Mixed layer depth dominates over upwelling in regulating the seasonality of ecosystem functioning in the Peruvian Upwelling System

. The Peruvian Upwelling System hosts an extremely high productive marine ecosystem. Observations show that the Peruvian Upwelling System is the only Eastern Boundary Upwelling Systems (EBUS) with an out-of-phase relationship of seasonal surface chlorophyll concentrations and upwelling intensity. This "seasonal paradox" triggers the questions: (1) what is the uniqueness of the Peruvian Upwelling System compared with other EBUS that leads to the out of phase relationship; (2) how does this uniqueness lead to low phytoplankton biomass in austral winter despite strong upwelling and ample nu- 5 trients? Using observational climatologies for four EBUS we diagnose that the Peruvian Upwelling System is unique in that intense upwelling coincides with deep mixed layers. We then apply a coupled regional ocean circulation-biogeochemical model (CROCO-BioEBUS) to assess how the interplay between mixed layer and upwelling is regulating the seasonality of surface chlorophyll in the Peruvian Upwelling System. The model recreates the "seasonal paradox" within 200 km off the Peruvian coast. We conﬁrm previous ﬁndings that deep mixed layers, which cause vertical dilution and stronger light limitation, mostly 10 drive the diametrical seasonality of chlorophyll relative to upwelling. In contrast to previous studies, reduced phytoplankton growth due to enhanced upwelling of cold waters and lateral advection are second-order drivers of low surface chlorophyll concentrations. This impact of deep mixed layers and upwelling propagates up the ecosystem, from primary production to export efﬁciency. Our ﬁndings emphasize the crucial role of the interplay of the mixed layer and upwelling and suggest that surface chlorophyll


Introduction
The Peruvian Upwelling System (PUS) hosts a disproportionally productive ecosystem and supports 10% of the world's fishing yield while covering only 0.1% of the ocean area . As one of the Eastern Boundary Upwelling Systems 20 (EBUS), upwelling-favorable winds bring up cool, nutrient-rich waters to the surface, supporting high primary production and

Analyses approaches
To assess the seasonal variance of phytoplankton biomass concentration in each grid box (C) we analyze with the budget of the phytoplankton biomass and how its tendency is driven by physical versus biological processes: PHY represents the physical processes including advection and mixing. The BIO term stands for the biological processes, 85 namely primary production P P , consumptive mortality, natural mortality, exudation and sinking. We analyse in detail the 3 https://doi.org/10.5194/bg-2021-113 Preprint. Discussion started: 10 May 2021 c Author(s) 2021. CC BY 4.0 License. drivers of P P : P P is determined by phytoplankton concentration (C) and the growth rate (J) P P = C · J(N, P AR, T ) where the growth rate J is related to light availability for photosynthesis (PAR: photosynthetically active radiation), temperature (T) and nutrients (N: nitrate, nitrite and ammonium). The growth rate here is a multiplicative function of the light-, temperature-90 or nitrogen-related growth factors. To quantify the limitation experienced by phytoplankton within the mixed layer L mld , it is calculated from each growth factor (L (PAR) , L (T) and L (N) ) using phytoplankton concentration (C) within the mixed layer as a weight (Eq. 3). Light-, temperature-or nitrogen-related growth factors that each phytoplankton cell experiences are computed online.
For the analysis, we attribute the seasonal change of the average phytoplankton biomass concentration (C mld ) within the mixed layer to the change of the integrated phytoplankton content within the mixed layer (B mld ), and the change of the volume of the mixed layer (V mld ). With the chain rule and V 2 >> V ∆V , we approximate a discrete change of the mixed layer tracer concentration (∆C mld ) with To assess the relative contributions we then divide by C mld = B mld · V −1 mld to obtain ∆C mld which allows attributing a decrease of the mixed layer phytoplankton concentration C mld to a decrease of the phytoplankton biomass B mld or an increase of the mixed layer volume V mld and vice versa.

