A series of deep-moored sediment trap deployments was instigated in 1997 by the Australian SAZ program (Trull et al., 2001b) and now continues as a component of the Australian Integrated Marine Observing System Southern Ocean Time Series (Trull et al., 2010; Shadwick et al., 2015). Two sites representative of a large proportion of the SAZ and PFZ were occupied quasi-continuously for the decade 1997–2007. Both sites were located along the 140∘ E longitude: station 47∘ S was set on the abyssal plain of the central SAZ, whereas station 54∘ S was placed on a bathymetric high of the Southeast Indian Ridge in the PFZ (Fig. 1, Table 1). Additionally, two other sites were instrumented over a 1-year period, beneath the SAF (site 51∘ S, 1997–1998) and within the southern Antarctic Zone (AZ) (site 61∘ S, 2001–2002). Here, we present data from the 47∘ S 1000 m trap between 1999 and 2001 (2-year record) and from the 54∘ S 800 m trap between the following years: 1997–1998, 1999–2000, 2002–2004 and 2005–2007 (6-year record). Biogenic particle flux data of sites 47, 51 and 54∘ S for the first year deployment (1997–1998) and of site 61∘ S for the year 2001–2002 have already been published in Trull et al. (2001a) and Rigual-Hernández et al. (2015), respectively.

All traps were MacLane Parflux sediment traps: conical in shape with a 0.5 m2 opening area and equipped with a carrousel of 13 or 21 sampling cups. Cup rotation intervals were established based on anticipated mass fluxes. The shortest intervals corresponded with the austral summer and autumn ranging typically between 4.25 and 10 days, whereas the longest intervals were 60 days, corresponding with winter (Table 2). Each trap was paired with an Aanderaa current metre and temperature sensor. The 250 mL collection cups were filled with a buffered solution of sodium tetraborate (1 g L-1), sodium chloride (5 g L-1) and mercury chloride (3 g L-1) in unfiltered deep seawater from the region (collected at 1200 m depth, 49∘17′ S, 153∘58′ E). Full details of the mooring designs can be found in Bray et al. (2000) and Trull et al. (2001a).

Current speeds largely influence the efficiency with which sediment traps collect the particles sinking in the water column (Baker et al., 1988; Yu et al., 2001). The threshold of current velocity above which sinking particles are no longer quantitatively sampled is not well known, but has been suggested to be around 12 cm s-1 (Baker et al., 1988). Average current speeds for the whole sampling interval at the trap levels were lower than 11 cm s-1 for both sites and showed little seasonal variability (Bray, unpublished results, available online at imos.org.au). Therefore, these mild conditions seem to be sub-critical for any strong concerns over collection efficiencies. Additionally, radioisotope analyses of material from the first year deployment by Trull et al. (2001) provide some extra insights to assess the collection efficiency of the traps. The 230Th flux/production ratios for the 1997–1998 deployment were 0.6 ± 0.1 and 0.7 ± 0.1 for the 47 and 54∘ S traps, respectively. These values suggest that some degree of under trapping is likely to have occurred at both sites. However, as these values are almost identical for both traps, it can be assumed that the trapping efficiency did not account for the observed latitudinal variations in the magnitude of the particle export between sites. Taking into consideration all the above and the fact that the assessment of trapping efficiency from 230Th alone is fraught with uncertainties (Trull et al., 2001a; Buesseler et al., 2007), trap fluxes were not corrected for possible under trapping in the present study.

A detailed description of the methodology used for the determination of the flux intensity and composition of settling particles for the first mooring deployments in 1997–1998 can be found in Bray et al. (2000) and Trull et al. (2001a). After recovery, sediment trap cups were allowed to settle before supernatant was drawn off with a syringe for salinity, nutrients and pH measurements. The remaining sample slurries were sieved through a 1 mm sieve and then split into 10 fractions using a rotary splitter (McLane, Inc.). Three of these splits were filtered onto Nucleopore filters (0.45 pore size), removed from the filter as a wet cake of material, oven-dried at 60 ∘C and ground in a mortar. This material was used to determine the dry mass flux and the major components of the flux (particulate inorganic carbon (PIC), POC and biogenic silica). PIC was determined by closed system acidification with phosphoric acid and coulometry. Particulate total carbon (PC) was determined by unacidified combustion using a carbon-hydrogen-nitrogen (CHN) elemental analyzer. POC was calculated from PC by substraction of PIC. Total silicon and aluminium contents were estimated by HF–HNO3 microwave digestion and inductively coupled plasma emissions spectrometry following the methodology described by Bray et al. (2000). Biogenic silica was determined from total silica by subtracting lithogenic silica estimated by assuming a lithogenic Al : Si mass ratio of 3.42 (Taylor, 1964). These methods for PIC and POC/particulate organic nitrogen (PON) were used for all subsequent years, with very slight modifications: (i) the wet cake method was replaced by drying prior to removing the material from the filter, (ii) in some years sieving and filtering was done at sea and the samples were frozen on the filters until dried upon returning to land. The silica methods varied more strongly over time: (i) for deployments beginning in 1998, 1999 and 2000, the use of HF in the digestion was replaced by high temperature combustion with lithium borate in a graphite crucible and HNO3 digestion to determine total silicon and aluminium; (ii) biogenic silica for these years (and retroactively for 1997) was calculated using the updated estimate for the lithogenic Al : Si mass ratio of 3.83 (Taylor and McLennan, 1985); (iii) from 2001 onwards, total silica was not measured, instead hot alkaline digestion and colorimetry was used to estimate biogenic silica directly (following the method of Quéguiner, 2001).

