Resolving ecological feedbacks on the ocean carbon sink in Earth system models

The Earth’s oceans are one of the largest sinks in the Earth system for anthropogenic CO2 emissions, acting as a negative feedback on climate change. Earth system models predict, though, that climate change will lead to a weakening ocean 10 carbon uptake rate as warm water holds less dissolved CO2 and biological productivity declines. However, most Earth system models do not incorporate the impact of warming on bacterial remineralisation and rely on simplified representations of plankton ecology that do not resolve the potential impact of climate change on ecosystem structure or elemental stoichiometry. Here we use a recently-developed extension of the cGEnIE Earth system model (ecoGEnIE) featuring a trait-based scheme for plankton ecology (ECOGEM), and also incorporate cGEnIE's temperature-dependent remineralisation (TDR) scheme. This 15 enables evaluation of the impact of both ecological dynamics and temperature-dependent remineralisation on the soft-tissue biological pump in response to climate change. We find that including TDR strengthens the biological pump relative to default runs due to increased nutrient recycling, while ECOGEM weakens the biological pump by enabling a shift to smaller plankton classes. However, interactions with concurrent ocean acidificationcarbonate chemistry cause opposite sign responses for the carbon sink in both cases: TDR leads to a smaller sink relative to default runs whereas ECOGEM leads to a larger sink. 20 Combining TDR and ECOGEM results in a net strengthening of the biological pump and a small net reduction in carbon sink relative to default. These results clearly illustrate the substantial degree to which ecological dynamics and biodiversity modulate the strength of the biological pumpclimate-biosphere feedbacks, and demonstrate that Earth system models need to incorporate more ecological complexity in order to resolve climate-biosphere feedbackscarbon sink weakening.


Introduction 25
Oceans absorb about a quarter of anthropogenic carbon dioxide emissions, drawing down around 2-3 GtCPgCyr -1 in recent decades (Ciais et al., 2013;Friedlingstein et al., 2019;Gruber et al., 2019). The mechanisms of carbon sink processes are well understood: solubility (dissolution) and biological (soft tissue and hard carbonate) pumps gradually transfer carbon to the deep ocean where it remains on timescales of several centuries to millennia (Broecker and Peng, 1982). However, increasing ocean temperature as a result of global warming could potentially lead to a weakening of this ocean carbon sink (Arora et al., 2013;30 Ciais et al., 2013). The global carbon sink uptake rate was observed to decline by ~0.91%yr -1 between 1959 and 2012, of which approximately 40% is estimated to be due to sink feedback responses of sink processes (nonlinear carbon-cycle responses to CO2 and carbon-climate coupling) with the oceans playing a large role (Raupach et al., 2014). The combined effect of future feedbacks on both land and ocean carbon sinks reduce the RCP4p54.5-compatible anthropogenic carbon budget by ~157 ± 76 GtCPgC (Ciais et al., 2013). 35 This sink weakening might therefore act as a positive feedback on anthropogenic warming (Steffen et al., 2018). However, many of the Earth system models (ESMs) used to make these carbon sink projections do not incorporate sufficient ecological complexity to fully resolve these feedbacks, including for the ocean the impact of both warming and acidification on metabolic dynamics, ecosystem structure, and changing nutrient stoichiometry (Ciais et al., 2013). Of the ten ESMs from the Coupled 40 Model Intercomparison Project Phase 5 (CMIP5) used for carbon sink projections in the fifth assessment report of the Intergovernmental Panel on Climate Change (IPCC AR5), only one resolves the impact of warming on organic carbon remineralisation, three resolve different plankton sizes, and three resolve changing nutrient usage ratios (discussed in Background below), all of which critically influence the biological pump in a warming ocean. While there have been some improvements in the next generation CMIP6 ESMs, most still use a fixed remineralisation parameterisation for exported 45 organic carbon and feature broad size classes rather than a full spectrum of plankton size classes.
In this study we investigate changes in the biological pump in response to climate change and ocean acidification using ecoGEnIE, an ESM of intermediate complexity (EMIC) with more complex biogeochemistry and ecosystem dynamics than present in most CMIP5 ESMs. The ecoGEnIE model allows temperature-dependent remineralisation (TDR), greater 50 biodiversity via size trait-based plankton ecology, and flexible elemental stoichiometry. This combination allows the impact of metabolic and ecological dynamics on the biological pump and the ocean carbon sink in response to climate change to emerge, while the choice of an EMIC makes such additional complexity computationally tractable. We simulate a suite of historical and future climate change scenarios and assess the impact on the ocean carbon sink of replacing the default Fixed Profile Remineralisation (FPR) remineralisation parameterisation with the TDR temperature-dependent scheme and/or the 55 replacing the original NPZD-basedparameterised biogeochemistry module with ecoGEnIE's new explicit trait-based plankton ecology scheme.
This manuscript is structured as follows. In section 2 we give detailed background on the role of the biological pump, how it may be affected by climate change and ocean acidification, and to what extent current Earth system models resolve these 60 effects. In section 3 we describe the ecoGEnIE model and our experimental setup. In section 4 we describe the results of our experiments, focusing on the contrasting results for biological pump strength and the ocean carbon sink across the different model configurations. And finally in section 5 we discuss the implications and limitations of our results.

Background
The primary driver of a weakening ocean carbon sink in response to anthropogenic climate change is the reduced CO2 65 dissolution capacity of warmer water (i.e. a weaker solubility pump), but changes in the biological pump modulate this physicochemical process by affecting the vertical partitioning of carbon within the ocean., In general Earth system models project a weakening of the biological pump as a result of ocean stratification reducing nutrient availability , but a reduced efficacy of the biological pump due to increased marine bacterial respiration has also been suggested as an important factor in past warm episodes (Boscolo-Galazzo et al., 2018;John et al., 2014a;Olivarez Lyle and Lyle, 2006). 70

[Figure 1][Figure 1]
The biological pump describes the fixation and export of carbon and nutrients from the surface to the poorly ventilated deep ocean by biological activity. The vast majority of this organic matter is remineralised as it sinks and is later gradually returned 75 in dissolved form to surface waters by ocean upwelling (Figure 1). The formation and export of calcium carbonate shells (Particulate Inorganic Carbon; PIC) also forms part of the biological pump, but hereafter we focus on the soft-tissue biological pump as it is the dominant driver of surface carbon export . but a reduction in the efficacy of the biological pump due to an increase in marine bacterial respiration has also been suggested as an important factor in past warm episodes (Boscolo-Galazzo et al., 2018;John et al., 2014a;Olivarez Lyle and Lyle, 2006). Despite this, the ecological dynamics 80 affecting the soft-tissue pump has had less attention in plankton model development than resolving calcifier and silicifier plankton shells .

