The PlankTOM10 dynamic green ocean model is a global ocean biogeochemistry model that includes plankton ecosystem processes based on the representation of 10 PFTs and their interactions with the environment. PlankTOM10 incorporates six autotrophic and four heterotrophic PFTs: picophytoplankton (pico-eukaryotes and non N2-fixing cyanobacteria such as Synechococcus and Prochlorococcus), N2-fixers (Trichodesmium and N2-fixing unicellular cyanobacteria), coccolithophores, mixed phytoplankton (e.g. autotrophic dinoflagellates and chrysophytes), diatoms, colonial Phaeocystis, bacteria (here used to subsume both heterotrophic Bacteria and Archaea), protozooplankton (e.g. heterotrophic flagellates and ciliates), mesozooplankton (predominantly copepods), and crustacean macrozooplankton (euphausiids, amphipods, and others, called “macrozooplankton” for simplicity; Fig. 1). Gelatinous macrozooplankton are not included in the model. Diversity within groups is not considered, and the physiological parameters for each PFT are the same everywhere in the ocean, although some are dependent on environmental conditions (i.e. nutrients, light, food, temperature).

The current version of the PlankTOM10 model was developed from the model of Buitenhuis et al. (2013a), using the strategy for regrouping PFTs described by Le Quéré et al. (2005). It does not include new equations for growth and loss terms compared with previous versions of the PlankTOM model, but it includes an additional trophic level in the zooplankton PFTs (i.e. macrozooplankton). Parameterisations are based on more data related to the vital rates of individual PFTs, where new information was available. Previous studies have shown that model results are highly sensitive to PFT growth rates (Buitenhuis et al., 2006, 2010), and considerable effort was made to constrain these rates using observations from LaRoche and Breitbarth (2005), Bissinger et al. (2008), Buitenhuis et al. (2008, 2010), Sarthou et al. (2005), Schoemann et al. (2005), Rivkin and Legendre (2001), Hirst and Bunker (2003), and Hirst et al. (2003).

The complete set of model equations and parameter values are provided in the Supplement. Here, we describe the elements that are most important for the analysis of the Southern Ocean and the strategy used to determine parameter values for PFT growth and loss processes.

PlankTOM10 simulates the growth of ten PFTs in response to environmental conditions. The PFT biomasses are produced by the model for each grid box based on the growth and loss term equations presented in the Supplement. The model includes three detrital pools: large and small particulate organic matter, and semi-labile dissolved organic matter. The sinking rate of large particles is based on the mineral (ballast) content of particles following Buitenhuis et al. (2001), while the sinking rate of small particles is constant at 3 m d-1. The model includes full cycles of carbon (C), oxygen (O2), and phosphorus (P), which are assimilated and released by biological processes at a constant ratio of 122 : 172 : 1 (Anderson and Sarmiento, 1994). Phytoplankton and particulate organic matter have a variable Fe / C ratio, while zooplankton and bacteria have a fixed ratio of 2 × 10-6, which is lower than the minimum phytoplankton Fe / C ratio (Schmidt et al., 1999). Zooplankton and bacteria relelase excess iron. The model also includes a full cycle of silica (Si) and calcite (CaCO3) as in Maier-Reimer (1993), and simplified cycles for Fe and nitrogen (N). CO2 and O2 are exchanged with the atmosphere using the gas exchange formulation of Wanninkhof (1992). The Fe cycle is represented as in Aumont and Bopp (2006). Iron is deposited with dust particles using the monthly fields of Jickells et al. (2005), the Fe content of dust is assumed to be 3.5 % everywhere. We assume an Fe solubility from dust of 1 % (Jickells et al., 2005). Iron is also delivered to the ocean via river fluxes following the outflow scheme of da Cunha et al. (2007) with 95 % sedimentation in estuaries. Dissolved inorganic nitrogen (DIN) is the sum of nitrate and ammonium. The N : P ratio of organic processes is set to the Redfield ratio of 16 : 1. N2-fixers can use N2 and thus have access to unlimited N from the atmosphere.