A total of 138 samples were processed for siliceous microplankton analysis. Each split was refilled with distilled water to 40 mL, from which 10 mL were subsampled and buffered with a solution of sodium carbonate and sodium hydrogen carbonate (pH 8) and stored at 4 ∘C in the dark for future calcareous nannoplankton analysis. The remaining 30 mL were treated with potassium permanganate, hydrogen peroxide and concentrated hydrochloric acid following the methodology used by Romero et al. (1999). Three slides per sample were prepared and mounted using the standard decantation method outlined by Bárcena and Abrantes (1998). This method produces random settling of the diatom valves for quantitative microscopic purposes. Siliceous microplankton analysis was carried out on permanent slides (Norland optical adhesive 61 mounting medium; refractive index: 1.56) of acid-cleaned material. Qualitative and quantitative analysis were done at x1000 and x400 magnifications using an Olympus BH-2 compound light optical microscope with phase-contrast illumination. In order to properly characterise the diatom assemblages, a target of 400 diatom valves was counted per sample. Owing to the strong seasonality in diatom production, some cups collected very low numbers of diatom valves. For these samples a compromise between number to be counted and time spent had to be reached but the number of valves counted was never less than 100 with the exception of cup no. 6 of year 2000–2001, and cup no. 14 of year 1999–2000 at the 47∘ S site which were not considered for relative abundance calculations due to their negligible diatom content. The resulting counts yielded estimates of specimens m-2 d-1 according to Sancetta and Calvert (1988) and Romero et al. (2009), as well as relative abundances of diatom taxa.

All diatom and silicoflagellate specimens were identified to the lowest taxonomic level possible. Radiolarians were only identified to group level. Scanning electron microscope imagery was used on selected samples to verify taxonomic identifications made with the light microscope. Taxonomy followed modern concepts in Hasle and Syvertsen (1997). The resting spores of members of the subgenus Hyalochaete of the genus Chaetoceros were identified only at group level due to a lack of morphological criteria. The differentiation between Pseudo-nitzschia lineola and Pseudo-nitzschia turgiduloides was often difficult due to their state of preservation in the samples; therefore, they were grouped under the category Pseudo-nitzschia cf.lineola in this study. A species or group of species of the genus Thalassiosira larger than 20 µm, highly dissolved and with radial to fasciculated areolation were grouped together under the name Thalassiosira sp. 1. Several small Thalassiosira species with similar morphological features were assembled together under Thalassiosira trifulta group following Shiono and Koizumi (2000). Due to the gradational nature of the morphology between Thalassiosira gracilis var. gracilis and T. gracilis var. expecta, both varieties were grouped together under the name T. gracilis group following the recommendations of Crosta et al. (2005).

In order to enable comparison with other sites, annual flux estimates are provided in Table 3. These were obtained by assuming that total mass flux outside of the sampling period was constant and by linearly interpolating values for the small gaps (i.e. 8.5–17 days intervals) during the productive season. No attempt was made to annualise the relative contribution of the diatom taxa, and therefore average values of the integrated diatom assemblage for whole sampling interval are provided in Table 3.

In order to investigate the covariability between the main diatom taxa along our sediment trap records, we conducted separate principal component analyses (PCA) for each site using of Statistica 7.0^® software. PCA analysis is a statistical technique that reduces the information brought by a high number of independent variables into a smaller set of dimensions (factors) with a minimum loss of information. Only species and taxonomic groups with relative contributions > 1 % for the entire sampling period were considered in the analysis, i.e. thirteen taxa from site 47∘ S and nine taxa from site 54∘ S. The relative contribution of these groups of species was recalculated for each sample and then a log transformation (log x+1) was applied in order to normalise the distribution of the data. Diatom groups were then determined using a Q-mode factor analysis of the samples with a maximised variance (VARIMAX) rotation.

The Shannon's diversity index (Shannon, 1949) was used to document latitudinal diversity trends across sites (Table 4).

Weekly SSTs for the decade 1997–2007 were derived from the IGOSS NMC (the Integrated Global Ocean Services System Products Bulletin, National Meteorological Center; Reynolds et al., 2002) database, each value is a weekly composite of data collected within the area 48.5–45.5∘ S × 130–150∘ E for the 47∘ S site and 55.5–52.5∘ S × 130–150∘ E for the 54∘ S site (Figs. 2a and 3a). Sea-viewing Wide Field-of-View Sensor (SeaWiFS) satellite-derived chlorophyll a and photosynthetically available radiation (PAR) estimates were obtained from NASA's Giovanni online data system (Acker and Leptoukh, 2007) for the same area used for the SST estimates (Figs. 2a and 3a).

Primary productivity values (mg C m-2 d-1) for all the sites were obtained from the Ocean Productivity website (www.science.oregonstate.edu/ocean.productivity/index.php), which provides estimates of net primary productivity derived from SeaWiFS satellite data by the standard vertically generalized production model (VGPM; Behrenfeld and Falkowski, 1997) and the carbon-based production model (CbPM; Behrenfeld et al., 2005).

