The Fram Strait is positioned between Greenland and Svalbard and is the only deepwater connection between the Nordic seas and the central Arctic Ocean, with a sill depth of approximately 2,600 m (Soltwedel et al., 2016). There are two main current systems in the Fram Strait (Figure 1). The WSC heads north along the shelf slope on the eastern part of the strait and transports warm, nutrient-rich Atlantic waters to the central Arctic Ocean (Beszczynska-Möller et al., 2012). On the western side of the strait, the East Greenland current transports cold, surface polar waters and ice southward (Smedsrud et al., 2017). It has been estimated that 90% of sea ice exported from the Arctic Ocean to the subarctic North Atlantic occurs through the Fram Strait (Serreze et al., 2006; Haine et al., 2015) such that the regional sea ice characteristics are representative of the Arctic basin as a whole (Krumpen et al., 2016; Belter et al., 2021). HAUSGARTEN is a Long-Term Ecological Research (LTER) in the Fram Strait established by the Alfred Wegner Institute, Helmholtz-Centre for Polar and Marine Research (AWI) (Soltwedel et al., 2016). As part of the LTER, AWI has maintained sediment trap moorings at some stations. In the present study, we report data from long-term 200 and 2,300 m sediment trap data from two stations (HGIV and HGN) subjected to regional variations in ice concentration (Wekerle et al., 2018).

Sinking particles were collected with modified cone-shaped automatic Kiel sediment traps (K/MT, sampling area ∼0.5 m²), each equipped with 20 collecting cups. All moorings were deployed and recovered on board RV Polarstern during summer expeditions to the Fram Strait. Sediment traps were deployed at two mooring locations positioned at approximately 79°N, 4°E (HGIV) and 79.5°N, 5°E (HGN). A total of 29 annual sediment trap deployments were collected during the study period: twelve at HGIV-200 m, nine at HGIV-2300 m, four at HGN-200 m, and four at HGN-2300 m (Supplementary Table S1). The traps were programmed to collect at high temporal resolution (7–17 days) during the productive summer season (May to September) and in longer intervals (up to 2 months) during winter. The sample collection cups were filled with filtered sterile North Sea water that was adjusted to a salinity of 40 psu through the addition of NaCl. The hypersaline seawater medium was poisoned with HgCl₂ to achieve a final concentration of 0.14% (Bauerfeind et al., 2009).

Temperature and salinity sensors from Sea-bird Electronics Inc. (SBE37 and SBE16) were deployed on moorings at HGIV and HGN. The temperature measurements used in the present study are taken from instrumentation deployed at 250 m on the HGIV mooring, the level of the Atlantic Water layer flowing into the Fram Strait (Beszczynska-Möller et al., 2012).

After the recovery of the moorings, the samples were stored in a refrigerator until further processing in the laboratory. The samples were split using a wet splitting procedure (von Bodungen et al., 1991). Before splitting of samples, swimmers >0.5 mm were manually removed under a dissecting microscope at a magnification of 20–50. Pteropods were removed from the samples and the total number of individuals enumerated, differentiating between sinkers and swimmers. Organisms with no clearly visible organic content were regarded as sinkers, whereas others with intact wings and bodies were classed as swimmers. For the calculation of pteropod aragonite, shells were rinsed to remove salt and mercury chloride and weighed following precombustion at 500°C for 24 h to remove the residual organic material. A weighted mean was calculated for the shells and used to calculate the CaCO₃ content of those pteropods classified as sinkers. Full details are provided in the study of Bauerfeind et al. (2014).