Model assessment 105
The model is evaluated based on averages over the focus region with observational data in monthly resolution. The correlation coefficient between the model simulation and observations, the root mean square error (RMSE) and the normalized standard deviation (SD) of the observations relative to the model results are shown in a Taylor diagram as a summary of the evaluation (dotted line; SeaWIFS (diamond) and MODIS (square)) and in situ data (digitized from Pennington et al. (2006, star) and Echevin et al. (2008, cross)).
but overestimates the nitrate compared to WOA. Cruise data  shows that the overestimation could arise from WOA failing to capture the high surface nitrate concentration in the coastal region with strong upwelling. A comparison of the simulated and observed seasonal cycle of surface chlorophyll in the focus region ( Fig. 1d) reveals that modelled chlorophyll generally follows the seasonal trend of satellite and in situ data with the amplitude of the seasonal cycle falling in between satellite and in situ data. Overall, the model shows reasonably good agreement with observational data on a seasonal scale, 120 sufficiently supporting an investigation of the seasonal paradox with CROCO-BioEBUS.
3.1 Anticorrelation of chlorophyll and upwelling: The seasonal paradox only appears in the Peruvian system Compared with other EBUS, the Peruvian system is unique in that it shows a clear anti-correlation between surface chlorophyll concentration and upwelling intensity on a seasonal scale, with lowest chlorophyll concentrations when upwelling is most 125 intense (Fig. 2a, R 2 = 0.71; Chavez and Messié, 2009). While the surface chlorophyll in the Benguela system does not feature a strong seasonality, surface chlorophyll follows upwelling intensity closely in the California (R 2 = 0.92) and Canary (R 2 = 0.88) systems, suggesting that upwelling of nutrient-rich waters fuels the chlorophyll increase in these two systems.
Indeed, comparatively low surface nitrate concentrations indicate that nitrate is used up and potentially limiting phytoplankton growth throughout the year in the California system, and for about half the year in the Canary system (Fig. 2b). For the 130 remainder of the year, in the Canary system enhanced surface nitrate concentrations are positively correlated with chlorophyll, suggesting that phytoplankton is stimulated by enhanced nitrate availability. In contrast, the Benguela (R 2 = 0.63) and the Peruvian system (R 2 = 0.90) feature replete surface nitrate over most of the year. With higher nitrate concentrations correlated with lower chlorophyll, this suggests that nitrate is not limiting.
In the Peruvian system, a strong relationship exists between deepening mixed layers and decreasing chlorophyll ( and to a lesser extent in the Benguela upwelling (R 2 = 0.41), suggesting that increasing temperatures are stimulating phytoplankton growth. The Benguela system also upholds the trend of increasing chlorophyll following the increasing temperature, though the correlation is not significant. As for MLD, the Canary system does not reveal a strong seasonal SST variance albeit features comparatively high SSTs throughout the year.
Strikingly, the Peruvian system is the only one of the four EBUS where high upwelling coincides with deep MLD (Fig. 2e,

150
R 2 = 0.79). The Canary system features a seasonality of upwelling but shallow mixed layers throughout the year, while the Benguela system does not vary much in terms of upwelling while varying in terms of mixed layer depth. In the California system, the relationship of upwelling and mixed layer depth is opposite to that of the Peruvian system, with the highest upwelling into the shallowest mixed layers.
Given the paradox that strong upwelling in the Peruvian system occurs at the time of the yearly chlorophyll minimum, it is intuitive that the concurrent deep mixed layers offset the positive impact of upwelled nutrients. In other words, more nutrients only have a strong local positive effect if concentrations are low / would be low otherwise. If concentrations are elevated, adding more nutrients will have a weak impact. We will look further into the interplay of the seasonality of mixed layers and upwelling in the Peruvian system in the following.
3.2 Modelled phytoplankton biomass, dissolved inorganic nitrogen, upwelling and the MLD in the Peruvian system 160 We use a regional ocean circulation model coupled to a marine biogeochemical model (CROCO-BioEBUS) to further analyse when upwelling is relatively weak (Fig. 3a,b). Less nitrogen is available within a shallow MLD compared with the rest of the year. In austral winter (July to September), when upwelling brings up ample nitrogen into the deep mixed layer, surface phytoplankton concentration is lowest.

