The Regional Oceanic Modeling System ROMS; is a high-resolution, free-surface, terrain-following coordinate ocean model that solves the primitive equations considering Boussinesq and hydrostatic assumptions. The Biogeochemical model for Eastern Boundary Upwelling Systems (BioEBUS), which was derived from a N2P2Z2D2 model from , is coupled online to the physical part .

In this study the coupled ROMS–BioEBUS model has the same configuration as in . It contains a small, high-resolution domain forced by a larger coarse-resolution domain, using the AGRIF (adaptive grid refinement in Fortran) 2-way nesting procedure. The small inner grid has a horizontal resolution of 1/12∘ spanning from about 69 to 102∘ W and from 5∘ N to 31∘ S (Fig. ). The large outer grid spans from 69 to 120∘ W and from 18∘ N to 40∘ S with a resolution of 1/4∘. The biological processes occur in three time steps of 900 s for each physical time step of 2700 s in both domains. The two domains have 32 sigma layers with a vertical resolution of less than 5 m at the surface and decreasing to around 500 m at a maximum depth of 4500 m.

The coarse-resolution domain temperature, salinity and horizontal velocity are forced at the lateral boundaries with monthly climatological (1990–2010) SODA reanalysis . Both domains are forced at the surface with 1/4∘ resolution wind velocity fields from QuikSCAT and monthly heat and freshwater fluxes from COADS . At the lateral boundaries of the coarse-resolution domain, the biogeochemical model is forced with monthly nitrate and oxygen values from CARS and surface chlorophyll from SeaWiFs . Phytoplankton and zooplankton boundary conditions are derived from a vertical extrapolation of the chlorophyll data. Detailed information about the boundary and initial conditions and validation of the model is available in .

The evolution of a biological tracer in time is represented by Eq. (). On the right-hand side of the equation, the first term represents the advection with the velocity vector u. The eddy and molecular diffusion is represented by the second and third terms, where Kh is the horizontal diffusion coefficient and Kz the vertical diffusion coefficient. The last term is a source-minus-sink (SMS) term due to biological processes. The full set of equations and detailed explanation about each process are available in . 1∂Ci∂t=-∇⋅uCi+Kh∇2Ci+∂∂zKz∂Ci∂z+SMS(Ci)

The BioEBUS model is adapted for the biogeochemical processes specific to the low-oxygen conditions of EBUSs, with some processes being oxygen dependent see. It has two compartments of phytoplankton, two compartments of zooplankton and two compartments of detritus. The zooplankton and phytoplankton groups are divided into two size classes (small and large). There is not an explicit size parameter in the model; however, the compartments aim at representing the main ecological functions of the plankton community falling within each group. Hence, small phytoplankton (PS) represents organisms smaller than about 20 µm that require low nutrients such as flagellates, while large phytoplankton (PL) represents larger organisms such as diatoms. Similarly, small zooplankton (ZS) simulates the role of a zooplankton community smaller than about 200 µm, such as ciliates, and large zooplankton (ZL) represents a community larger than 200 µm, such as copepods. The two size classes of detritus (small DS and large DL) are produced from phytoplankton and zooplankton mortality and by release of unassimilated grazing material. The model also contains three compartments of dissolved inorganic nitrogen (NH4+, NO2- and NO3-), dissolved organic nitrogen (DON), oxygen (O2) and nitrous oxide (N2O).

The BioEBUS model includes oxygen-dependent remineralisation processes, which are divided into ammonification, nitrification and denitrification, as well as anammox, and are based on the formulations by . N2O is a diagnostic variable for model output, and its production does not affect the concentration of the other variables. It is based on the parameterisation of , which relates the production of N2O to the consumption of O2 from decomposition of organic matter in oxic and suboxic conditions. O2 concentrations depend on primary production, zooplankton respiration, nitrification and remineralisation. This model includes gas exchange of O2 and N2O with the atmosphere.

As noted above, this model was already validated against oxygen, nitrate and chlorophyll . As a complement, we here compare the large zooplankton compartment of the model, averaged from January to March, with observational data collected on RV Meteor cruise M93. The samples obtained during this cruise include day- and nighttime hauls with a Hydro-Bios multinet (nine nets, 333 µm mesh) between 10 February and 3 March 2013 on a transect off the Peruvian coast (≈12∘ S; see Fig. f), capturing the vertical and horizontal gradient in zooplankton concentration. Samples were size-fractionated by sieving and processed according to the ZooScan method . Observations included crustaceans, chaetognaths and annelids greater than 500 µm. For model comparison we converted the observation from nighttime hauls to dry biomass according to and further to nitrogen units as suggested by . A detailed description of the zooplankton processing is provided by . Only night observations were compared since our model does not include diel vertical migration.