85
After organic carbon is fixed in the surface euphotic layer by phytoplankton and some is consumed by zooplankton, Particulate Organic Carbon Matter (POMC) begins to be remineralised by detritivorous bacteria as it falls through the water column as POMC rain. Most POMC is remineralised to Dissolved Organic Matter (DOM) near the surfacewithin the epipelagic mixed layer (~0-200m) where the nutrients released are rapidly recycled into 'regenerated' production (Dugdale and Goering, 1967) and in the mesopelagic zone (~200-1000m) below, but up to 4-12 GtCPgCy -1 of Particulate Organic Carbon (POC) reaches 90 the deepleaves the surface ocean (Ciais et al., 2013;Dunne et al., 2007;Henson et al., 2011Henson et al., , 2012Mouw et al., 2016a). This remineralisation profile follows a power law distribution, with a rapid geometric decline in export flux from the base of the mixed layer to a small asymptotic flux by ~1000m. Once in the poorly ventilated deep ocean, where it the surviving POM (most of which is subsequently remineralised to DOM) becomes part of the long-term carbon sinkremains on centennial-tomillennial timescales before being eventually returned to the surface by upwelling, while a tiny fraction of mostly recalcitrant 95 POM is buried in sediment and so sequestered on geological timescales (Ciais et al., 2013;Dunne et al., 2007;Henson et al., 2011Henson et al., , 2012b.
Theis simplified representation of plankton ecology and the biological pump shown in Figure 1 forms the basis of many marine biogeochemical models, such as the one-size fixed-trait phyto-and zooplankton classes in the common NPZD (Nutrient-100 Phytoplankton-Zooplankton-Detritus) scheme (Friedrichs et al., 2007;Kwiatkowski et al., 2014). This approach misses many important biogeochemical processes though, prompting the development of 'dynamic green ocean models' which introduce multiple Plankton Functional Types (PFTs) with differentiated biogeochemical roles (Aumont et al., 2003;Quere et al., 2005).
Although significant progress has been made since IPCC AR5 in optimising biogeochemical and ecological parameterisations in both NPZD and dynamic green ocean models using novel data assimilation and statistical techniques (Chien et al., 2020;Frants et al., 2016;Kriest, 2017;Kriest et al., 2020;Niemeyer et al., 2019;Sauerland et al., 2019;Schartau et al., 2017;Yao 110 et al., 2019), Critically, bothneither approach NPZD and dynamic green ocean models also fail tofully accounts for allometric effects in biogeochemistry. Cell size distribution and elemental stoichiometry being theare dominant traits controlling plankton ecosystem function and total production (Finkel et al., 2010;Guidi et al., 2009) and projections indicate that the fraction of large phytoplankton projected towill increase with nutrient availability and decrease with warming (Mousing et al., 2014).
Plankton size also has a substantial effect on POC export efficiency, with observations and models suggesting that although 115 smaller plankton favour a greater proportion of POC being exported from the surface layer this POC dominated by small, slow sinking particles degrades more rapidly in the mesopelagic zone (Leung et al., 2021;Mouw et al., 2016b;Omand et al., 2020;Weber et al., 2016). Trait-based plankton models have been proposed to cover this allometric gap, based on simulating generic ecosystem rules using key functional traits such as size rather than specific taxonomic identity, allowing ecosystem structure, biodiversity, and biogeography to emerge without being parameterised (Bruggeman and Kooijman, 2007;Follows et al., 2007;120 Harfoot et al., 2014;McGill et al., 2006). These ecosystem models still do not enable better understanding of Earth system feedbacks though because they have not been systematically incorporated in to ESMs and so do not capture wider biogeochemical and large-scale physical dynamics.
Most biogeochemical models feature fixed phytoplankton stoichiometry, often following the canonical Redfield ratio for 125 C:N:P of 106:16:1 or similar (Martiny et al., 2014;Redfield, 1934). However, real organisms can deviate substantially from this ratio, depending on cell size, functional group. and environmental conditions, with the Redfield ratio only emerging on a wider scale (Finkel et al., 2010). Climate change and ocean acidification are expected to substantially change ecosystem composition and nutrient availability, while increasing temperatures and CO2 concentrations have a direct impact on nutrient assimilation (Martiny et al., 2016;Riebesell et al., 2009). The C:P ratio has also been observed to 130 increase with decreasing P availability as phytoplankton increased their P usage efficiency, which could help maintain production and therefore export despite expansion of low-nutrient oligotrophic zones ('oligotrophication') (Galbraith and Martiny, 2015). It is therefore likely that stoichiometry of POC may change in response to ocean warming and acidification, with potential knock-on effects for the efficacy of the biological pump as a whole . Despite this, flexible stoichiometrywith nutrient uptake by phytoplankton depending on current availability and their current cell quotais rarely 135 incorporated in ocean biogeochemistry models .
Metabolic processes are also temperature-dependent, and so ocean temperature partly determines many marine biogeochemical patterns (Hoppe et al., 2002;Laws et al., 2000;Regaudie-De-Gioux and Duarte, 2012). For every 10 o C increase in temperature, gross primary productionphotosynthesis in any location is expected to increase by up to 100% (represented by a Q10 factor of 140 1-2), while average community respiration is expected to increase by between 100 and 200% (Q10 = 2-3) (Bendtsen et al., 2015;Boscolo-Galazzo et al., 2018;Eppley, 1972;Pomeroy and Wiebe, 2001;Regaudie-de-Gioux and Duarte, 2012;Sarmento et al., 2010). If warming-induced increases in respiration rates rise faster than production rates, organic matter will be remineralised more quickly, raising shoaling the remineralisation depth (the e-folding point at which most ~63% of POC is remineralised) higher up in the water column (Boscolo-Galazzo et al., 2018;John et al., 2014a;Kwon et al., 2009) and may 145 also reduce transfer efficiency within the mesopelagic zone (Fakhraee et al., 2020;Weber et al., 2016). One might expect this to reduce carbon export overall as less carbon makes it out of the surface ocean, but increased remineralisation also allows more nutrients to be recycled back into the surface, potentially resulting in more regenerated production (Leung et al., 2021;Segschneider and Bendtsen, 2013;Taucher and Oschlies, 2011). Even only a small shift in the remineralisation depth has could have a significant potential impact on atmospheric CO2, potentially acting as a positive climate feedback mechanism., For 150 example, with a global deepening of 24m (for example as a result of cooling) of 24m reduceding CO2 by 10-27 ppm in one model (Kwon et al., 2009), and so can potentially act as a positive climate feedback mechanism. Although the biological pump itself does not act as a carbon sink in long-term equilibrium (as exported carbon is returned to the surface by upwelling on millennial timescales), a change in biological pump strength could create a transient carbon sink if it enables a higher equilibrium accumulation of carbon in the deep ocean. 155 Other processes that affect the biological pump and remineralisation will also be impacted by climate change. Ocean stratification is projected to increase, as surface warming increases the temperature gradient (Ciais et al., 2013;Riebesell et al., 2009). This reduces the nutrient flux from deep to surface waters, potentially leading to an expansion of low-nutrient oligotrophic zones (hereafter referred to as 'oligotrophication') in subpolar lower latitude surface waters (Bopp et al., 2005;160 Sarmiento et al., 2004). Oligotrophication is normally assumed to leads to lower overall productivity in productive regions, but there is also evidence that warming will not substantially affect productivity in existing oligotrophic regions where production is already limited (Richardson and Bendtsen, 2017). and Oligotrophic regions may also be more productive than expected due to continued sub-surface production in deep chlorophyll maxima, but most ESMs do not resolve this phenomenonthat the depth rather than intensity of stratification determines productivity (Richardson and Bendtsen, 2019). The 165 reduction in nutrient supply may also favour smaller plankton that can better cope with warmer and oligotrophic conditions, resulting in a shift in ecosystem dynamics and function (Beaugrand et al., 2010;Bopp et al., 2005;Finkel et al., 2010). Reduced mixing rates along with surface warming also results in ocean interior deoxygenation, leading to an expansion of oxygen minimum zones, reduced nitrogen availability due to increasing denitrification, and increased phosphate release from affected sediments (Ciais et al., 2013;Keeling et al., 2010;Stramma et al., 2008). 170 The organic biological pump may also be affected by ocean acidification through shifting ecosystem composition, altered nutrient availability, and stoichiometricy effects (Ciais et al., 2013;Nagelkerken and Connell, 2015;Riebesell et al., 2009;Tagliabue et al., 2011). Acidification may increase the C:N uptake ratio and decrease the N:P uptake ratio, potentially making production more efficient (Riebesell et al., 2007;Tagliabue et al., 2011). , and Acidification could also lead to reduced particle 175 ballastingthe hypothesised process by which POC sticks to denser falling PIC protects associated POC and so increases POC exportby reducing the supply of PIC and therefore reducing the efficiency of POC export (Armstrong et al., 2001;Klaas and Archer, 2002). However, the overall effect of ocean acidification feedbacks remains uncertain (Doney et al., 2020), and many of these processes are not resolved by ESMs. Furthermore, the human-driven loss of organisms higher up the food chain as a result of overharvesting and habitat degradation has a considerable yet poorly quantified effect on the biological 180 pump (Pershing et al., 2010). Many of these factors influence and/or are influenced by both the magnitude of primary production and the remineralisation depth. [

