Role of zooplankton in determining the efficiency of the biological carbon pump

Abstract

The efficiency of the ocean's biological carbon pump (BCP${}_{\text{eff}}$ – here the product of particle export and transfer efficiencies) plays a key role in the air–sea partitioning of CO${}_{\mathrm{2}}$. Despite its importance in the global carbon cycle, the biological processes that control BCP${}_{\text{eff}}$ are poorly known. We investigate the potential role that zooplankton play in the biological carbon pump using both in situ observations and model output. Observed and modelled estimates of fast, slow, and total sinking fluxes are presented from three oceanic sites: the Atlantic sector of the Southern Ocean, the temperate North Atlantic, and the equatorial Pacific oxygen minimum zone (OMZ). We find that observed particle export efficiency is inversely related to primary production likely due to zooplankton grazing, in direct contrast to the model estimates. The model and observations show strongest agreement in remineralization coefficients and BCP${}_{\text{eff}}$ at the OMZ site where zooplankton processing of particles in the mesopelagic zone is thought to be low. As the model has limited representation of zooplankton-mediated remineralization processes, we suggest that these results point to the importance of zooplankton in setting BCP${}_{\text{eff}}$, including particle grazing and fragmentation, and the effect of diel vertical migration. We suggest that improving parameterizations of zooplankton processes may increase the fidelity of biogeochemical model estimates of the biological carbon pump. Future changes in climate such as the expansion of OMZs may decrease the role of zooplankton in the biological carbon pump globally, hence increasing its efficiency.

Introduction

The biological carbon pump plays an important role in regulating atmospheric carbon dioxide levels (Kwon et al., 2009; Parekh et al., 2006). Phytoplankton in the surface ocean convert inorganic carbon during photosynthesis to particulate organic carbon (POC), a fraction of which is then exported out of the upper ocean. As particles sink through the interior ocean they are subject to remineralization by heterotrophs, such that only a small proportion of surface-produced POC reaches the deep ocean (Martin et al., 1987). The efficiency of the biological carbon pump (BCP${}_{\text{eff}}$; defined here as the proportion of surface primary production that is transferred to the deep ocean; Buesseler and Boyd, 2009) therefore affects the air–sea partitioning of CO${}_{\mathrm{2}}$ (Kwon et al., 2009). Greater understanding on the controls of this term may consequently result in more accurate assessments of the BCP's role in the global carbon cycle.

One approach to determine BCP${}_{\text{eff}}$ over long timescales (millennia) is by assessing the relative proportions of preformed and regenerated nutrients, i.e. the fraction of upwelled nutrients that is removed from surface waters by biological uptake (Hilting et al., 2008). However, to assess BCP${}_{\text{eff}}$ over much shorter timescales (days to weeks) we use the definition of Buesseler and Boyd (2009), where BCP${}_{\text{eff}}$ is the product of particle export efficiency (PE${}_{\text{eff}}$, the ratio of exported flux to mixed layer primary production) and transfer efficiency ($T_{\text{eff}}$, the ratio of deep flux to exported flux). Using these two parameters together allows a more in-depth analysis of the biological processes involved and thus the assessment of the role of zooplankton in setting BCP${}_{\text{eff}}$. Additionally, the attenuation coefficients Martin's $b$ (Martin et al., 1987) and the remineralization length scale $z^{\ast }$ (Boyd and Trull, 2007) are useful to quantify how rapidly exported POC is remineralized in the mesopelagic zone.