The total mass and bulk component (biogenic silica, carbonate and POC) fluxes for both traps are shown in Figs. 2b and 3b and listed in Table 2. Annual total mass flux at ∼ 1 km depth was the lowest at station 47∘ S (14 ± 2 g m-2 yr-1; 2-year average ± standard deviation) and the highest at station 54∘ S (24 ± 13 g m-2 yr-1; 6-year average ± standard deviation; Table 3). BSi flux followed a similar latitudinal trend with lower fluxes at 47∘ S (1 ± 0 g m-2 yr-1) compared to 54∘ S (12 ± 9 g m-2 yr-1). Carbonate export exhibited less variability between sites, with values somewhat higher at 47∘ S (10 ± 3 g m-2 yr-1) than those measured at 54∘ S (7 ± 3 g m-2 yr-1, respectively). Interestingly, despite the strong latitudinal differences in the magnitude of the mass fluxes, POC export was very similar at both stations (1.0 ± 0.1 and 0.8 ± 0.4 g m-2 yr-1, for 47 and 54∘ S, respectively).

In terms of relative abundance, the biogenic silica fraction represented 57 % of the mass flux at the 54∘ S site, whereas its contribution dramatically dropped to 7 % at the 47∘ S station. Calcium carbonate and POC accounted for 70 and 7.3 % at the 47∘ S site, respectively, and 21 and 3 % at the 54∘ S station (Table 3). These differences were primarily driven by the northward decrease in the biogenic silica fluxes. The BSi : PIC mole ratios decreased northward mirroring the latitudinal variations of the particle composition, from 5.0 at station 54∘ S to 0.2 at station 47∘ S (Table 3; Fig. S1 in the Supplement). The POC : BSi followed an opposite pattern with 0.3 at 54 and 4.9 at 47∘ S.

The seasonality of the total mass flux at station 47∘ S during the 2-year record showed a period of enhanced particle export in spring and secondary peaks in summer and autumn (Fig. 2b). The highest fluxes were registered in November–December 2000 (92–176 mg m-2 d-1), March 2001 (105 mg m-2 d-1) and October 1999 (90 mg m-2 d-1). Total mass flux at the 54∘ S site was strongly seasonal with maximum values occurring during the late spring–summer and very low export prevailing through the autumn and winter months. The late spring–summer export maxima were as short as 3 months and often showed a bimodal distribution (e.g. 1997–1998, 1999–2000, 2002–2003; Fig. 3b). The highest total mass fluxes at this site were collected during December–January 1999 (511–724 mg m-2 d-1), January 2006 (418 mg m-2 d-1), February 2003 (397 mg m-2 d-1) and January 1998 (396 mg m-2 d-1).

The biogenic silica flux at the 47 and 54∘ S site was composed of diatoms, silicoflagellates, radiolarians and a handful of skeletons of the dinoflagellate Actiniscus pentasterias. Diatom fluxes were 1 order of magnitude higher than those of silicoflagellates and radiolarians at the 47∘ S site, and 1 and 3 orders of magnitude higher, respectively, at the 54∘ S site. Consistent with the biogenic silica flux, diatoms were most numerous in the 54∘ S site with an annual flux of 31 ± 5.5 × 108 valves m-2 yr-1 (6-year average ± standard deviation) compared to 0.5 ± 0.4 × 108 valves m-2 yr-1 (2-year average ± standard deviation) of the 47∘ S site.

Total diatom-valve flux at the 47∘ S site (Fig. 4a) showed a less pronounced seasonality than that observed at 54∘ S (Fig. 5a) and exhibited a weak correlation with the total mass (r= 0.37, n= 30) and BSi (r= 0.42, n= 29) fluxes. Diatoms occurred in the greatest numbers during November 2000 (1.6 × 106 valves m-2 d-1), February–March 2001 (0.4–0.8 × 106 valves m-2 d-1) and October 1999 (0.4 × 106 valves m-2 d-1).

At station 54∘ S, total diatom-valve flux was highly seasonal and followed a similar pattern to that of the total mass (r= 0.66, n= 108) and BSi fluxes (r= 0.68, n= 108). These correlations are high despite the biases associated with our diatom-valve counting technique which does not allow for quantification of small valve fragments. In particular the high diatom-valve fragmentation observed during the productive period of 1999–2000 reduced the correlations between diatom-valve flux and total mass and BSi fluxes. In fact, the latter correlations increased significantly after excluding the 1999–2000 data (r= 0.85, n= 88 and r= 0.87, n= 88, respectively).

The spring–summer diatom bloom often exhibited two peaks of enhanced export separated by a period of lower flux (e.g. 1997–1998, 1999–2000, 2002–2003; Figs. 5a and 7). During the productive period of 2006–2007, the diatom bloom exhibited one single peak during which the largest diatom fluxes of the record were registered (up to 100 × 106 valves m-2 d-1 in January 2007; Fig. 5a). Secondary diatom flux maxima were registered in January 1998 (71 × 106 valves m-2 d-1), December 2002 (65 × 106 valves m-2 d-1) and December 1999 (52 × 106 valves m-2 d-1). We noticed that during the 1999–2000, summer bloom the high BSi fluxes were not coupled with a proportional increase of the diatom valves (Fig. 5a). The higher degree of fragmentation observed on these samples could be attributed to either a more intense grazing pressure by the zooplankton community that year or by a higher fragmentation of the valves during the sample preparation due to the presence of abundant numbers of weakly silicified diatoms (e.g. species of the genus Pseudo-nitzschia) which are more prone to break during the sample processing (Rembauville et al., 2015).