185
Despite these known influences on the biological pump, many of the CMIP5 Earth system models (ESMs) used for the IPCC AR5's ocean carbon sink projections did not incorporated many orfew if any of these biogeochemical processes (Ciais et al., 2013;Schwinger et al., 2014). One study (Segschneider and Bendtsen, 2013) quantified the impact of including TDR, modifying the CMIP5 model MPI-ESM and its marine biogeochemistry model HAMOCC5.2, and projected an ~18 GtCPgC reduction in ocean carbon uptake by 2100 under high emission scenario RCP8p58.5. However, only one out of ten CMIP5 190 ESMs featured non-fixed POC remineralisation profiles by enabling TDR (CanESM2) (Table 1), with most instead prescribing a fixed attenuating remineralisation profile with vertical POC flux following modern ocean observations (sometimes called the 'Martin Curve' (Bendtsen et al., 2015;Dunne et al., 2007;Martin et al., 1987)). Additionally, NPZD-type models cannot fully resolve the potential impact of climate change or ocean acidification on ecosystem structure, biodiversity, and plankton size shifts as they do not resolve allometric or stoichiometric effects. Only four of the ten CMIP5 ESMs featured multiple 195 PFTs with different ecosystem functions beyond a simple NPZD scheme. Of these, only three account for plankton size in some way, and only three featured at least partially flexible stoichiometry (e.g. nutrient quotas and optimal allocation) that allow potential changes in nutrient utilisation in response to changing environmental conditions to be resolved (Kwiatkowski et al., 2018;.

200
The next generation of CMIP6 ESMs for IPCC AR6 are currently in the process of completion, so insufficient results are available for use as comparison in this study. These models show some improvements in these regards, with five models reporting an increase in the number of explicit or implicit PFT or bacteria classes, three models introducing more variable stoichiometry (although one model has instead reduced flexible stoichiometry), and two models introducing more than one sinking POC classes (Séférian et al., 2020). Despite these improvements, the CMIP6 models still only feature broad size classes 205 rather than a full spectrum of plankton size classes, only three have fully flexible stoichiometry, and most still use a fixed remineralisation profile for exported POC. Investigating changes in the biological pump in response to the physical and chemical perturbations of climate change and ocean acidification therefore requires an ESM with more complex biogeochemistry and ecosystem dynamics than present in the CMIP5these ESMs.

The ecoGEnIE model
ecoGEnIE is an extension of cGEnIEthe carbon-centric Grid Enabled Integrated Earth system model, an EMIC based on a modular framework efficiently resolving ocean circulation, biogeochemistry, and optional deep-sea sediment that has been simplified to focus on long-term carbon cycle featuring modules for 3D ocean circulation, 2D energy-moisture balance atmosphere, simplified thermo-dynamic sea ice, optional ocean sediments, and a comprehensive ocean biogeochemistry 215 module with phosphorus (in the form of phosphate, PO4) as the main limiting nutrient (Ridgwell et al., 2007;Ridgwell and Schmidt, 2010). cGEnIE has been used in many previous studies of climate-carbon cycle interactions in both modern (Tagliabue et al., 2016) and palaeo applications (Gibbs et al., 2016;John et al., 2014a;Meyer et al., 2016;Monteiro et al., 2012;Norris et al., 2013;Ridgwell and Schmidt, 2010). The default cGEnIE configuration uses a fixed remineralisation profile similar to the Martin curve (Martin et al., 1987;Ridgwell et al., 2007), but includes an optional temperature-dependent 220 remineralisation scheme which has previously been used to explore the biological pump in warm palaeo oceans (John et al., 2014b). An updated tuning of this scheme which also couples TDR with temperature-dependent export production is currently being developed (Crichton et al., 2020), but was not available at the time of this study. EMICs such as cGEnIE have lower spatiotemporal resolution than more comprehensive ESMs based on atmosphere-ocean general circulation models and so are limited in their physical realism, but they are also less computationally expensive and thus well-suited for investigating more 225 complex biogeochemical dynamics and performing efficient simulations of longer timescales or multiple scenarios (Claussen et al., 2002;Ward et al., 2018).
cGEnIE's climate model (C-GOLDSTEIN) features 3D reduced physics (frictional geostrophic, non-eddy resolving) ocean circulation model coupled to a 2D energy-moisture balance model of the atmosphere and a dynamic-thermodynamic sea-ice 230 model (Edwards and Marsh, 2005;Marsh et al., 2011). C-GOLDSTEIN is configured on a 36 x 36 equal area horizontal grid (each cell being 10 o in longitude and varying from ~3.2 o to 19.2 o in latitude), has 16 logarithmically-spaced vertical layers, and 96 time steps per year. The horizontal and vertical transport of heat, salinity, and biogeochemical tracers is calculated via a combined parameterisation for isoneutral diffusion and eddy-induced advection. cGEnIE also features a comprehensive ocean biogeochemistry module (BIOGEM) with phosphorus (in the form of phosphate, PO4) and iron as the co-limiting 235 nutrients (Ridgwell et al., 2007;Ridgwell and Schmidt, 2010;Ward et al., 2018). Organic matter production and export is parameterised in BIOGEM as a function of nutrient availability and following a fixed dissolved to particulate organic matter (DOM:POM) ratio, while CaCO3 production and export is parameterised by a saturation state-dependent particulate inorganic to organic carbon (PIC:POC) rain ratio. BIOGEM by default uses a fixed remineralisation profile similar to the Martin curve for the sinking labile fractions of both POC and PIC (Martin et al., 1987;Ridgwell et al., 2007), but includes an optional 240 temperature-dependent remineralisation scheme which has previously been used to explore the biological pump in warm palaeo oceans (John et al., 2014b). An updated calibration of this scheme which also couples TDR with temperature-dependent export production was also recently developed (Crichton et al., 2021) and is the version (cGEnIE.muffin v0.9.13) used in this paper.

