the formation of sticky TEPs, more stable in oxidation processes than the organic matter (Pakulski and Benner 1994), could also contribute to the presence of larger particles in the sediment trap. These TEPs were observed in the filters because of their relatively good preservation after the procedure described to eliminate the organic matter. The oceanic deposition of a Saharan dust event in surface waters produces an increment of lithogenic particles, favouring an intensification of the POC and TEP ballast effect (Mari et al. 2017). These TEPs are a key factor in the formation of new aggregates, which could be unspoiled during the settling process through the mixed layer, and consequently accelerate the sinking velocity of the POC (Reid et al. 2003, Neuer et al. 2004, de Leeuw et al. 2014).

Lithogenic particles are denser, and act as ballast, supporting the vertical transport of biogenic aggregates to the deep ocean (Armstrong et al. 2002). These particles could be incorporated into organic aggregates, acting as ballast increasing sinking, and enhancing the carbon exported to the mesopelagic zone (see van der Jagt et al. 2018). Conversely, these dragging processes force a reduction in the sinking velocities of the lithogenic particles. In this sense, we observed a six-fold reduction in the vertical velocity compared with the theoretical Stokes settling velocities (50 m day^–1 instead of 275 m day^–1). Physical processes in the water column, such as mixing and turbulence, could also promote a slowdown of sinking particles, but the Canary Current velocity (about 4300 m day^–1; Sangrà et al. 2009) could be neglected in the settling calculations due to the free-drifting line design of this sediment trap. Maximum vertical velocities in this area of the ocean were estimated in oceanic eddies. The vertical upward velocity generated by cyclonic eddies shed by Gran Canaria Island was w=7.5 m day^–1 (Barton et al. 1998), though this value might represent a slight reduction of about 3% in the sinking velocity of lithogenic particles.

The three-day delay between the lithogenic flux and the dust peaks was taken into account to calculate the settling particles rate from the ocean surface to 150 m depth, yielding a value of 50 m day^–1. This value is equal to that observed by Jackson et al. (2015) in the upper 100 m of the ocean along the southwest coast of the United States and in the range of the maximum velocities registered in a mesocosms experiment at 15 m depth in Corsica, of about 24 to 86 m day⁻¹, corresponding to organic-mineral aggregate populations (>61.2 μm; Bressac et al. 2012). A settling estimation rate was also made close to Cape Verde at 1000-2500 m depth, resulting in this case of 150 m day^–1 (Bory and Newton 2000). This value is thus three times the velocity observed in our study. Berelson (2002) obtained an increase of the settling rate by a factor of two- to ten-fold between 100 and 2000 m depth, which was explained due to the formation of aggregates. Our values for the epipelagic zone could thus be reasonable, since the formation of aggregates should be lower in upper layers than aggregation during their long downward sinking, entering the meso- and bathypelagic zones. Both lithogenic and biogenic fluxes seem to increase with depth (Table 1). This intensification of fluxes with depth can also be explained as lateral advection (Alonso-González et al. 2010b, Freudenthal et al. 2001, Sprengel et al. 2002). However, lithogenic fluxes at similar depths also show significant changes depending on local conditions and factors such as trophic status, continental sediment supply, seasonality, upwelling and/or ocean currents. Recent efforts to standardize lithogenic flux methods demonstrated a discrepancy of more than one order of magnitude in these values (Anderson et al. 2016).

The coincidence of peaks in biogenic (POC and chitin) and lithogenic fluxes (Fig. 4) suggests a similar settling velocity of these particles, probably due to the aggregation effect of the stable TEPs. Accordingly, the observed quantitative relationship between dust and lithogenic, chitin and foraminifer fluxes (Fig. 9) might reflect their ballast-balance interaction. The highest sinking rates of the marine snow were observed in dust-rich ocean regions off NW Africa, but could not be explained by carbonate content (Fischer and Karakas 2009). Lithogenic particles (of Saharan dust origin) may play a considerable role in the POC-TEP-carbonate biogenic aggregate fluxes, as observed in the present study. Coupled with the occurrence of lithogenic-biogenic fluxes, this finding could contribute to more accurate prediction in ocean carbon pump models.

The lithogenic, POC and foraminifera fluxes did not seem to vary between day and night. Lithogenic flux was regular in time, but POC and foraminifera fluxes appeared to be more irregular. Conversely, the chitin flux was larger at night, the greatest difference being with the largest flux (Fig. 6). Despite the extraordinary amount of chitin produced annually in the oceans (Souza et al. 2011), few studies about the chitin flux can be found in the literature. Montgomery et al. (1990) study the chitin flux measured in the mixed layer of the subarctic Pacific study (4.4-1.3 mg m^–2 day^–1), much lower than that measured in the Canary Basin.