In terms of diatom assemblage composition, the occurrence and fractional contributions of all the diatom taxa found at the 47 and 54∘ S study sites, as well as at 61∘ S (Rigual-Hernández et al., 2015), are provided in Table 4. The diatom sinking assemblage at station 47∘ S was more diversified (H′ for the entire sampling period = 2.48) than those found south the SAF (H′ = 1.86 at the 54∘ S; H′ = 1.04 at the 61∘ S) consisting of 79 species or groups of species. The most abundant species was Fragilariopsis kerguelensis, which represented 43 % of the integrated assemblage for the entire sampling period (Fig. 4). Subordinate contributions to the diatom assemblage were made by Azpeitia tabularis (10 %), Thalassiosira sp. 1 (4 %), Nitzschia bicapitata (4 %), resting spores of Chaetoceros spp. (subgenus Hyalochaetae; 3 %), Thalassiosira oestrupii var. oestrupii (3 %), Hemidiscus cuneiformis (3 %) and Roperia tesselata (3 %; Fig. 4). A total of 77 taxa were identified at the 54∘ S site (Table 4). F. kerguelensis was also the dominant species, contributing up to 59 % of the diatom assemblage for the whole sampling period (Fig. 5). Secondary contributors correspond to Pseudo-nitzschia cf. lineola (8 %), Pseudo-nitzschia heimii (5 %), Thalassiosira gracilis group (4 %), Fragilariopsis pseudonana (3 %), Fragilariopsis rhombica (2 %) and Thalassiosira lentiginosa (2 %; Fig. 5).

The PCA for the 47∘ S site identified four components containing 64 % of the total variance, whereas that of the 54∘ S site required three components to describe 79 % of the information of diatom data (Table 5). Figure 6 shows the position of the species on the first two PCA axes for the 47 and 54∘ S sites. Together with the species, we plotted total and major components mass fluxes.

The first component of the PCA for the 47∘ S site accounted for 19 % of the variance. The centric species A. tabularis and H. cuneiformis (Fig. 6) had a positive loading on factor 1 and exhibited their highest relative abundance during spring and summer (Fig. 4). Factor 2 explained 19 % of the variance and was dominated by F. kerguelensis, T. oestrupii var. oestrupii and Thalassiosira sp. 1. F. kerguelensis maintained a relatively constant contribution to the diatom assemblages during the whole sampling interval with a tendency to peak in late summer and autumn together with T. oestrupii var. oestrupii. None of the factors of the PCA of the 47∘ S site were significantly correlated with the biogenic particle fluxes (Fig. 6a and Table 6a).

At the 54∘ S site, the first component (48 % of the total variance) was highly correlated with the bulk components of the flux (Fig. 6b and Table 6b) and individualises two groups of diatom species. High-positive factor loadings characterise the bloom-forming Pseudo-nitzschia cf. lineola, F. rhombica, F. pseudonana and N. directa and the cool-open-ocean diatom T. gracilis group. The relative contribution of these species peaked during the productive season (Fig. 5) and showed a strong positive correlation with all the components of the flux (Fig. 6b and Table 6b). Therefore, diatom species characterized by a high-positive first factor loading can be defined as the “high-export group”. In contrast, a high-negative factor loading on the first PCA axis was attributed to F. kerguelensis, which peaked during winter and autumn, coinciding with very low particle fluxes. Pseudo-nitzschia heimii was the only species with a high positive factor loading on the second PCA axis (Fig. 6b and Table 5) and its relative abundance peaked mainly from mid-summer to autumn. With the exception of year 2002-2003, a consistent diatom species succession was consistently observed over the growth season at the 54∘ S site (Figs. 5b and 7). During those years with a double peak diatom sedimentation bloom, the first maximum (November to early December) was always dominated by F. kerguelensis and by other large and heavily silicified diatoms, such as T. lentiginosa. During the second peak (in late December to early February), the relative contribution of Pseudo-nitzchia cf. lineola and small Fragilariopsis species increased sharply, representing together up to 50 % of the diatom assemblage in January 2000 (Figs. 5b and 7). Even during year 2006–2007, when the diatom sedimentation bloom exhibited a single maximum, a similar succession can be discerned within the peak.

The contrasting latitudinal variations in the composition and magnitude of the particle fluxes along the 140∘ E transect reflect the physicochemical and biological characteristics of the different zonal systems sampled by the traps. Relatively low BSi and diatom export measured in the mesopelagic waters of the SAZ (Fig. 8a and Table 3) are consistent with the low-to-moderate diatom biomass accumulation in the surface layer of this region (Kopczynska et al., 2001; de Salas et al., 2011). Low silicic acid (Bowie et al., 2011a) and iron levels (Sedwick et al., 2008; Bowie et al., 2009; Mongin et al., 2011), together with light limitation, as a result of cloudiness (Bishop and Rossow, 1991) and deep summer mixed layers (70–100 m; Rintoul and Trull, 2001), are considered the main factors responsible for the reduced diatom production in the SAZ. Moreover, the low BSi : PIC mole ratios measured by the traps (< < 1; Table 3) illustrate the relatively low contribution of diatoms to the particle flux export to the ocean interior. Low diatom export fluxes and BSi : PIC mole ratios are characteristic of carbonate-dominated and low-productivity regimes (Honjo et al., 2008) and typical of much of the circumpolar SAZ (Honjo et al., 2000; Trull et al., 2001a).

