A missing link in the carbon cycle: phytoplankton light absorption under RCP scenarios

. Marine biota and biogeophysical mechanisms, such as phytoplankton light absorption, have attracted increasing attention in recent climate studies. Under global warming, the impact of phytoplankton on the climate system is expected to change. Previous studies analyzed the impact of phytoplankton light absorption under prescribed future atmospheric CO 2 concentrations. However, the role of this biogeophysical mechanism under freely-evolving atmospheric CO 2 concentration and future CO 2 emissions remains unknown. To shed light on this research gap, we perform simulations with the EcoGEnIE 5 Earth system model and prescribe CO 2 emissions out to 2500 following the four Extended Concentration Pathways (ECP) scenarios, which for practical purpose we call RCP scenarios. Under all RCP scenarios, our results indicate that phytopankton light absorption weakens the biological carbon pump while it increases the surface chlorophyll, the sea surface temperature, the atmospheric CO 2 concentrations and the atmospheric temperature. Under the RCP2.6, RCP4.5 and RCP6.0 scenarios, the magnitude of changes due to phytoplankton light absorption is similar. However, under the RCP8.5 scenario, the changes in the 10 climate system are less pronounced due to temperature limitation of phytoplankton concentration, highlighting a reduced effect of phytoplankton light absorption under strong warming. Additionally, this work highlights the major role of phytoplankton

tions indicate that the abundance of phytoplankton biomass has decreased due to global warming. For instance, oceanographic measurements from 1890 to 2010 reveal that chlorophyll concentration has declined over more than 62% of the ocean surface (Boyce et al., 2014). Additionally, Polovina et al. (2008) indicate that between 1998 and 2006, low surface chlorophyll areas have expanded by 15% on a global scale although their results might not be exclusively attributed to climate change due to 25 their short time series (Henson et al., 2010;Schlunegger et al., 2020). Using an ocean-color database spanning 6 years, Mc-Clain et al. (2004) show that the oligotrophic waters expand in the Northern hemisphere while the expansion in the Southern hemisphere is much weaker. Complementing these observations, modeling studies have also investigated the future changes in net primary production due to anthropogenic warming. For instance, a CMIP6 model-ensemble study indicates a decrease in depth-integrated primary production of 2.99±9.11% by the end of the 21st century under the high emission scenario SSP5-8.5 30 (Kwiatkowski et al., 2020). However, this estimate is rather imprecise due to incomplete understanding and insufficient observational constraints; thus the projections of primary production changes show large uncertainties (Tagliabue et al., 2021).
Furthermore, using a coupled ocean-biogeochemistry model, Couespel et al. (2021) highlight a decrease in net primary production of 12% after a linear increase in atmospheric temperature reaching +2.8°C by the end of the 21st century. These changes in phytoplankton abundance, distribution and biogeography have consequently an impact on the role of phytoplankton light 35 absorption.
Different modeling studies investigate the effect of phytoplankton light absorption under global warming. It is suggested that the decrease in phytoplankton abundance will increase ocean clarity and lead to a lower biological increase of sea surface temperature (SST). A reduction of phytoplankton-induced oceanic warming could thus counteract in part the warming associated 40 with climate change (Patara et al., 2012). To study the effect of phytoplankton light absorption in a warming scenario, Sonntag (2013) modified the oceanic forcing by increasing the sea surface temperature for the whole model domain by 3°C. Taking into account phytoplankton light absorption, surface phytoplankton concentrations are enhanced and the maximum SST increase is 0.4°C compared to a present-day scenario (Sonntag, 2013). Furthermore, Paulsen (2018) uses an Earth system model of high complexity to perform simulations under a transient increase of 1% of atmospheric CO 2 per year. With phytoplankton https://doi.org/10.5194/egusphere-2023-921 Preprint. Discussion started: 24 May 2023 c Author(s) 2023. CC BY 4.0 License. ner, 2017;Paulsen, 2018) has been investigated. However, using an Earth System model of intermediate complexity, Asselot et al. (2022) study how atmospheric temperature is affected by phytoplankton light absorption. To do so, the authors compare the changes in air-sea heat versus air-sea CO 2 exchange due to this biogeophysical mechanism. They conclude that phytoplankton light absorption mainly affects the climate system via air-sea CO 2 exchange. Therefore, prescribing atmospheric 60 CO 2 concentrations for global warming simulations blurs the real effect of this biogeophysical mechanism. As a consequence, rather than prescribing the atmospheric CO 2 concentrations, we are interested in the effects of phytoplankton light absorption under future CO 2 emissions on a long timescale. To address this question we apply the EcoGEnIE Earth system model  and force the atmosphere with CO 2 emissions following the four Representative Concentration Pathways (RCP) scenarios used by the Intergovernmental Panel on Climate Change (IPCC) for their Fifth Assessment Report (Moss et al.,65 2010).