Ariza et al. (2015) observed large zooplankton (>1 mm) to show a clear day/night difference of biomass (D/N= 0.63±0.19) in the epipelagic layer as a consequence of the diel vertical migration of copepods, euphausiids and chaetognaths. This is a 59% increase in the biomass of large organisms in the upper layers of the ocean at night, and a quite significant increase in crustacean density passing the depth of the sediment trap every night. The explanation for this increase of chitin at night must be related to the vertical migration of zooplankton. We removed large swimmers (see material and methods) but not smaller sizes; however, we observed small chitin pieces of these crustaceans, probably pieces that were broken when crustaceans were eaten through trophic interaction at night. The origin of these small pieces of chitin in the sediment trap remains unknown, but further research should be undertaken to ascertain their role as a flux or, simply, as contamination.

The Saharan dust plume in early April 2011 produced an increase of one and two orders of magnitude in the lithogenic and chitin fluxes, respectively. Rather than biogenic production, the considerable flux increase could be due to mineral particles promoting organic-inorganic aggregation, increasing the sinking velocity of these aggregates and providing physical protection against the remineralization of the labile organic matter component (van der Jagt et al. 2018). The measured magnitudes of particle fluxes at 150 m depth in the Canary Current mixed layer (Table 1) were not balanced, the chitin flux clearly predominating (96%) over far smaller foraminifer-carbonate (3%) and lithogenic (0.3%) fluxes. Neuer et al. (2004) also showed that the carbonate biogenic flux was in the same range of values as those observed in the present study; the lithogenic flux was lower in the present study (Table 1), but the lithogenic/(lithogenics+dust) ratio was similar in the two studies (0.1). This finding suggests that in this oceanic region, as expected, the inputs of lithogenic fluxes are restricted to Saharan dust, without any significant effect from fluvial input. According to the data presented by Maeda et al (2007), the ratio for the NW Pacific Ocean was higher (0.6; Table 1), due to the important fluvial contribution to the lithogenic flux in this area.

Opal fluxes (diatoms and radiolarian) were not detected by our image processing but the values measured by Neuer et al. (2004) in this area and at this depth were low (0.21 mg m^–2 day^–1). It seems that diatoms tend to sink rapidly to the seafloor as large chains (Berelson 2002), and therefore might not contribute to marine snow formation. Calcium carbonate organisms are therefore the predominant biogenic flux collected in sediment traps in the Canary region (Fischer et al. 1996, Abrantes et al. 2002, Alonso-González et al. 2010a) and also further south (Korte et al. 2016).

The low export of biogenic matter characteristic of the Canary Basin waters due to their oligotrophy (Neuer et al. 2004) could also accelerate export because of the link to the lithogenic fluxes. POC flux north of the Canary Archipelago was on average 4.5±2.9 mgC m^–2 day^–1 at 200 m depth (Helmke et al. 2010), while Ariza et al. (2015) found a value of 11.9±5.8 mgC m^–2 day^–1 at 150 m depth after the late winter bloom in these waters. These slightly higher values were due, in part, to the larger value observed three days after the dust deposition. As stated above, lithogenic material can be incorporated into organic aggregates, acting as ballast, increasing sinking, and enhancing the carbon exported to the mesopelagic zone (van der Jagt et al. 2018).

Finally, a comparison of lithogenic and biogenic fluxes in other regions, at similar sampling depths, such as in the Panama Basin (Honjo et al. 1982) and the NW Pacific (Maeda et al. 2007), showed much higher values (11 and 12 mg m^–2 day^–1, respectively) than those observed here in the Canary Basin, but the lithogenic/(lithogenics+biogenic) ratio was similar to that of these studies (0.1-0.4). This generally small range could be a consequence of the well-known micronutrient effect (i.e. iron) of the lithogenic minerals (Jickells et al. 2005, Hernández-León et al. 2007, Álvarez-Salgado and Arístegui 2015), controlling biological production in relation to micronutrient availability. Bressac et al. (2014) did not find significant correlations between primary production and the POC flux in the framework of dust seeding mesocosm experiments, and thus showed that POC flux was instead explained by dust seeding, estimating that 42% to 50% of POC fluxes were associated strictly through processes of aggregation, and most probably sorption, of the lithogenic particles.