Subsamples of the remaining material were filtered for the subsequent analyses of bulk components (von Bodungen et al., 1991). The subsamples for total particulate matter (TPM) measurements were filtered onto pre-weighed GF/F filters (nominal pore size: 0.7 μm; 25 mm ø), rinsed with distilled water to remove salt, dried at 60°C, and weighed on a microbalance. The subsamples for CaCO₃ measurements were filtered onto additional pre-weighed GF/F filters (nominal pore size: 0.7 μm; 25 mm ø) and treated like TMF filters, after which the filters were soaked in 0.1N HCl to remove inorganic CaCO₃, rinsed and dried at 60°C, and weighed again on a microbalance, with the carbonate fraction representing the difference between the TPM and dissolved CaCO₃. The subsamples for POC measurements were filtered onto GF/F filters (nominal pore size: 0.7 μm; 25 mm ø, pre-combusted at 500°C for 4 h), then soaked in 0.1N HCl for the removal of inorganic carbon, and dried at 60°C. POC measurements were conducted on a Carlo Erba CHN elemental analyzer and should be considered as minimum values as they were not corrected for the dissolution of organic material in the collecting cups. The subsamples for biogenic particulate silica measurements were filtered on cellulose acetate filters (pore size: 0.8 μm), processed using the wet-alkaline method (pre-treated 12 h at 85°C in an oven), and extracted for 2 h at 85°C in a shaking water bath. The flux of biogenic matter representing TPM was calculated according to the following equation: F b i o = 1.8   x   F P O C + F C a C O 3 + F O p a l , (1) where F_bio, F_POC, F_CaCO3, and F_Opal indicate the mass fluxes of biogenic matter, POC, CaCO₃, and Opal, respectively. The conversion of POC to POM requires a scaling factor, which ranges from 1.8–2.3 in the literature (Boyd and Trull, 2007). In the present study, we used a lower estimate of 1.8 and it should thus be considered as a conservative estimate of POM. The calculations of residual flux and its approximation to lithogenic flux are similar to those employed and validated in previous studies (Jickells et al., 1998; Salter et al., 2010). Previous studies comparing the minimum (1.8) and maximum (2.1) scaling factors for POC to POM occasionally yield negative residual fluxes (Salter et al., 2010), and therefore, we selected the minimum value. The factor of 2.1 for pbSi is a conversion that accounts for the atomic mass of elemental Si and SiO₂.

The subsamples for biomarker analyses of brassicasterol (24-methylcholesta-5, 22E-dien-3β-ol), campesterol (24-methylcholest-5-en-3β-ol), β-sitosterol (24-ethylcholest-5-en-3β-ol), and IP₂₅ were extracted with dichloromethane/methanol (1:1, v/v) and dichloromethane using separating funnels. For the quantification of lipid compounds, the internal standards, 7-hexylnonadecane, C₃₆ n-alkane, and androstanol (5α-androstan-3β-ol, 20 μL/sample), were added prior to further analytical steps. The different compounds (IP₂₅ and sterols) were separated via open column chromatography (SiO₂) using n-hexane (for the hydrocarbons) and ethyl acetate/n-hexane (20:80 v/v for sterols) as an eluent.

The individual sterols were silylated with 500 μL BSTFA (bis-trimethylsilyl-trifluoroacet-amide) at 60°C for 2 h. After extraction with hexane, analyses were carried out by gas chromatography–mass spectrometry (GC–MS) using an Agilent 6850 GC (30 m HP-5 MS column, 0.25 mm inner diameter, and 0.25 μm film thickness) coupled to an Agilent 5975 C VL mass selective detector. The GC oven was heated from 60°C to 150°C (rate: 15°C/min) and at 3°C/min to 320°C (20 min isothermal) for sterols and at 10°C/min to 320°C (15 min isothermal) for IP₂₅. Helium was used as a carrier gas. Individual compound identification was based on comparisons of their retention times with that of reference compounds and on comparisons of their mass spectra with the published data (Boon et al., 1979; Volkman, 1986; Belt et al., 2007). IP₂₅ was quantified using its molecular ion m/z 350 in relation to the abundant fragment ion m/z 266 of 7-hexylnonadecane and by means of an external calibration curve (R² = 0.9989) to balance the different responses of the used ions (for further details, see the work of Fahl and Stein (2012)). Brassicasterol, campesterol, and ß-sitosterol were quantified as trimethylsilyl ethers using the molecular ions m/z 470, m/z 472, and m/z 486, respectively, in relation to the molecular ion m/z 348 of androstanol.

Subsamples for measurements of ∂¹³C were filtered on pre-combusted GF/F filters. The filters were then treated with HCL fume (24 h) to remove inorganic carbonate and were rinsed with distilled water after the treatment, dried at 60°C, and stored in a desiccator until measurement. The stable carbon isotope ratio in organic matter on the filters was determined with an isotope ratio mass spectrometer (Delta Plus Thermo-Finnigan) coupled to an elemental analyzer (NC 2500 Carlo Erba) at the Humboldt Innovation—GmbH, Berlin, and elemental contents (weight % C and N) were determined on the same sample. Isotope ratios are expressed in the δ notation with Vienna Pee Dee Belemnite (VPDB) as a reference standard for carbon. Stable isotope values are expressed in ‰ deviation relative to the standard, according to the conventional formula δ¹³C = [R_sample/R_standard − 1] × 10³, where R is the ratio ¹³C/¹²C.