Biomass dilution by the deepening mixed layer
Dilution of phytoplankton in deepening winter mixed layers is a key driver behind the seasonality of surface phytoplankton 170 concentration. Within the research area, the MLD shows a seasonal variation with the shallowest mixed layer in austral summer (around 10 m), and the deepest mixed layer in austral winter (around 45 m). Phytoplankton is vertically well mixed within the mixed layer throughout the year (Fig. 3c). In austral winter, in the 'deep-mixing' regime, phytoplankton is evenly distributed over a relatively deep mixed layer, diluting phytoplankton biomass. Accordingly, phytoplankton biomass concentrations in the mixed layer, and along with it at the surface, decrease. Hence, seasonal mixed layer deepening and shoaling alone is an 175 important factor in driving phytoplankton concentrations at the ocean surface as observed for instance based on satellite images.
While dilution causes a decrease of winter surface phytoplankton biomass, it explains only part of the observed biomass decrease: the decline persists, even though attenuated, if we integrate phytoplankton over the mixed layer (Fig. 3b). The phytoplankton concentration at the surface (and within the mixed layer) declines by around 70% while declines by around 30% for MLD-integrated biomass between late April and late July (shaded area in Fig. 3 & 4, hereafter referred to as the 180 decline phase). The decline of surface phytoplankton concentrations can be attributed to the decline due to the increase of the mixed layer volume ∆V (dilution effect, see Eq. 5) and a decrease of the biomass ∆B within the mixed layer (through local biological and physical processes, see Eq. 5). During the decline phase, ∆V contributes slightly more than half to the concentration change while ∆B contributes slightly less than half. That is, the dilution effect due to the deepening mixed layer in the decline phase amplifies the decline of surface biomass concentrations by about a factor of two. Yet, dilution can not fully 185 explain the low phytoplankton biomass in conditions of ample supply of nitrogen, MLD-integrated biomass still declines by around 35%.  3.4 Biological and physical processes change total biomass within the mixed layer

Disentangling physical and biological processes
Besides dilution due to the deepening mixed layer, the imbalance of a series of biological and physical processes during the 190 decline phase also diminishes phytoplankton concentrations. To disentangle their contributions to the decline of phytoplankton concentration without the complication of the dilution effect, we next analyze the change of phytoplankton biomass integrated over the mixed layer (Fig. 4a) and its drivers, that is the mixed layer budget of phytoplankton biomass (Eq. 1, Supplementary   Fig. C2). We separate biological processes (e.g. primary production, grazing from zooplankton, natural mortality, exudation, sinking) and physical processes (mixing, advection and entrainment) that affect the integrated biomass (Fig. 4b). Throughout the year, the net biological flux is positive, acting as a source for MLD-integrate phytoplankton biomass, while the net physical flux is negative, acting as a sink. The balance of these terms determines if the total biomass within the mixed layer decreases or increases. Most biological and physical processes decrease from the start (t1) to the end (t2) of the decline phase ( Fig. 4cd). While mortality is larger at t2 than t1, it is only due to a sudden increase by the end of the decline phase (Fig.C2). The more rapid decrease of net biological source relative to net physical loss leads to biomass reduction during the decline phase 200 (Fig. 4b).
The biological and physical processes that promote a decline of the MLD-integrated biomass are a reduction of primary production and entrained phytoplankton, along with an enhanced phytoplankton offshore transport ( Fig. 4e-f). Among the biological processes, the reduction of primary production as the only source process overpowers the weakening of biological sink processes and significantly promotes the biomass decline. Within the physical processes, the net effect of lateral and ver-205 tical transport of phytoplankton biomass (advection) is picking up during the decline phase, leading to a net offshore export of phytoplankton biomass. In addition, while phytoplankton is entrained into the mixed layer from below when mixed layers deepen, the rate of entrainment decreases during the decline phase. The decreasing rate is due to decreasing phytoplankton concentrations at the bottom of mixed layers as mixed layers deepen. All other biological and physical processes act to oppose the decline of phytoplankton biomass, such as an attenuation of mixing out of the mixed layer over the decline phase. Details 210 regarding the two major contributors, primary production and advection, are presented in the following sections.