The growth rate parameters for the ten PFTs in PlankTOM10 are based on a compilation of growth rates as a function of temperature (Sect. 2.2). Phytoplankton PFT growth rates are also limited by light and inorganic nutrients (P, N, Si, and Fe) using a dynamic photosynthesis model that represents the two-way interaction between photosynthetic performance and Fe / C and Chl / C ratios (Buitenhuis and Geider, 2010). Light limitation is constrained by the slope of the photosynthesis-irradiance curve (α) and the maximum Chl / C ratio (θmax). We could not distinguish PFT-specific values for α (Geider et al., 1997) and used a mean value of 1.0 mol C m2 (g Chl mol photons)-1 for all PFTs. Observed θmax for diatoms are systematically higher than those of other PFTs (Geider et al., 1997). There are too few direct observations to parameterize θmax for other PFTs, so we fitted the observations (Geider et al., 1997) for θmax to the maximum growth rate (μmax) presented in that paper. The fit showed θmax increasing with growth rate (n= 19, p= 0.02). We thus used a θmax higher than average for Phaeocystis and diatoms, and a lower than average θmax for N2-fixers.

We used a two-step approach to define the nutrient limitation parameters, which are not well constrained by observations. Firstly, we assigned initial PFT-specific half-saturation values to each phytoplankton PFT based on literature-derived values, using the value for a similar-sized PFT when PFT-specific information was not available. We then examined the covariations of surface Chl concentrations with the limiting nutrient concentrations as shown in Fig. 3, and adjusted the magnitude of the half-saturation parameters of phytoplankton PFT to approximately fit the observations, keeping the ratios of k half values between phytoplankton PFTs approximately the same as the initial ratios. With this approach, we use the observed k half values as an initial starting point but tune the model to match the emerging properties highlighted in Fig. 3.

Initial values for the half-saturation concentrations of P (kP) and N (kN) for phytoplankton growth rates were based on observations. For N2-fixers, coccolithophores and diatoms, the half-saturation values for growth were computed using the half-saturation values of uptake reported in Riegman et al. (1998), LaRoche et al. (2005), and Sarthou et al. (2005) multiplied by the minimum/maximum N : C ratio (0.33) to account for the acclimation of nutrient saturated vs. nutrient limited growth (Morel, 1987). For picophytoplankton, reported values for the half saturation extend over 3 orders of magnitude. We assigned low half-saturation values as these organisms grow even under very low nutrient conditions (Timmermans et al., 2005). For mixed phytoplankton, we assigned a value intermediate between picophytoplankton and diatoms. For Phaeocystis, we used half-saturation values that characterise colonies (Schoemann et al., 2005). The selected set of parameter values shown in Fig. 3 is reported in Table 2.

Iron uptake was computed using a cell quota model (Buitenhuis and Geider, 2010; Geider et al., 1997), where the Fe uptake by phytoplankton PFTs is explicitly regulated by the light conditions. The three parameters needed are the minimum, the maximum and the optimal Fe quotas. The minimum and maximum quotas were set at the same value of 2.5 and 20 µmol Fe / mol C for all PFTs based on the analysis of Buitenhuis and Geider (2010). The optimal quota was set to the minimum quota plus 2∗μ20max based on Sunda and Huntsman (1995) for all PFTs. In addition, phytoplankton PFTs also respond to the concentration of Fe in water which is parameterised with a half-saturation constant. The half saturation of Fe uptake (kFe) is lower for picophytoplankton (Timmermans et al., 2005) than other phytoplankton, and higher for N2-fixers (LaRoche and Breitbarth, 2005) and diatoms (Sarthou et al., 2005). Intermediate values for kFe have been reported for the other phytoplankton PFTs (Le Vu, 2005; Schoemann et al., 2005). The selected set of parameter values after adjustments produces no systematic covariation between Chl and Fe, as observed (Fig. 3, Table 2).

The half-saturation parameters of zooplankton grazing rate were initially based on the relationship between metabolic rates and body volume of Hansen et al. (1997). We used the same approach as for nutrient limitation of the phytoplankton PFTs, and adjusted the half-saturation parameters for grazing based on the observed covariations between surface Chl concentrations and zooplankton biomass (Fig. 3). The selected set of parameter values that approximately fit the observed covariations in Fig. 3 is reported in Table 2.