In both the model and observations, concentration of large zooplankton is greatest in the surface and decreases with depth (Fig. ). At the surface, modelled concentrations are 1 order of magnitude larger than observations at almost all stations (Fig. ). Only at station d do observations reach 1 mmolNm-3, while the model exhibits maximum values close to 4 mmolNm-3. At most stations, the distribution of modelled concentrations is similar to that of observed concentrations in the surface layer (upper 100 m), although model estimates are consistently higher. This is also the case when comparing against a different dataset of observations (see Supplement). Below 100 m, however, model estimates are consistently lower than the observations, which is in particular evident at the deep offshore stations (Appendix , Fig. a and b). Zooplankton in our model does not consume detritus or bacteria; small zooplankton feeds on phytoplankton, and large zooplankton feeds on small zooplankton and on phytoplankton. Therefore, in contrast to observations, its presence is not expected in deep water. In summary, the model matches the observed spatial pattern of zooplankton distribution but tends to overestimate zooplankton concentration in the surface layer and to underestimate it in the mesopelagic depths. Possible reasons for this mismatch will be discussed in Sect. .

To mimic changes in grazing pressure on zooplankton due to small-pelagic-fish biomass fluctuations, we followed the approach by and varied the mortality rate of each zooplankton compartment by ±50 % in comparison to the reference scenario described by . Thereby, an increase in mortality assumes a large consumption of zooplankton by fish, while a decrease in mortality assumes fewer fish. Because the model does not include an explicit compartment for fish, it is assumed that all zooplankton biomass consumed by fish becomes part of the detritus pool via immediate fish mortality and defecation. In reality, a fraction of the biomass is extracted from the system by the fishing industry, predation by sea birds that defecate over land and migrations.

Our model has two zooplankton compartments. In order to explore the different roles of large zooplankton as top predator and small zooplankton as grazer and prey, we performed four experiments, in which we varied the respective mortality rate of large and small zooplankton (0.05 and 0.025 [mmolNm-3]-1 d-1, respectively) by ±50 %:

A_high with 1.5×μZL

The ETSP is highly dynamic at temporal and spatial scales, with nutrient-rich cold water near the coast of Peru and oligotrophic regions offshore. Therefore, we analysed three different regions: the “full domain” without boundaries (F), an “oligotrophic” region (O) offshore and the “coastal upwelling” region (C) near the Peruvian coast (Fig. ). Since the upwelling system of Peru is quite heterogeneous with lots of mesoscale processes, we restricted the C region to the very coastal upwelling area, where the concentration of large phytoplankton is high, up to about 40 to 50 km offshore. Region O was picked to be as far as possible from the nutrient-rich areas along the Equator and along the coast but apart from the domain boundary to avoid boundary effects. For most of our analysis, percentage relative differences between the reference scenario and the other scenarios were calculated. In addition, we analyse the development of plankton succession from the coast of Peru towards the open ocean at 12∘ S (Fig. , white line). Because plankton concentrations are negligible below 100 m depth and we are mainly interested in the plankton community, we focus most of our analysis on the upper water layers. For our analysis we therefore integrate or average plankton concentrations over the upper 100 m or, in the case of shallower waters, down to the seafloor. Also, all analyses in our study take into account only annual averages. However, we recognise that there is high temporal variability in the NHCS (see Appendix  and Supplement).

In this section we describe the concentrations of the inorganic and organic compartments in our model reference scenario, averaged between a 0 and 100 m depth (Fig. ) unless otherwise specified. Oxygen concentrations increase offshore, with an average of 226.6 mmolO2m-3 in the oligotrophic region (O) compared to 69.7 mmolO2m-3 in the coastal upwelling region (Fig. ). The concentration decreases further below 100 m, reaching an average of 5.3 mmolO2m-3 between 100 and 1000 m in the coastal upwelling region (C). Nitrate is the most abundant nutrient all over the domain, ranging from 0.6 in O to 21.7 mmolNm-3 in C. On the other hand, ammonium and nitrite are only 0.8 and 3.2 mmolNm-3 in C, respectively (Fig. ). Please refer to for a further in-depth analysis of biogeochemical tracers in the reference scenario.