PE${}_{\text{eff}}$ varies proportionally to primary production, although uncertainty exists as to whether the relationship is inverse or positive (Aksnes and Wassmann, 1993; Cavan et al., 2015; Henson et al., 2015; Laws et al., 2000; Maiti et al., 2013; Le Moigne et al., 2016). Potential controls on PE${}_{\text{eff}}$ include temperature (Henson et al., 2015; Laws et al., 2000), zooplankton grazing (Cavan et al., 2015), microbial cycling (Le Moigne et al., 2016), mineral ballasting (Armstrong et al., 2002; François et al., 2002; Le Moigne et al., 2012), or large export of dissolved organic carbon (Maiti et al., 2013). $T_{\text{eff}}$ and POC attenuation coefficients describe how much of the exported POC reaches the deep ocean and how much of it is remineralized. Essentially the attenuation of POC with depth is determined by the sinking rates of particles and how rapidly the POC is turned over (Boyd and Trull, 2007). However, these factors themselves are controlled by various other processes such as ballasting by minerals (François et al., 2002; Le Moigne et al., 2012), epipelagic community structure (Lam et al., 2011), temperature (Marsay et al., 2015), lability of the particles (Keil et al., 2016), and zooplankton diel vertical migration (Cavan et al., 2015). Therefore, it is unlikely that any single factor controls BCP${}_{\text{eff}}$.

The role of zooplankton in controlling the efficiency of the BCP is often overlooked, with greater focus on factors such as biominerals for ballasting (De La Rocha and Passow, 2007) or microbial respiration (Herndl and Reinthaler, 2013). Nevertheless, zooplankton have the potential to significantly impact the biological carbon pump as they can consume and completely transform particles (Lampitt et al., 1990). Grazing by zooplankton results in POC either passing through the gut and being egested as a faecal pellet, being respired as CO${}_{\mathrm{2}}$ or fragmented into smaller particles through sloppy feeding (Lampitt et al., 1990). Further, zooplankton can undergo diel vertical migration, feeding on particles at night in the surface and egesting them at depth during the day (Wilson et al., 2013). Consequently, a significant proportion of POC may escape remineralization in the upper mesopelagic zone (Cavan et al., 2015), where recycling of POC is most intense (Martin et al., 1987).

In this study we combine observations (made using Marine Snow Catchers, MSCs) and model output to investigate the role of zooplankton in setting the efficiency of the biological carbon pump in three different oceanic regions: the Atlantic sector of the Southern Ocean (SO), the Porcupine Abyssal Plain (PAP) site in the temperate North Atlantic and the equatorial tropical North Pacific (ETNP) oxygen minimum zone (OMZ). The ecosystem model used here, MEDUSA (Yool et al., 2013), was chosen as it separates particle fluxes into slow and fast sinking groups. Additionally, the only interactions of zooplankton with particles in MEDUSA are through the production of particles (faecal pellets) and by grazing on slow sinking particles only. Here we compare various indices of BCP${}_{\text{eff}}$ between the observations and model to infer the role of zooplankton in controlling BCP${}_{\text{eff}}$.

Methods

Site description

Three very different sites were chosen in this study: the Atlantic sector of the SO (45–65${}^{\circ }$ S, 20–70${}^{\circ }$ W), the Porcupine Abyssal Plain (PAP) site in the temperate North Atlantic (49${}^{\circ }$ N, 17${}^{\circ }$ W), and the equatorial tropical North Pacific (ETNP) OMZ (13${}^{\circ }$ N, 91${}^{\circ }$ W) (Fig. 1). The SO accounts for $\sim$ 20 % of the global ocean CO${}_{\mathrm{2}}$ uptake (Park et al., 2010; Takahashi et al., 2002) and is a large high-nutrient, low-chlorophyll region, in part due to limited iron availability (Martin, 1990). Nevertheless, iron from oceanic islands and melting sea ice can cause intense phytoplankton blooms, which may lead to high POC export (Pollard et al., 2009). In the temperate North Atlantic seasonality is high, with phytoplankton blooms occurring in spring and summer (Lampitt et al., 2001). The region contributes disproportionally to global export, accounting for 5–18 % of the annual global export (Sanders et al., 2014). In the ETNP region a strong OMZ persists where, between 50 and 1000 m depth, dissolved oxygen concentration can fall below 2 $\mathrm{\mu }$mol kg${}^{-\mathrm{1}}$ (Paulmier and Ruiz-Pino, 2009). In OMZs the low oxygen concentrations may lead to a high transfer efficiency of POC flux potentially due to reduced heterotrophy (Devol and Hartnett, 2001; Hartnett et al., 1998; Keil et al., 2016; Van Mooy et al., 2002).