The Representative Concentration Pathways scenarios
The RCP scenarios describe possible future climate systems adopted by the IPCC (Moss et al., 2010) depending on the volume of greenhouse gases emitted in the next years ( Figure 1). Originally, there were four RCP scenarios, namely RCP2.6, RCP4.5, RCP6.0 and RCP8.5, labeled after a net enhancement of radiative forcing at the beginning of the 22 nd century (2.6, 4.5, 6.0 70 and 8.5 W/m 2 , respectively). These scenarios are consistent with socio-economic assumptions and associated greenhouse gas emissions. They comprise a stringent mitigation scenario (RCP2.6), two intermediate scenarios (RCP4.5 and RCP6.0) and a high greenhouse gas emissions scenario (RCP8.5). The RCP scenarios only span the 2005-2100 period but this study is conducted on a multi-century timescale to understand the long term climate response. As a consequence, our study requires data beyond 2100. We therefore use the Extended Concentration Pathways (ECPs) designed by stakeholders and scientific 75 groups and spanning the 2100-2500 period (Meinshausen et al., 2011). Similar to RCP2.6, the ECP2.6 represents a strong mitigation scenario including negative CO 2 emissions from 2100 to 2500. For the ECP4.5 and ECP6.0, the atmospheric CO 2 emissions start to decrease in the 21 st century while for ECP8.5 this decrease happens at the end of the 22 nd century. For practical purposes, here, referring to the RCP scenarios indicate the period between 1765 and 2500.

80
The ESM used in this study is called EcoGEnIE  and is an association between a new ecosystem component (ECOGEM) and a previous model named cGEnIE . EcoGEnIE is an ESM of intermediate complexity (EMIC) (Claussen et al., 2002) and due to the limitation of such a model, we focus on the qualitative assessments rather than on quantitative estimates of our results. Moreover, cGEnIE is widely used to study past climate systems and the carbon cycle over geological timescales (Gibbs et al., 2016;Meyer et al., 2016;Greene et al., 2019;Stockey et al., 2021). EcoGEnIE was 85 already used to analyze the role of marine phytoplankton in the warm early Eocene period  and to explore the relationships between plankton size, trophic complexity and the availability of phosphorus during the late Cryogenian (Reinhard et al., 2020). We use the same configuration as described in Asselot et al. (2021). This model contains components related to climate processes, including ocean dynamics, marine biogeochemistry, marine ecosystem, atmospheric circulation and sea-ice dynamics ( Figure 2). We do not consider a dynamical land scheme, thus the surface land temperature is equal 90 to the surface atmospheric temperature. For this study, we modify the ecosystem component and the oceanic component to implement phytoplankton light absorption.