Phyto- and zooplankton are generally absent in the deep water. In the surface layer between a 0 and 100 m depth (Fig. ), phytoplankton is clearly favoured by nutrient-rich coastal upwelling, where total phytoplankton reaches 0.93 mmolNm-3 on average, compared to 0.25 mmolNm-3 in the oligotrophic region (Fig. ). Total detritus follow the concentration trend of plankton, with 0.26 in C and only 0.03 in O.

When zooming into the coastal region (Fig. ), large phytoplankton exhibits a sharp peak near the coast which drops offshore. Moving further offshore, large zooplankton peaks at the decline in the large phytoplankton peak, followed by increased concentrations of small phytoplankton. Given the (Ekman-driven) transport of surface waters (Fig. ) this spatial pattern might be interpreted as a form of succession as the water is advected offshore. In general, modelled concentrations of large zooplankton are high not only in the coastal upwelling region (Sect. ) but also in large parts of the domain (Fig. ), except for in the oligotrophic south-western region (Fig. ).

The growth rate of phytoplankton is limited by temperature, nutrients and light (see Eq. ). Hence, the spatial pattern of plankton near the coast can be explained by the competitive advantage of large phytoplankton in deep water due to its steeper initial slope of the P–I curve (see Appendix , Table ), eutrophic conditions in the nutrient-rich upwelling water and relatively low predation due to the lack of large zooplankton. This opens a loophole for large phytoplankton to grow in the upwelling waters. As water is transported offshore (Fig. ), large zooplankton starts to grow and grazes on large phytoplankton. More oligotrophic sunlit conditions even further offshore favour small phytoplankton growth at the surface. Therefore, this first analysis reveals spatial segregation and succession from the coast to offshore waters. These patterns are caused by the model's parameterisation of plankton groups and their mutual interactions.

When changing zooplankton mortality, the inorganic variables (Fig. ) are not noticeably affected by the experiments.

Plankton responds very similarly to changes in mortality in experiments A and B (Fig.  and Appendix , Fig. ). Phytoplankton and large zooplankton follow the same direction of response in all regions: concentrations of large zooplankton decrease in the high mortality scenario and increase in the low mortality scenario, as could be expected. Large phytoplankton responds inversely to large zooplankton, evidencing a top-down control of its main grazer. On the other hand, the response of small phytoplankton is inverse to that of large phytoplankton. In contrast, small zooplankton shows an asymmetric response to changes in mortality, as it mainly decreases in the low mortality scenarios but responds only weakly in the high mortality scenarios (Fig.  and Appendix , Fig. ).

The spatial plankton distribution along the transect (Fig. ) remains the same when zooplankton mortality changes, but the absolute concentrations of each compartment change. In all scenarios large phytoplankton peaks close to the coast. When large zooplankton concentrations are reduced because of its higher mortality in experiment B_high, the large phytoplankton peak increases (Fig. , right). Similarly, large phytoplankton decreases with lower zooplankton mortality, due to higher grazing of zooplankton on phytoplankton (Fig. , left). This pattern is also similar in experiments A. Because the largest effects occur in the very productive coastal region (C), in the following section we narrow our analysis to this domain.

An increase in zooplankton mortality causes only small changes in total primary production in the coastal upwelling region (C), but the partitioning between the two phytoplankton groups changes (Fig. ). In particular, total primary productivity of the system is increased by 3.9 % in B_high and reduced by 5.5 % in B_low. Large phytoplankton is the dominant group. However, its productivity increases by about 19 % in B_high and decreases by 22 % in B_low; i.e. its changes are much more pronounced than the overall phytoplankton response. Because small phytoplankton shows an inverse response in production, this dampens the change in total primary production. Thus, a low zooplankton mortality favours small phytoplankton and its growth, and a high mortality favours large phytoplankton; changes in both phytoplankton groups result in a weak response of total primary production.