Map showing study areas. Blue rectangle is location of sites in the Southern Ocean, red is the North Atlantic Porcupine Abyssal Plain, and orange the equatorial North Pacific oxygen minimum zone.

[Figure omitted. See PDF]

Observations

Particles were collected using MSCs (Riley et al., 2012) from the three oceanic sites as shown in Fig. 1. In total 27 stations were sampled: 18 in the SO, 5 at PAP, and 4 in the ETNP (Table S1 in the Supplement). MSCs have the advantage of being able to separate particles intact into two groups dependent on their sinking rate: fast (> 20 m d${}^{-\mathrm{1}}$) or slow (< 20 m d${}^{-\mathrm{1}})$. MSCs were deployed below the mixed layer depth (MLD), which was determined as the depth with the steepest gradient of salinity and temperature, and usually occurred between 20 and 70 m (Table S1). The shallowest MSC was deployed 10 m below the MLD and another 100 m deeper than this for the SO (Cavan et al., 2015) and the PAP site. In the ETNP MSCs were also deployed deeper into the water column to a maximum depth of 220 m.

Fast and slow sinking particles were collected from the MSC following the protocol by Riley et al. (2012). Images of fast sinking particles were taken to estimate the equivalent spherical diameter (ESD) of the particles and ESD converted to POC mass via conversion factors, CFs (Alldredge, 1998; Cavan et al., 2015). Two different CFs were used, one for phytodetrital aggregates (PDAs, Eq. 1) and one for faecal pellets (FPs, Eq. 2): $$\begin{array}{}& & \text{phytodetrital aggregates}m=1.09\times V^{0.52},& & \text{faecal pellets}m=1.05\times V^{0.51},\end{array}$$ where $m$ is mass of POC and $V$ volume of the particle. Very few published studies exist comparing size of particles to carbon content, and those that do tend to focus on FPs (Manno et al., 2015). We chose to use the Alldredge (1998) CFs because it allows comparison with other published studies (e.g. Ebersbach and Trull, 2008; Guidi et al., 2007; Laurenceau et al., 2015; Riley et al., 2012, Belcher et al., 2016), describes the fractal nature of particles and is an upper ocean study (50 % of our observations lie close to the depth ($\sim$ 20 m) of the Alldredge, 1998, particles; Table S1). Another important point is that the MSCs allow collection of particles relatively undamaged or unaltered compared to sediment traps (other than gel traps), making the use of conversion factors more reliable as particle shapes reflect those measured in situ (Romero-Ibarra and Silverberg, 2011).

To test the robustness of the Alldredge (1998) CFs we compared the resulting POC mass with that calculated using the CFs from Manno et al. (2015). This study was done in the SO and only focussed on FPs; hence, we only tested the similarity using our SO data where FPs comprised most (> 60 %) of the particle flux (Fig. S1). Manno et al. (2015) found a linear relationship between FP size and POC content for cylindrical FPs (0.018 mg C mm${}^{-\mathrm{3}})$. We calculated the total fast sinking POC mass (FP $+$ PDA) using the Manno CF for FPs and Alldredge CFs for the PDAs (remembering that FPs dominated flux) and compared these to just using the Alldredge CFs (Fig. S2a). The slope of the regression between the two is very close to 1 at 0.96, showing neither CF under- or overestimates POC relative to the other. There was no statistical difference ($t$ test, $t=0.25$, df $=$ 77.23, $p$ value $=$ 0.80) between the mean masses of the two methods, with Manno CFs producing a mean mass of POC per sample of 11 $\mathrm{\mu }$g C and Alldredge CFs producing a mean of 8.4 $\mathrm{\mu }$g C. Therefore, we conclude that using the Alldredge CFs is justified and using one general CF allows comparisons between our different study regions.

