We constructed food webs for the WSC region of the eastern Fram Strait (Figure 1) using published data for the late spring/summer of 2003 (before the warming event), 2006 (during), and 2010 (after), or as close as possible to the time period (but always within 1 year). The same model structure was used for each time period (Figure 2). Each web comprised 42 flows that represented carbon flows between two compartments, or from one compartment to a sink (Table 1). The structure of the food webs was based on the assumption that sizes of the producers and consumers were major determinants of the trophic dynamics of these systems, i.e., small grazers are restricted to small algae. Choices of compartments and trophic relationships were a compromise between achieving biological reality and keeping the total number of flows in the system reasonable. The living components included two phytoplankton compartments, three zooplankton compartments, one compartment for small planktivorous fish, one for cod, and one compartment for heterotrophic bacteria. The phytoplankton were divided into “small” (0.2 to ~10 μm; assumed to be mainly picophytoplankton, coccolithophores, Phaeocystis sp., and small autotrophic flagellates) and “large” (> 10 μm; mainly diatoms and larger dinoflagellates; Kilias et al., 2014; Nöthig et al., 2015). Zooplankton size classes were the microzooplankton (20–200 μm; ciliates and flagellates), the mesozooplankton (200 to ~1,000 μm; mainly small copepods) and macrozooplankton (chaetognaths, euphausiids, and Calanus copepods >1,000 μm; Bamstedt et al., 1991; Hop et al., 2006; Blachowiak-Samolyk et al., 2007; Calbet, 2008; Pasternak et al., 2008; De Laender et al., 2010; Svensen et al., 2011; Monti and Minocci, 2013). Small planktivorous fish were assumed to be mainly capelin and herring but this compartment also includes carnivorous zooplankton, such as amphipods (Wassmann et al., 2006; Dalpadado et al., 2016). The top predator in the system was cod (Wassmann et al., 2015)

All living compartments contributed to a labile dissolved organic carbon (DOC) pool through excretion and to the detrital pool through mortality or defecation. Sloppy feeding was implicitly included as excretion to DOC. Detritus was transformed to DOC by chemically- or bacterially-mediated dissolution (Jumars et al., 1989). All living compartments lost carbon by respiration. Other forms of mortality (e.g., viral lysis or natural cell mortality) were implicitly included in flows to detritus and DOC. All non-respiratory losses from the system were represented by flows to an “external” compartment that served as a mathematical closure term. These losses included particulate organic carbon (POC) export by detrital settling, DOC loss by advection, and removal of cod through the fishery or via consumption by higher trophic levels.

We used published data from Fram Strait to calculate input (“known”) values for 7 of the 42 carbon flows: small and large phytoplankton primary productivity, bacterial productivity, microzooplankton grazing on small phytoplankton and flagellates, microzooplankton grazing on large phytoplankton, ingestion rates for the small fish and ingestion rates for cod (Table 2). Total primary productivity rates were taken from remote sensing estimates by Arrigo et al. (2008), Arrigo and van Dijken (2011, 2015) and daily production was estimated assuming annual primary production was evenly distributed over the period of open water. Contributions of flagellates vs. large phytoplankton to total primary productivity were assumed to be proportional to their size-specific contributions to biomass and were calculated from chlorophyll a measurements and phytoplankton community composition data of Nöthig et al. (2015, their Table 2). Accordingly, small phytoplankton and flagellates constituted 20% of the total phytoplankton biomass for the before models and 80% were considered large phytoplankton; Phaeocystis sp. and other flagellates accounted for 97 and ~50% of the small phytoplankton biomass for the during and after models, respectively.

Conversions from chlorophyll a to carbon units were done using an average C:chl ratio of 53 (g:g) to avoid seasonal biases (Svensen et al., 2011). The C:chl ratio of 53 for phytoplankton was chosen as a compromise between spring and summer values and those for small and large phytoplankton cells as shown by mesocosm experiments in which C:chl ratio of diatoms, dinoflagellate and mixed composition were 95, 45, and 60, respectively (Svensen et al., 2011; Spilling et al., 2014). Similarly, phytoplankton in the Fram Strait in May and August 2014 had a C:chl ratio of 41 (Marit Reigstad, personal communication). The ratio of 53 was used when converting from chl a estimates in the field (in mg chla m⁻²) to phytoplankton carbon (mg C m⁻²). A different C:chl ratio would either increase or decrease phytoplankton biomass in the model constraints but has no effect where phytoplankton biomass is an unknown.