Experiment B_high exhibits the highest total plankton biomass in the upper 100 m of the upwelling system (112.6 mmolNm-2, Fig. ), which is mostly concentrated in the large phytoplankton compartment (59.43 mmolNm-2). In this experiment the main pathway of nitrogen transfer to large zooplankton is via its grazing on large phytoplankton (6.77 mmolNm-2d-1). As mortality decreases, small phytoplankton and large zooplankton gain biomass. Large phytoplankton grazing remains the main nitrogen source for large zooplankton. However, large zooplankton consumption of small phytoplankton is almost 3 times higher in B_low than in B_high (Fig. ). Thus, a reduction in mortality causes a switch in the diet of large zooplankton, from mainly large phytoplankton to a diet that consists of more than one-quarter small phytoplankton.

Small zooplankton biomass decreases by ∼0.4 mmolNm-2 (about 5 % of the reference value) in B_low, but it only increases by 0.15 mmolNm-2 (about 2 %) in B_high (Fig. ). Despite the changes in small zooplankton biomass, the consumption of its biomass by large zooplankton remains approximately the same in all experiments, resulting in a higher proportional biomass loss of small zooplankton in scenario B_low. Hence, predation by large zooplankton as well as competition for food negatively affects small zooplankton. Under high mortality conditions, the availability of large phytoplankton as food increases (Fig. ). However, small phytoplankton, the preferred prey of small zooplankton, declines as explained above. Such antagonistic effects on small zooplankton buffer its response in this scenario.

In this section, we describe the response of the nitrogen loss due to mortality, also referred to as the quadratic mortality term (μZi⋅[Zi]2; see Eqs.  and ). The integrated μZi⋅[Zi]2 in the reference scenario was provided in Sect. , Fig. . In experiment B, μZS⋅[ZS]2 and μZL⋅[ZL]2 exhibit different behaviours. In the coastal upwelling region, both μZS⋅[ZS]2 and μZL⋅[ZL]2 increase in B_high and decrease in B_low (Fig. ). However, the relative response of μZL⋅[ZL]2 is mild and fluctuates between ±30 % outside the oligotrophic area (Fig. ). In contrast, μZS⋅[ZS]2 exhibits a clear relative increase all over the domain in B_high, and a decrease in B_low. The moderate response of large zooplankton loss can be attributed to the combined effects of changes in zooplankton concentration ([ZL]) and changes in the mortality parameter (μZL), which we varied by ±50 % in this study. Large zooplankton concentration increases when μZL is decreased, and vice versa. This opposite trend buffers the effect of a change in μZL. On the other hand, small zooplankton concentration ([ZS]) changes in the same direction as μZS, due to the combined effects of changes in its concentration, due to grazing pressure exerted by large zooplankton and competition for food with this group, and changes in μZS.

To summarise, increasing and decreasing zooplankton mortality by 50 % generates a rearrangement of the plankton ecosystem; however, the overall changes in the large zooplankton loss are not as high as would be expected from a change in the mortality rate alone. This might buffer the system once the biogeochemical model is coupled to a model of higher trophic levels.

An increasing need for the development of end-to-end models has generated interest in using results of biogeochemical models as forcing for higher-trophic-level models (fish, macroinvertebrates and apex predators) seefor a review. In a one-way coupling set-up, the biomass of plankton available as food for higher trophic levels has been adjusted during calibration of the latter, reducing the amount of plankton that is available for fish consumption e.g.. However, for two-way coupling set-ups, this adjustment of the available plankton biomass could buffer the effect of higher trophic levels on lower trophic levels e.g.. Biogeochemical models can produce a wide range of output depending on their parameter values and their non-linearity. For example, a quadratic zooplankton mortality exacerbates the reduction in zooplankton biomass when concentrations are very high and prevents its extinction at very low concentrations. In addition, the multiple-resource form of the Holling type-II grazing function allows the predator to modify its grazing preference towards the most abundant prey . Finally, noted that coupled physical and food web models can transition from equilibrium to chaotic states under even small changes in their parameters. Few studies have aimed to understand such behaviour and examined the sensitivity of the model to parameters . Our model study was partly motivated by the uncertainty associated with the zooplankton mortality. Indeed our model showed that a small alteration in the mortality parameter (small compared to the wide range of values that have been used for this parameter in different biogeochemical models) can strongly affect the mass flux within the simulated ecosystem. Hence, there is an increasing need for accurate plankton representation in biogeochemical models without dismissing other compartments, such as nutrients or oxygen. Nevertheless, lack of data for validation especially of higher trophic levels is a common problem for biogeochemical models of the northern Humboldt Upwelling System . Oxygen, chlorophyll and nitrate in our model have been evaluated previously . Here we presented the first attempt to compare the large zooplankton compartment of the ROMS–BioEBUS ETSP configuration with mesozooplankton observations.