The higher diatom-valve fluxes and BSi export at the 54∘ S site (Table 3; Fig. 8a) agrees well with previous studies of the PFZ surface waters south of Tasmania, which reported relatively large and heavily silicified diatoms as major contributors to the phytoplankton biomass (Kopczynska et al., 2001; de Salas et al., 2011). Higher levels of silicic acid (Smith Jr et al., 2000), colder summer surface waters and shallower mixed winter layers than those of the SAZ (Rintoul and Trull, 2001) are most likely the main factors responsible for the greater prevalence of diatoms in this region. As a result of the enhanced diatom production and the drop in the abundance of calcifying phytoplankton (Findlay and Giraudeau, 2000; Honjo et al., 2000), BSi : PIC mole ratios of the settling material at this site shift to > 1 (Table 3).

Further south, at station 61∘ S in the southern AZ, Rigual-Hernández et al. (2015) documented an annual diatom flux 1 order of magnitude greater than that measured at the 54∘ S site (243 × 108 valves m-2 d-1 at 2000 m; Table 3 and Fig. 8a). The corresponding BSi export was as large as 65 g m-2 y-1, a value very similar to that reported in the AZ south of New Zealand by Honjo et al. (2000; 57 g m-2, station MS-4; Fig. 1). These very high BSi fluxes are arguably the largest BSi exports ever measured in the world ocean (Honjo et al., 2008). Due to the upwelling of Circumpolar Deep Water (CDW) at the Antarctic Divergence, the surface waters of the southern AZ exhibit very high silicate concentrations (up to 70 mmol Si m-3; Pollard et al., 2006) which enhance diatom growth at the expense of other phytoplankton groups (Mengelt et al., 2001; Selph et al., 2001). These high diatom export values are consistent with the large accumulation of diatom remains in the surface sediments between the polar front (PF) and the winter sea ice edge that encircles Antarctica, the so-called diatom ooze belt (Burckle and Cirilli, 1987). This diatom ooze belt constitutes the single most important sink for silica in the world ocean (DeMaster, 1981; Ledford-Hoffman et al., 1986; Tréguer et al., 1995; Tréguer, 2014).

The species occurrence observed along the 140∘ E transect is consistent with previous reports on diatom assemblage composition in the surface waters (Kopczynska et al., 1986; Kopczynska et al., 2001; de Salas et al., 2011) and sediments (Armand et al., 2005; Crosta et al., 2005; Romero et al., 2005) of the Australian sector of the Southern Ocean, and provide evidence, once again, that the frontal systems represent natural physical boundaries for phytoplankton species distribution (Boyd, 2002).

Overall, the diatom assemblage registered at the 47∘ S site is typical of the SAZ and differs significantly from those found in the PFZ and AZ (Table 4). The SAZ represents a “buffer zone” between the subtropical gyres to the north and the polar waters to the south which results in a highly diverse diatom community as highlighted by the highest H′ (2.48; Table 4) of the study transect. The occurrence of the warm water taxa H. cuneiformis, Fragilariopsis doliolus, Nitzschia kolaczeckii and T. lineata (Romero et al., 2005; Venrick et al., 2008) is restricted to this station, and therefore, these species appear as good indicators for the southward migration of the warmer, saltier and nutrient-poor water masses of the SAZ into the ACC. Moreover, the stark increase in the abundance of the open-ocean diatoms A. tabularis, N. bicapitata, R. tesselata and Thalassiosira oestrupii north of the SAF suggest the preference of these species for warmer waters (Hasle and Syvertsen, 1997; Romero et al., 2005).

The sinking diatom assemblage registered at the 54∘ S site is characteristic of the ACC waters and largely defined by the dominance of F. kerguelensis. The relative abundance of F. kerguelensis at the PFZ (59 %) represents a transitional value between that of the AZ (80 %) and that of the SAZ (43 %). This strong latitudinal gradient mirrors its distribution in the surface sediments, which has been previously tied to summer SST (Crosta et al., 2005; Esper et al., 2010). However, other potentially important influences such as mixed layer depth, seasonality, and iron and silicate abundance also exhibit latitudinal gradients, and therefore may also influence the distribution of this species. Peak abundances of Pseudo-nitzschia species along the 140∘ E transect are observed in the PFZ (Table 4) and are consistent with previous studies that described this genus as a major contributor to the bulk phytoplankton biomass in the ACC waters (e.g. Kopczynska et al., 2001; Smetacek et al., 2002; de Salas et al., 2011). Moreover, it is worth noting that P-n. heimii, together with other large diatoms (e.g. Thalassiothrix and Proboscia), have been reported to be major contributors of a SCM consistently observed between 53 and 58∘ S along 140∘ E (Kopczynska et al., 2001; Parslow et al., 2001). Navicula directa also showed maximum abundances at the PFZ site with values ∼ 5 %. This species has been traditionally described as a benthic-dwelling species (Scott and Marchant, 2005 and references therein) with an affinity for sea ice conditions (Armand, 1997). However, its persistent presence throughout the 6-year record and similar seasonal flux pattern to that of other well-known open-ocean species of the ACC, such as Thalassiosira gracilis group (r= 0.8, n= 108; Fig. 5b), point to a pelagic distribution of this species. This concept agrees well with Kopczynska et al. (1986) and Waite and Nodder (2001), who documented Navicula populations of considerable abundance in areas remote from coastal and sea ice influence in the Australian sector.