Ocean, atmosphere and sea-ice representation
The oceanic component is a 3D frictional-geostrophic oceanic component (GOLDSTEIN) that calculates the horizontal and vertical redistribution of heat, salinity and biogeochemical elements (Edwards and Marsh, 2005). The horizontal grid (36 × 36) 95 is uniform in longitude and uniform in sine latitude, giving ∼3.2°latitudinal increments at the equator increasing to 19.2°in the polar regions. This horizontal grid has been employed as the standard resolution to study the global carbon cycle (Cameron et al., 2005). Furthermore, we consider 32 vertical oceanic layers, increasing logarithmically from 29.38 m for the surface layer to 456.56 m for the deepest layer. The model underestimates the upwelling in the northeastern Atlantic, Arabian Sea and polar regions  while it overestimates low-latitude upwellings (Ridgwell et al., 2007). However, on a global scale, 100 Marsh et al. (2011) show that the model simulates realistic upwelling.
The atmospheric component (EMBM) is closely based on the UVic Earth system model (Weaver et al., 2001). It is a 2D model, where atmospheric temperature and specific humidity are the prognostic variables. Heat and moisture are horizontally transported by winds and mixing. The incoming shortwave radiation at the top of the atmosphere depends on the planetary albedo, which varies as a function of latitude and time of the year to account for the effects of changes in solar zenith angle.

105
The net longwave radiation represents ∼45% of the total atmospheric energy balance while net shortwave radiation represents ∼25%. The radiative forcing associated with changes in atmospheric CO 2 concentrations is considered in the calculation of outgoing planetary longwave (Q P LW ). Higher atmospheric CO 2 concentration leads to higher amount of Q P LW being trapped in the atmosphere. Furthermore, the parameterization for Q P LW is taken from Thompson and Warren (1982) and depends on the surface relative humidity and atmospheric temperature through a logarithmic dependency. Precipitation instantaneously 110 removes all moisture corresponding to the excess above a relative humidity threshold. Wind velocities are prescribed following the annual average data of Trenberth (1989) and a constant and dimensionless land surface drag coefficient is set to 1×10 −3 (Weaver et al., 2001).
The sea-ice component (GOLDSTEINSEAICE) solves the fraction of the ocean surface covered by ice within a grid cell and computes the average sea-ice thickness (Edwards and Marsh, 2005). A diagnostic equation is solved for the ice surface 115 temperature. Growth or decay of sea ice depends on the net heat flux into the ice (Hibler, 1979;Semtner, 1976). Sea-ice dynamics consists of advection by surface currents and diffusion. The sea-ice component acts as a coupling module between the ocean and the atmosphere, where heat and freshwater are exchanged and conserved between these three modules.

Ocean biogeochemistry component
The biogeochemical module (BIOGEM) represents the transformation and spatial redistribution of biogeochemical tracers 120 (Ridgwell et al., 2007). The state variables are inorganic nutrients and organic matter. Organic matter is partitioned into dissolved and particulate organic matter (DOM and POM). The model includes iron (Fe) and phosphate (PO 4 ) as limiting nutrients but similar to Asselot et al. (2021), we do not consider nitrate (NO − 3 ) here. Furthermore, BIOGEM calculates the air-sea CO 2 and O 2 exchange. These fluxes depend on the gas transfer velocity, the water density, the concentration of dissolved gas in the ocean surface, the solubility coefficient calculated from Wanninkhof (1992), the concentration of gas in the atmosphere, and 125 the fraction of the ocean covered by sea ice (Ridgwell et al., 2007).