Slow sinking and suspended particles were filtered onto ashed (400 ${}^{\circ }$C, overnight) GF/F filters and run in a HNC elemental analyser to determine POC mass. Sinking rates were estimated for fast sinking particles in the SO and at PAP by placing particles into a measuring cylinder filled with in situ seawater and timing how long it took each particle to pass a discrete point (Cavan et al., 2015). At the ETNP a FlowCAM was used to measure fast particle sinking rates (Bach et al., 2012). All slow sinking particle rates were calculated using the SETCOL method (Bienfang, 1981). Fluxes (mg C m${}^{-\mathrm{2}}$ d${}^{-\mathrm{1}})$ were calculated by dividing the mass of POC (mg) by the area of the MSCs (m${}^{\mathrm{2}})$ and the sinking time of the particles (d) (Cavan et al., 2015). Primary production (PP) was estimated from 8-day satellite-derived data using the Vertically Generalized Productivity Model (Behrenfeld and Falkowski, 1997) applied to MODIS data.

Model output

The ecosystem model MEDUSA (Yool et al., 2013) was used for this study as it distinguishes detrital fluxes in two pools: fast and slow sinking. In MEDUSA, fast sinking particles are assumed to sink more rapidly than the time step of the model and are remineralized instantaneously at all vertical levels with the flux profile determined by a ballast model (Armstrong et al., 2002). Slow sinking particles sink at 3 m d${}^{-\mathrm{1}}$ and remineralization is temperature dependent, with zooplankton grazing on slow sinking particles but not on the fast sinking particles. Zooplankton DVM is not parameterized. Primary production is modelled as non-diatom and diatom production, which is summed to give the total depth-integrated primary production. The model was run in hindcast mode at $\mathrm{1}/\mathrm{4}$${}^{\circ }$ spatial resolution and output saved with a 5-day temporal resolution. The model output was extracted at the same locations and times as the observations were made and averaged over 12 years (1994–2006) to give the climatological seasonal cycle. The model outputs fluxes of particulate organic nitrogen (mg N m${}^{-\mathrm{2}}$ d${}^{-\mathrm{1}})$ which are converted to POC (mg C m${}^{-\mathrm{2}}$ d${}^{-\mathrm{1}})$ using the Redfield ratio (Redfield, 1934).

Data manipulation

For both the observations and the model output the fast and slow sinking fluxes were summed to calculate the total sinking POC flux. Model output was available at fixed depths of 100 and 200 m, which introduces an offset with our at-sea observations (Table S1). This study is therefore assessing BCP${}_{\text{eff}}$ in the upper ocean only. Parameters calculated to test the efficiency of the biological carbon pump were the percentage contribution of fast and slow sinking particles to the total sinking flux, particle export efficiency (PE${}_{\text{eff}}$), the attenuation of flux with depth expressed as $b$ and $z^{\ast }$, and transfer efficiency ($T_{\text{eff}}$).

PE${}_{\text{eff}}$ is the proportion of surface-produced primary production (PP) that is exported out of the mixed layer (observations) or at 100 m (model) and is calculated by dividing the exported flux by PP. To estimate the attenuation of flux over the upper mesopelagic zone the exponents $b$ (Martin et al., 1987) and $z^{\ast }$ (Buesseler and Boyd, 2009) were calculated, where fluxes at the export depth and 100 m below were used for observations and fluxes at 100 and 200 m from the model. The $b$ exponent is dimensionless and generally ranges from 0 to 1.5, with low values indicating low attenuation, and thus low remineralization, and higher values representing high attenuation and remineralization. The $z^{\ast }$ (m) exponent is the remineralization length scale, or the depth by which only 37 % of the reference flux (here at the export depth) remains. Thus, a large $z^{\ast }$ suggests low attenuation and low remineralization of the particle flux. The $T_{\text{eff}}$ is another parameter that represents how much flux reaches the deeper ocean and hence is not remineralized. This is simply calculated by dividing the deep flux (125–220 m in observations and 200 m in model) by the export flux. All indices are dimensionless apart from $z^{\ast }$, which is in metres.

Results and discussion

Comparison of fluxes

We compare model output with satellite-derived estimates of primary production (PP), POC export, and deep (150–300 m) fluxes in the upper ocean (Fig. S3). Overall, modelled PP compares well compared to satellite-derived estimates, with a strong positive correlation between the two ($p$ < 0.001, $r^{\mathrm{2}}=0.84$, Fig. S3a), although the model slightly overestimates PP. When comparing the total sinking export fluxes and total deep fluxes, most points lie below the 1 : 1 line, suggesting that the model is overestimating POC flux (Fig. S3b and c).