Although in many aspects the composition of the diatom assemblage at the 61∘ S site was similar to that of station 54∘ S, there were some qualitative and quantitative differences. As a result of the southward increase in the relative abundance of F. kerguelensis, the diversity (H′) and the relative contribution of most of the secondary constituents of the diatom assemblage at 61∘ S exhibited lower values than at 54∘ S (Table 4). For example, Pseudo-nitzchia species that represented cumulatively 13 % of the integrated assemblage, dropped to < 1 % at the 61∘ S site. Navicula directa followed a similar pattern with maximum abundances at 54∘ S (5 %) and negligible fluxes at 61∘ S. It is possible, however, that other factors, such as selective grazing or ecological constraints, may also account for the lower contribution of these species in the AZ.

Taking into account that diatoms are, by far, the main contributors to the BSi production at the 54∘ S site, and that the BSi fraction, in turn, dominated the total mass flux, the strong correlation between diatom-valve and mass fluxes (r= 0.85; n= 88) suggests that the particle export at the PFZ is mainly mediated by diatoms. In contrast, at the 47∘ S site, the silica-poor content of the particles and the low correlation between diatom-valve and mass fluxes (r= 0.37; n= 30) indicates a minor role for diatoms in regulating the export in the SAZ. These results underscore the contrasting role that diatoms play in the controls on the flux north and south of the SAF (Trull et al., 2001a; Ebersbach et al., 2011).

The less defined seasonal pattern and lower amplitude of the diatom fluxes observed at the 47∘ S site (Fig. 4) are a reflection of the different algal community north of the SAF, dominated by non-siliceous phytoplankton (Odate and Fukuchi, 1995; Kopczynska et al., 2001; de Salas et al., 2011). For both years of our study, the highest annual diatom export events coincided with the onset of the biomass accumulation in the surface waters, indicating that diatoms responded rapidly to the enhanced light levels (Fig. 2a) and to the formation of a stable and shallow mixed layer (Rintoul and Trull, 2001). However, unlike the chlorophyll a concentration that gradually increased throughout the spring, diatom export rapidly returned to winter values most likely caused by the depletion of the winter silicate and/or iron stocks (Lannuzel et al., 2011). This seasonal pattern is characteristic of the SAZ and other silicate-poor environments, where diatoms typically bloom at the beginning of the successional sequence and then are replaced by other functional groups (Margalef, 1978; Balch, 2004; Alvain et al., 2008; Rigual-Hernández et al., 2013). The increase in the diatom and BSi fluxes from January to early March 2001 suggests the export of a second diatom bloom that year. South of Tasmania the SAZ exhibits a complex physical structure with frequent wind mixing events (Yuan, 2004) and fronts meandering and forming eddies that can reach the trap location (Rintoul and Trull, 2001; Herraiz-Borreguero and Rintoul, 2011). Thus, it is likely that one of these mechanisms injected nutrients into the surface layer of the 47∘ S site fuelling diatom production and allowing the “reset” of phytoplankton succession. In terms of population dynamics, the seasonal succession of species at the 47∘ S site was not as clearly expressed as in station 54∘ S and none of the diatom species seem to play an important role in the export controls of any of the components of the flux as indicated by the results of the PCA (Fig. 6a). F. kerguelensis exhibited fairly constant relative abundances throughout the record suggesting little competition for resources with other diatom species. The temperate-to-warm water species H. cuneiformis and A. tabularis showed their maximum contribution at times of maximum diatom export which suggests that these species are the first to respond to nutrient supply in the surface waters in this region.

The annual export maxima of total mass and diatom-valve flux at 54∘ S were separated into two peaks for most of the years (Figs. 5a and 7). A similar double peak feature of the particle bloom has been previously reported in the AZ of the Pacific (Honjo et al., 2000; Grigorov et al., 2014) and Atlantic (Fischer et al., 2002). Honjo (2004) speculated that such a double peak structure may be due to a break in primary production caused by a temporary depletion of a limiting nutrient, while Grigorov et al. (2014) attributed the drop in the diatom flux between two periods of enhanced export to a storm event that mixed the diatom biomass out of the surface layer. The lack of accompanying in situ measurements of nutrient concentration and mixed layer depth precludes the direct assessment of these possibilities.

The initial diatom population size, species-specific physiological traits and selective grazing pressure are crucial factors determining which diatom species dominates or co-dominate an individual bloom (Assmy et al., 2007; Assmy et al., 2013; Boyd, 2013). The chain-forming F. kerguelensis is one of the most abundant diatom species in ACC waters (e.g. Laubscher et al., 1993; Bathmann et al., 1997; Smetacek et al., 2002) and has been reported to represent up to 90 % of the summer diatom populations in the AZ (Gall et al., 2001). The high relative contribution of F. kerguelensis throughout our record is consistent with these latter studies and suggests the presence of a large seeding population of this species before the onset of the bloom. These large initial seed stocks, together with the effective mechanical protection of its robust frustule (Hamm et al., 2003) against the heavy copepod grazing pressure of the ACC (Pollard et al., 2002; McLeod et al., 2010) are most likely the main factors determining the dominance of F. kerguelensis during the growth season. The increase in the relative abundance of the lightly silicified Pseudo-nitzschia cf. lineola and small Fragilariopsis species during the second part of the bloom (Figs. 5b and 7) is consistent with the observations of Kopczynska et al. (2001) who reported F. pseudonana and P-n. lineola dominating the diatom assemblages in the PFZ waters south of Tasmania during late summer. Assmy et al. (2007) reported large numbers of P-n. lineola during the last stages of the fertilisation experiment EisenEx, indicating the capacity of this diatom to outcompete other taxa under iron-limiting conditions. Moreover, small Fragilariopsis and Pseudo-nitzschia species are known to produce an iron-storage protein (ferritin) that allow them to undergo more cell divisions than other open-ocean diatoms under low iron concentrations (Marchetti et al., 2009). We speculate, that due to these particular physiological traits Pseudo-nitzschia and small Fragilariopsis species may gain a competitive advantage under the environmental conditions during the last stages of the diatom bloom (i.e. low silica and iron concentrations, and enhanced PAR), enabling such species to escape grazing and/or outcompete other diatoms. However, this scenario does not account for our observations in 2002–2003, when Pseudo-nitzschia and small Fragilariopsis species exhibited higher relative contribution in the first seasonal export peak (Figs. 5b and 7). This exceptional seasonal flux peak remained unexplained and likely due to other environmental conditions not captured by our study.