In this study, we make use of the catchment area of sediment traps HGIV and HGN deployed in 200 m and 2,300 m depths that was determined by Wekerle et al. (2018). The catchment area of the four traps was estimated with a Lagrangian particle tracking algorithm. The particles were released once per day during the time period 2002–2009 at the depth of the trap, advected with a constant sinking velocity of 60 m d^-1 and the daily mean horizontal velocity field of an ocean sea ice model, and tracked until they reached the sea surface. A sinking velocity of 60 m d^-1 was selected based on the measured settling rates of marine aggregates collected in the Fram Strait (Wekerle et al., 2018). The velocity field used for back tracking was taken from a FESOM1.4 setup configured for the Fram Strait, with 1 km mesh resolution in the region 20°W-20°E/76°N-82.5°N (Wekerle et al., 2017). Particle positions at the sea surface were binned into a spatial grid and then divided by the total number of particles to determine the fraction of collected particles originating from each grid box. This resulted in a climatological two-dimensional probability distribution for the particle origin.

We calculated a time series of environmental parameters representative for the catchment areas, such as sea ice and chlorophyll-a concentrations, by weighting the concentrations with the probability distribution for the particle origin. The monthly mean sea ice concentration provided by CERSAT has been used, which is available for years 2000–2013. It is based on 85 GHz SSM/I brightness temperatures, applying the ARTIST Sea Ice (ASI) algorithm, and is available on a 12.5 × 12.5 km grid. Chlorophyll-a concentration was computed from the ESA Ocean Color CCI Remote Sensing Reflectance (merged, bias-corrected) data. It has a spatial resolution of 1 km² and also covers years 2000–2013. More details can be found in the study of Wekerle et al. (2018). The temperature of the AW layer flowing into the Fram Strait was taken from mooring instrumentation deployed at 250 m in the core of the AW layer.

We computed the stratification regimes present in the four catchment areas based on daily mean salinity fields and mixed layer depth (MLD; defined as the shallowest depth where the deviation from surface density exceeds 0.03 kg m^-3) from the ocean–sea ice model FESOM1.4. Here, an update of the model described by Wekerle et al. (2017) has been used, with the atmospheric force replaced by the ERA-interim dataset and the simulation now covering the time period 1990–2018. As in the study by von Appen et al. (2021), we distinguished between three cases: the unstratified regime (US) is defined by MLD>50 m, the mixed layer regime (ML) is defined by MLD< 50 m and a salinity difference between surface and 100 m (DS) below 1, and the meltwater regime (MW) is defined by MLD<50 m and DS >1. Each of the three criteria was computed every day during the time period 2000–2014. As described for sea ice concentration and chlorophyll-a, the different stratification regimes were weighted with the probability distribution for particle origin to obtain an estimate for the catchment areas.

Residual fluxes were calculated as the difference between TPM flux and the analytically determined biogenic fluxes according to the following equation: F L i t h = F R e s = F T P M − F P O M + F O p a l + F C a C O 3 , (2) where F_Lith is the assumed flux of the lithogenic matter, F_Res is the residual mass flux, F_TPM is the total particulate mass flux, F_POM (F_POC * 1.8) is the particulate organic matter flux, F_Opal (pbSi * 2.1) is the flux of SiO₂, and F_CaCO3 is the calcium carbonate flux. Equation 1 shows the explanation of the scaling factors.

Sediment trap data were wrangled to calculate the flux values for each individual month by taking into consideration the variable opening periods between different deployments and seasons. In those instances, when multiple opening periods occurred within the same month, the daily flux values were weighted based on the proportional coverage comprising each sampling cup. F m o n t h = F x × I x + F y + I y n , (3) where F_month is the daily flux over a given month (mmol m⁻² d⁻¹), F _x is the daily flux (mmol m⁻² d⁻¹), and I _x is the sampling interval (days) of sediment trap sampling cup x. F _y and I _y correspond to sampling cup y within the same month, and n (= I _x + I _y ) is the number of days in a month. Annual fluxes were calculated by integrating the daily flux values from each month. In the case that some days were missing from a given month, data were extrapolated based on the daily flux values for the given month. Deployment coverage periods are documented in Supplementary Table S1.