Factors limiting primary production
Primary production changes due to variations of both the growth factor and the biomass (Eq. 2). The growth factor (calculated as in Eq. 3, Fig. 5a) combines the effects of light, temperature and nitrogen on phytoplankton growth. It shows a clear decrease 215 (around 30%) during the decline phase. Optimal phytoplankton growth conditions are reached in March despite the low dissolved inorganic nitrogen (DIN) conditions, with the warmest and brightest environment. The lowest growth rate occurs just after the decline phase despite relatively high nitrogen concentrations due to limiting light and temperature conditions. Strong light limitation experienced by phytoplankton in combination with low temperatures slows down the growth during the decline phase. Light conditions for phytoplankton growth are best in March when the water is rather stratified and worsens 220 over the decline phase to a minimum in August when the water column is mixed deepest. The light-related growth factor declines by 17% during the decline phase and would decrease the growth factor by around 60% in the absence of other limiting factors (estimated based on Eq. 3). The decreasing temperature is the second most important factor to slow the growth rate in the decline phase. The temperature-related growth factor reaches its maximum by March, similarly to the light-related growth factor, and its minimum by October. The temperature-related growth factor declines by 12% and would decrease the growth 225 factor by around 40% in the absence of other limiting factors during the decline phase. In contrast, the seasonality of the growth factor due to nitrogen shows the opposite seasonality compared to the total growth factor. Clearly, light and temperature regulate primary production and override the effect of nitrogen supply during the decline phase. Therefore, while light is the dominant mechanism that reduces productivity towards winter, we find that temperature plays a relevant secondary role. ML-integrated phytoplankton biomass decline. pp stands for primary production; graz for consumptive mortality; mort for natural mortality; exu for exudation; sink for sinking; mix for mixing; adv for advection and entr for entrainment.
Stronger light and temperature limitation during the decline phase are due to deeper mixing and stronger upwelling, respec-230 tively (Fig. 5b-c). While upwelling intensity is approximately correlated with deep mixed layers occurring when upwelling is high, the peak of upwelling happens just after the deepest mixed layers. The variation of MLD-averaged light limitation is correlated (R 2 = 0.92) with the change of MLD. As phytoplankton is evenly distributed within the mixed layer, deeper MLD means more phytoplankton is exposed to a relatively lower light condition on average in the decline phase, with a minimum in  upwelling intensity is highest. In the decline phase, cold waters are upwelled into the mixed layer at a higher rate, further damping phytoplankton growth in addition to the limiting light conditions. Reduced winter surface solar radiation and heat loss to the atmosphere, also play a role in the seasonality of the light and temperature growth factors, respectively (Fig. C3), yet of much less importance (not shown).

Enhanced upwelling and offshore transport of phytoplankton
An enhanced advective loss of mixed layer phytoplankton is a second-order process promoting the decrease of MLD-integrated phytoplankton biomass during the decline phase. Like nutrients, phytoplankton biomass is affected by the seasonality of upwelling and offshore export of waters. Phytoplankton growing at the bottom of the mixed layer is being upwelled and later pushed offshore along with the phytoplankton that is growing in the mixed layer. During the decline phase, the upwelling 245 and offshore transport of water increases and more of what is produced in coastal waters is exported offshore: 4% of primary production is lost via advection by the end of the decline phase compared to 2% gained at the beginning. This greater loss of biomass due to divergent advection is mainly caused by stronger upwelling in the decline phase (Fig. C4).

Discussion
4.1 Mixed layer depth drives surface phytoplankton biomass seasonality in the Peruvian upwelling system 250 The regional ocean circulation-biogeochemical model that we use successfully reproduces the "seasonal paradox", with seasonal out-of-phase surface chlorophyll concentration and upwelling intensity, as derived from observations. As shown in the results, the low surface chlorophyll concentration in high upwelling conditions in austral winter is constrained by a combined effect of MLD-driven processes and upwelling-driven processes. Under high upwelling conditions in austral winter, phytoplankton is diluted over a deeper mixed layer, leading to a decrease of the mixed layer, and likewise, surface phytoplankton 255 concentrations by over 50%. Also, phytoplankton growth is slowed down as they experience deteriorating light and temperature conditions. On top of that, strong upwelling pushes phytoplankton offshore.
Several previous studies have also focused on the possible reasons behind the seasonal paradox in the PUS. Echevin et al. (2008) also suggested based on model results that relative low surface chlorophyll concentration in austral winter off the Peruvian coast is generated by the combined effect of dilution and deteriorating light limitation with deepening mixed layers.