Ecosystem community component
The marine ecosystem component (ECOGEM) represents the marine plankton community and associated interactions within the ecosystem . The biological uptake in ECOGEM is limited by light, temperature and nutrient availability.
Phytoplankton is allowed to flexibly take up nutrients according to availability. The production of dead organic matter is 130 a function of mortality and messy feeding. The surface production is then distributed along the water column as a depthdependent flux. To achieve this, the flux is partitioned between POM which is predominantly remineralized below 590 m deep, and DOM which is remineralized above 590 m deep. This particular depth value has been calibrated against observations following the ensemble Kalman filter method (Ridgwell et al., 2007). In ECOGEM, the sinking speeds of organic matter are constant. The model assumes that photosynthesis is a Poisson function of irradiance and that phytoplankton growth is limited 135 though this function (Geider et al., 1998;Moore et al., 2001). The phytoplankton growth model requires NO − 3 to simulate chlorophyll synthesis but we do not consider this nutrient in our study. As a consequence, the nitrate biomass is equal to the phosphate biomass times the standard Redfield ratio of 16 . Nutrient uptake is a Michaelis-Menten function and phytoplankton growth is limited by a minimum function of internal nutrient status. Plankton biomass and organic matter are subject to processes such as resource competition and grazing before being passed to DOM and POM. The ecosystem is divided 140 into different plankton functional types (PFTs) with specific traits. Each PFT can be sub-divided into size classes with specific size-dependent traits. Yet we incorporate only two PFTs: one phytoplankton and one zooplankton species. Phytoplankton is characterized by nutrient uptake and photosynthesis whereas zooplankton is characterized by predation traits. Zooplankton grazing depends on the concentration of prey biomass and prey size, predominantly grazing on preys that are 10 times smaller than themselves. The model considers nutrients (DIC, PO 4 and Fe), plankton biomass and organic matter (POM and DOM) as 145 state variables. However, plankton biomass is not subject to transport by oceanic circulation. ECOGEM considers a dynamic photoacclimation (Geider et al., 1998) where chlorophyll-to-carbon ratio is regulated as the cell attempts to balance the rate of light capture by chlorophyll with the maximum potential rate of carbon fixation. Phytoplankton biomass can only be lost via grazing and mortality. Plankton mortality is reduced at very low biomass such that plankton cannot become extinct. The production of alkalinity is coupled to phytoplankton uptake of phosphate via a fixed linear ratio, meaning that alkalinity 150 increases while phosphate is consumed. The exports of calcium carbonate (CaCO 3 ) and alkalinity are scaled to the export of POC via a spatially uniform value which is modified by a thermodynamically based relationship with the calcite saturation state. The dissolution of CaCO 3 below the surface is treated in a similar way to that of POM.

Temperature limitation
Metabolic processes of photosynthesis, nutrient uptake and zooplankton predation are all driven through the same exponential 155 temperature limitation term . The temperature limitation scheme is given by Eq. 1: where γ T is the temperature limitation, A is the temperature sensitivity (0.05°C −1 ), T is the sea surface temperature and T ref is the reference temperature. A reference temperature of 20°C is used because most experimentally determined metabolic rates are made at this temperature (Behrenfeld and Falkowski, 1997;Goldman, 1977;Rhee and Gotham, 1981). Photosynthesis is light 160 limited, which results in a sub-exponential growth rate. Yet zooplankton predation disproportionately increases and nutrient uptake disproportionately decreases with increasing temperature, leading to limitation of photosynthesis and thus limitation of chlorophyll when temperatures exceed ∼20°C (Appendix A1).

Phytoplankton light absorption
In the original model version , light was only absorbed by phytoplankton. Following Asselot et al. (2021),  . The vertical light attenuation scheme is given by Eq. 2: where I(z) is the radiation at depth z, I 0 is the radiation at the surface of the ocean, k w is the light absorption by clear water (0.04 m −1 ), k Chl is the light absorption by chlorophyll (0.03 m −1 (mg Chl) −1 ) and Chl(z) is the chlorophyll concentration it is a downward flux from the sun to the surface of the ocean. Phytoplankton changes the optical properties of the ocean through phytoplankton light absorption, causing a radiative heating and changing the heat distribution in the water column (e.g. Wetzel et al., 2006;Anderson et al., 2007;Sonntag, 2013). We implement phytoplankton light absorption into the model (Eq. 3) following the scheme of Hense (2007) and Patara et al. (2012): ∂T /∂t denotes the water temperature change only associated with radiative heating, c p is the specific heat capacity of water, ρ is the ocean density, I is the solar radiation incident at the ocean surface, and z is depth. We assume that the whole light absorption heats the water (Lewis et al., 1983).