Flux stability indices (FSIs) were calculated by ranking each month within a sampling year in descending order and assessing the accumulation of monthly flux as a function of year day. FSI was determined as the year day when 50% of the cumulative annual flux occurred (Lampitt et al., 2001).

Modeled particle trajectories (sinking velocity = 60 m d^-1) arriving in sediment traps at HGIV and HGN predominantly originated from southeast of the moorings (Figure 2). At both mooring positions, the catchment area was larger for 2,300 m deployments than 200 m deployments. Average particle trajectory lengths were 38 km and 45 km at HGIV-200 m and HGN-200 m compared to 195 km and 193 km at HGIV-2300 m and HGN-2300 m, respectively.

The sediment trap datasets were used to construct two annual records of bathypelagic particle flux, corresponding to those obtained during high-temperature/low-ice (HGIV-2005) and low-temperature/high-ice (HGN-2007) (Figures 11A,D). Comparison of these two end-members in fluxes and catchment area conditions allowed us to explore the impact of varying ice and temperature on particle flux parameters to the deep ocean. HGN-2007 catchment areas were characterized by higher ice cover (Figure 11A) and a greater proportion of meltwater regimes (Figure 11G), and HGIV-2005 catchment areas were characterized by higher temperatures (Figure 11D). Seasonal progression and chlorophyll concentrations were similar in both end-member situations (Figure 11B). Annually integrated fluxes of POC, PIC, and pbSi were enhanced 1.5, 4, and 4 times, respectively, in the high-ice HGN-2007 end-member (Table 4). Monthly flux values showed that these biogenic flux components in HGN-2007 were notably elevated from February to May (Figures 11E,H,K). The Si:POC ratios were consistently higher in the high-ice catchment area. In contrast, the POC:PIC ratios were notably lower under enhanced ice cover (Figure 11I), resulting in an annual ratio of 2 at HGN-2007, compared to 5 at HGIV-2005 (Table 4).

The primary objective of the present study was to compare mesopelagic and bathypelagic fluxes in two locations characterized by different environmental conditions in sediment trap catchment areas. Satellite-derived ice coverage shows that both mooring positions are located in a marginal ice zone (MIZ). Despite considerable interannual variability over the observation period, monthly and annual ice cover was consistently higher at HGN than HGIV (Figure 3; Supplementary Figure S1), although only significantly different in mesopelagic catchment areas (Table 3). The more northerly position and, thus, northward-displaced catchment areas at HGN (Wekerle et al., 2018) resulted in a higher proportion of particles originating in surface waters intersecting the pathway of southward (transpolar drift) ice export from the Siberian shelves (Krumpen et al., 2019). Variable ice export and subsequent melting in the Fram Strait influence the prevailing stratification regimes (Figure 4). Building on previous work (von Appen et al., 2021), we employed an eddy-resolving sea ice ocean model (Wekerle et al., 2017) to estimate the proportion of sediment trap catchment areas that were represented by MW and ML stratification regimes. Consistent with higher ice coverage, HGN was characterized by a higher proportion of MW regimes, with significant interannual variability corresponding to changes in sea ice export through the Fram Strait (Min et al., 2019). Fundamental differences in the ice properties of sediment trap catchment areas were used as a basis to examine the impact of variable ice cover, temperature, and MW regimes on mesopelagic and bathypelagic flux characteristics.

An increase in phytoplankton biomass has been reported in the Fram Strait over the observation period (2000–2013). Increases in satellite-derived chlorophyll concentrations (Chl) at HGIV were statistically significant in both mesopelagic and bathypelagic catchment areas (Tables 1, 2), and long-term trends are significantly correlated at HGN and HGIV (Supplementary Figure S1). Previous studies in the Fram Strait have reported increasing Chl concentrations from both in situ (Nöthig et al., 2015) and remote sensing data (Cherkasheva et al., 2014; Wekerle et al., 2018). The warming of AW entering the Fram Strait (Beszczynska-Möller et al., 2012) may explain increasing Chl concentrations (Nöthig et al., 2015), potentially mediated through enhanced sea ice melt and stratification (Cherkasheva et al., 2014). Validation of this trend in sediment trap catchment areas from different depths provides an opportunity to examine the resulting impact of this increasing productivity on POC fluxes. Our data show that annual POC fluxes have not increased in line with Chl concentrations at either mesopelagic or bathypelagic depths. In contrast, there was some indication of decreasing values of POC flux, although these are not statistically significant (Tables 1, 2).