At the surface, zooplankton concentrations simulated by our model in the reference scenario are 1 order of magnitude higher than observations at most stations. However, sampling in the upper 10 m depth may be impacted by water disturbance by the ship adding additional errors to the measurements. The match to observations improves with depth. Modifying the mortality rate by +50 % (-50 %) produced only a change of -12 % (+19 %) in large zooplankton concentration, indicating that either the induced changes in mortality rate were not large enough or this parameter is not overly influential in improving the model fit to observations. Systems with a non-density-dependent, or linear, mortality rate respond to perturbations in a “reactive” way, as defined by , drifting away from equilibrium, in contrast to systems with density-dependent closure terms which tend to buffer the perturbations . Therefore, we might have expected a stronger impact if we had manipulated the linear closure term of the model, or metabolic losses (see Sect. ), rather than the quadratic term. For an average nitrogen flux due to large zooplankton mortality of 2.2, 3.1 and 3.52 mmolNm-2d-1 and large zooplankton integrated concentrations of 46.36, 38.82 and 34.17 mmolNm-2 in the coastal upwelling region (see Sect. ), the mortality rate would be 0.04, 0.08 and 0.1 d-1 in scenarios B_low, reference and B_high, respectively. In all cases the values are lower than the 0.19 d-1 estimated by for copepods in the field at 25 ∘C. The closest scenario to observations is B_high where the mortality rate is only about half of the estimate by . This is also the scenario that better resembles mesozooplankton observations, since it exhibits the lowest concentrations. On the other hand, the mortality rate estimated for the reference scenario (0.08 d-1) is closer to the 0.065 d-1 estimation by at 5 ∘C. This is considerably lower than the temperature in our modelled region (see Fig. ). Note, however, that zooplankton mortality in our model does not depend on temperature.

Some part of the mismatch between model and observations might be related to how both data types are generated. Therefore, a direct comparison between model and observations has to be viewed with some caution. In our model, large zooplankton acts as a closure term which is adjusted to balance the biomass and nitrogen flux to other compartments and does not resemble a specific set of species. Its parameters (maximum grazing rate, feeding preferences, etc.) are meant to represent larger, slow-growing species with a preference for diatoms. As such, they might not be directly comparable to the observed groups. The observations, on the other hand, are susceptible to sampling errors such as net avoidance and do not cover the whole taxonomic and size spectrum of mesozooplankton. For instance, no gelatinous organisms are accounted for, which may account for an important fraction of the wet biomass ; only mesozooplankton greater than 500 µm is considered in the sampling ; and fragile organisms, such as Rhizaria, are not quantitatively sampled by nets . Therefore, the observations might be biased low in comparison to the model. Furthermore, a lack of an explicit size term in the model limits a direct comparison with observations because these depend on the mesh size of the sampling net. Finally, given the above-mentioned rather pragmatic parameterisation especially of zooplankton growth and loss rates, it is very likely that the model could be improved by a tuning or calibration exercise that targets a good match between observed and simulated zooplankton concentrations. Both the mortality estimation by and the zooplankton observations in the field suggest that further tuning of the model should lean towards higher mortality rates. Nevertheless, this may require the further tuning of other parameters. Despite the complexity of the model, the considerable uncertainty in model parameters and the sparsity of observations that can constrain these parameters, this is a complex task see e.g.. Therefore, we have refrained from this effort for the present but aim at providing a better-calibrated model in the future.

The spatial variability between different profiles of zooplankton is greater in the observations than in the model, and the variability in concentrations within each single profile is much larger than the differences between the modelled mortality scenarios (Fig. ). Several sources of variability are not accounted for in the model as it only simulates the most relevant processes in the system. We employ a climatological model which aims at simulating the average dynamics over several years, dismissing interannual variability. Furthermore, we here compare a 3-month average from January to March, while observations provide only a snapshot of a highly dynamical system. In addition, we only compared our simulated zooplankton against night observations because in our model zooplankton is always active at the surface. In reality, zooplankton is known to perform diel vertical migrations (DVMs), which could increase the export flux to the deep ocean . The lack of DVM could affect the export of organic matter to greater depths and therefore the biogeochemical turnover at the surface. Zooplankton likely also experiences lower mortality at depth ; however, off Peru these benefits might be counterbalanced by reduced oxygen availability and the concurrent metabolic costs. These obstacles for comparing zooplankton models with observations had already been described by more than 4 decades ago: (a) “the zooplankton is a very heterogeneous group, defined operationally by the gear used for capture rather than by a discrete position in the food web” . (b) Zooplankton is irregularly distributed in space, not necessarily following physical features. (c) Adult stages of some zooplankton groups perform vertical migrations .