260
In addition, Messié and Chavez (2015) find that more severe iron limitation under low light condition could also be one of the reasons behind low primary production under high upwelling conditions. According to results from culture experiments, phytoplankton iron demand would increase under light limitation conditions (Sunda and Huntsman, 1997). Based on observations, Friederich et al. (2008) suggest that winds in the winter high-upwelling conditions favor curl driven upwelling, which would draw more offshore iron-deficient waters to the surface. On the contrary, a model study (Albert et al., 2010) finds that 265 stronger wind-curl driven upwelling actually is recruiting more nutrient-rich water from a shoaling coastal undercurrent, thus contributing to enhancing surface chlorophyll concentrations. We cannot assess the role of iron in regulating the seasonality of phytoplankton biomass because our biogeochemical model does not simulate iron (see Sect.B). Nevertheless, our study confirms the importance of vertical redistribution of biomass and light limitation due to vertical mixing. Here, we emphasize the importance of the deepening mixed layers in the decline phase.

Upwelling into deep mixed layers: A unique feature of the Peruvian upwelling system and its implications
As we just argued in the previous paragraphs using the differences of the seasonalities of MLD and upwelling in the Peruvian system. The upwelling of nutrient rich waters happens when growth conditions are the worst, i.e. light availability is lowest due to deep mixed layers. Also, the upwelled waters are comparatively cool and in deep mixed layers may warm up relatively slowly. At the same time, the upwelling will charge the deep mixed layers with nutrients that allow for growth once the mixed 275 layer shallows. That is, the upwelling into deep mixed layers in the Peruvian may precondition high summer phytoplankton production.
In contrast, in the California system nutrients are upwelled into the shallowest mixed layers. While this nutrient supply coincides with shoaling mixed layers and associated improved light conditions and reduced dilution, it does not result in as high phytoplankton concentrations as for the Peruvian system. This supports that nutrient limitation is important, as the supply 280 of nutrients to shallow mixed layers by upwelling appears insufficient to relieve nutrient limitation. We speculate that the charging of deep mixed layers as in the Peruvian system is more efficient in supplying nutrients to the euphotic zone compared to upwelling into shallow waters as in the Californian system, possibly warranting further studies. Also, if nutrients are being upwelled into deep mixed layers and allow the onset of a bloom, zooplankton standing stock might be low and take a while to catch up and eventually reduce phytoplankton biomass. On the contrary, if nutrients are being upwelled into shallow mixed 285 layers, zooplankton standing stock is probably already high and zooplankton can act on the spot to limit phytoplankton biomass.
While the Canary and Benguela systems lack a seasonality in MLD and phytoplankton, respectively, we mark a few aspects that may point towards a role of MLD also in these systems. Given that the Canary system does not feature a substantial seasonal MLD variability, it is intuitive that it follows the seasonality of upwelling intensity more strongly compared to the other EBUS.
While mixed layer conditions do not modulate the seasonality of phytoplankton they may contribute to Canary high phyto-290 plankton concentrations insofar as mixed layers are shallow throughout the year, creating favorable light conditions. Finally, the Benguela system features a rather constant upwelling through the year into varying mixed layers. The non-responsiveness of phytoplankton to the varying MLD hypothetically could be due to compensating effects of deepening mixed layers that dilute phytoplankton and deteriorate light conditions, but the same time are accompanied by enhanced supply of nutrients that are mixed up from below. The higher surface nitrate concentrations in conditions of deep mixed layers in the Benguela system 295 could be interpreted this way. Enhanced growth due to higher nutrient availability would then offset, yet not completely, the worsening light conditions and dilution. Finally, the strong response of phytoplankton in the Peruvian upwelling system to mixed layer depth would then be due to the unique situation of upwelling into deep mixed layers.
Again, other factors may also play a role in regulating phytoplankton in the EBUS next to nutrients, dilution and light associated with upwelling and MLD (see also Results section; Messié and Chavez, 2015), including the advection of biomass 300 and regulation by temperature that varies with upwelling (see Results). In addition, Lachkar and Gruber (2011) suggest that a longer residence time because of a wide shelf and weak mesoscale activity may also promote phytoplankton growing. Also, next to iron supply from the shelves and upwelling of source waters, Fung et al. (2000) found that atmospheric deposition of iron varies between EBUS.
The interplay of mixed layer depth and upwelling that leads to the seasonal paradox in the PUS is further propagating up the food chain and modulates the trophodynamics. In austral summer, when the shallow mixed layer along with the addon effect from upwelling supports the highest phytoplankton biomass and primary production, it provides an ideal feeding place for zooplankton. In contrast, in winter zooplankton is facing a food shortage, less efficient grazing due to dilution and transports due to enhanced upwelling. Similar to the spatial match-mismatch observed for phytoplankton and top predator in 310 Benguela system (Grémillet et al., 2008), mesozooplankton with its slower growth rate may also be affected by upwelling, with phytoplankton thriving near the coast and zooplankton concentrated further offshore. In our model, mesozooplankton is responsible for the major part of the export in the coastal upwelling region. During the productive season, the faecal material of mesozooplankton accounts for close to 100% of the sinking matter which is in good agreement with what is found in Stukel et al. (2013) for the California system.

