The “generalist” group represents all unicellular plankton including bacteria, phytoplankton, phagotrophs, and mixotrophs (organisms which combine trophic strategies, such as photosynthesis, diffusive uptake of nutrients and predation on smaller cells), but excluding diatoms. This group therefore encompasses a wide range of unicellular organisms, such as heterotrophic and photosynthetic bacteria, flagellates, dinoflagellates, and ciliates, while diatoms are simulated as a separate group (explained in the following section). The generalists are all modeled as potential mixotrophs that acquire carbon and nutrients through a combination of osmotrophy (taking up dissolved organic carbon), autotrophy (photosynthesis), and/or phagotrophy. Whether a given size group represents bacteria, phytoplankton, etc., depends on the principal mode of carbon acquisition, which is determined by a combination of cell size and the environment. The trophic strategies of each size class are therefore an emergent property and vary dynamically over space and time. The generalist spectrum is thus a trait-based model with cell size being the trait.

Large diatoms, that thrive in eutrophic environments, have historically been correlated with food webs that lead to major fisheries . Thus, the inclusion of diatoms in the model is critical for the estimation of fish production potential. Diatoms are simulated as a separate group because of their importance in marine systems (they account for about 40 % of marine primary production ), and their distinct characteristics that affect their physiology and predation mortality. Diatoms enclose a vacuole that increases their volume, allowing them to have larger physical size than other cells of the same biomass. In the model, this trait allows them to increase their volume per carbon without additional nutrient demands allowing them to have a higher per-carbon diffusive uptake than non-vacuolated cells. Further, diatoms are engulfed in a silicate shell, which makes silicate another limiting nutrient. The hard silica shell also reduces predation pressure on the diatoms . The rigidity of the hard silica shell deprives the cell of the plasticity needed to engulf prey and excludes diatoms from eating prey. Diatoms therefore cannot perform phagotrophy and only obtain nutrients through diffusive uptake. The silicate shell also induces growth limitation of diatoms by dissolved Si, besides light and nitrogen.

Multicellular zooplankton are the key link between lower and higher trophic levels, as they ingest unicellular plankton and are themselves prey for fish. Among the different zooplankton taxa in the ocean, copepods are the most abundant metazoans . Copepods undergo size changes throughout their life stages and thereby occupy different ecological niches. These ontogenetic niche shifts substantially impact population dynamics . For instance, copepod growth rates are strongly correlated with spring bloom intensity . To account for these size changes, each population of copepods is represented by a size spectrum that spans from the offspring size to the adult size.

Each population of copepods is characterized by two traits: the adult size and their feeding strategy (active or passive). In this way, the model represents the large differences in adult size observed in the oceans . Modelling the diversity of copepods with these two main traits makes it possible to represent the entire copepod community in the global ocean with just a few populations that span the trait space of adult size and feeding modes.

Dead organic matter (detritus) is represented by a size spectrum of particulate organic matter (POM) that includes dead organisms, cell fragments from lysis, and fecal pellets produced by multicellular plankton. POM may be represented by a size spectrum to enable the calculation of sinking velocities for different POM sizes. However, in the simulations presented here we simulated just a single size class of POM.

Predation is the central process that connects the plankton size classes in NUM. Each size class i is characterized by its mass mi (either cell mass or body mass in µg carbon) and the organisms in the class prey on smaller organisms mprey with a log-normal preference function: 1ϕklm,mprey=pklexp⁡-12σk2ln⁡mβkmprey2, where β is the preferred predator:prey mass ratio, σ is the width, and pkl is the preference of the predator group, i.e. copepods, k towards the prey group l, i.e., diatoms (all parameters are specified in Tables , , and  in the supplementary). As each size class represents a finite range of sizes (with m representing the geometric mean; see Fig. ), the actual preference is the average of the encounter over all prey and predator masses, so the average encounter θij between two size classes i (predator) and j (prey) becomes a more complex function (Appendix ).