The short and vigorous summer particle export, consistently observed during our 6 year record at the 54∘ S trap is characteristic of high latitude systems (e.g. Honjo et al., 2000; Fischer et al., 2002; Pilskaln et al., 2004) and can contribute up to 66 % of the annual POC export to 800 m in just 2 months (e.g. 1999–2000). Therefore, these large summer pulses of POC are responsible for a major proportion of the variability in carbon sequestration from the atmosphere in the PFZ. The mechanism is primarily through the increase in the overall flux, because the fractional POC content was not observed to increase during high flux periods. For example, % POC for the year 1999–2000 ranged between 1.2 % and 3.7, and maximum relative abundances occurred at times of relatively low fluxes (Fig. 3b).

The strong positive correlation between factor 1 and POC fluxes at the 54∘ S site (Table 6) indicates an intimate association between high relative abundances of the high-export group species and pulses of POC export. As a specific example of this, during January 1999 and December 2000, when the highest contribution of the high-export group was noted (55–60 % of the total diatom flux; Fig 7), the PFZ sediment trap registered the largest POC fluxes of the record (up to 23 mg m-2 d-1; Fig. 7). Interestingly, these observations of elevated POC flux coincide with significantly lower summer SSTs than other years.

All the members of the high-export group have been previously reported as important components of both natural and iron-fertilised blooms in the Southern Ocean (Bathmann et al., 1997; Waite and Nodder, 2001; Smetacek et al., 2002; Assmy et al., 2007; Quéguiner, 2013; Grigorov et al., 2014; Rigual-Hernández et al., 2015). The increase in the relative abundance and fluxes of these species during the growth season indicates that they respond opportunistically to the enhanced light levels, most likely undergoing cycles of rapid biomass build-up followed by mass mortality and sinking in the form of aggregates (Smetacek et al., 2004; Green and Sambrotto, 2006; Assmy et al., 2013). This concept is supported by the recent findings of Closset et al. (2015) who documented an increase in the particle sinking speeds at the 54∘ S site during the summer 1999–2000 of up to at least 35 m d-1, a value that falls within the range of previous estimates for marine snow sinking rates (Turner, 2002; Trull et al., 2008; Laurenceau-Cornec et al., 2015). Moreover, other regionally relevant PFZ studies (Ebersbach et al., 2011; Grigorov et al., 2014) concluded that aggregates are the principal form of particle export during the growth season. Taken together, our data and these studies strongly suggest that aggregate formation is a widespread mechanism of the summer bloom in the open-ocean waters of the ACC.

We speculate that the massive development of high-export group diatoms during the growth season facilitates the formation of aggregates in the upper water column, which results in an increase in sinking rates and POC fluxes. Aggregates, and particularly diatom flocs, are rich in exopolymers that increase their effectiveness at scavenging particles they have collided with (Alldredge and McGillivary, 1991; Passow and De La Rocha, 2006). Therefore, it is possible that the formation of aggregates during the diatom bloom facilitated the scavenging of other particles (including phytoplankton chains and cells, biominerals and detritus), leading to the co-sedimentation of the major components of the flux (i.e. calcium carbonate, silica and organic carbon). This scavenging mechanism is consistent with previous laboratory observations made by Passow and De La Rocha (2006) and can explain the increase of the sinking rates during the growth season as well as the positive correlation between factor 1 and all bulk components of the flux (Fig. 6b, Table 6b).

Since most of the members of the high-export group are of relatively small in size and weakly silicified, it is unlikely that these species accounted for the major fraction of the BSi export during the summer bloom for most of the years. In contrast, the thick-shelled F. kerguelensis is a more compelling candidate to be responsible for the bulk of the BSi export, because despite the fact that its relative abundance exhibited the lowest values of the record during summer, its valve fluxes always were the highest during this season. In terms of carbonate export, the correlation between factor 1 and carbonate flux is not as strong as with the rest of the components of the flux but still high (Fig. 6b, Table 6b), indicating that highest relative contribution of the high-export group diatoms is also associated with high carbonate export. Although speculative, it is possible that the formation of aggregates during the diatom bloom also facilitated the scavenging of at least the fine fraction of the carbonate (mainly coccoliths; Ziveri et al., 2007; Iversen and Ploug, 2010) which would have led to the co-sedimentation of the BSi, POC and carbonate fractions. However, other seasonal ecological influences are also likely to be involved, given that the contribution of larger carbonate particles in the form of foraminifera tests is also increased in summer (King and Howard, 2003).