Model setup and simulations
We use the same model setup and parametrization as described in Asselot et al. (2021), with 32 oceanic vertical layers, primary  (Table 1). For the simulations without phytoplankton light absorption k Chl = 0 m −1 (mg Chl) −1 meaning that light is only attenuated by k w (Eq. 2). We run the simulations with prescribed global CO 2 emissions, which are the sum of the fossil, industrial and land-use related CO 2 emissions ( Figure 1). Moreover, all simulations include ECOGEM and are forced with the same constant flux of 195 dissolved iron into the ocean surface (Mahowald et al., 2006). We compare the yearly-averaged outputs of the year 2500. pendently of the RCP scenario, Figure 3 shows that our increases in SAT are in agreement with the global mean warming of Zickfeld et al. (2013) and lay in between the model ensemble minimum and maximum values. Thus, our model setup is suitable to study climate change.

Results
In this section, we are interested in resolving the effects of phytoplankton light absorption and the relative differences between To compare the strength of the biological carbon pump between our simulations, we look at vertical fluxes of POC in the water column. In our study, these fluxes are originally described by an exponential decay. However, to compare the vertical POC fluxes we compute them via a Martin curve (Martin et al., 1987), which is often used as a diagnostic tool for comparison. Even if our modeled POC fluxes follow an exponential decay function, using a Martin's curve function to compare these fluxes is a reasonable assumption because these two functions give similar vertical POC profiles in our model (Ridgwell, 2001). The  Table 2). The enhanced surface remineralization with phytoplankton light absorption is due to the higher amount of organic matter, generated by the higher primary production at the ocean surface ( Table 2). The biological pump is therefore weaker with phytoplankton light absorption meaning that more inorganic matter, such as nutrients are located in the surface of the ocean. We look at the distribution of chlorophyll at the ocean surface because this climate variable directly affects the heat distribution along the water column through phytoplankton light absorption. On a global scale, independently of the RCP scenario, phytoplankton light absorption leads to an increase of chlorophyll at the ocean surface ( Figure 4). This increase is due to two mechanisms. First, phytoplankton light absorption leads to a weaker biological pump ( Table 2). As a consequence, more labile organic matter lays in the ocean surface, increasing the remineralization and thus the surface nutrient concentrations. Second, 230 phytoplankton light absorption leads to a differential heating between the surface and bottom of the ocean. The ocean surface experiences a stronger heating than the ocean bottom due to the direct effect of phytoplankton light absorption at the surface while heat is slowly transported and redistributed at the bottom of the ocean by oceanic circulation. As a consequence, the pressure gradient along the water column is strengthened and the upward vertical velocity is enhanced (Appendix D1), bringing more nutrients at the ocean surface. The increased surface nutrient concentrations (Appendix E1) via these two mechanisms  indicating that phytoplankton light absorption always leads to a global increase of surface chlorophyll and SST.
The regional patterns of surface chlorophyll changes due to phytoplankton light absorption are similar between the RCP scenarios ( Figure 5). The largest differences of chlorophyll occur in the high latitudes. Such as, between the simulations RCP8.5-LA 245 and RCP8.5, the maximum increase of 0.4 mgChl/m 3 occurs in the northern polar region (Figure 5d). This pronounced chlorophyll response in the high latitudes is explained by two mechanisms: First, chorophyll is not subject to transport and therefore cannot be redistributed in the mid-latitudes. Second, light availability is enhanced for phytoplankton growth due to the decrease of sea-ice. For instance, the global sea-ice area decreases by 13% between RCP8.5-LA and RCP8.5, thus increasing light availability for phytoplankton growth. The upwelling and mid-latitude regions show a higher chlorophyll concentration 250 with phytoplankton light absorption. These regional patterns are due to enhanced vertical velocity caused by the differential