Our findings indicate that despite the general pattern of increasing productivity in the Fram Strait, there is a corresponding decrease in the apparent efficiency of the biological carbon pump. Logically, this must be related to changes in plankton dynamics regulating the export of organic carbon out of the surface layer. At mesopelagic depths, a decrease in pbSi and an increase in PIC fluxes were observed, the latter mediated primarily through an increase in pteropod aragonite (Table 1). From 1998 to 2011, summertime surface plankton assemblages have shifted from a dominance of diatoms Thalassiosira, Chaetoceros, and Fragilariopsis spp. to a dominance of Phaeocystis pouchetti, pico-, and nanoflagellate communities (Nöthig et al., 2015) following a well-documented warming anomaly (Beszczynska-Möller et al., 2012). Similarly, the analysis of zooplankton in sediment trap samples indicates an increase in the abundance of North Atlantic amphipod species Themisto compressa (Kraft et al., 2013; Schröter et al., 2019) and an increasing occurrence of the pteropod Limacina retroversa (Bauerfeind et al., 2015; Busch et al., 2015) and general shifts in zooplankton community composition (Ramondenc et al., 2022). Changes in phytoplankton communities, as a result of warming, have been previously linked to decreasing pbSi fluxes in the mesopelagic at HGIV (Bauerfeind et al., 2009; Lalande et al., 2013), and our data confirm that changes in pbSi are observed further north in the Fram Strait at HGN. Here, we document that decreases in Si:POC and Si:PIC ratios were statistically significant at mesopelagic depths (Table 1). These stoichiometric shifts suggest a decrease in the contribution of diatoms to POC flux, and a decrease relative to calcifying organisms may partly explain a weakened biological carbon pump in spite of increasing surface productivity.

Contrary to mesopelagic depths, decreases in pbSi fluxes were not observed at bathypelagic depths since there was no significant change in pbSi fluxes, Si:POC, or Si:PIC ratios over the observation period (Table 2). The much lower concentration of diatom biomarkers (Figure 10) and elevated Si:POC ratios in the bathypelagic in comparison to the mesopelagic catchment areas imply significant attenuation of diatom-derived organic carbon with depth. However, concentrations of diatom-specific biomarkers in the bathypelagic are 3% of those in the mesopelagic depths, compared to 0.3% for dinoflagellates, indicating diatom-derived carbon is more effectively transferred to the deep ocean than that of dinoflagellates. Higher pbSi fluxes are observed at bathypelagic than mesopelagic depths, thus probably mostly representing aged siliceous material derived from lateral advection of shelf sediments and released from ice melt. Unfortunately, these particularities of particle dynamics in the Fram Strait render it difficult to correctly interpret how declining mesopelagic fluxes of siliceous communities impact the associated delivery of diatom-derived organic carbon to benthic communities.

The strong increase in pteropod aragonite to CaCO₃ fluxes over the observation period at mesopelagic depths was not evident at bathypelagic depths (Table 2). Sediment trap data (Bauerfeind et al., 2014; Boeuf et al., 2019; Dong et al., 2019), sediment analysis (Sulpis et al., 2022), and modelling studies (Buitenhuis et al., 2019) show that pteropod aragonite can reach and impact bathypelagic depths. Our data in the Fram Strait show that annually integrated pteropod aragonite only accounted for 10% of CaCO₃ fluxes at 2,300 m compared to approximately 50% at 200 m. The strong dissolution of aragonite with depth and associated breakdown of organic matter (Oakes et al., 2019) implies a minor contribution to bathypelagic POC flux on annual scales. However, both mesopelagic and bathypelagic results show high seasonal pteropod aragonite fluxes during winter (Figure 8) that may represent a source of POC to the benthos during a time of otherwise low flux (Bauerfeind et al., 2014). Enhanced fluxes of calcifying zooplankton serve to strengthen the carbonate counter-pump (parameterized through the POC:PIC ratio) and reduce the overall efficiency of ocean carbon sequestration (Salter et al., 2014; Manno et al., 2018). The inclusion of pteropod aragonite at mesopelagic depths reduces the POC:PIC molar ratio by 3–6, but only 0.3 at bathypelagic depths (Table 3). The increasing trend of pteropod fluxes observed in the Fram Strait, thus, has a negligible effect on the strength of the carbonate counter-pump in the deep ocean due to the dissolution of aragonite during sinking.