The massive sedimentation of giant diatoms characteristic of the SCM shade flora (e.g. Thalassiothrix and some rhizosolenids) in autumn and winter (the so-called fall dump) has been hypothesised to contribute to a substantial fraction of the annual carbon export in the PFZ (Kemp et al., 2006; Kemp and Villareal, 2013; Quéguiner, 2013). At the 54∘ S site, the highest fluxes and relative contribution of the deep dwellers Thalassiothrix antarctica and Proboscia were recorded between the end of the productive period and winter (Fig.  5); however their contribution to the total diatom assemblage was always low (< 3.6 %) and their flux pulses were not coupled with significant increases in POC export (Fig. 7). Thus, our data do not provide evidence that the sedimentation of these species was associated with a “fall dump” in the PFZ south of Tasmania. However, it has been suggested that sediment traps do not act as good samplers of large mats formed by these long diatoms (Kemp et al., 2006), and therefore their mass sedimentation during autumn and/or winter could have been missed by our sampling technique.

The depth at which the organic carbon is re-mineralised to CO2 by zooplankton and bacteria determines the timescales during which carbon is sequestered from the atmosphere (Yamanaka and Tajika, 1996; Smetacek et al., 2012). In the SAZ and the PFZ, the fraction of organic carbon recycled within the winter mixed layer (> 400 m in the SAZ and between 150 and 200 m in the PFZ; Rintoul and Bullister, 1999; Rintoul and Trull, 2001) would re-equilibrate with the atmosphere within months, whereas only the comparatively smaller portion that reaches deeper layers will remain in the ocean interior for centuries or longer timescales (Trull et al., 2001a). Thus, from the perspective of carbon sequestration, the POC fluxes measured by the traps reported in this study are probably of greater importance than those re-mineralised at mid-depths.

Despite the fact that total mass fluxes in the PFZ at ∼ 1 km were twofold larger than those of the SAZ, the annual POC export was almost identical in both regions (Fig. 8a), implying that particles sinking out of the mixed layer in the SAZ were relatively POC rich (Trull et al., 2001a; Ebersbach et al., 2011). Taking into account that gross primary production is similar in the two zones, or perhaps somewhat lower in the PFZ (Fig. 8b; Lourey and Trull, 2001; Cavagna et al., 2011; Westwood et al., 2011), our results challenge the notion that for a given similar level of production, diatom-dominated ecosystems export greater amounts of carbon to the deep ocean than ecosystems dominated by smaller, non-siliceous phytoplankton (Buesseler, 1998; Boyd and Newton, 1999; Laws et al., 2000). Trull et al. (2001) hypothesised that the similar POC export at both sites could be due to either (1) a more efficient repackaging of carbon for deep transport by the zooplankton community in the SAZ than in the PFZ or (2) to the fact that the silicate-rich particles exported in the PFZ may experience stronger loses of organic carbon at mesopelagic depths than do the carbonate-rich particles of the SAZ. Results from of the SAZ-sense programme (Bowie et al., 2011b) taken together with the data presented in this study provide key information to assess these hypotheses.

Analysis of the flux size spectra at the 47 and 54∘ S sites by Ebersbach et al. (2011) during January and February 2007 revealed that the vertical export at both stations was dominated by heavily processed particles, mainly faecal aggregates with a slight shift towards smaller particles within the PFZ due to abundant chains of diatoms sinking individually or as part of unconsolidated aggregates. Although the latter study was limited to a short observational period, the results of Ebersbach et al. (2011) suggest that zooplankton grazing had a similar impact on the control of particle export at both sites, and therefore the first hypothesis seems unlikely.

On the other hand, our data show that only a few diatom species, particularly F. kerguelensis, dominate the particle export in the silicate-rich and iron-limited waters of the PFZ and AZ. Most of these species are known to significantly increase their BSi : PON and BSi : POC ratios under iron deficiency resulting in the thickening of its already robust frustule (Takeda, 1998; Hoffmann et al., 2007). Furthermore, recent findings from the European Iron Fertilisation Experiment (EIFEX; Smetacek et al., 2012) illustrated that the cellular content of a large fraction of the F. kerguelensis stock outside and inside the patch was recycled in the surface layer, resulting in the disproportional sinking of empty frustules to the deep ocean (Assmy et al., 2013). Assmy et al. (2013) concluded that due to these particular traits, F. kerguelensis and other exceptionally robust diatoms, such as Thalassiosira lentiginosa and Thalassionema nitzschioides, preferentially sequester silicon relative to carbon in the iron-limited waters of the ACC. This concept is consistent with our findings in the open waters of the Australian sector south of the SAF, and would help to explain the low POC content and POC : BSi ratios of the particles registered at meso- and bathypelagic depths by our PFZ and AZ traps.

Significantly, comparisons of our results (Fig. 8a) with satellite and in situ measurements of primary production (Fig. 8b) suggest that high BSi sedimentation rates should be interpreted as a proxy for iron-limited diatom assemblages (Hutchins and Bruland, 1998; Takeda, 1998; Assmy et al., 2013) rather than for high primary production. This conclusion raises corresponding caution to previous studies that suggest that higher BSi fluxes in the past refer to a stronger biological carbon pump (Anderson et al., 2009; Sigman et al., 2010).