The ecoGEnIE extension 245
The current cGEnIE version (cGEnIE.muffin) has recently been extended to ecoGEnIE (v.1.0) by incorporating a new scheme for plankton ecology (ECOGEM), replacing cGEnIE's implicit, flux-based parameterisation biogeochemistry module BIOGEM with an explicitly resolved and temperature-sensitive trait-based ecosystem module . In contrast to BIOGEM, biomass is now explicitly resolved, with each plankton population subject to ecophysiological processes including nutrient uptake (subject to quota saturation), photosynthesis and oxygen production (subject to light limitation, 250 photoacclimation, and seasonal light attenuation within a variable mixed layer depth), predation (subject to prey-switching, prey refugia, and prey assimilation), and mortality. Many of these processes are temperature-sensitive (nutrient uptake, photosynthesis, and predation) or size-dependent (maximum photosynthetic and nutrient uptake rates, nutrient affinities, cell carbon quotas, maximum prey ingestion rates, and DOM fraction). In this configuration of ecoGEnIE (v.1.0) there are two plankton functional types (PFTs) available: phytoplankton (with nutrient uptake and photosynthetic traits enabled) and 255 zooplankton (with predation traits enabled), with further classes such as calcifiers and silicifiers to be made available in future.
Explicitly resolving biomass also allows introduces a lag between environmental forcing and ecosystem response, allowing seasonal cycles and transient behaviour in POC production and export to emerge . In this case, As size is the dominant trait controlling plankton biogeochemical function and response to warming (Finkel et al., 2010;260 Mousing et al., 2014), and so each PFT (here only phytoplankton and zooplankton, but further classes such as calcifiers and silicifiers will be available in future) is further split into 8 size classes ranging from 0.6μm to 1900μm. Zooplankton graze on all potential prey subject to availability with an optimum predator:prey length ratio of 10. This allows a better representation resolution of biodiversity within the model relative to models without size classes, with the ecosystem capable of shifting to a different structure in response to environmental forcing. ECOGEM also includes flexible stoichiometry rather than being fixed 265 to the canonical Redfield Ratioratio (of C:N:P = 106:16:1 or similar (Martiny et al., 2014;Redfield, 1934)), allowing dynamic usage of nutrients in response to warming, ocean acidification, and nutrient availability to also be resolved (Boscolo-Galazzo et al., 2018;Martiny et al., 2016;. Dissolved Organic Matter (DOM) production is also explicit in ECOGEM and so allows a variable and plankton size-dependent POM/:DOM ratio, variations in which may have a significant impact on primary production in oligotrophic regions (Richardson and Bendtsen, 2017) and would result in reduced POM 270 export with a shift to smaller plankton classes.
Although using an EMIC such as cGEnIE/ecoGEnIE allows for greater ecological resolution, it introduces different limitations.However, cGEnIE/ecoGEnIE has relatively coarse ocean layerspatial resolution (36 x 36 equal area horizontal grid and 16 ocean layers of variable thickness) and temporal resolution (every ~4 days for C-GOLDSTEIN, every ~8 days for 275 BIOGEM, and every ~0.4 days for ECOGEM), and so is not able to sufficiently fully resolve spatial circulation and ecological patterns, vertical POC distribution, or the dynamics that potentially link stratification and deep chlorophyll maxima in oligotrophic regions Bendtsen, 2017, 2019). Subtle differences in spatial resolution and physical framework representations can have a substantial impact on circulation patterns, which could affect plankton community structure and the residence time of exported nutrients and carbon (Pasquier and Holzer, 2016;Sinha et al., 2010). Currently only two PFTs are 280 available in ecoGEnIE (phytoplankton and zooplankton, with PIC export set as saturation state-dependent ratio of POC), limiting the extent to which hard pump dynamics involving calcifiers and silicifiers can emerge in our results. ecoGEnIE has not yet been fully recalibrated to the modern ocean and does not perform quite as well against observational data for key biogeochemical tracers (DIC, ALK; PO4, O2) as cGEnIE , but the results are still broadly similar (reproducing approximately 90% of the global variability in DIC, more than 70% for PO4, O2, and ALK, and more than 50% 285 for surface chlorophyll, and broadly captures vertical distributions of these tracers).  and In this study we focus primarily on the global biological pump response rather than its spatial patterns, and are also particularly concerned with surface DIC and its relation to ocean carbon sink dynamics, and so this configuration isare sufficient for this global analysis of the global biological pump response.

Experimental setup 290
We assess the differing impacts of replacing cGEnIE's Fixed Profile Remineralisation (FPR) parameterisation with its Temperature-Dependent Remineralisation (TDR) scheme (John et al., 2014b) and replacing cGEnIE's original parameterised NPZD-based biogeochemistry BIOGEM module (BIO) with ecoGEnIE's trait-based ECOGEM module (ECO) . We test each new element both separately and in combination, analysing four cGEnIE/ecoGEnIE configurations: • ECO+TDR is ecoGEnIE (ECO) and the alternative Temperature-Dependent Remineralisation scheme (TDR) We use the global POC export flux (GtCPgCyr -1 ) from the surface layer (fixed in cGEnIE/ecoGEnIE as the top 80.8m of the ocean, compared with ~100m in some studies (Martin et al., 1987)) as our measure of biological pump strength and compare 305 cumulative changes betweenup to the years 2000 and 2100 CE, and also quantify cumulative changes in the ocean carbon sink for each configuration through the air-to-sea CO2 flux. We calculate cumulative changes in biological pump and ocean carbon sink capacity for the policy-relevant timescale of the 21 st Century CE (Table 2), but the biological pump results are also shown up to 2500 CE ( Figure 2).