Zooplankton food preferences were assigned based on predator–prey size ratio (Table 3), as there were insufficient data to determine these parameters directly across the range of zooplankton and phytoplankton considered here. This approach assumes that protozooplankton generally have a high preference for bacteria and a low preference for diatoms, that mesozooplankton have a higher preference for protozooplankton and a low preference for N2-fixers and bacteria, and macrozooplankton have a lower preference for N2-fixers, picophytoplankton and bacteria than other groups. Although some data were available to characterise grazing on Phaeocystis spp. (Nejstgaard et al., 2007), it is not used specifically here because it required knowledge on the life forms of Phaeocystis in situ. We assume that all zooplankton graze on organic particles (Table 3) but prefer to graze on other PFTs. The weighing factors influenced primarily the biomass of the prey and predators, but had little influence on their geographic distribution. We thus used the model results on biomass (Table 4) to guide the size of the relative preferences among PFTs for each grazer.

The gross growth efficiency (the part of grazing that is incorporated into biomass) was defined based on the mean across available observations: 0.21 for bacteria (data from Rivkin and Legendre, 2001), and 0.29, 0.25, and 0.30 for protozooplankton, mesozooplankton and macrozooplankton, respectively (data from Straile, 1997). Respiration and mortality parameters were based on observations from Buitenhuis et al. (2010) for protozooplankton, Buitenhuis et al. (2006) for mesozooplankton, and Moriarty (2013) for macrozooplankton. The temperature-dependence of respiration and mortality was fitted to all data as for the growth rate (Sect. 2.2), except for the mortality of macrozooplankton and mesozooplankton. There are nine observations on macrozooplankton mortality and we tuned this term based on the resulting biomass. The fitted relationship for the mortality of mesozooplankton was reduced by a factor of ∼ 2 to account for the explicit mortality from macrozooplankton represented in the model. This correction preserves the temperature-dependence of mortality, but it recognises that explicit grazing by macrozooplankton already takes place in the model, which does not represent the grazing by other organisms (e.g. salps, fish larvae). In total, grazing accounts for 2/3 to 3/4 of the mortality of mesozooplankton (Hirst and Kiorboe, 2002).

The most important trait that distinguishes the various PFTs is the rate at which they grow under different conditions (Buitenhuis et al., 2006, 2010). We compiled maximum growth rates as a function of temperature (Table 1). We fit an exponential growth relationship to the observations by optimising the relation μT=μ0×Q10T/10 where T and μT are the observed temperature and associated growth rate, μ0 is the growth at 0 ∘C, and Q10 is the derived temperature-dependence of growth (Table 1). The parameter values for μ0 and Q10 were estimated by minimising the error, quantified as the least squares cost function Σ((μT-μobsT)/μobsT)2. Normalising to observations helps ensure a good fit of μT in cold waters where growth rates are low. We used exponential growth, rather than a temperature-optimal growth, to avoid biases caused by the lack of observations for some PFTs at low or high temperatures. The p-value of a linear regression between observations and the exponential fit (Table 1) provides a measure of how well the relationship is constrained by the observations. The fit assigns equal weight for all the data, rather than following the 99 % quantile (e.g. Eppley, 1972; Bissinger et al., 2008) to provide a better representation of the mean community for each PFT.

Growth rate parameters estimated with this method are well constrained (p values < 0.05) for seven of the ten PFTs, including all of the heterotrophic PFTs (Table 1). There are insufficient data to provide significant constraints on the growth rates of N2-fixers (p= 0.76), and some uncertainty in the growth data for coccolithophores (p= 0.06) and Phaeocystis (p= 0.23; Table 1). However, the growth of N2-fixers is less than that of other phytoplankton PFTs (Fig. 2), and the fitted relationship produces μT less than that of other PFTs despite these uncertainties. An exponential function may not be appropriate for growth rates of coccolithophores and Phaeocystis (Schoemann et al., 2005). The growth rate of coccolithophores was overestimated at low temperatures due to high growth rates at 20∘ C and the absence of observations for temperatures below 5 ∘C. We reduced the fitted growth rate of coccolithophores linearly to 0 below 10 ∘C to match the observed reduced coccolithophore biomass in cold regions (O'Brien et al., 2013).