Observed particles

In all three regions particles were classified as PDAs or FPs using the images taken. PDAs were of a similar aesthetic nature in all three areas (Fig. 2) consisting of unidentifiable (to phytoplankton species level) detrital material. In the SO FPs were either from krill which form long chains of pellets (Fig. 2b) or copepods (Wilson et al., 2008). At PAP and in the ETNP, only copepod FPs were observed (Fig. 2d and f). FPs dominated particle flux in the SO but PDAs dominated at PAP and in the ETNP (Fig. S1).

Particle images from the three different regions: the Southern Ocean (a, b), the PAP site (c, d), and the ETNP (e, f). Panels (a, c, e) are phytodetrital aggregates and panels (b, d, f) are faecal pellets. Panel (b) is a chain of krill pellets from the SO and panels (d, f) are copepod pellets. Scale bars are 0.5 mm for the SO and PAP images (a–d) and 0.1 mm for the ETNP (e, f). A stereomicroscope was used in the SO, a compound microscope at PAP, and a FlowCAM in the ETNP, giving rise to the different background colours and shades.

[Figure omitted. See PDF]

Export production

The traditional view of export production is that as PP increases, so does POC export out of the mixed layer (Laws et al., 2000). However, recent analyses from the SO observe the opposite relationship, that an inverse relationship between PE${}_{\text{eff}}$ and PP exists (Cavan et al., 2015; Maiti et al., 2013; Le Moigne et al., 2016). We find that for fast sinking particles the model shows PE${}_{\text{eff}}$ increases with PP (Fig. 3a) according to a power law function ($p$ < 0.001, $r^{\mathrm{2}}=0.6$), while the observations show an inverse relationship (logarithmic function, $p$ < 0.001, $r^{\mathrm{2}}=0.4$), even when including sites outside of the SO. This inverse relationship was preserved for the SO when including conversion factors of Manno et al. (2015) (Fig. S2b).

Primary production against particle export efficiency (PE${}_{\text{eff}}$) for (a) fast sinking, (b) slow sinking, and (c) total sinking particles. Blue circles are Southern Ocean, red squares PAP, and orange triangles equatorial Pacific. Filled circles and solid black lines show model output and open circles and dashed lines are observations. All fitted lines are statistically significant to at least the 95 % level (see text for details).

[Figure omitted. See PDF]

However, for the slow sinking particles the model shows an inverse relationship between PP and PE${}_{\text{eff}}$, similar to that seen in the observations for the fast sinking particles (power law function, $p$ < 0.001, $r^{\mathrm{2}}=0.97$, Fig. 3b). Potential reasons for an inverse relationship between PP and PE${}_{\text{eff}}$ include the temporal decoupling between primary production and export (Salter et al., 2007), seasonal dynamics of the zooplankton community (Tarling et al., 2004), or grazing by zooplankton (Cavan et al., 2015; Maiti et al., 2013; Le Moigne et al., 2016). As previously mentioned, one of the differences between the fast and slow sinking detrital pools in the model is that slow sinking particles are grazed on by zooplankton and fast sinking are not. Thus, when zooplankton graze on particles in the model an inverse relationship between PE${}_{\text{eff}}$ and PP exists and when zooplankton grazing is not accounted for, the opposite occurs. This highlights the importance of zooplankton in determining the efficiency of the BCP.

The observed slow sinking PE${}_{\text{eff}}$'s were generally very low (< 0.05) and thus had little influence on the PE${}_{\text{eff}}$ for total sinking POC flux, which also had a non-linear inverse relationship with PP ($p$ < 0.001, $r^{\mathrm{2}}=0.4$, Fig. 3c). It is important to note that high values of PP (> 1000 mg C m${}^{-\mathrm{2}}$ d${}^{-\mathrm{1}})$ were only present at PAP, and that the SO had the greatest range of PP, so drives a large part of the inverse relationship. Therefore, measuring PE${}_{\text{eff}}$ in other regions with large PP ranges is fundamental to see whether this relationship holds outside the sites from this study.