Sea surface temperature
Due to changes in surface chlorophyll, we expect variations in SST. Our results highlight that under the RCP2.6, RCP4.5 and RCP6.0 scenarios, phytoplankton light absorption increases the SST by ∼0.6°C (Figure 4). These assessments are higher than previous global estimates, giving a global SST increase of 0.33-0.5°C (Wetzel et al., 2006;Patara et al., 2012;Asselot et al., 260 2021). This stronger increase in SST is caused by higher increases in surface chlorophyll compared to previous assessments.
For the RCP8.5 scenario, phytoplankton light absorption only increases SST by 0.23°C. This lower increase in SST is due to the lower increase in global surface chlorophyll under this scenario. The regional patterns of SST changes due to phytoplankton light absorption are similar between the simulations but the magnitude of changes differs ( Figure 6). Independently of the RCP scenario, even if the polar regions experience a high increase in chlorophyll, they also experience the lowest increase of SST.

265
This is due to the underestimated oceanic circulation in these regions, which is due to the coarse grid resolution, limiting the heat redistribution. For instance, between the simulations RCP4.5-LA and RCP4.5, the minimum increase of 0.03°C occurs in the Southern Ocean. Even in the regions where small differences in surface chlorophyll occur, such as the subtropical gyres, we find high SST increases. The missing spatial patterns between chlorophyll and SST can be explained by the model setup.
Chlorophyll is not subject to transport while physical quantities, such as heat, are transported by oceanic currents. Therefore,

270
heat is smoothly redistributed around the globe. reasonable primary production and nutrient fields but not to get future atmospheric CO 2 concentrations. As a consequence, with this configuration, the model is known to simulate low atmospheric CO 2 concentrations (Asselot et al., 2021(Asselot et al., , 2022. Because we are more interested in qualitative assessment rather than quantitative estimates, such limitation does not affect the main findings of our study. Independently of the RCP scenario, the atmospheric CO 2 concentration increases with phytoplankton light absorption (Figure 7). For the RCP2.6, RCP4.5 and RCP6.0 scenarios, phytoplankton light absorption increases the 280 atmospheric CO 2 concentration by ∼20% while a previous study indicates an increase of 10% (Asselot et al., 2021). However, Asselot et al. (2021) do not prescribe CO 2 emissions, neglecting their effect on the atmospheric CO 2 concentration. For the RCP8.5 scenario, the atmospheric CO 2 concentration increases by 8% only, which is due to the lower increase in chlorophyll and SST. The increase in atmospheric CO 2 concentrations with phytoplankton light absorption is mainly due to the higher SST de-285 creasing CO 2 solubility, and thus increasing the oceanic CO 2 outgassing (Asselot et al., 2022). The reduced solubility pump enhances the ocean-to-atmosphere CO 2 flux by ∼10%. In contrast, the changes in the biological and carbonate pump enhance the air-sea CO 2 fluxes by <1%. However, the temperature dependence of solubility could not explain the changes in atmospheric CO 2 concentration in steady state, suggesting that the effect is transient. For the first three RCPs scenarios (but not RCP8.5), the change in atmospheric CO 2 concentration driven by phytoplankton light absorption follows a roughly linear de-290 pendence on the baseline concentration for that RCP (Figure 8). The rate of CO 2 uptake is roughly proportional to baseline concentration for the first three RCPs scenarios but is reduced for RCP8.5 because of the smaller effect of phytoplankton light absorption on SST. To validate this inference, we continue our simulations for another 1000 years with no further CO 2 emissions (Appendix C1). These additional simulations indicate that CO 2 differences decrease through time, converging towards the far smaller steady-state difference previously highlighted by Asselot et al. (2021).