A large change in particle composition was observed between mesopelagic and bathypelagic depths at both sites (Figure 5). Most notable is a strong shift from biogenic to lithogenic flux components at 2,300 m. Observed decreases in the proportion of POM with depth are consistent with open ocean remineralization length scales (Martin et al., 1987; Dunne et al., 2007). The decrease in total CaCO₃ content was largely due to the loss of pteropod-derived aragonite, a form more susceptible to dissolution (Orr et al., 2005; Dong et al., 2019; Sulpis et al., 2022). The average and annual fluxes of residual material were an order of magnitude higher at depth, indicating that the observed increase in bathypelagic mass fluxes may be largely accounted for by a source of lithogenic material (Table 3). The annual fluxes of all biogenic mineral components were greater in the bathypelagic, notably for pbSi, somewhat contrary to what might be expected from depth-dependent remineralization of biogenic opal (Nelson et al., 1995; Van Cappellen et al., 2002) and calcite (Milliman, 1993; Sulpis et al., 2021) during sinking. Deep-ocean sediment trap studies frequently report higher fluxes in the bathypelagic depth (Conte et al., 2001; Lampitt et al., 2001; Fischer et al., 2020), including the Fram Strait (Lalande et al., 2016), potentially due to higher horizontal current velocities and decreased trapping efficiency in upper-ocean sediment trap deployments (Gardner, 1980; Gust et al., 1992; Lalande et al., 2016) and potential modification by swimmer activity (Lee et al., 1988; Micheals et al., 1990), and the latter is typically higher in mesopelagic rather than bathypelagic traps (Ramondenc et al., 2022). Local resuspension from sediments could enhance fluxes in the deeper traps, depending on the thickness of the turbid bottom boundary layer. In the present study, sediment traps were typically deployed >150 m from the bottom such that significant resuspension seems to be unlikely, although we cannot account for it completely.

The enhanced lithogenic and biogenic fluxes in the bathypelagic zone of the eastern Fram Strait have previously been linked to lateral transport from the Svalbard region or the Barents Sea with the intermediate or deep nepheloid layer (Lalande et al., 2016), supported by enhanced fluxes of terrestrial and plankton biomarkers. The simulation of sediment trap catchment areas indicates that few particles originated from shallow areas (0–500 m), which encompasses the Svalbard Shelf, suggesting the WSC could serve as a barrier for shelf particles reaching the central Fram Strait (Wekerle et al., 2018). However, the downslope transport of resuspended shelf sediments deeper than 200 m (Sanchez-Vidal et al., 2015) may be laterally advected to the bathypelagic in the Fram Strait, contributing to enhanced fluxes. In addition, ice melting in the Fram Strait may release lithogenic and biogeochemical material entrained during suspension freezing, as has been documented on the Siberian shelves (Dethleff and Kempema, 2007), and enhanced fluxes of terrestrial biomarkers in the presence of ice (Lalande et al., 2016).

The contributions of terrigenous material to organic carbon arriving in the bathypelagic areas are notoriously difficult to constrain. There is a clear increase in the lithogenic contribution to TPM fluxes from mesopelagic to bathypelagic depths (Figure 6), which should correspond to an increase in associated terrigenous organic matter. Previous studies have attempted to estimate the provenance of POC from the mass balance of ∂¹³C signatures (Hedges et al., 1988; Goñi et al., 2000). Isotope mass balance approaches require a clear constraint of isotopic end-members. Terrigenous POC (∂¹³C_Terr) values from carbon (C3) fixation of Arctic plants are reasonably well constrained at values of −26−28‰ (Goñi et al., 2000; Hop et al., 2006; Sanchez-Vidal et al., 2015). However, marine POC (∂¹³C_Mar) values in the Arctic Ocean are highly dynamic (−18−26‰) due to variable phytoplankton community composition, ice cover, and CO₂ concentrations in surface waters (Schubert and Calvert, 2001; Zhang et al., 2012). For consistency with a previous study on the Spitsbergen continental shelf, we assigned a ∂¹³C_Mar value of −21‰ and ∂¹³C_Terr value of −27‰ (Sanchez-Vidal et al., 2015). Using a two-end linear mixing model (Hedges et al., 1988; Goñi et al., 2000) and average ∂¹³C from this study, we estimated terrigenous contributions to POC at 56–60% in the mesopelagic and 43–39% in the bathypelagic areas.