The available food for size class i is the sum over all other size classes weighted by the preference (θij): Fi=∑jθijBj where Bj is the biomass concentration of prey (µg C L⁻¹). The actual consumption is limited by a mass-specific maximum consumption rate: jFmax (d⁻¹): 2jF.i=ϵFfijFmax.iwithfi=aF.iFiaF.iFi+jFmax.i, where aF.i is the mass-specific clearance rate of the predator (L(µg C d)⁻¹) and ϵF is the assimilation efficiency. The predation results in a predation mortality on the prey size class j: 3μpr.j=∑iθijjF.iBiϵFFi.

In addition to the predation mortality inflicted by larger plankton, the largest size classes are also exposed to predation by higher trophic levels that are not explicitly represented in the model, such as fish. The exposed size classes are those larger than mHTL (µg C), described by the selectivity function pHTL: 4pHTL,i=11+mi/mHTL-2. Here we use a value of the mHTL=1 µg C (≈500 µm) as the size of prey of forage fish in the size range 1–10 g wet weight Chap. 2. Further, in the standard setup presented here, we constrain the higher trophic level (HTL) mortality to act only on copepods. If a setup is run without copepods the HTL mortality would represent the predation by the copepods on the unicellular organisms. The higher trophic level mortality can either be a constant or proportional to the biomass concentration in the size class: 5μHTL,i=μHTL,0pHTL,iconstantμHTL,0pHTL,iBiln⁡1/zi``quadratic”. The density dependent mortality is termed “quadratic” to follow common nomenclature, though the mortality is in reality linear and it is only the loss term (mortality multiplied by biomass) that is quadratic. The quadratic formulation of mortality has the practical advantage that it helps stabilize the dynamics in the simulation – running without a quadratic loss term results in competitive exclusion among copepod groups and limited co-existence. We therefore use the quadratic mortality formulation in this capacity. The term 1/ln⁡(1/zi) is a correction for the number of size classes used. Size classes are evenly distributed on a logarithmic size grid and in that case the number of size classes in a fixed size range is proportional to this correction term, where zi is the ratio between the lower and the upper size range in each size class. It should be noted that higher trophic level mortality does not respect the increased prey vulnerability due to motility that predation mortality follows. This assumption follows from higher trophic level organisms being visual predators and the hydrodynamic signals from prey motility do not affect the detection of prey .

The unicellular spectra can be either generalists or diatoms. The only differences between the two spectra are that diatoms have a vacuole, that they cannot perform phagotrophy, and that they have a lowered predation risk (p<1) (as explained above). The present description of diatoms with a fixed vacuole size is a simplified version of the description with dynamic vacuoles in . The full set of equations and parameters for the unicellular model are given in Tables  and .

The cell model describes the encounter and uptake of resources and metabolism of an individual plankton cell as modulated by the cell's size (mass) m (µg C). The size of the cell defines traits, and traits in combination with the environment define its trophic strategy, whether it is an osmo-heterotroph (a bacterium feeding mainly on DOC), a light- or nutrient-limited phototroph, or, for the generalists, a mixotroph (combining osmo-, photo-, and phagotrophy) or a heterotroph (living mainly on phagotrophy) .

Regarding their morphology, we assume that cells are spherical and consist of a cytoplasm surrounded by a membrane of thickness δ (Fig. ). The diatom cells additionally have a vacuole that occupies a fraction vvac of the cell volume and is surrounded by an additional inner membrane. The radius of a cell with mass m is approximately (assuming δ≪r) (Eq. ): 6r=34πmρ11-vvac1/3, where ρ is the density of the cytoplasm (µg C µm⁻³). A fraction ν of the cell's mass is used for the membrane(s), and the remainder 1-ν is mass available for active metabolism. The fraction of the cell mass that is membranes for a generalist cell is : 7ν=3δr, and for a diatom given in Eq. (). The cell model describes three processes: encounter with resources, uptake of resources and metabolism, and mortality.