We used relationships between observed concentrations of Chl and both inorganic nutrients (e.g. NO3, PO4 and Fe), and zooplankton biomasses (protozooplankton, mesozooplankton and macrozooplankton; Fig. 3) to provide additional constraints on model parameters. Specifically, we used observations for in situ NO3 and PO4 concentrations from the World Ocean Atlas 2009; in situ Fe concentration data from Tagliabue et al. (2012); protozooplankton biomass data from Buitenhuis et al. (2010); mesozooplankton biomass data from Buitenhuis et al. (2006); macrozooplankton biomass data from Atkinson et al. (2004) and Moriarty et al. (2013). All the data were binned into 1 × 1∘ grid boxes. Most observations are for the surface ocean. Mesozooplankton and macrozooplankton data are from depth-integrated tows of typically 200 m depth and may underestimate surface concentrations (by a factor 1.5–2 based on our model simulations). All data are monthly except for mesozooplankton, which are seasonal. Chl concentration is from SeaWiFS satellite averaged over 1998–2009 and interpolated to the same grid. The model output was averaged over the same time period, and sampled for the same month and on the same grid box as the observations. The data intervals were chosen to include approximately the same number of grid boxes, except for macrozooplankton where the lowest interval was set to 0–0.05 µmol C L-1 because of the large number of grid boxes with very low macrozooplankton concentration. Ten concentration intervals were used for the nutrients (Fig. 3).

Chlorophyll concentrations covary with NO3 concentrations at < 3 µmol L-1, and with PO4 in the range 0.3–0.5 µmol L-1 (Fig. 3; Spearman ranked correlations for data in the 25–75 % interquartile range give r= 0.72 for NO3 and r= 0.73 for PO4). These relationships are consistent with our understanding of the growth limitation of phytoplankton in the subtropics, where NO3 and PO4 concentrations are low. There is no observed covariation between Chl and Fe concentration (r= -0.16). The strongest covariations are between Chl and protozooplankton at concentrations < 0.6 µmol C L-1 (r= 0.83) and between Chl and mesozooplankton at concentrations < 0.3 µmol C L-1 (r= 0.77). There is no covariation between Chl concentration and macrozooplankton biomass (r= -0.19; Fig. 3). We use these relationships to tune the growth limitation parameters in the model, so that the functional relationships between Chl and nutrients or zooplankton are close to the observed relationships overall.

PlankTOM10 is coupled to the Ocean General Circulation Model (OGCM) NEMO version 3.1 (NEMOv3.1). We used the global configuration (Madec and Imbard, 1996), which has a resolution of 2∘ of longitude and a mean resolution of 1.5∘ of latitude, with enhanced resolution up to 0.3∘ in the tropics and at high latitudes. The model resolves 30 vertical levels, with 10 m depth resolution in the upper 100 m. NEMOv3.1 calculates vertical diffusion explicitly and represents eddy mixing using the parameterisation of Gent and McWilliams (1990). The model thus generates its own mixed-layer dynamics and associated mixing based on local buoyancy fluxes and winds. NEMOv3.1 is coupled to a dynamic-thermodynamic sea-ice model (Timmermann et al., 2005).

PlankTOM10 is initialised from observations of dissolved inorganic carbon (DIC) and alkalinity from Key et al. (2004), O2 and nutrients from Garcia et al. (2006a, b), and temperature and salinity from the World Ocean Atlas 2005 (Antonov et al., 2006; Locarnini et al., 2006). Fe is initialised with a constant concentration of 0.6 nmol Fe L-1 north of 30∘ S and 0.2 nmol Fe L-1 in the Southern Ocean, consistent with observations (Parekh et al., 2005; Tagliabue et al., 2012). The PFTs equilibrated within 3 years and were not influenced by initialisation. The model is forced by daily winds and precipitation from the ECMWF interim reanalysis (Simmons et al., 2006) from 1989 to 2009. Results for standard simulations are averaged over 1998–2009. A series of sensitivity tests are presented for the model parameters that influence the key results the most.

To understand the interaction pathways among ecosystems, biogeochemistry and climate, we developed a simplified version of the model that included only six PFTs (PlankTOM6) (Fig. 1). PlankTOM6 is identical to PlankTOM10, except that the growth rates of N2-fixers, mixed phytoplankton, Phaeocystis, and macrozooplankton are zero, and the mortality of the mesozooplankton is increased to account for the lack of macrozooplankton predation until the point when primary production is at its maximum. Given the otherwise similar model structure, parameters, initialisation and simulation protocol, comparison of results from PlankTOM6 and PlankTOM10 provides information on the specific roles of zooplankton dynamics in the model.