315
Both primary production and export are found to be determined by the mixed layer dynamics and foodweb structures (Ducklow et al., 2001;Turner, 2015;Steinberg and Landry, 2017). The efficiency of the export, defined as the ratio of export to primary production, is also related to trophodynamics. Export efficiency depends, as mentioned above, on the composition of the exported material. In the PUS, it is positively correlated with net primary production (Fig. C5). Mesozooplankton produces fast-sinking large detritus, which enhances the export efficiency during the productive season. In both the Peruvian and Cali-320 fornia systems, high primary productivity coincides with a shallow MLD. What is different in the California system compared to the Peruvian system is that in the former the time of shallow mixed layers and high primary productivity is also the time of strongest upwelling. Strong upwelling and offshore transport may lead to the above mentioned spatial mismatch between phytoplankton and mesozooplankton, because mesozooplankton grows more slowly. This may then result in comparatively low large detritus despite high primary production. Therefore, high primary production may not necessarily tie in with high 325 mesozooplankton biomass and subsequent high export. Indeed, Kelly et al. (2018) observed that export efficiency is negatively correlated with net primary productivity in the California system. However, they suggested that this negative correlation arises from a seasonal decoupling of export and particle production through long-lived, slowly sinking particles that would introduce a temporal lag of mesozooplankton production and export to depth. In addition, Henson et al. (2019) find a negative correlation between export efficiency and primary productivity on a global scale. They imply in their study that it is not just the phyto-330 plankton community but also the foodweb structure important to export efficiency. We suggest that the role of the interplay of the mixed layer and upwelling in EBUS and ecosystem functioning is closely linked and warrants further examination.

Potential change under global warming
Our findings suggest that for an assessment of the response of the EBUS to climate change, it is important to consider the potential change of the interplay between the mixed layers and upwelling dynamics. Phytoplankton will inevitably be influenced 335 by climate change, responding to the changes in the biotic and abiotic environment. Impacts in a changing climate will arise from changes of stratification and upwelling that further lead to shifting growth conditions due to changes of light, temperature and nutrient (Behrenfeld, 2014). Previous observations indicate that waters near the coast have cooled since the 1950s, possibly due to an increase in upwelling (Gutiérrez et al., 2011). A recent regional modelling study (Echevin et al., 2020) projects an intensive surface warming along with a weakening of wind-driven upwelling. While studies typically focus on changes in upwelling, our results suggest that MLD changes maybe even more relevant. Assuming that winds will never cease entirely and hence there will always be some upwelling that recharges the mixed layer with nutrients, shoaling of the mixed layer caused by intensive surface warming may dominate the response of phytoplankton and ecosystem functioning. A shoaling MLD may release the phytoplankton from the strong dilution and light limitation in austral winter, along with better light and temperature condition in austral summer, leading to the expectation of an attenuation of the seasonal paradox in future. While modulation 345 of growth conditions due to changing temperatures and nutrient supply from changing upwelling also may play a role, Echevin et al. (2020) simulations suggest that surface chlorophyll overall will increase in the PUS due to global warming.

Conclusions
In summary, CROCO-BioEBUS performs well compared to observational data and successfully reproduces the "seasonal paradox" with an out-of-phase relationship of surface chlorophyll and upwelling intensity in the Peruvian coastal waters. The into the deepest mixed layers, whose combined impacts lead to the seasonal fluctuation of surface chlorophyll concentration.
We find that the seasonal variability of phytoplankton further propagates up the food chain and affects the trophodynamics, 355 and ultimately export efficiency. Therefore, a more thorough understanding of the interactions behind the mixed layer and upwelling dynamics along with the food web processes will help to better project how coastal upwelling ecosystems, and in particular the Peruvian system, may vary under climate change.
Code availability. CROCO and BioEBUs models are available at http://www.croco-ocean.org Data availability. The model data used in this paper are available via the corresponding author  Figure A1. Bathymetry of the "parent" (1/4 • resolution) and "child" (1/12 • resolution) domains. White lines near the coast highlight the focus region.