The higher contribution of terrigenous-derived organic carbon in the mesopelagic derived from isotope mass balance are inconsistent with the observed increase in lithogenic fluxes in the bathypelagic resulting simply from lateral advection of terrigenous material (Lalande et al., 2016). However, isotope mass balance calculations were in agreement with biomarker data, typically showing higher organic carbon-normalized concentrations of terrigenous biomarkers (campesterol and ß-sitosterol) in the mesopelagic compared to the bathypelagic. It remains elusive to fully characterize the contribution of terrigenous carbon to the bathypelagic in the Fram Strait. The relative contributions of laterally advected and ice-derived terrigenous matter to POC flux in the deep Fram Strait and their potential role in fueling benthic carbon demand (Wiedmann et al., 2020), thus, remain ill constrained.

One of the main objectives of the present manuscript was to examine the impact of varying ice cover on bathypelagic fluxes. To address this objective, we compiled flux records from two end-members of sediment trap catchment area conditions in the Fram Strait: 1) –Ice end-member: samples from HGIV in 2005, a year with enhanced temperatures of AW, previously characterized as a warm anomaly (Beszczynska-Möller et al., 2012; Lalande et al., 2013), considered representative of future “Atlantification” scenario in the eastern Arctic Ocean (Polykov et al., 2017). 2) +Ice end-member: a sample from HGN in 2007, a year with higher ice-area export through the Fram Strait (von Appen et al., 2021) and site with typically smaller distances to the ice edge (Wekerle et al., 2018; Fadeev et al., 2021). Due to truncated flux records, it was necessary to extrapolate daily flux averages within months for 15 days in July 2005 and 25 days in August 2005 at HGIV and 10 days in June 2007 and 17 days in July 2007 at HGN. Differences in bathypelagic deployment depths may also serve to confound comparisons of the end-members. These limitations should be taken into account in the following data interpretation.

In the +Ice end-member, ice concentrations were higher throughout the entire year resulting in a dominant melt water (MW) regime (>50% of catchment area) already present in early spring (Figures 11A,G). This led to spring peaks in biogenic flux components, with almost half of the annual flux (1.7 g POC m^-2, Figure 11E) arriving at the seafloor by the end of May, compared to only 20% over the same period in the–Ice end-member (0.6 g POC m^-2). The timing of POC fluxes to the deep is considered to be an important aspect of pelagic benthic coupling that impacts benthic communities (Wolff et al., 2011; Ruhl et al., 2020; Wiedmann et al., 2020). Peak Chl concentrations occurred 1 month prior in the +Ice end-member (Figure 11B), concomitant with maximum coverage of meltwater in the catchment area. It is widely observed that ice melt stratification in marginal ice zones can initiate and support phytoplankton blooms (Wu et al., 2007; Cherkasheva et al., 2014). Average ∂¹³C isotope values were enriched from February to May compared to the July–November period (Figure 11J). The onset of peak bathypelagic carbon flux occurred in March, prior to any notable increase in remote sensing chl concentrations, although it is known that low light and high ice concentrations at this time may hamper reliable remote sensing data during this period (Wekerle et al., 2018). Discrepancies between chlorophyll concentrations and bathypelagic carbon flux are also observed in open ocean systems such as the North-eastern Atlantic (Salter, 2008; Lampitt et al., 2010) and remain enigmatic. In the Arctic, light penetration through leads in fragmented ice can stimulate under-ice phytoplankton blooms (Assmy et al., 2017) that are obscured from detection by remote sensing.