Contribution of fast and slow sinking POC fluxes

Particles naturally sink at different rates, with one operational definition being that slow sinking particles sink at < 20 m d${}^{-\mathrm{1}}$ and fast sinking particles at > 20 m d${}^{-\mathrm{1}}$ (Riley et al., 2012). Most sediment traps cannot separately measure fluxes of fast and slow sinking particles and are unlikely to capture much of the slow sinking flux due to their deployment in the lower mesopelagic and bathypelagic zones (Buesseler et al., 2007; Lampitt et al., 2008). Slow sinking particles may sink too slowly and be remineralized too quickly to reach the deep ocean unless they are formed there. Hence, the MSC is a useful tool to analyse the two sinking fluxes separately.

In both the model and the observations, the slow sinking flux was consistently smaller than the fast sinking flux and generally only contributed < 40 % of the total flux (Fig. S4). However, in the model the proportion of slow sinking flux always decreases with depth (Fig. S4a–c), whereas observations at the PAP site showed the proportion of slow sinking fluxes increased with depth (Fig. S4e). Increases in slow sinking particles with depth must be from the fragmentation of larger fast sinking particles either abiotically (Alldredge et al., 1990) or from sloppy feeding by zooplankton (Lampitt et al., 1990) or advection or mixing (Dall'Olmo et al., 2016). Sloppy feeding results in zooplankton fragmenting particles into smaller particles resulting in a larger surface area to volume ratio increasing colonization by microbes and thus remineralization (Mayor et al., 2014). Zooplankton do not graze on fast sinking particles in the model; hence, neither sloppy feeding nor abiotic fragmentation are represented (Yool et al., 2013). This likely explains why the contribution of slow sinking particles can only decrease with depth in the model, unlike the observations in which slow sinking particles may increase with depth.

Attenuation of POC with depth

The attenuation of POC through the water column describes how quickly POC fluxes are remineralized, with a high attenuation indicating high POC remineralization. We used the parameters $b$ (Martin et al., 1987) and $z^{\ast }$ (Boyd and Trull, 2007) to describe the attenuation of flux with depth. A recent study suggests POC remineralization is temperature dependent (Marsay et al., 2015); hence, we compared the attenuation coefficients with temperature. Calculated mean $b$ and $z^{\ast }$ values for total (fast $+$ slow) sinking POC from the model were similar at all sites (Fig. 4a and b) with no correspondence with temperature, even though slow sinking particles are remineralized as a function of temperature in the model. Hence, slow sinking $b$ and $z^{\ast }$ increase and decrease respectively with temperature (Table S2). The observations (for total sinking particles) show a non-linear relationship with temperature that deviates away from the Marsay et al. (2015) regression, such that remineralization increases (high attenuation) at temperatures greater than 13 ${}^{\circ }$C. The variability is much greater in the observations than the model, a feature that is consistent across all indices (Fig. 4a and b). Apart from at the ETNP, where the model and observations agree, the observations consistently show slower POC attenuation compared to the model. The active transfer of POC to depth via diel vertical migration (DVM) of zooplankton (Wilson et al., 2008) may contribute to the observed slower rates of POC attenuation. Cavan et al. (2015) showed that high SO $b$ values were a result of DVM, a process not parameterized in the MEDUSA model. Although active transfer via DVM is a complex process that may be difficult to model, it is potentially important to include in biogeochemical models, as it has been shown to account for 27 % of the total flux in the North Atlantic (Hansen and Visser, 2016).

Total sinking POC attenuation coefficients (a) $b$ and (b) $z^{\ast }$ with temperature. Blue circles are Southern Ocean, red squares PAP, and orange triangles equatorial Pacific. Filled points show model output and open points are observations. Solid line is Marsay et al. (2015) regression. Error bars are standard error of the mean and only plotted on the observations as the error is too small in the model. See Table S2 for attenuation coefficients of fast and slow sinking particles.