Copepods represent multicellular zooplankton and the copepod community consists of different populations. Each population is characterized by the adult body-mass and the feeding mode (active or passive). The spectrum for the population resolves the life stages of copepods, from nauplii to adults, following the description in : 13dBidt=Jin,i+jout.i+gi-μpr.i-μ0-μHTL,iBi,i∈multi where multi = {all multicellular groups}. Here Jin (µg C L⁻¹ d⁻¹) is the mass flux into the size class, which equals the mass flux out of the previous class, except for the first class where it is the reproductive flux. The other fluxes are “pure” rates in units of 1 per day: jout.i is the flux out of a size class, gi is the biomass accumulation rate, and μ0 is background mortality. The fluxes and the mortalities are calculated from the consumption by predation (Eq. ) and the predation mortality (Eq. ), and are detailed in Table .

Particulate organic matter is produced by losses from plankton (Fig. ). The POM has the same C:N ratio as the plankton, meaning that POM does not transport silicate. Instead, all silicate that should have gone to POM is considered lost to the deep sea and therefore also lost from the model. POM is generated from several sources (Fig. ): (1) a fraction 1-γ2 of cells dying from lysis; (2) multicellular plankton produce POM from sloppy feeding, fecal pellets (incomplete assimilation) and from the losses due to inefficient reproduction (egg mortality); (3) a fraction γPOM=1-γHTL of the losses from higher trophic level mortality is assumed to be fecal pellets routed to POM. Particulate organic matter is lost from remineralization with a rate rPOM and from grazing by copepods: 14dBidt=∑j1-γ2μ2.j+1-γFJFloss,j+1-γHTLμHTL,jBi+∑S∈multiadults1-ϵRgSBS-rPOM+μpr.iBi,i∈POM, where index “S” indicates the last (adult) size class in each copepod population. Parameters are given in Table .

The dissolved state variables are taken up through diffusive encounter and replenish with the losses from the plankton groups (Fig. ). The equations for the dissolved phases of nitrogen, DOC, and silicate are: 15dNdt=∑i∈uni-jN,i+jNloss,i+jpassive,i+γFjFloss,i+γ2μ2,iBi+γHTLμHTL∑i∈multiBi+rPOM∑i∈POMBi1ρC:N16dCdt=∑i∈uni-jDOC,i+jpassive,i+jLloss,i+γFjFloss,i+γ2μ2,iBi 17dSdt=∑i∈diatoms-jSi,i+jSiloss.i.i+jpassive,i+γ2μ2,iBi1ρC:Si. We have checked that all nutrients are conserved within an error close to numerical noise in Eqs. ()–(), while accounting for losses of silicate to the deep sea.

Temperature T (°C) influences the physical and the metabolic processes of all groups in the model. The influence of temperature is described via Q10 corrections to the parameters from a reference temperature of 10° as: 18Q(T)=Q10T-10°/10°. The only physical process influenced by temperature is the diffusion rate of dissolved matter (DOC, Si, and nitrogen) towards the cell, which has a Q10=1.5 . This means that the parameter αN is corrected by temperature (Table ). The other temperature correction is on metabolic rates by Q10=2. The affected metabolic rates are: the basal metabolic rate (jR for unicellulars and kR for multicellulars), the maximum synthesis rates (αmax and h respectively), and the remineralization rate of POM rPOM. Applying temperature corrections on the individual processes means that the temperature response of division rate and growth are emergent and typically do not follow a Q10 relation .