310
Each configuration is run both under its default published calibration and recalibrated to result in the same preindustrial global biological pump strength (POC export of ~7.5 GtCPgCy -1 and PIC export of ~1 GtCPgCy -1 ) and similar global mean total Dissolved Inorganic Carbon (DIC), and Alkalinity (ALK), and surface DIC speciation relative to the cGEnIE/ecoGEnIE default configurationsBIO+FPR and observational data (see Supplementary Table S1 & Figures S1-S4418). The configurations were recalibrated to have as similar a carbon cycle as possible in order to make the results easily comparable 315 across the configurations, while POC export was chosen as the primary calibration constraint as the main variable being analysed. However, some differences remain between the recalibrated configurations as well as with the observational data.
The main difference between them is a higher POC sedimentation rate in the ecoGEnIE ECO configurations as a result of specifying a higher recalcitrant fraction (from ~5% to ~32-35%) in order to counter much high POC export in ecoGEnIE.
Although this recalcitrant fraction and the resulting POC rain rate is unrealistically high compared to observations, in cGEnIE 320 recalcitrant POC remains inert until sedimentation and so does not directly interact with the rest of the carbon cycle.but as POC rain in non-sediment module-endisabled configurations of cGEnIE is returned as deep ocean DIC and nutrients upon reaching the sea floor, meaning total ocean DIC is still conserved, this is acceptable and biological pump perturbations on shorter sub-overturning timescales (<500-1000y) will not significantly affect surface DIC within that time. Optimising for equivalent POC export also leads to surface carbonate concentration ([CO3]) being reduced in BIO+TDR compared with the 325 default calibration, leading to a reduced carbonate buffer for the ocean carbon sink in these runs. To constrain the impact of these calibrations on our results, we also present default calibration results in the Supplementary Material and discuss the differences in our Results.
Each model configuration is spun-up for 10,000 years and restarted at 0 CE (10000 Holocene Era, HE), and then forced in 330 emissions mode from 1765 CE with combined historical and future CMIP5 RCP total CO2 emission scenarios (3PD, 4p54.5, 6p06.0, and 8p58.5, corresponding to low, moderate, high, and very highsevere emission scenarios respectively; 3PD used instead of RCP2.6 to allow for long-term simulation beyond 2100 CE) extended through to 2500 CE in order to assess multicentennial dynamics (Meinshausen et al., 2011). We use the global POC export flux (GtCyr -1 ) from the surface layer (fixed in cGEnIE/ecoGEnIE as the top 80.8m of the ocean) as our measure of biological pump strength and compare cumulative changes 335 between the years 2000 and 2100 CE, and also quantify cumulative changes in the ocean carbon sink for each configuration through the air-to-sea CO2 flux. We calculate cumulative changes in biological pump and ocean carbon sink capacity for the policy-relevant timescale of the 21 st Century CE (Table 2), but the biological pump results are shown up to 2500 CE ( Figure   2).

Physical Climate Response
In its default configuration (BIO+FPR) cGEnIE projects surface air temperature warming of 1.

Biological Pump Strength
[ Our results show that the biological pump weakens under almost all scenarios and configurations, but adding TDR and traitbased plankton ecology with flexible stoichiometry has strong and opposite impacts on relative biological pump strength.  Figure S47). This is in line with past projections of a 7.2% decline in surface POC export under SRES A2 (warming levels between RCPs 6.0 and 8.5) during the 21 st century in a EMIC with an 365 NPZD biogeochemistry module (Taucher and Oschlies, 2011), and a selection of CMIP5 ESMs declining by between ~6 and ~19% under RCP8.5 during the 21 st century predominantly in the lower latitudes Cabré et al., 2015). In the modelcGEnIE this is primarily driven by stratification resulting in reduced surface nutrient concentrations and decreased primary production in high-productivity subpolar low-latitude waters (Figure 3, left) in line with previous model results (Bopp et al., 2005;Ciais et al., 2013;Crichton et al., 2021;Riebesell et al., 2009;Sarmiento et al., 2004). In contrast, there is an 370 increase in production in high-latitude waters, where mixing the mixed layer is already so much deeper than cGEnIE's surface layer (mostly >>100m, versus cGEnIE's ~81m surface layer; Supplementary Figure S48) that stratification and decreased mixing actually increases productivity by more effectively confining phytoplankton nutrients within the euphotic zonecGEnIE's surface layer. This partially (matchesing theoretical expectations in which stratification drives increased polar productivity by confining phytoplankton within the euphotic zone; (Riebesell et al., 2009), but the mechanism driving this 375 effect in cGEnIE is different as plankton are confined to the surface layer.