There is difficulty in obtaining reliable and consistent data in high latitude environments due to the effort and cost of such studies. The data from Nöthig et al. (2015) as well as other field campaigns are based on cruises of a few weeks length, in the June and July time period, with the exception of estimates of fish abundance, provided by year. In this way, it is possible to compare year-to-year summer variability. Data from cruises from other times of the year (either April–May or August–September) were not included in this study.

Bacterial productivity values were calculated by multiplying bacterial abundance data by a C-specific production of 0.109 ± 0.89 d⁻¹ (L. Seuthe, personal communication) an integrated over 0–45 m during cruises to NW Spitsbergen in 2014 (n = 7), and a biomass of 10 fg C cell⁻¹ (Fukuda et al., 1998). Microzooplankton grazing rates were estimated from Verity et al. (1999, 2002) in the Barents Sea and Calbet et al. (2011) in the Fram Strait. For the before model, we assumed that microzooplankton grazing would be higher on small phytoplankton and flagellates (0.2 d⁻¹) than on the larger phytoplankton (0.05 – 0.1 d⁻¹) based on Strom et al. (2001). In contrast, for the models of during and after the warm period, when Phaeocystis sp. dominated the phytoplankton community, we assumed that microzooplankton grazing rates on small phytoplankton and Phaeocystis sp. were also low (0.05 d⁻¹ for during and 0.1 d⁻¹ for after) and no more than 8% of the phytoplankton standing stock based on previous studies (see also Caron et al., 2000; Calbet et al., 2011). After the warm period, grazing of microzooplankton on non-Phaeocystis sp. was 0.2 d⁻¹. Ingestion rates for the small fish and cod were calculated from annual fish biomass in ICES ASWG 2014 and a conservative C-specific ingestion rate of 0.017 d⁻¹ for cod (range of 0.017–0.057 d⁻¹, De Laender et al., 2010) and an average C-specific ingestion rate of 0.04 d⁻¹ for capelin and herring (range of 0.01–0.1 d⁻¹, Ajiad and Pushchaeva, 1992; Megrey et al., 2007).

Sources of biomass for compartments are detailed in Table 3; these data are used to formulate the constraints used to set bounds on the flows predicted by the model (see Section Inverse Analysis below) (Table 4). The conversion factor of 0.132 was used to estimate carbon from wet weight in fishes (Sakshaug et al., 1994). Microzooplankton biomass was estimated from cell counts by the conversion factors of Verity and Lagdon (1984) and Menden-Deuer and Lessard (2000).

The linear inverse modeling approach of Vézina and Platt (1988) was used in conjunction with the Monte Carlo solutions approach of Donali et al. (1999) for estimating the range of values for all flows in our constructed food webs. Model code was run in Matlab R2011b and was kindly provided by Dr. Nathalie Niquil (Centre National de la Recherche Scientifique, Caen, France). The approach taken assumes that biomass in any compartment is in steady state, i.e., the total flows entering any compartment are equal to the flows leaving it without any accumulation or decrease (with the exception of the “external” compartment), although modifications to the approach can be made to accommodate non-steady state scenarios by allowing residual flows to balance the system (e.g., Richardson et al., 2003).

As described above, data from the scientific literature were used to formulate 7 input equations. Combined with the 10 mass balance equations (one for each compartment; see Table 5), there were 17 equations available to describe the system with 42 flows. We reduced the number of possible solutions for this underdetermined system by applying a set of biological constraints (provided in Table 4). Allometric constraints based on published relationships incorporated available biomass data and provided upper and lower bounds on the rates and efficiencies of biological processes. For example, the respiration of all phytoplankton was constrained to be at least 5% but no more than 30% of the gross primary productivity (GPP) (Vézina and Platt, 1988). Growth efficiencies were assumed to be 25–50% of ingestion for the zooplankton groups (Straile, 1997). Bounds on assimilation efficiencies for all grazers were 50–90% of ingestion for the microzooplankton (Vézina and Platt, 1988; Straile, 1997) and 50–80% for the macrozooplankton (Straile, 1997). We also set a lower bound on the macrozooplankton grazing such that they consumed at least 30% of the large phytoplankton productivity (based on Wassmann et al., 2006). Other constraints are detailed in Table 4. We used temperatures of 3.5, 5, and 4°C for the before, during and after models, respectively (Beszczynska-Moller et al., 2012).