The core NUM model library solves the state-variable equations (Eqs. –) and can be embedded into a full circulation model. The library provides three simulation environments: chemostat (steady state and seasonal), water-column, and global. All environments, except the steady state chemostat, are implemented based on the transport matrix method (TMM) . The transport matrix method is a computationally cost-efficient offline numerical scheme to simulate ocean biogeochemical tracers, where a sparse matrix represents the advective-diffusive transport of a passive tracer. The transport matrix approach is a reliable alternative to conventional ocean general circulation models that reproduces reasonably well both the mean spatial and seasonal patterns of biogeochemical tracers in the ocean . The NUM library employs the monthly resolved MITgcm transport matrices in a coarse resolution of 2.8°×2.8° and 15 vertical layers , and a higher resolution of 1°×1° with 21 vertical layers (ECCO) . Both setups provide monthly transport matrices with a transport time step of 0.5 d. Sinking of POM is resolved with an implicit first-order upwind scheme Chap. 19. For the global and water-column simulations, the state-variable equations are solved with a forward Euler scheme with a simple predictor-corrector step to avoid negative values of the dissolved tracers, and a time step of 0.1 d. The bottom boundary conditions are closed for all living biological state variables and for DOC. POM is allowed to sink into the bottom. Nutrients (nitrogen and silicate) are nudged towards a fixed concentration at the bottom: 19ut+1=ut+κΔtΔzubottom-ut, where u is the state variable (N or S), t is the time step, Δt is the transport time step (0.5 d), Δz is the thickness of the bottom layer, κ=1/365 m d⁻¹ is the relaxation rate, and ubottom is the bottom concentration taken from World Ocean Atlas climatologies .

The chemostat environment models the upper mixed layer which mixes with an unresolved deep layer, with a mixing rate d (d⁻¹) . The deep layer concentrations of nutrients are fixed, Ndeep and Sdeep, and the concentration of DOC is zero. Unicellular organisms are mixed out of the upper layer, while the multicellular organisms are assumed to stay in the upper layer. POM is mixed out of the surface layer but also sinks out with a rate v/M, where v (m d⁻¹) is the sinking velocity and M (m) the thickness of the mixed layer. The governing equations are solved with the internal matlab Rosenbrock 2nd order stiff solver: 20dBidt=fi(L,T,B)+dBdeep.i-Bi︸fornutrientsanduni.-vMBi︸forPOM, where the first term fi(L,T,B) is the right-hand-side of the state-variable equations (Eqs. –), which depends on light L, temperature T and the set of all the state variables B. Finally, the chemostat can be run in a seasonal environment where light, temperature, and mixing rate d vary with time and are extracted from the transport matrix. In high latitude environments it is necessary to disable the mixing of unicellular plankton to the deep layer to allow them to survive the winter period.

The model framework allows for a flexible combination of the four groups. A model setup requires at least one unicellular group (generalists or diatoms), and may include POM and a number of copepod groups. The default NUM model setup used here involves all four groups. Each group contains a number of size classes and the model results depend, to some degree, on the number of size classes used in the unicellular groups (Fig. a and b) and in the copepod groups (Fig. c and d). Further, it depends on the number of active-copepod and passive-copepod groups (Fig. e and f). The number of size classes and copepod groups in the default NUM model setup is chosen such that the overall results, in terms of size spectra (Fig. a, c and d) and ecosystem function (Fig. b, d and f), do not change considerably upon the addition of more classes or groups. The setup includes 10 size classes of generalists, 10 size classes of diatoms, two passive copepod groups (adult sizes 0.2 and 5 µg C; prosome lengths 320 and 1040 µm), and three active copepod groups (1, 32, and 1000 µg C; prosome lengths from 584 and 7200 µm), where each group is discretized in 6 size-classes representing the growth in body mass. The number of state variables adds up to a total of 54.

Running the NUM model from the matlab interface proceeds in four steps: (1) setting up the simulation type (i.e., choosing which size spectra groups to simulate), (2) setting up the simulation environment (chemostat, water column, or global), (3) running the model, and (4) plotting the output:

All parameters are encapsulated in the |p| structure, which is passed to the simulation (in this example a global simulation). The simulation returns the entire output in the |sim| structure which is passed on to other functions for analysis or plotting. These two structures are documented on the github site and all functions are documented in their |help| pages.