To summarise, some biases and mismatches between model and observations remain; given the uncertainties and episodic nature associated with the observations and their correspondence to their model counterparts, further studies will be necessary to more precisely calibrate the model. For a complete model evaluation, however, the small zooplankton compartment should also be evaluated against microzooplankton samples. The high mortality scenario, B_high, is the one that is closest to the observations, due to producing the smallest concentrations of large zooplankton at the surface. However, changing this parameter was obviously not enough to match the observations. In fact, in our model an increase of 50 % in the mortality rate produced only an increase of 14 % (0.4 mmolNm-2d-1) in large zooplankton mortality loss (see Sect. ) because of the high non-linearity of the model. Indeed, potential changes in zooplankton losses of 5 mmolNm-2d-1, derived by fluctuations in anchovy stocks and grazing pressure (see Sect. ), point towards much larger values for the mortality rate. An even stronger increase in the zooplankton mortality rate (e.g. , applied a 5-times-larger value), along with a subsequent adjustment of other parameters, may be necessary to approach observed values. In addition, complementary observations with other sampling methods could provide a better estimation of mesozooplankton concentrations for tuning the model.

Our model study showed the strongest response of the ecosystem to changes in the mortality rate in the highly productive coastal upwelling. Here, the response of the model ecosystem was mainly driven by large zooplankton. This can be concluded from the close similarity of model solutions A_high and B_high, as well as of A_low and B_low (see Appendix ). The mortality term for small zooplankton played a lesser role; in addition to the direct effect of the mortality rate, this compartment was also affected by grazing by, and competition with, large zooplankton. Large zooplankton fluctuations due to mortality directly affected large phytoplankton through grazing. Small phytoplankton, on the other hand, was affected by grazing but also by competition with large phytoplankton. Changing the mortality rate produced little effect on the mass loss of large zooplankton due to mortality; however, it altered the nitrogen pathways along the trophic chain and ultimately the concentrations of most plankton groups, albeit in different ways, depending on the direction of change. Under conditions of high zooplankton mortality the food chain is dominated by nitrogen transfer from large phytoplankton to large zooplankton, the classical food web attributed to highly productive upwelling systems . When zooplankton mortality is reduced, large zooplankton increases its consumption of small phytoplankton, taking over the role of small zooplankton.

In our model, large zooplankton has a competitive advantage by feeding on its competitor, small zooplankton, a strategy that was also found to evolve in simple ecosystem models as an advantageous alternative to direct competition . We find that under low mortality conditions, this advantage increases. The importance of competition is further evidenced in the changes in small phytoplankton concentrations in the coastal upwelling region. These were partly driven by changes in the availability of resources arising from fluctuations in large phytoplankton concentrations, which constitute the dominant group in the coastal upwelling. Natural selection, competitive exclusion and different resource utilisation strategies, together with bottom-up forcing by the physical processes in the environment, can shape the plankton community in global models and indicate bottom-up effects on the phytoplankton community. On the other hand, showed that variable zooplankton predation can increase phytoplankton diversity by opening refuges for less competitive phytoplankton groups and thus exert top-down effects. In our study, the biological interactions between two phytoplankton groups, mainly competition for resources (bottom-up), are additionally affected in a top-down manner by changes in zooplankton concentrations.

The processes driving the ecosystem response in our regional study are dominated by trophic interactions among the size classes of phytoplankton and zooplankton. We found a top-down-driven response affecting mainly the plankton compartments of the model. The direction of the total zooplankton and total phytoplankton change is determined by the large zooplankton and large phytoplankton groups. Small zoo- and phytoplankton buffer the response when they present opposite trends to their larger counterparts (Table ). In the following, we will compare our results with a similar study by .