A2 Adjustment of biogeochemical model parameters 365
The parameter setting is the same as in José et al. (2017), with only a few biological parameters adjusted to make the ecology (phyto-and zooplankton biomasses, productivity) better fit observational data. The changed parameters along with value ranges from literature are listed in Table A1 and will be further explained below.
Here, we assign a higher mortality rate for large phytoplankton to simulate the potential impact of virus infection during bloom conditions (Suttle, 2005). Simulated phytoplankton biomass and its seasonality has been calibrated and evaluated against 370 chlorophyll concentration data from MODIS monthly climatology data (https://oceancolor.gsfc.nasa.gov/). Nitrate has been evaluated based on WOA and cruise data while simulated MLD has been validated against the ARGO mixed layer database (Holte et al. (2017), http://mixedlayer.ucsd.edu/).

B1 Surface chlorophyll concentration
The large-scale spatial pattern of annual average surface chlorophyll of the monthly climatology of MODIS data and CROCO-BioEBUS are similar (Fig. B1), with higher chlorophyll concentrations in coastal regions and lower concentrations offshore (note that chlorophyll is shown in log-scale). The satellite data features a higher cross-shore chlorophyll concentration gradient compared to the model simulation. The model's overestimation of the low offshore chlorophyll and hence weaker cross-shore 380 gradient potentially is due to the lack of iron limitation in the model. Apart from that, the model is also not able to correctly capture the alongshore pattern (Fig. B1), i.e. it misses two observed high surface chlorophyll concentration patches between 8 • S to 10 • S and 12 • S to 14 • S (Bruland et al., 2005). Within a 200 km band near the coast, both satellite data and the model simulation show a similar seasonality with maximum chlorophyll concentrations exceeding 4 mg /m 3 from March to April and minimum concentrations around 2 mg /m 3 in August. In general, simulated surface chlorophyll concentrations agree reasonably Figure B1. Annual mean surface chlorophyll concentration (in log(chl (mg m −3 ) −1 )) distribution of (a) MODIS and (b) CROCO-BioEBUS.
White lines highlight the focus region.

B2 Surface nitrate concentration
The simulated surface nitrate distribution shows the same seasonality as observations from the World Ocean Atlas (WOA; Garcia et al., 2019) (Fig. B2). The simulated surface nitrate concentration in the coastal region is biased high compared to the 390 WOA data. This may be partly due to the WOA data failing to capture high-nitrate concentrations due to coastal upwelling.
This notion is supported by nitrate concentration data from a cruise in austral summer that show nitrate concentrations in the coastal region are high compared to the model data.

B3 Mixed layer depth
We validate the simulated MLD against the gridded ARGO mixed layer dataset (Holte et al. (2017), http://mixedlayer.ucsd.edu/) 395 both in terms of spatial pattern and seasonal variability within the research area (Fig. B3). The annually averaged spatial distribution of MLD within the research area presents the same features as ARGO: shallower MLD in the coastal region (around 20 m) and deeper MLD in the offshore region (around 80 m). The simulated seasonal variability of MLD within the research region generally follows the seasonal trend of the Argo data. The water column within the research region is most stratified in February to March and most deeply mixed in August. Although simulated MLD in austral winter is somewhat deep, the 400 simulated MLD are largely within the range of the ARGO data.

B4 Sea surface temperature
The simulated SST has been validated against monthly climatological MODIS data in terms of both spatial pattern and seasonal variability within the research area (Fig. B4). The annually averaged spatial distribution of SST is well simulated by the model.
The model successfully captures the cold coastal upwelled water as well as slightly warmer water masses further offshore.

405
The simulated SST seasonality within the research region generally follows the seasonal trend of the observations, with a cool bias of less than 1 o C. The surface waters within the research region are warmest in February to March matching the modelled/observed shallowest mixed layers and coldest from August to October. In general, the simulated SST matches the observations well both in terms of spatial pattern and seasonal variation.