Surface atmospheric temperature
Due to higher greenhouse gases concentrations, the atmospheric temperature increases with phytoplankton light absorption ( Figure 9). For the RCP2.6, RCP4.5 and RCP6.0 scenarios, the global increase in SAT is ∼0.8°C, which is higher than previous model estimates indicating a zonally-averaged SAT increase of 0.2-0.45°C (Shell et al., 2003;Patara et al., 2012;Asselot et al., 2021). However, compared to our model setup, Shell et al. (2003) use an uncoupled ocean-atmosphere model, neglecting  The regional patterns of SAT changes due to phytoplankton light absorption are similar among the RCP scenarios but the magnitude of changes differs ( Figure 10). The polar regions experience a strong increase in SAT, with the highest values occurring in the Southern Ocean. For instance, comparing the simulations RCP4.5-LA and RCP4.5, the maximum increase of 1.6°C occurs in the Southern Ocean (Figure 10b). This maximum value is due to the rather coarse grid resolution in the high latitudes. This estimate is again higher than previous local estimates (Shell et al., 2003;Patara et al., 2012;Asselot et al., 310 2021) for the same reasons described above. Furthermore, around the rest of the globe, heat is redistributed smoothly in the atmosphere.  surface chlorophyll, which in turn leads to a warming of the ocean surface. Furthermore, the higher CO 2 concentration with 320 phytoplankton light absorption leads to an enhanced greenhouse gas effect. As a consequence, the radiative forcing increases, warming the ocean surface as well. Under the RCP2.6, RCP4.5 and RCP6.0 scenarios, phytoplankton concentration is not strongly limited by temperature. As a result the impact of phytoplankton light absorption on the climate system is similar between these RCP scenarios. However, under the RCP8.5 scenario, the effect of phytoplankton light absorption on the climate system is reduced. This is due to the model setup where a SST higher than ∼20°C limits net phytoplankton concentration 325 (Appendix A1). This limit is exceeded in all the simulations but RCP8.5-LA and RCP8.5 are the only ones where the average SST exceeds 20°C (Appendix F1). Phytoplankton concentration is thus limited by temperature and the difference of chlorophyll between RCP8.5-LA and RCP8.5 is weaker than between the other simulations ( Figure 4). The response of the climate system to phytoplankton light absorption is therefore weaker under the RCP8.5 scenario. Our findings indicate that the effect of phytoplankton light absorption is smaller under high greenhouse gas emissions compared to reduced and intermediate green-

330
house gas emissions. In agreement with Patara et al. (2012), this study indicates that a severely warmer world increases ocean clarity and slows down the warming due to phytoplankton light absorption. However, the reduced effect of phytoplankton light absorption under the RCP8.5 scenario may not be as strong if phytoplankton were able to adapt to higher temperature in our model setup.

335
For the first time, using EcoGEnIE , we investigate the impact of phytoplankton light absorption under prescribed CO 2 emissions following the RCP scenarios on a multi-century timescale. However our model setup has limitations that must be overcome to improve our quantitative estimates. Most notably, our version of the model must be tuned to fit the projected atmospheric CO 2 concentrations under global warming scenarios. For instance, for the simulations following the RCP2.6 scenario, the final atmospheric CO 2 concentrations and SSTs are lower than pre-industrial levels. This is due 340 to the negative emissions for this scenario and the underestimation of the atmospheric CO 2 concentrations with our model setup (Asselot et al., 2021(Asselot et al., , 2022. As detailed previously, primary production is allowed until the sixth oceanic layer and the model has not been tuned in this configuration yet. The lower levels under the RCP2.6 scenario compared to the preindustrial levels are not an issue for our study because we exclusively focused on the effect of phytoplankton light absorption rather than on the differences between the simulations and the pre-industrial state. Furthermore, we switch on ECOGEM 345 and the RCP emission forcings at the same time. We know from previous work (Asselot et al., 2021), that switching on ECOGEM decreases the atmospheric CO 2 concentration, thus our simulations contain an effect of both drift and emissions.
However, the drifting effect is identical between simulations and therefore balances out when comparing simulations. The model inter-comparison against Zickfeld et al. (2013) suggests that this drifting effect is not an issue because the response to ECOGEM is fast enough that most of the adjustment happen in the first 200 years of simulation. With our model setup 350 we demonstrate that phytoplankton light absorption increases local SST by 0.4-1.1°C depending on the scenario considered.
These estimates are lower than previous observations showing a local increase of SST by 0.95-4.5°C (Kahru et al., 1993;Capone et al., 1998;Wurl et al., 2018 (Anderson et al., 2021). We cannot rule out that the strong phytoplankton concentration limitations in our simulations RCP8.5-LA and RCP8.5 will also occur if more PFTs were considered. Depending on the model and on the region of interest, the future of primary production is highly uncertain.