, in their sensitivity study, modified the zooplankton mortality by ±50 % in a global biogeochemical model with a spin-up time of 300 years. Their model has only one zooplankton size class with a quadratic mortality term. For their analysis, evaluated three regions: the whole domain, also referred to here as “global”; the region from 20∘ S to 20∘ N which in this coarse-resolution model is mainly an oligotrophic region and is referred to as “tropics”; and the region south of 40∘ S where nutrient concentrations are high and is named “Southern Ocean”. After the model spin-up, global zooplankton biomass changes by about -(+)5 %, and phytoplankton biomass by about +(-)1 % in the high (low) mortality scenario their Fig. 2. In contrast, in our study, total zooplankton averaged over the full domain decreases (increases) by -11 % (10 %), and total phytoplankton changes by about +(-)6 % in the high (low) mortality scenario (Fig.  and Table ). Hence, the effects of changing zooplankton mortality follow the same trend in both studies, but they are slightly milder in the study by . At the regional scale, the responses in the study of have a feedback from lower trophic levels, either to phytoplankton in the nutrient-replete region or all the way to nutrients in the oligotrophic region. Therefore, the biomass of both zooplankton and phytoplankton is ultimately bottom-up affected. This is evidenced by a change in zooplankton and phytoplankton concentrations in the same direction their Fig. 3. On the other hand, although our study also exhibits a feedback from phytoplankton to zooplankton, the strongest driver remains the top-down predation of zooplankton on phytoplankton (Table ).

The regional differences between our study and that by can likely be explained by the different model structures and experimental set-ups, namely the number of phyto- and zooplankton compartments, different timescales considered for model simulation and analysis, and the spatial domain: while applied a global biogeochemical model with just one size class of phyto- and zooplankton, simulated until near a steady state, our regional model study applies a more complex biogeochemical model with a strong seasonal cycle (see Appendix ), simulated for only 30 years and constantly forced at the boundaries. Further, the short few-year timescale of our model simulations might have prevented the effects of changed zooplankton mortality from propagating to deeper water layers which contain the largest concentrations of nutrients. Finally, the region modelled in our study is spatially already very dynamic at the mesoscale resolution, as evidenced by well-defined plankton spatial succession from the coast of the continent towards the open ocean. The phytoplankton bloom, which develops closest to the coast and then is offset while water is transported offshore, can be explained by an imbalance between sources and sinks, triggered by changing environmental conditions. For example, applied the concept of “loopholes”, proposed by to explain fish productivity and recruitment success, to phytoplankton: according to their concept, phytoplankton blooms are formed when environmental conditions open a loophole in the plankton community of a mature ecosystem. Then, particularly phytoplankton species that are able to escape microzooplankton predation are those that will take advantage of the loophole and bloom . Our model results suggest that similar processes occur. Low concentrations of large zooplankton allow large phytoplankton to bloom near the coast. While the water is advected offshore, zooplankton growth and grazing offset the bloom. Observations by also reported spatial succession with large diatoms abundant in the coastal upwelling region being replenished by nanophytoplankton offshore. However, they propose silicate as the limiting nutrient offsetting the diatoms' peak, which is not present in our model. Furthermore, such succession is not unique to the NHCS but a characteristic feature of upwelling regions . We note that in the present configuration the BioEBUS model does not include any temperature-dependent zooplankton grazing rate. We expect that, if this were implemented, the loophole for phytoplankton growth in the cold waters of the coastal upwelling region would even be widened, amplifying the spatial succession we observed. However, temperature might also affect zooplankton metabolism, with colder temperatures decreasing its loss rates, which could in turn mute these effects again. On the other hand, the global model by does not resolve mesoscale processes. Furthermore, while they divided their study into the tropics, as an oligotrophic region, and the Southern Ocean, as an upwelling region, the upwelling system off Peru in the eastern tropical South Pacific is a nutrient-rich area. For all of this, we based our comparison on similarities in the nutrient concentration (high nutrients, oligotrophic and whole domain), rather than on geographic overlap.

In , the zooplankton loss due to mortality is not provided. However, it can be calculated from the zooplankton concentration and mortality rate. Assuming integrated zooplankton biomasses at year 300 their Fig. 2 of 98, 93 and 89 Tg N in the low mortality, reference and high mortality scenarios, respectively; mortality parameters of 0.03, 0.06 and 0.09 (mmolNm3)-1d-1; and a quadratic mortality term, then there is a difference of -44.5 % and 37.4 % in the zooplankton loss due to mortality between the low and reference scenario and between the high and reference scenario, respectively. As shown in Sect. , the mortality rate in our study is also smaller than the ±50 % changes that would be expected from a change in the mortality parameter of ±50 % (see Sect. ). The non-linearity of both global and regional models seems to reduce the effect of changes in the mortality parameter on zooplankton loss.