[Figure omitted. See PDF]

The strong alignment of the modelled and observed attenuation at the ETNP is likely because of the lack of particle processing by zooplankton, by design in the model and naturally in OMZs. The daytime depth of vertically migrating zooplankton is reduced in OMZs due to low dissolved oxygen concentrations (Bianchi et al., 2013), which at the ETNP reach < 2 $\mathrm{\mu }$mol kg${}^{-\mathrm{1}}$ by 120 m. Further, the population of zooplankton below this depth is almost non-existent in OMZs (Wishner et al., 2013) and those that are there feed on particles at the surface, not in the OMZ core (Williams et al., 2014). Thus, zooplankton consumption and manipulation of particles is greatly reduced in OMZs and is non-existent in the MEDUSA model.

Efficiency of the biological carbon pump

To calculate BCP${}_{\text{eff}}$ (proportion of mixed layer primary production found at depth, here 150–300 m) we replicated the BCP${}_{\text{eff}}$ plots of Buesseler and Boyd (2009) by plotting PE${}_{\text{eff}}$ against transfer efficiency ($T_{\text{eff}}$) for fast, slow, and total sinking particles (Fig. 5). According to the observations, the SO had the highest total sinking BCP${}_{\text{eff}}$ at 40 %, similar to the maximum observed by Buesseler and Boyd (2009) in the North Atlantic. The SO observations showed a higher BCP${}_{\text{eff}}$ than the model by about 10 % across all sinking fluxes (Fig. 5). This difference was largely due to a very high $T_{\text{eff}}$ (> 1) estimated from observations, which implies fluxes increased at depth. This could be due to active fluxes by vertically migrating zooplankton, possibly krill (Cavan et al., 2015). Active fluxes could account for high observed $T_{\text{eff}}$ in the slow sinking particles, as well as fragmentation of larger particles at depth (Mayor et al., 2014).

Efficiency of the biological carbon pump for (a) fast, (b) slow and (c) total sinking particles. Particle export efficiency (PE${}_{\text{eff}}$) is plotted against transfer efficiency ($T_{\text{eff}}$). Contours represent BCP${}_{\text{eff}}$ (proportion of primary production reaching depths of 150–300 m). Blue circles are Southern Ocean, red squares PAP and orange triangles equatorial Pacific. Filled points show model output and open points are observations.

[Figure omitted. See PDF]

$T_{\text{eff}}$ is often thought to be controlled by the dominant phytoplankton group in the upper ocean, which is linked to the ballasting hypothesis (Francois et al., 2002; Henson et al., 2012). $T_{\text{eff}}$ for total POC flux estimated by the model (Fig. 5c) is roughly the same in all three regions, even though the MEDUSA model is capable of altering the ratio of diatom to non-diatom PP (Yool et al., 2013), as would be expected when comparing the SO and the equatorial Pacific. Diatoms often dominate the SO (Salter et al., 2007), whilst the ETNP is dominated by pico- and nanophytoplankton (Puigcorbé et al., 2015), and at the PAP site a range of phytoplankton species are found from diatoms to smaller cyanobacteria and dinoflagellates (Smythe-Wright et al., 2010).

In this study we observe the opposite trend in $T_{\text{eff}}$ compared to Henson et al. (2012) with the SO exhibiting high $T_{\text{eff}}$ (> 1) and the ETNP the lowest ($\sim$ 0.5). This could be due to the depth range over which $T_{\text{eff}}$ is calculated as here it is relatively shallow (< 200 m) compared to the 2000 m range in Henson et al. (2012). For instance, Henson et al. (2012) estimate $T_{\text{eff}}$ of $\sim$ 0.4 in the ETNP, very close to the calculated values in this study, which, due to the hypothesized lack of zooplankton interactions with particles in the deep OMZs, could remain unchanged if calculated at 2000 m depth. However, in the SO zooplankton influence particle transfer heavily, and thus if we had observed $T_{\text{eff}}$ at 2000 m, the observed $T_{\text{eff}}$ in the SO may have been much lower, and thus our results from the SO would conform to the finding that at high latitudes PE${}_{\text{eff}}$ is high and $T_{\text{eff}}$ is low (Henson et al., 2012). This highlights the potential complex interactions between particles and zooplankton in the upper mesopelagic zone which may be missed in deep ocean particle studies.