Application of constraints reduces the range of possible solutions, but does not provide a unique solution. The Monte Carlo approach of Donali et al. (1999) (see also the review by Niquil et al., 2012) calculates 10,000 possible solutions for each set of flows, thus we were able to calculate both an average and a standard deviation for each flow in the food web.

After food webs were constructed for the before, during and after warm water event, the structure and function of each web was assessed using EcoNetwork analysis software (available at https://www.cbl.umces.edu/~ulan/ntwk/network.html; see also Ulanowicz and Kay, 1991; Ulanowicz, 2004). Michaels and Silver (1988), Ducklow et al. (1989), and McManus (1991) used earlier versions of this program to examine flows of nitrogen and energy, respectively, through microbial food webs to higher trophic levels in planktonic systems. We chose a key index from the output, input/export vectors, to calculate how much of each input flow (i.e., primary production of the two phytoplankton groups) eventually was exported through the three possible routes of export (via cod, detritus, or DOC). This calculation allowed us to break down the export flows into the relative contributions by the flagellate (small) vs. diatom (large) phytoplankton.

The prediction of the proliferation of small cells in a future warmer ocean is usually associated with flagellates, whereas larger cells are presumed to be diatoms (Li et al., 2009, but see Wright et al., 2010). In the WSC, the transition of phytoplankton communities exposed to elevated temperatures can be more complex: the prymnesiophyte Phaeocystis sp. is a flagellate that can form both large colonies as well as single cells (Rousseau et al., 2000). In the WSC, the warming event was associated with a proliferation of this microalga from the summer of 2005 onwards, i.e., different clades of Phaeocystis sp. were dominant during and after the warm water event from 2005 to 2007 (Nöthig et al., 2015). This species is not new to the Fram Strait, commonly found in the region in spring and summer (Smith, 1987; Hegseth and Tverberg, 2013; Saiz et al., 2013). P. pouchetti slowly decreased in concentration after 2007 with an average 45% concentration after the warm water event. A major consequence of this shift from diatoms to Phaeocystis sp. is the change in grazing pressure; Phaeocystis sp. experiences less grazing than other flagellates (Caron et al., 2000; Strom et al., 2001; Calbet, 2008; Calbet et al., 2011).

Blooms of single-cell Phaeocystis sp. are of widespread distribution, with individual cells of 3–7 μm (Vernet et al., 1996; Kozlowski et al., 2011; Metfies et al., 2016) although it is generally considered that Phaeocystis sp. blooms in its colonial form, with colonies in excess of 100 μm and up to 1000 μm (Schoemann et al., 2005; Lasternas and Agusti, 2010). Single-cell Phaeocystis sp. in the eastern Fram Strait was observed in 2012 after the warm event in the <3 μm phytoplankton size fraction (Metfies et al., 2016). The proportion of P. pouchetii in colonial or in single-cell form at the WSC in 2005–2007 is unknown; both forms were modeled in this study (Stelfox-Widdicombe et al., 2004).

The model provides realistic estimates of phytoplankton growth rates, approximated from C-specific primary production. Flagellates grew at an average 0.2, 0.13, and 0.28 d⁻¹ and large cells at an average 0.21, 0.39, and 0.28 d⁻¹ within the surface layer before, during and after the warm water event, respectively. These rates were not measured, the biomass originated from the field (Table 3) and primary production from the model output (Table 1), which carry the inherent approximation that both phytoplankton size fractions have equal photosynthetic efficiency.