[Figure 3][Figure 3]
Adding TDR (BIO+TDR) leads to a substantially different result than the default cGEnIE configuration, with a far smaller biological pump weakening under RCP4p5 of only ~-0.33% under RCP4.5 and ~2.1% under RCP8.5 by 2100 CE, and eventually a net strengthening after 2100 (Figure 2). This occurs in the model because adding TDR results in an initial decrease in biological pump strength with warming as more POC is remineralised within the surface layerwith warming, but thiswhich also leads to a shallower remineralisation depth and an increase in nutrient recycling and regenerated production in the surface 385 layer. While an increase in nutrient recycling within the surface layer can lead to an increase in production by reducing nutrient loss, it does not directly lead to an increase in export as well as it is the reduction in export driving the increase in production.
Only a new allochthonous source of nutrients would allow sustained increases in both production and export (Dugdale and Goering, 1967;Laws, 1991;Laws et al., 2000) warming-induced shoaling of the remineralisation depth has been modelled to reduce POC export (Kwon et al., 2009). However, we find that a secondary effect of the remineralisation depth shoaling is to 390 increase PO4 concentrations just in the layer below the mixed productive surface layer (cGEnIE layers 2-3, ~81-283m) from remineralisation that would otherwise have occurred deeper in intermediate waters (Supplementaryorting Figure S4919). This in turn leads to increased allochthonous PO4 input to the surface layer through mixing, which is sufficient to lead to an elevated baseline in new production and POC export in warmer subpolar lower-latitude waters (Figure 4a) and stimulate a relative increase in POC export with further warming. This result is consistent with previous modelling, which has shown that shoaling 395 of the remineralisation depth in a common biogeochemical model leads to increased POC export (Kwon et al., 2009) a previous TDR-enabled EMIC, which foundand that including TDR in an EMIC resulted in increased Net Primary Production and a marginally smaller decrease in POC export under RCP8p58.5 (Taucher and Oschlies, 2011). A recent update to cGEnIE's TDR scheme (Crichton et al., 2021) also found a similar result, with historical warming resulting in a ~0.3% decline in POC export with TDR activated versus ~2.9% without. In contrast, in higher latitudes including TDR leads to a lower baseline POC 400 export than with FPR (Figure 4a), as colder waters result in a deep remineralisation depth and less PO4 returned to the surface layer.  Figure S50), as warming and stratification leads to oligotrophication in subpolar lower latitude waters which favours smaller plankton size classes, and is in line with previous observational and modelling studies (Finkel et al., 2010;Riebesell et al., 2009). Smaller 410 taxa produce more DOM than POM (Finkel et al., 2010) and so Tthe shift to smaller plankton classes lower in the food chain in warmer regions decreases overall POC and PIC export (Figure 4b), increases the rapidity of carbon cycling within the surface ocean, while the shift to smaller taxa lower in the food chain extends the number of trophic levels and so reduces the efficiency and productivity and biomass of the whole ecosystem (Supplementary Figure S51) (Riebesell et al., 2009), and so decreases overall POC export (Figure 4b). This decline is sufficient to counteract the negative feedback of the shift to smaller 415 particles increasing surface nutrient recycling due to shallower remineralisation (Leung et al., 2021). Activating ECOGEM also enables flexible stoichiometry, but the effect of this is difficult to disentangle from that of multiple size classes as well.
However, some patterns and trends can be seen. The preindustrial POM export C:P ratio lies above the standard Redfield ratio of 106:1 across most of the ocean outside the Southern Ocean, reaching ~200:1 in equatorial upwelling regions and the global mean closely matching recent observations of 163:1 (Supplementary Figure S52) (Martiny et al., 2014). By 2100 CE this ratio 420 increases across almost the entire ocean, especially along the Antarctic Polar Front and in the Arctic Ocean (Supplementary Figure S53). This indicates that the amount of carbon exported for every unit of phosphorus increases with warming in response to stratification, reducing surface phosphorus loss and so partly ameliorating the decline in carbon export.
Without stratification and nutrient restriction, higher temperatures in a previous ecoGEnIE study resulted in an increase in 425 export production and mean cell size despite an overall decrease in biomass . Although the increase in phytoplankton nutrient usage (which is temperature-dependent in ecoGEnIE) boosted small phytoplankton production in their study, this increase was assimilated by zooplankton grazing (which is also temperature-dependent). This allowed larger phytoplankton to compete against small phytoplankton with higher nutrient affinities, and resulted in increased particulate export from larger phytoplankton and inefficient zooplankton feeding despite lower overall ecosystem biomass (Ward et al., 430 2014). When nutrient fluxes were increased without higher temperatures, increases in small phytoplankton biomass were again limited by zooplankton grazing and made larger phytoplankton more competitive, but unlike warming alone higher nutrient fluxes facilitated both elevated total ecosystem production and export. In our ecoGEnIE results though higher temperatures are accompanied by both stratification and reduced nutrient flux in low-latitudes, resulting in an overall shift to smaller phytoplankton dominance despite warming allowing greater phytoplankton nutrient usage and grazing. The consequent 435 reduction in grazing and large phytoplankton abundance in turn accentuates the decline in POC export from declining low latitude ecosystem biomass, leading to a greater decline in POC export than BIO+FPR and BIO+TDR.
In contrast to low-latitudes, in most high-latitude waters biomass increases while mean cell size and export decline (Supplementary Figures S50, S51 & S54), and along the Antarctic Polar Front biomass decreases, mean cell size is stable or 440 increases, and POC export increases. The latter is because warming in nutrient-rich upwelling regions allows for increased zooplankton and larger phytoplankton abundance (Supplementary Figures S55-60) and therefore leads to restrained total biomass due to grazing coupled with increased export. In non-upwelling polar regions such as the western Arctic where nutrients are limited but unlike in low-latitudes warming-induced stratification does not restrict nutrient flux further, warming preferentially boosts smaller phytoplankton (6µm vs. 19µm) which along with a commensurate decline in dependent 445 zooplankton (19µm) and top-down grazing pressure leads to increased overall biomass but lower export. In the eastern Arctic this process is not as apparent due to the interference of Atlantic meridional overturning slowdown resulting in a moderate reduction in formerly elevated nutrient availability. This leads to a reduction in medium relative to small phytoplankton classes (19µm vs. 1.9 & 6µm) and a commensurate shift to smaller zooplankton classes (6 & 19µm vs. 60µm), and therefore relatively stable biomass and mean cell size coupled with reduced export 450 Adding both trait-based plankton ecology and TDR (ECO+TDR) produces a complex result, with the weakening effect of adding ECO on the biological pump partly counteracting the strengthening effect of adding TDR. The overall effect is a moderate net weakening of the biological pump by ~7.98% by 2100 CE under RCP4.5 and ~12.3% under RCP8.5 (Figure 2), as decreasing plankton size and POC export in subpolar lower latitude waters due to adding ECO reduces the capacity for 455 nutrient recycling to increase as a result of adding TDR (Figure 4c). The combined effect of ECO+TDR relative to BIO+FPR in this model is therefore an additional ~1.49% weakening of the biological pump by 2100 CE relative to pre-industrial under across the RCPs4p5 (Figure 3, rightFigure 5), resulting in ~9-118.2 GtCPg less POC being exported by the biological pump in this model during the 21 st Centuryby 2100. In all configurations and scenarios the changes in the biological pump continue past 2100 CE, and in many cases only begin to stabilise after several hundred years (Figure 2). 460 Using the published or default calibrations for each configuration instead of our recalibrations results in the same overall pattern of TDR ameliorating and ECO amplifying the biological pump weakening with warming, but with a reduced weakening for ECO+FPR and ECO+TDR, greater long-term strengthening for BIO+TDR, and a smaller rather than greater net weakening in uncalibrated ECO+TDR than for BIO+FPR (Supplementary Figure S61). However, these calibrations have substantially 465 different baseline biological pumps, with higher POC export in the uncalibrated configurations with TDR and/or ECO. High production and export leads to differing initial ecosystem structure and therefore amplified effects on remineralisation when POC export changes with warming, which acts as a confounding factor when comparing their responses.