Even though the PAP site had the highest PP, the BCP${}_{\text{eff}}$ was lowest (< 15 %). There were also large differences (up to 15 %) in the BCP${}_{\text{eff}}$ between the model and the observations at the PAP site driven by large discrepancies in PEeff. Observations of fast sinking PE${}_{\text{eff}}$ were much lower than predicted by the model (Fig. 5a), which we suggest could result from active grazing and fragmentation of fast sinking particles by zooplankton. $T_{\text{eff}}$ of fast sinking particles was low and consistent with model predictions, suggesting that active transfer via DVM (not parameterized in the model) plays a relatively minor role at the PAP site. Therefore, mineral ballasting (Armstrong et al., 2002), which drives $T_{\text{eff}}$ in the model, may be the main driver of $T_{\text{eff}}$ at PAP. The modelled and observed slow sinking BCP${}_{\text{eff}}$ was similar at PAP ($\sim$ 1 %) despite a large difference in $T_{\text{eff}}$ (Fig. 5b). Fragmentation of fast to slow sinking particles (not included in the model) at depth could explain the difference in slow sinking $T_{\text{eff}}$.

Finally, the BCP${}_{\text{eff}}$ for the ETNP is very similar between the model and observations for all sinking fluxes (Fig. 5). The similarity in the BCP${}_{\text{eff}}$ here echoes the similarity shown for POC attenuation with depth. This reiterates our hypothesis that the model and observations agree on BCP${}_{\text{eff}}$ only in areas of the global ocean where processing of particles by zooplankton is reduced due to very low dissolved oxygen concentrations.

Conclusions

We have used observations and model output from the upper mesopelagic zone in three contrasting oceanic regions to assess the influence of zooplankton on the efficiency of the biological carbon pump. We separately collected in situ fast and slow sinking particles, which are also separated into discrete classes in the MEDUSA model. The model has limited processing of particles by zooplankton with only slow sinking detrital POC being grazed upon.

Our results highlight the crucial role that zooplankton play in regulating the efficiency of the biological carbon pump through (1) controlling particle export by grazing, (2) fragmenting large, fast sinking particles into smaller, slower sinking particles, and (3) active transfer of POC to depth via diel vertical migration. Comparisons of the model and observations in an OMZ provide strong evidence of the importance of zooplankton in regulating the BCP. Here extremely low dissolved oxygen concentrations at depth reduce the abundance and metabolism of zooplankton in the mid-water column. Thus, the ability of zooplankton to degrade or repackage particles is vastly reduced in OMZs, and as such it is here that the model, with limited zooplankton interaction with particles, shows the strongest agreement with observations.

We recommend that grazing on large, fast sinking particles and the fragmentation of fast to slow sinking particles (either via zooplankton or abiotically) is introduced into global biogeochemical models, with the aim of also incorporating active transfer. Future changes in climate such as the expansion of OMZs may decrease the role of zooplankton in the biological carbon pump globally, increasing its efficiency and hence forming a positive climate feedback.

Data availability

Data are available for the PAP cruise at 10.1594/PANGAEA.833364 (Cavan and Belcher, 2014) and for the SO cruise by request through the British Oceanographic Data Centre, http://www.bodc.ac.uk (Cavan and Sanders, 2014).

The Supplement related to this article is available online at doi:10.5194/bg-14-177-2017-supplement.

Acknowledgements

We would like to thank all participants and crew on cruises JR274, JC087, and JC097. Thanks to Annike Moje for running all POC samples in Bremen, Germany. Thanks also to Andrew Yool for providing the MEDUSA model output. Data from the Southern Ocean are stored in the British Oceanographic Data Centre and for the PAP site at PANGAEA online. Support for this research was provided by NERC through the NC PAP programme, Ocean Acidification programme (NE/H017348/1), and NERC standard grant NE/E01559X/1. Edited by: G. Herndl Reviewed by: two anonymous referees