When diatoms dominated (i.e., cells > 10 μm), the model predicted that phytoplankton were consumed by meso- and macrozooplankton herbivores, with a major carbon flow from fecal pellets to detritus and eventual export (i.e., sedimentation, Figure 2). Carbon also flowed to detritus and microzooplankton, accounting for the rest of the large cells, but the contribution was minor based on the small grazing pressure of microzooplankton on diatom blooms (Figure 3, Sherr and Sherr, 2009). The inverse model suggested that when Phaeocystis sp. dominated pelagic photosynthetic communities, e.g., by contributing up to 97% of the autotrophic community, phytoplankton carbon goes directly to detritus, and to a lesser extent to microzooplankton and DOC production (Figure 3). Mesozooplankton fed mostly on the detrital carbon, originating from phytoplankton sinking and fecal pellets, while a large proportion of the detritus was also exported. In this way, the mesozooplankton role in the food web increased when flagellates dominated but macrozooplankton role stayed rather constant, as in the case of microzooplankton. Total export out of the system decreased by 15%, due to a 35% diminution in POC export while DOC export increased (Figure 4). When the phytoplankton community bounced back to more diatoms, but still with ~45% Phaeocystis sp., overall export of particulate carbon also recovered while DOC contribution remained high (Figure 4). Additionally, the increase in microzooplankton and bacterial abundance during and after the warm water period indicate an increase in substrate, as expected from Kirchman et al. (2009a,b). The modeled sedimentation flux, where the mixed phytoplankton community composed of diatoms, Phaeocystis sp. and nanoflagellates exported as much carbon as diatoms alone, was also expected from the sediment trap data of Lalande et al. (2013) that showed that fluxes remained the same and only the quality changed. However, these model predictions are novel and will be discussed further.

What the inverse model in this study provides is a picture of possible carbon pathways within Arctic food webs that could explain how to maintain an important contribution of phytoplankton carbon to deep water in the absence of a diatom bloom. Export of carbon originating from non-diatoms is not surprising (Figure 4). Picoplankton and flagellates have been observed in sedimenting matter for the last 30 years: Synechococcus was a major contributor to fluxes to the deep sea (Lochte and Turley, 1988) and has been detected in the South Pacific (Waite et al., 2000), tropical Pacific (Stukel et al., 2013) and in equatorial Pacific, Atlantic and Indian Oceans (e.g., Lampitt et al., 1993). High abundance of Synechococcus has been recently reported for the eastern Fram Strait, included here as cells <10 μm (Paulsen et al., 2016). Similarly, the ubiquitous Micromonas sp. in Arctic waters has been detected in this region by sediment traps, and associated with increased ²³⁴Thorium adsorption in the Central Arctic (Charles Bachy, pers. commun., Roca-Marti et al., 2016). Great quantities of Phaeocystis sp. were observed at the ocean bottom in the Ross Sea, at >500 m depth (DiTullio et al., 2000). In the Fram Strait, highest export was observed during a Phaeocystis sp. bloom, equivalent to the export efficiency, or % of primary production that sediments, by diatoms (Le Moigne et al., 2015). In the Gulf of St. Lawrence, the export efficiency increased from 10% during the diatom bloom to 10–5% in post-bloom conditions (Rivkin et al., 1996). In the Canadian Arctic, flagellates were associated with high export ratios of 0.38–0.69 (Lapoussiere et al., 2013). Similarly, the inverse model in the Fram Strait predicts an export efficiency of 51, 44, and 53.5% before, during and after the warm water event (Table 1). In this way, Phaeocystis sp. and flagellates can fuel the biological pump, transferring an important proportion of surface primary production to depth.