In summary both studies show a similar global response to changes in zooplankton mortality, driven by zooplankton preying on phytoplankton. Two zooplankton and phytoplankton size classes present opposite trends in our studies, buffering the overall response. Nevertheless, the relative changes in total zooplankton and total phytoplankton are on the same order of magnitude as in and even slightly higher. Regionally different feedbacks operated in the two models, possibly due to the specific set-up of each study, spin-up time and resolution. Finally, the relative change in the zooplankton loss due to mortality is smaller than the expected ±50 % in both studies.

In our study, we changed the zooplankton quadratic mortality, which could be regarded as the effect of a predator targeting highly aggregated zooplankton populations, or whose concentration closely follows that of zooplankton. This can be viewed as a way to parameterise the effect of changing fish abundance on the biogeochemistry of the system. In this case, a low zooplankton mortality implies fewer small pelagic fish (such as anchovies and sardines), while a high zooplankton mortality implies a higher abundance of such fish. Further, our experiments are based on two different assumptions: one where small pelagic fish feed only on large zooplankton (experiments A) and one where they feed on and affect the mortality of both large and small zooplankton (experiments B).

The diet of anchovy is still under debate. While previous studies had considered that anchovies feed mainly on phytoplankton, concluded that anchovies feed on zooplankton, especially euphausiids and copepods. Furthermore, the diet of anchovy seems to be more flexible than previously considered . For instance, the anchovy collapse in 1972 was correlated with a shift from a population feeding mostly on phytoplankton to a southern population feeding on zooplankton . On the other hand, small zooplankton groups such as ciliates have been reported as a minor component of anchovy diet (Table 5 in ). Thus, experiments A are more likely to resemble the fluctuations in anchovy populations. On the other hand, sardines, with their finer gill rakers, obtain most of their nutrition from microzooplankton . Although currently sardines are not as abundant as anchovies off Peru, historically they have also built up large concentrations and strong fluctuations over time have been observed . Thus, when also considering sardine populations and feeding modes, experiments B (simultaneous mortality change in both large and small zooplankton) might be more appropriate to parameterise the effects of changed fishing mortality on lower trophic levels.

Although the quadratic mortality rate is constant over the entire domain, the fractional loss rate by zooplankton mortality (μZi×Zi; d-1) varies over the domain because of changes in zooplankton concentration. This might mimic spatially variable grazing pressure by fish. However, our experimental set-up might be too simple to investigate the detailed response of predator–prey relationships. We partly tried to avoid conclusions that are too general by focusing our analysis on the coastal upwelling region off Peru, since anchovies are highly concentrated in this region . In addition, we neglected any feedback effects between zooplankton and their predator fish. A more detailed model set-up, as, for example, in coupled biogeochemical–fish models , would help to elucidate the specific trophic interactions in this region and their response to environmental changes and changes in fishing pressure.

Finally, we note that the parameterisation of fish predation in our study implies that all organic matter is transferred from zooplankton straight to the detritus compartment and then remineralised. In the real world, it is stored as fish for some time until the fish defecate or die . In an equilibrium state this would not be a problem since there is a constant turnover of nutrients. In a non-equilibrium state, the time that nutrients spend as part of larger animals' biomass would increase the gap between nutrient consumption by phytoplankton and their replenishment, affecting phytoplankton growth rate and potentially the blooming timing. However, this should not be a problem in the coastal upwelling region because nutrients are highly concentrated here. This is not the case for the oligotrophic region although, in our study, this region presents the weakest response. Furthermore, small pelagic fish concentrate mainly in the highly productive upwelling region rather than in the oligotrophic waters offshore. On the other hand, fish and larger animals are highly mobile and do not constantly drift by advection as nutrients and plankton do. Therefore, migrations transport nutrients and organic matter in and out of the region in a horizontal and vertical fashion. Such nutrient dynamics over time and space driven by higher trophic levels can be further explored either in explicitly coupled end-to-end models or in an implicit way similarly to in our study.