All parameters of the plankton groups are based on first principles or cross-species comparisons . While these parameters are uncertain (explored by ), they are not expected to vary spatially. However, some of the extrinsic parameters either have no first-principle arguments or data, and/or are varying spatially. These extrinsic parameters together mold the overall function of the community: the global average light extinction coefficient by water kw without feedback from plankton, the average POM sinking velocity v, and the higher trophic level mortality μHTL.0. These variables conjointly determine the total biomass and production of the global model: lower light extinction increases gross primary production, lower sinking velocity decreases the remineralisation depth and thereby increases the amount of primary production. Decreasing the mortality by higher trophic levels increases the copepod biomass, which influences the unicellular spectrum through a trophic cascade (Fig. ). In nature, the values of these three variables vary on a global scale, but here they are kept constant. Consequently, we must calibrate to find reasonable global average values. The calibration algorithm minimizes the error between modeled and observed values, based on the following steps:

First we compute the mean error between in situ data Qobs to modeled output Qmodel for three metrics: picoplankton (i=1), POC (i=2), copepod biomass (i=3) in varying space and time for each metric, such that the number of available measurements of each metric i is ni, and each distinct measurement is j. This calculation is represented in the first addend in Eq. (). The datasets are described below. The datasets used in this part of the algorithm are: (1) picophytoplankton biomass (organisms with a diameter less than 2 µm) between 0 and 10 m depth from water samples between 1997 and 2014, across all latitudes mainly in the Atlantic Ocean (Fig. ) ; (2) Particulate Organic Carbon (POC) biomass of all plankton and detritus particles with a radius between 0.35–15 µm from the GO-POPCORNv2 dataset from cruises (2011–2020) across all major ocean basins, spanning from 70° S to 55° N (Fig. ); (3) copepod biomasses across the Atlantic Meridional Transect cruises . 21ε=16∑i=131ni∑jnilog⁡Qmodel.ijQobs.ij+∑k=13log⁡PNPPmodel.kPNPPobs.k. Second, we select three distinct locations in the ocean that correspond to an oligotrophic, a eutrophic and a seasonally varying environment to compare the mean error between the annual averages of satellite derived NPP and modelled NPP (PNPPmodel): an oligotrophic location (22° N, 158° E), a eutrophic location (5° S, 5° E), and a seasonal location (60° N, 40° W). At these sites, the annual averaged NPP over the entire water column from CAFE is approximately 200, 1000, and 500 mg C yr⁻¹, respectively. This comparison with NPP appears in the second addend of Eq. (). PNPPobs corresponds to annual averages (2002–2021) from satellite observations processed with the CAFE model, selected due to its better performance on a global average relative to other satellite-derived NPP models .

The net primary production is calculated in the model as the amount of carbon that is fixed by photosynthesis minus what is used for respiration: 22PNPP=∑i∈unimax0,jL.i-jresp.iBi. This definition appears simple but there are two important comments. First, all mixotrophic metabolic costs are included in the respiration (jresp) that is subtracted from the fixed carbon (jL), including the costs of feeding and DOC uptake. This is different from the usual thinking about primary production, which implicitly assumes that primary production only occurs by obligate phototrophs. Second, it should be noted that the definition does not include all primary production. Some carbon is lost to DOC during the uptake process (Fig. b), which can fuel osmotrophic biomass production. The DOC part of primary production can contribute a substantial portion of measured primary production . Adding this potential carbon production into the definition can be achieved by replacing jL with jLϵL-1. However, we refrain from including DOC NPP, which may result in underestimating by the model compared to measured NPP at oligotrophic sites.

After the initial calibration of the three parameters we evaluate the performance of the NUM model using a second set of observations. For the performance evaluation we use observations of nano- and microplankton from three Atlantic Meridional Transects cruises , a meta-analysis of global mesozooplankton biomass distribution , and annual averages of surface nutrients from the World Ocean Atlas . To validate the global diatom distributions we use the Copernicus Global Ocean Colour satellite product . This product provides the surface chlorophyll divided between groups. We use the ratio between the total chlorophyll and the diatom chlorophyll as a proxy for the ratio between modeled phytoplankton and diatom biomasses. This use assumes that the biomass:chlorophyll ratio is roughly similar between diatoms and other phytoplankton at the same site, but allows for variation in the ratio between sites. However, the satellite observations only capture the first few meters of the surface, whereas the top layer in the global model is a mean over the top 50 m. This evaluation data-set will therefore be inaccurate in the areas with a deep chlorophyll maximum.