Carbon export in the model originated from either diatoms or detritus that during the warm period is overwhelmingly dominated by Phaeocystis sp. carbon (Figure 5). By which processes can flagellates contribute to export out of the surface layer? High biomass, stickiness, and presence of ballast all correlate with increased phytoplankton sedimentation by coagulation (e.g., Passow and Alldredge, 1999; Jouandet et al., 2014). In general, diatoms and coccolithophores are considered to sink faster than other phytoplankton and their silicon frustule (opal) or carbonate coccoliths are assumed to act as ballast for phytoplankton sinking and zooplankton fecal pellets, activating the biological pump (Armstrong et al., 2001; Klaas and Archer, 2002; Ploug et al., 2008). Ballast for phytoplankton could originate also from intracellular carbohydrates, minerals or carbonate precipitated within sea ice (Richardson and Cullen, 1995; Iversen and Ploug, 2010). Phaeocystis sp. blooms are reported to have very high sinking rates (Wassmann et al., 1990; DiTullio et al., 2000 but see Schoemann et al., 2005). DMSP, known to be elevated in Phaeocystis sp., has recently been suggested as ballast for this species (Lavoie et al., 2015, but see Boyd and Gradmann, 2002). Coagulation of cells in turbulent environments, in particular species with a sticky surface as observed in senescent Phaeocystis sp. blooms, generates marine snow; this process is considered a widespread venue of removing cells from the upper ocean (Passow and Wassmann, 1994; Logan et al., 1995). Mucus webs of pteropods are also known to be an efficient transport vehicle for pico-plankton particles (Noji et al., 1997). High stickiness in Arctic phytoplankton is expected; diatoms excrete large amounts of polysaccharides (Myklestad, 1995) and Arctic phytoplankton, both diatoms and flagellates, can excrete as much as 70% of their daily primary production as DOC (Vernet et al., 1998; Matrai et al., 2007; Poulton et al., 2016), which can be considered a source of stickiness (Schoemann et al., 2005). Marine snow is part of the detrital carbon, and is difficult to detect and quantify. TEP (transparent exo-polymers) is believed to comprise most of the marine snow and its sinking speed is also related to size, porosity and ballast usually provided by its constituents (Passow, 2002; Bach et al., 2016). For example, porosity of marine snow is lower when flagellates dominate in comparison with diatom-rich aggregates, thus providing another mechanism by which non-diatom aggregates can export carbon (Bach et al., 2016).

The changes in carbon export predicted by the model when Phaeocystis sp. dominated agree in large extent to the observations from sediment traps in the eastern Fram Strait. Flux of (POC) at 179–280 m depth from 2002 to 2008, with 20 sampling cups per year collecting material from 59 days in winter and 7 days in summer, showed POC sedimentation associated with biogenic silica (bSi) pulses (Lalande et al., 2013). Before the fall of 2004, these pulses occurred in spring (April to June), sometimes associated with the ice edge and in the late summer (August to October) due to atmospheric heating of the upper water column. Before 2004 the pulses ranged from 30 to 50 mg C m⁻² d⁻¹ and 10–30 mg bSi m⁻² d⁻¹. From late 2004 to the summer of 2008, during the warm water event, the consistency of the spring and late summer bSi pulses disappeared, with a few peaks in sedimentation in either May or August (~10 mg bSi m⁻² d⁻¹) remaining and the rest of the time sedimentation was <5 mg bSi m⁻² d⁻¹ (Figure 2f, Lalande et al., 2013). The pulses of POC remained unchanged throughout this period, both in magnitude and time of the year (Figure 2e, Lalande et al., 2013).

Any differences between model predictions and sediment trap data on sedimentation rates are expected, as export in the model represents carbon loss out of the surface layer (<100 m) while the sediment traps were deployed at ~250 m depth. In the field, the changes in flux of bSi correlated with other important changes in the nature and quality of the sedimenting matter: lower fecal pellet carbon and an increase in the amount of small fecal pellets, attributed to dominance of smaller zooplankton (Lalande et al., 2013). The inverse model predicts the change in quality of sedimenting matter can be attributed to the dominant phytoplankton community, diatoms vs. Phaeocystis sp. and to the changes in trophic pathways in the food web (Figure 5, see Section Discussion. for a detailed discussion on carbon flow through the modeled food web). The predicted changes in the biomass of the different compartments are reflected in higher respiration during and after the warm water event for small phytoplankton, mesozooplankton (small copepods) and bacteria and lower respiration from large phytoplankton (Figure 6).

Our model results present an alternative hypothesis that warming and flagellates could bring increased heterotrophy to the Arctic, expressed as the ratio of respiration to primary production (Table 1, Figure 6). The paradigm that flagellates will decrease sedimentation corresponds to a scenario of higher retention of organic carbon in surface waters and higher respiratory losses (Forest et al., 2010; Vaquer-Sunyer et al., 2010). Similarly, predictions of lower sedimentation are associated with an activation of the microbial loop (Kirchman et al., 2009b). High export in absence of diatoms in the Arctic is also possible as shown in this study, or when dinoflagellates replace diatoms (Rivkin et al., 1996); high sedimentation by flagellates has been observed in the field even when diatoms are abundant in surface waters (Amacher et al., 2013). A high export is possible if a link between the classical and microbial food webs develops through the consumption of microzooplankton and detritus by copepods.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.