The data show systematic patterns in growth rates that differ among PFTs. The growth rates of all PFTs increase with increasing temperature, but not to the same extent (Fig. 2). The growth rate of phytoplankton PFTs increases with PFT size, from 0.15 d-1 for N2-fixers to 1.87 d-1 for Phaeocystis, and the growth rate of heterotrophic PFTs decreases with size, from 1.22 d-1 for bacteria to 0.19 d-1 for macrozooplankton (Table 1). The sign of the relationship between growth rate and size between phytoplankton PFTs is the opposite of the sign of this relationship within specific PFTs, including diatoms (Sarthou et al., 2005), picophytoplankton (Chen and Liu, 2010) and coccolithophores (Buitenhuis et al., 2008).

PlankTOM10 reproduces the main characteristics of observed surface Chl, with high concentrations in the high latitudes and low concentrations in the subtropics, higher Chl concentration in the Northern compared to the Southern Hemisphere, and in the South Atlantic compared to the South Pacific Ocean (Fig. 4). The global biogeochemical fluxes simulated by PlankTOM10 are generally below or at the low end of the range of observed values (Table 4), with global primary production of 42.4 PgC yr-1, export production of 7.6 PgC yr-1, export of CaCO3 and SiO2 of 0.4 PgC yr-1 and 2.9 PgSi yr-1, respectively, and N2 fixation of 165 TgN yr-1.

PlankTOM10 produces distinctive geographical distributions of carbon biomasses among PFTs (Fig. 5). About a third of the phytoplankton biomass occurs as picophytoplankton, followed in descending abundance by diatoms and Phaeocystis, mixed phytoplankton, coccolithophores and N2-fixers (Table 4). This distribution is broadly consistent with observations (Buitenhuis et al., 2013b), but the simulated phytoplankton biomass is generally on the low side of the observational range, which is consistent with the results of the global biogeochemical fluxes. The simulated biomass of coccolithophores is overestimated (i.e. 0.077 compared with 0.001–0.032 PgC), although CaCO3 export is underestimated, suggesting either that the model calcification or aggregation rates are too low or that zooplankton calcifiers contribute significantly to CaCO3 export.

The model underestimates bacterial biomass by a factor of 10 compared with observations. This possibly reflects the fact that the model only represents highly active bacteria and a substantial fraction of observed biomass is from low activity and ghost cells. The model underestimates protozooplankton by a factor of 1.5–5 (in absolute value) or 2–3 (as a fraction of total biomass value) compared to observations (Table 4). This discrepancy could be caused by the underestimation of bacterial biomass, as bacteria are an important source of food for protozooplankton. The simplified representation of the range of protozooplankton grazers in a single PFT representing both heterotrophic nanoflagellates and microzooplankton could also play a role. Simulated mesozooplankton biomass is only slightly below the observed range, while simulated macrozooplankton biomass is within the observed range, although the uncertainty here is large (0.010–0.64 PgC). Overall the balance is slightly skewed towards relatively more biomass than observed in the larger zooplankton (53 % compared to 3–47 %) compared to the smaller zooplankton groups (13 % compared to 27–31 %; Table 4).

The geographic distribution of each simulated PFT is also distinctive (Figs. 6–7). Satellite data products indicate that small phytoplankton (picophytoplankton and N2-fixers) are generally dominant in the tropics, haptophytes (coccolithophores and Phaeocystis) in mid to high latitudes, and diatoms in high latitudes (Alvain et al., 2005; Brewin et al., 2010). The simulated phytoplankton distribution generally matches the distribution inferred from satellite normalised radiance (Fig. 6), except in the temperate zones where observations suggest a balance between picophytoplankton and haptophytes and the model shows a dominance of haptophytes. PlankTOM10 also reproduces the locations of blooms of colonial Phaeocystis and coccolithophores (Fig. 7). The simulated geographic distributions of zooplankton PFTs are particularly distinctive, with protozooplankton abundant in the tropics and subtropics, mesozooplankton at high latitudes of both hemispheres, and macrozooplankton with high biomass in the North Pacific and South Atlantic and along the coasts (Fig. 5).