The calibration gave an optimal light extinction coefficient of 0.06 m⁻¹, a sinking velocity of 19 m d⁻¹, and a coefficient of higher trophic level mortality of 0.017 L(µg C d)⁻¹. The response of the model to changes in each calibration parameter is shown in Fig. . The calibrated model produces biomasses of picoplankton, POC, and copepods which are in the correct order of magnitude of the observations (Fig. a–c). Though the three parameters are calibrated to be constant in time and space, the model does produce latitudinal patterns in accordance with the observed patterns (Fig. d–f: elevated biomass at temperate latitudes (-50 and 50°) and around Equatorial upwelling. The modeled copepod biomass approximates observations at temperate latitudes, however, it underestimates copepod biomass in the Equator.

The model captures general global-scale biomass patterns, with higher concentrations of unicellular and multicellular plankton in high-latitude seasonal environments and equatorial upwelling areas and the lowest biomasses in oligotrophic gyres (Fig. ). These patterns are mirrored in satellite-derived NPP estimates (Fig. ). Satellite production of NPP is subject to a systematic bias , and further the panels b–e highlight the big variation between satellite observation products. Nevertheless, one unrealistic pattern in the model that clearly stands out is the very high biomass and production in the Southern Ocean.

Turning to the evaluation data from the Atlantic transects, we see that the model captures the right order of magnitude of nano- and micro-plankton biomasses summed together (Fig. ). In these data, increased biomass around equatorial upwelling is less than expected, which is also reproduced by the model. However, the model overestimates plankton biomass at higher latitudes compared to the AMT 14 cruise (Fig. d), where samples were collected during May. At the global scale, the simulated average macrozooplankton concentration (copepods, larvae, and krill with a prosome length > 2 mm) is of the correct order of magnitude compared to observations, although the correlation is weak (Fig. ).

The model resolves nitrogen in a simplified manner without accounting for its different forms (i.e. nitrate, nitrite, ammonia). We have chosen nitrate for the comparisons because it is considered to be the main nitrogenous compound utilized for primary production . The model captures the spatial patterns in annual mean nitrogen and silicate concentrations at the surface (Fig. ) with high values in the North Atlantic and Southern Ocean and lower at the oligotrophic gyres. In addition, it reproduces elevated concentration in equatorial upwelling areas. However, the modeled values are generally much lower than the observed values. This is likely due to very efficient uptakes of nutrients by unicellular plankton in the model, which leads to low limiting nutrient concentrations (low “R*” sensu, ). This effect was also observed in the analysis of the generalist level .

The model captures the high surface diatom : phytoplankton ratio in high latitudes but overestimates diatoms in mid latitudes (Fig. ). This concurs with observations in the Pacific and the Atlantic Oceans . The overestimation in mid latitudes could be an artifact of the satellite measurements only seeing the very surface waters, whereas the top layer in the simulations performed here is the upper 50 m. Further, data from the Continuous Plankton Recorder (CPR) survey shows that diatoms constitute a fraction up to about 50 % of the total phytoplankton biomass throughout the North Atlantic , in correspondence with the model simulations.

The unicellular NUM model's output differs from traditional biogeochemical plankton models: instead of having state variables of bacteria, phytoplankton, and zooplankton, the NUM model resolves generalists and diatoms. While the diatoms are obligate phytoplankton (including DOC uptake), the trophic strategies of the generalists are dynamic and an emerging property of the model. However, common terminology and most models explicitly resolve state variables of bacteria, phytoplankton, and zooplankton. To relate the model's output to the aforementioned groups, we can estimate bacteria, phytoplankton, and zooplankton biomass from the generalists and diatoms, by weighing each size class with the fraction of carbon assimilation from DOC (bacteria-like), photosynthesis (phytoplankton-like), and phagotrophy (zooplankton-like): 23Bbact=∑ijDOCjCBi24Bphyto=∑ijLjCBi25Bzoo=∑ijFjCBi, where jC=jDOC+jL+jF (Fig. a–c).