In the +Ice end-member, there was a corresponding spring peak in lithogenic fluxes (Figure 11L) which may have ballasted organic carbon sedimentation through an abiotic flux pump (Ittekkot, 1993; Salter et al., 2010). High pbSi fluxes and elevated Si:POC ratios seem to indicate that diatoms were an important component of spring bathypelagic organic carbon flux in the +Ice end-member (Figures 11C,H). However, CaCO₃ fluxes were also elevated, and Si:PIC ratios were low (Figures 11F,K). Negligible pteropod fluxes (Table 4) suggest that other calcareous plankton mediated deepwater PIC fluxes. Previous studies have highlighted the importance of cold-water-adapted coccolithophores (Coccolithus pelagicus) to deepwater PIC fluxes in the Fram Strait (Samtleben and Bickert, 1990). Similarly, foraminifera fluxes are typically higher in deeper sediment traps in the Fram Strait (Ramondenc et al., 2022) and biogenic calcite export from the sea-ice-associated species Neogloboquadrina pachyderma is of known importance in the Arctic (Tell et al., 2022).

In the –Ice end-member, peak POC fluxes occurred during summer with an approximate 1-month delay from peak Chl values, comparable to recent estimates from the ML regime at HGIV (von Appen et al., 2021). In the present comparison, the –ice catchment areas during summer were dominated by an ML regime, despite a regional peak in strong MW coverage. Summer POC fluxes (July to September) in the –Ice end-member occurred previously and were larger and more episodic than the +Ice end-member, accounting for 65% rather than 36% of annual POC export, broadly consistent with the previous findings (von Appen et al., 2021). Enhanced pbSi fluxes and high Si:POC and Si:PIC indicated that diatoms rather than calcareous plankton dominated summer bathypelagic POC fluxes in the –Ice end-member. The peak in lithogenic flux in the –Ice end-member was dissociated from the biogenic flux but corresponded to enhanced ice in the catchment area and to the inflow of warmest Atlantic water. This pattern could reflect the release of lithogenic material entrained from suspension freezing (Dethleff and Kempema, 2007) as sea ice exported through the Fram Strait (Smedsrud et al., 2017) in late summer comes in contact with warm AW. This process could also explain the higher annual lithogenic flux in the –Ice end-member with elevated AW temperatures than in the +Ice end-member (Table 4).

Annual bathypelagic POC flux was 50% higher in the +Ice end-member, implying a stronger, though slower (von Appen et al., 2021), and perhaps more efficient, biological carbon pump resulting from stratified MW regimes compared to more weakly stratified open water. Despite significant interannual variability and challenges of non-overlapping sediment trap deployments, a similar enhancement of deepwater POC flux was observed at HGN (larger proportion of stratified MW regimes) from comparing annual flux averages across the decadal time series (Table 3). In this regard, our findings are in contrast with a recent comparison of stratification regimes in the Fram Strait which showed higher annual POC fluxes in ML regimes (von Appen et al., 2021), although von Appen et al. (2021) could not account for the fate of direct sea ice algal production and export early in the year, which appears to be an important component of annual POC flux in +Ice end-member scenario. In this context, it is likely that distance to the ice edge in MIZ areas has an important effect on particle export. Enhanced export mechanisms in close proximity to locations where ice quickly melts result from enhanced sinking velocities (Fadeev et al., 2021), deposition of algal biomass from melting Arctic sea ice (Boetius et al., 2013; Assmy et al., 2017), and ballasting from cryogenic mineral phases (Wollenburg et al., 2018). It is also challenging to compare different years since variation in the source area of ice exported through the Fram Strait are likely to influence the transfer of ice-rafted material (Krumpen et al., 2019), which may be an important component of deepwater POC fluxes at HGN. Despite these complications, higher summertime POC flux in ML regimes and a more efficient biological carbon pump in MW regimes appear to be a consistent feature across our datasets in the Fram Strait.

The analysis of our dataset and information from previous studies (Sanchez-Vidal et al., 2015; Lalande et al., 2016) indicate that laterally advected materials may contribute to bathypelagic fluxes in the Fram Strait. This could confound interpretations related to changes in ice cover, although particle trajectory modelling shows relatively consistent catchment areas (Wekerle et al., 2018). Marginal ice zones in the Fram Strait arise through sea ice export (Krumpen et al., 2019) rather than retreating ice as encountered in the central Arctic Ocean. Nevertheless, differences in bathypelagic fluxes between HGIV and HGN and +Ice and–Ice end-member comparison offer some useful insights into the potential impacts of ongoing “Atlantification” and transition of MIZ in the future Arctic. Our data suggest modifications in quantity, biogeochemical composition, phenology, and episodicity of deepwater particle fluxes that may all change the nature of pelagic–benthic coupling.