The marine ecosystem as a whole appears to function realistically: mesozooplankton grazing on phytoplankton is somewhat overestimated relative to the 5.5 Pg yr estimated by Calbet (2001), so they have taken over the role of principal herbivores. Possibly the faster turnover rates of small copepods are overrepresented in the observational data on mesozooplankton, leading to a trophic position of mesozooplankton somewhat too low in the food chain. Export production, phytoplankton biomass and metazoan zooplankton biomass are realistic in the model, leading to realistic seasonal cycles, but the regenerated part of primary production is underestimated, concomitant with low protozooplankton biomass, which impacts the model on shorter timescales of days.

PlankTOM10 and PlankTOM6 generally produce similar results in surface Chl concentration, nutrient distribution, primary and export production (Fig. 8), except that PlankTOM6 fails to reproduce the observed low Chl concentration in summer in the Southern Ocean (Fig. 4; Sect. 3.4). The overall differences between the two models, quantified statistically using a Taylor distribution (Taylor, 2001), are less than 0.1 in either correlation or normalised standard deviation (Fig. 8). PlankTOM10 does slightly better than PlankTOM6 for the distribution of Chl, primary and export production, but slightly worse for the distribution of silica and nitrate, with similar performance for phosphate (Fig. 8). These differences are small in part because of the short duration of the simulations presented here (20 years), which allow equilibration of the ocean surface only. The models are generally similar also in their representations of the distribution of biomass among phytoplankton PFTs, with most of biomass being in picophytoplankton in both models (Fig. 9 and Table 4). However, PlankTOM6 allocates more biomass to protozooplankton compared to PlankTOM10, though PlankTOM6 is still at the low end of observed concentrations (Table 4).

The failure of PlankTOM6 to reproduce the observed low Chl concentration in the Southern Ocean during summer is further highlighted in Fig. 10, which shows the seasonal cycle of mean Chl for the Northern Hemisphere and the Southern Ocean, where it is most pronounced. In PlankTOM6, the seasonal cycles in the north and south are very similar, with the slightly lower concentrations in the Southern Ocean during summer caused by a slightly deeper summertime mixed-layer depth (29 m compared to 19 m). In contrast, in PlankTOM10, the seasonal cycle of Chl in the south is smaller and concentrations are always below those in the north, as is the case for observations. As PlankTOM6 and PlankTOM10 have identical physical environments (including mixed-layer depth), the north–south differences are due to ecosystem structure. In the following sections, we focus our analysis on the model parameters that influence the low Chl concentration in the Southern Ocean the most.

The observed phytoplankton biomass, including the low Chl concentrations in high-nutrient low-chlorophyll (HNLC) regions, reflects the balance between phytoplankton growth and loss. Phytoplankton growth rates vary with temperature, light, and nutrient supply, whereas losses result mainly from grazing by zooplankton, respiration, cell death, sinking to depth, and dilution by vertical mixing. Any process that reduces the net rate of increase in phytoplankton biomass (i.e. differences between growth and loss) may lead to low residual Chl concentration. For example, Platt et al. (2003a) showed that deep mixing by wind dilutes Chl in the surface layer and reduces the average irradiance experienced by the phytoplankton. This results in low growth rate and demand for nitrate, the conditions generally observed in HNLC regions. Here we further examine the consequences of high zooplankton-mediated grazing losses.

We use the north / south ratio in surface Chl concentration as a metric to quantify model performance, focusing on the Pacific Ocean where the contrast between the Northern Hemisphere and the Southern Ocean is most pronounced. This metric is simple and easy to quantify with data (geographic locations: boxes in Fig. 4). Satellite observations indicate a north / south Chl ratio of 2.16 ± 0.35 (1998–2009 mean ± 2 SD of annual values). To ensure that the ratio is not affected by potential biases in the SeaWiFS Southern Ocean data (Johnson et al., 2013), we also used in situ data from the World Ocean Atlas, which indicates a similar north / south Chl ratio of 2.0. This ratio is 1.72 ± 0.051 in the PlankTOM10 and 1.21 ± 0.074 in the PlankTOM6 simulations (Fig. 11). Controlling factors in this ratio are examined here through a set of sensitivity tests.