The copepod community is dominated by active copepods in higher latitudes (Fig. ) and by passive (ambushing) copepods in oligotrophic gyres. In equatorial regions the presence of active copepods varies in different oceans. In the Equatorial Atlantic active copepods occupy between 50 % and 60 % of the multicellular plankton community. Lastly, active copepods occupy 60 % to 80 % of copepods along boundary currents in the Pacific, namely the California Current east and Kuroshio Current west.

We compared runs of the water-column and seasonal chemostat model setups to the output from a global simulation at a high latitude location (Fig. ) that is seasonally stratified. We chose a seasonally stratified location because it exhibits the succession of oligotrophic and eutrophic conditions throughout the year. The seasonal dynamics of total biomass show similar dynamics in the global and water-column simulations (Fig. a and c): a bloom in spring throughout the water column, followed by a deeper production maximum in summer, and terminated by a smaller autumn bloom. The spring bloom of diatoms is terminated first by silicate limitation, and the generalist bloom is terminated a bit later by nitrogen limitation (Fig. a and b). Contributing to the demise of the bloom is the grazing by copepods, which become established over summer (Fig. f–h). The deep maximum is less well represented in the global simulation, with only 13 vertical levels, than the water column simulation with 21 levels.

The size spectra show an overall similar community structure in the global ocean, water column and chemostat simulations (Fig. b, d and e). The plankton biomass is generally higher in the chemostat simulations than in the global and water-column simulations. The physical mixing in the global and water-column models constitutes a loss term by exporting planktonic organisms to the deep water, where they die unless they are mixed back into the photic zone. In the seasonal chemostat this mixing does not occur. That healthy growing diatoms tend to have near-neutral buoyancy is well established . How they do it is also well established . It is only when cells become stressed (e.g., low nutrients) and die that they start to succumb to gravity . The modelling assumption that diatoms while growing are neutrally buoyant is relatively robust. The difference in the modelling of mixing losses between chemostat simulations and the two other environments should be kept in mind when using the environments. Nevertheless, the output of the water column illustrates the same dynamics as the global at the same site, indicating that the water column model – which is naturally much faster than a global simulation – performs well. It should be noted, though, that the water column simulations perform best in seasonal environments. In less seasonal environments, the absence of lateral advection appears to result in too low production (Figs.  and ).

The NUM library makes it possible to extract descriptions of the rates that drive the dynamics of each size class. Figure  shows the rates of the generalists, diatoms, and copepods from the chemostat simulation in Fig. d, with uptakes of resources (or prey) illustrated as positive rates and losses including mortalities as negative rates. Overall, the rates of uptakes and losses are much higher in the spring bloom (left column) than in the summer (right column). The spring panels portray strong predation pressure on the unicellular community (dashed red lines in Fig. a and b), indicating that the bloom is decaying at this time. The high predation is also reflected in high feeding rates of the copepod community (red lines in Fig. c and d). Therefore, the spring copepod community is acquiring biomass at the expense of the unicellular community. In contrast, during summer, the division and loss rates in the unicellular community are roughly equal (grey lines in Fig. e and f). This indicates a community in balance with little changes in biomass. Further, in the unicellular community, the figure shows how the trophic strategies change with cell size : for the smallest cells, the acquisition of carbon is governed by DOC uptake (magenta lines), and dissolved nutrients (blue lines), in particular in the spring bloom (Fig. a and b). Larger sizes of generalists combine phototrophy (yellow lines) for carbon uptake for respiration with predation (red lines) for nitrogen and carbon for biosynthesis (and possibly respiration). Larger sizes of diatoms are purely phototrophic (negligible uptake of DOC) and are limited by nitrogen and silicate uptakes during summer.

The regulation of uptakes by unicellular plankton (Sect. ) entails that the respiration budget is resolved explicitly among the carbon used for synthesizing new biomass and different respiratory costs of basal metabolism, uptakes of dissolved nutrients and carbon, specific dynamic action of feeding (digestion), and costs of synthesizing new biomass (Fig. ). In general, the synthesis of new biomass (the division rate) is one third of the total metabolic budget , while the respiratory costs are distributed among the different uptake costs, and a small portion is allocated to basal metabolism.