Ocean Carbon Sink Capacity
It has sometimes been implied in previous discussions of empirical and model results that a decrease in biological pump strength directly leads to a corresponding decrease in the ocean carbon sink capacity, as less POC is exported from the surface to deep ocean and so more CO2 remains in surface waters and therefore the atmosphere (Boscolo-Galazzo et al., 2018;John et al., 2014a;Olivarez Lyle and Lyle, 2006;Steffen et al., 2018). However, reduced POC export affects many other processes, 475 which results in a nonlinear relation between biological pump strength and the ocean carbon sink capacity that can lead to counter-intuitive outcomes (Gnanadesikan and Marinov, 2008;Kwon et al., 2009).  In our simulations, the relative strengthening of the biological pump when TDR is included actually leads to a net decrease in the ocean carbon sink capacity during the 21 st centuryby 2100 CE (Table 2Table 2 (Table 2). Beyond the 21 st century though, ECO+TDR eventually results in a net increase in ocean carbon sink capacity from the 22 nd century onwards ( Figure 5).
Including trait-based ecology using size classes therefore largely but not entirely offsets the impact on the ocean carbon sink 490 of also including TDR in this model during the 21 st century, and entirely offsets the TDR-induced sink decline after that. The model thus suggests that ecological dynamics increases the resilience of plankton ecosystem functioning against the pressures of climate change.
A decrease in POC particulate export does not automatically result in a decrease in the ocean carbon sink capacity in this model 495 as a result of interactions with carbonate chemistry and ocean acidification. Adding TDR results in greater production of both POC and PIC relative to BIO+FPR in sub-non-polar regions in response to warming, as described in the Section 4.2. Along with Iincreased respiration ratesphotosynthesis and CaCO3 formation this results in an initial net decrease in surface DIC and ALK, which through DIC speciation leads to a decrease in the concentration of dissolved carbonate ([CO3]), an increase in the concentration of surface dissolved CO2 ([CO2]), and decreased pH and carbonate saturation state (Ω) (as theoretically described 500 by Zeebe and Wolf-Gladrow, 2001). This increases the partial pressure of CO2 in surface waters (pCO2), therefore reducing the capacity for additional CO2 to dissolve from the atmosphere into the ocean. This effect on the air-to-sea CO2 flux gradually limits the total DIC content for the whole ocean and therefore the ocean carbon sink as a whole (see explanatory schematic in Supplementary Figure S62). Ocean acidification also concurrently increases surface pCO2 and decreases Ω and PIC production (Supplementary Figure S63), and so adding TDR results in a synergistic interaction with ocean acidification. Conversely, as 505 shown in Section 4.2 adding ECOGEM reduces total ecosystem POC and /PIC production with warming as a result of the shift to smaller plankton taxa, leading to higher surface DIC and ALK, increased surface [CO3] and Ω, decreased surface [CO2] and pCO2, and therefore increased air-to-sea CO2 flux and total ocean DIC in the long-term. Introducing ECOGEM and the resultant oligotrophication-induced plankton size shift therefore slightly counters the ocean acidification trend.

510
It is possible that the small differences in surface carbonate chemistry between the different model configurations have a confounding effect on our carbon sink results. BIO+TDR has ~7.5% lower baseline [CO3] than BIO+FPR (Supplementary   Table S1), which as discussed in Section 3.3 somewhat reduces carbonate buffering and so could explain a proportion of the simulated carbon sink weakening through a reduced solubility pump. Using the published or default calibrations of each configuration instead (Supplementary Figure S64) reduces the long-term sink strengthening effect by ECO and enhances sink 515 weakening by TDR relative to the recalibrated configurations, which results in a sustained net sink weakening with ECO+TDR relative to the recalibrations. However, these original calibrations have substantially different baseline biological pumps and carbonate chemistry, which act as confounding factors in their response. POC production and export is higher and more resilient against warming in the original TDR and ECO calibrations (Supplementary Table S1 and Figure S64), and therefore has a reduced impact on surface carbonate chemistry. In the ECO configurations [CO3] is also much weaker (~70-80 vs, ~106 520 μmol kg -1 in BIO+FPR) and [CO2] much higher (~40-50 vs. 24 μmol kg -1 in BIO+FPR) than in the recalibrations, resulting in substantially weaker carbonate buffering in the uncalibrated configurations. Both higher and lower [CO3] in the original BIO+TDR and ECO calibrations respectively are also associated with reduced carbon sink capacity relative to the recalibrations across all configurations, while in the recalibrated configurations ECO+FPR shows an increase in carbon sink capacity despite lower [CO3] than ECO+TDR. Together this indicates that [CO3] has a relatively minor impact on the sign and 525 magnitude of our carbon sink results.

Discussion
These initial results clearly illustrate the importance of incorporating multiple dimensions of ecological complexity within Earth system models in order to capture the impact of nonlinear climate-biosphere feedbacks, biodiversity, and ecological 530 resilience on the future dynamics of carbon sinks. However, although the introduction of either TDR or ECO leads to substantially different changes in POC export in response to warming, the impact on the overall ocean carbon sink is less pronounced. Our ecoGEnIE experiments simulate a modest decline in the ocean carbon sink capacity of around ~65 GtCPgC (~0.065 GtCPgCy -1 ) during the 21st Century under an RCP8p58.5 scenario when accounting for TDR. This can be compared to a previous estimate of a ~18 GtCPgC (~0.18 GtCPgCy -1 ) decline in ocean carbon sink capacity by 2100 in response to 535 RCP8.5 made using a much simpler NPZD-based ecosystem representation that differentiated silicifying plankton (Segschneider and Bendtsen, 2013), and to the 2018 ocean carbon sink uptake rate of 2.6±0.6 GtCPgCy -1 (Friedlingstein et al., 2019). This decline is partially countered when greater ecological complexity and flexible stoichiometry is introduced as well, with a shift to smaller plankton classes in response to oligotrophication leading to an ocean carbon sink reduction of only ~2.34 GtCPgC. Other processes that are not resolved in this configuration of ecoGEnIE could also substantially affect the biological 540 pump though, such as ballasting, calcifier-silicifier trade-offs, nitrogen cycle and stoichiometry-acidification feedbacks (Buchanan et al., 2019;Dutkiewicz et al., 2015;Landolfi et al., 2017;Riebesell et al., 2007;Somes et al., 2016;Tagliabue et al., 2011), and deep chlorophyll maxima (discussed more fully below), and on longer timescales redox-dependent feedbacks (Niemeyer et al., 2017;Watson, 2016). Limited physical resolution can also have significant impacts on biogeochemistry (Sinha et al., 2010), and so also limits our results, and fFurther work is required to assess their impact of these features on our 545 estimates.
Few of the ESMs used in CMIP5 sufficiently resolve marine ecology, instead relying on simple plankton ecosystems that are often highly parameterised with minimal or non-existent ecological and metabolic dynamics (Table 1). This reduces computational expense and so allows higher resolution of important physical processes, but comes at the price of poorly 550 resolving known biogeochemical and ecological feedbacks that could can substantially affect carbon partitioning (Anderson, 2005;Ward et al., 2018). To date, gains in computational power have largely been allocated to improved resolution and physical process representation, while despite recent progress biogeochemical parameters have remained too poorly constrained to allow greater biogeochemical complexity in high resolution ESMs. However, the development of trait-based ecological models could enable ESMs to include more complex marine biogeochemical modules without compromising the 555 high resolution representation of physical processes. An approach that focuses on functional traits and generic ecosystem rules potentially reduces the need for taxonomic-specific parameterisations and also allows better representation of allometric effects. This study suggests that it is timely for the research community to debate again where future gains should be focused, in order toDevelopments enable ESMs to include more complex marine biogeochemical modules without compromising the high resolution representation of physical processes. of This biogeochemical models with higher physical resolution would 560 also allow more accurate representation of fine-scale biogeochemical processes such as the interaction of stratification, the nutricline, and deep chlorophyll maxima in oligotrophic regions Bendtsen, 2017, 2019), which issues raised in has not been possible in this study that have not been possible to explore. EMICs with lower physical resolution can more readily incorporate ecological complexity though, and remain a crucial tool for further exploring these feedbacks in the interim (Chien et al., 2020;Frants et al., 2016;Kriest, 2017;Kriest et al., 2020;Niemeyer et al., 2019;Sauerland et al., 2019;Schartau 565 et al., 2017;Ward et al., 2018;Wilson et al., 2018;Yao et al., 2019).
In this study we focus on the dominant soft-tissue biological pump, but the variable response of plankton classes with different shell types to climate change and ocean acidification will also hasve an impact on the biological pump. For instance, silicifiers with opal-based shells such as diatoms thrive in nutrient-rich waters. Segschneider and Bendtsen (2013) found that the 570 increased nutrient recycling when TDR was introduced in their model initially drives an increase in diatom production and opal export in response to climate change. In their model, tThis eventually soon leads to silicate-depleted surface waters and suppressed diatom production, allowing a subsequent increase in calcifying plankton and PIC export instead. This has the effect of reducing surface alkalinity and increasing surface pCO2, which drives a substantial proportion of the large ocean carbon sink reduction in their analysis. Despite the likely importance of this 'hard-shell' mechanism, ecoGEnIE does not 575 currently allow independent representation of calcifiers and does not represent silicifiers at all, and so the potential impact of this mechanism is not resolved by our results. However, the model of Segschneider and Bendtsen (2013) does not feature traitbased size classes or flexible stoichiometry, which we have shown is critical important for determining the soft-tissue biological pump response. In order to fully compare our results it will be necessary to repeat our simulations with the silicifierenabled ECOGEM currently under development. Together, resolving plankton size classes, TDR, flexible stoichiometry, and 580 separate silicifier and calcifier functional types will allow the response of the marine biological pump to climate change to be more fully diagnosed.
Further development will also allow the potential impact of ballasting to be assessed. Using a different EMIC, (Kvale et al., 2015(Kvale et al., , 2019 found that adding ballasting alongside calcifier functional types mitigated the biological pump response to ocean 585 warming by facilitating increased calcifier production and therefore increasing nutrient export from the surface. In contrast, activating ballasting in ecoGEnIE without separating out a competitive calcifier functional type would likely result in greater surface layer remineralisation in scenarios with reduced PIC production. However, eEmpirical observations have suggested that the ballasting effect on the ocean carbon sink is weaker than has been hypothesised (Wilson et al., 2012), making ballasting unlikely to substantially alter our findings, but it would likely result in greater surface layer remineralisation in scenarios with 590 reduced PIC production.