365
For instance, using a suite of nine coupled carbon-climate ESMs under the RCP8.5 scenario, Laufkötter et al. (2015) show that primary production may increase, remain stable or decrease under global warming. Though we note that the simulations of Laufkötter et al. (2015) only went out to 2100, not to 2500 as in our extended simulations. Our results highlight that phytoplankton light absorption itself increases chlorophyll leading to more heat being trapped in the ocean surface.

Implication for Earth system models 370
The traditional view is that dominant carbon cycle uncertainties come from the terrestrial response to elevated atmospheric CO 2 concentrations. For instance, the net land emissions over the 1858-2008 period is estimated as likely (66% confidence) to lie in the range from 0 to 128 GtC (Holden et al., 2013). However, this work suggests that introducing biogeophysical mechanisms such as phytoplankton light absorption leads to major carbon cycle uncertainties. For instance, with our model setup, implementing phytoplankton light absorption increases the atmospheric carbon content by 79 GtC in RCP2.6 and by 375 258 GtC in RCP8.5. This study highlights a highly uncertain feedback on the carbon cycle that is missing from 50% of the CMIP6 models (Pellerin et al., 2020). Neglecting the effect of phytoplankton light absorption on the carbon cycle can lead to incomplete future climate projections, thus this biogeophysical mechanisms should be included by default in climate models.
Code availability. The code for the model is hosted on GitHub and can be obtained by cloning or downloading: https://zenodo.org/record/5676165.

380
The user-configuration files to run the experiments can be found in the directory "EcoGENIE_LA/genie-userconfigs/RA/Asselotetal_ESD".
Details of the code installation and basic model configuration can be found on a PDF file (https://www.seao2.info/cgenie/docs/muffin.pdf).
Finally, section 9 of the manual provides tutorials on the ECOGEM ecosystem model. of the manuscript.
Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. Our special thanks go to Jana Hinners, Isabell Hochfeld, Félix Pellerin, Maike Scheffold and Laurin Steidle for their valuable comments on the early version of this manuscript. This work was supported by the Center for Earth System Research and Sustainability (CEN), University of Hamburg, and contributes to the Cluster of Excellence "CLICCS -Climate, Climatic Change, and Society".  We base our ecosystem community on the one described by Ward et al. (2018). We only use 2 PFTs: one phytoplankton group and one zooplankton group (Appendix B1). We show that the complexity of the ecosystem does not have an important impact on the climate system compared to the effect of phytoplankton light absorption (Asselot et al., 2021). Therefore, for simplification, we reduce the ecosystem complexity. To investigate the substantially reduced ocean CO 2 uptake with phytoplankton light absorption, we continue our simulations for another 1000 years with no further CO 2 emissions. The difference in atmospheric CO2 concentrations between the simulations with and without phytoplankton light absorption decreases through time. This result evidences that large CO 2 differences are driven by a transient effect of reduced CO 2 uptake fluxes, consistent with reduced CO 2 solubility under phytoplankton light absorption warming.