Conclusions
The response of the biological pump to future climate change and its role in the ocean carbon sink is critical is important for projecting climate feedbacks and the future behaviour of the ocean carbon sink, but many of the most influential Earth system models fail to incorporate sufficient metabolic or ecological complexity for this to be fully resolved. In this study, we have 595 investigatede for the first time the impact of integrating both temperature-dependent remineralisation, size-based biodiversity, and flexible nutrient usage on the biological pump and ocean carbon sink in response to climate change. As expected, wWe fouind that while adding temperature-dependent remineralisation to an Earth system model of intermediate complexity (ecoGEnIE) results in a greater weakening of the ocean carbon sink as a result of climate change. However as expected, this actually results from a relative strengthening of the biological pump itself as a result of shallower nutrient remineralisation, 600 contrary to the common expectation that the direct effect of warming further amplifies a weakening of the biological pump.
Conversely, adding trait-based ecosystem dynamics instead results in an even weaker biological pump as a result of oligotrophication favouring smaller plankton, and in turn a larger ocean carbon sink. Finally, combining both of these features results in a smaller relative weakening of the biological pump and a modest reduction in the ocean carbon sink capacity.

605
Together, this implies that the biological pump positive feedback on climate change may be larger than CMIP5 models project, but is potentially less than some other more recent post-CMIP5 model projections (Segschneider and Bendtsen, 2013;Steffen et al., 2018). This study has primarily focused on the allometric aspects of dominant soft-tissue components of the biological pump, and the results clearly illustrate the substantial degree to which ecological dynamics and biodiversity can modulate the strength of climate-biosphere feedbacks. TThese complex relations require further analyses and validation, but at present 610 comparison of model studies is a challenge because today's ESMs take such different approaches and simplifications. Traitbased ecological modules that goGoing beyond simple biogeochemical traits could in future enable ESMs to include more ecological complexity without compromising the high resolution representation of physical processes, and incorporating more ecological complexity in Earth system models will allow feedbacks such as the marine biological pump to be more fully resolved in future. 615

Author Contributions
DIAM, SEC, & KR conceived of the study; DIAM designed the study, configured and ran the model, and performed the analyses; DIAM wrote the paper with input from SEC, KR, & JR.

Conflicts of Interest
The authors declare that they have no conflict of interest 625 Figure 1: Schematic illustrating the impact of warming on the soft tissue biological pump. On the left-side, under cooler pre-industrial conditions the cGEnIE's surface layer remains fairly well mixed with the deep ocean (green arrow), returning dissolved nutrients and carbon (DNut & DOC) from remineralisation of exported POC (red arrow), while some POC is remineralised above the remineralisation depth 920 (surface red arrows) partly within the surface layer. On the right-side, warming leads to a shift to dominance by smaller plankton as well as stratification leading to less mixing between the shallow and deep ocean, while shoaling of the remineralisation depth leads to greater recycling of nutrients and carbon close to the surface layer, combining to result in an overall reduction in POC export and sedimentation.

925
Results for RCP4p54.5 (pale bluesolid lines) and RCP8p58.5 (dark reddot-dashed lines) are shown for each of the configurations (BIO+FPR dottedblack; BIO+TDRdot-dashedblue; ECO+FPRdashedyellow; ECO+TDRsolidred), and the baseline POC export and the 21 st Century (used for cumulative POC flux and ocean carbon sink capacity calculations in Table 2) marked by the horizontal and vertical dotted lines respectively. Results for all emission scenarios are shown in Supplementary Figure S21.     Features critical for resolving biological pump dynamics of CMIP5 ESMs used to simulate ocean carbon sink projections in IPCC AR5. Details based on IPCC AR5 WG1 Table 6.11, Table 9.A.1, and cited literature. Note that there are some mismatches between number of functional groups reported in the literature and the IPCC description. Highlighted cells indicate the models with the most (green/darker**) or moderately (orange/lighter*) comprehensivebut not necessarily sufficientrepresentation of the relevant model feature.   3PD, 4p54.5, 6p06.0, and 8p58.5), illustrating the relative changes in biological pump strength and ocean carbon sink capacity respectively. Colours and shading designate strengthening (green/darker) or weakening (red/lighter) of the biological pump and ocean carbon sink relative to the default cGEnIE (the BIO+FPR configuration).