Calibration of key temperature-dependent ocean microbial processes in the cGENIE.muffin Earth system model

Temperature is a master parameter in the marine carbon cycle, exerting a critical control on the rate of biological 10 transformation of a variety of solid and dissolved reactants and substrates. Although in the construction of numerical models of marine carbon cycling, temperature has been long-recognised as a key parameter in the production and export of organic matter at the ocean surface, it is much less commonly taken into account in the ocean interior. There, bacteria (primarily) transform sinking particulate organic matter into its dissolved constituents and thereby consume dissolved oxygen (and/or other electron acceptors such as sulphate) and release nutrients, which are then available for transport back to the surface. Here we present and calibrate a 15 more complete temperature-dependent representation of marine carbon cycling in the cGENIE.muffin Earth system model, intended for both past and future climate applications. In this, we combine a temperature-dependent remineralisation scheme for sinking organic matter with a biological export production scheme that also includes a temperature-dependent limitation on nutrient uptake in surface waters (and hence phytoplankton growth). Via a parameter ensemble, we jointly calibrate the two parameterisations by statistically contrasting model projected fields of nutrients, oxygen, and the stable carbon isotopic signature (δC) of dissolved 20 inorganic carbon in the ocean, with modern observations. We find that for the present-day, the temperature-dependent version shows as-good-as or better fit to data than the existing tuned non-temperature dependent version of the cGENIE.muffin. The main impact of adding temperature-dependent remineralisation is in driving higher rates of remineralisation in warmer waters and hence a more rapid return of nutrients to the surface there -stimulating organic matter production. As a result, more organic matter is exported below 80m in mid and low 25 latitude warmer waters as compared to the standard model. Conversely, at higher latitudes, colder water temperature reduces the rate of nutrient supply to the surface as a result of slower in-situ rates of remineralisation. We also assess the implications of including a more complete set of temperature-dependent parameterisations by analysing a series of historical transient experiments. We find that between the pre-industrial and the present day, in response to a simulated air temperature increase of 0.9°C and ocean warming of 0.12°C (0.6°C in surface waters and 0.02°C in deep waters), a reduction in 30 POC export at 80m of just 0.3% occurs. In contrast, with no assumed temperature-dependent biological processes, global POC export at 80m falls by 2.9% between the pre-industrial and present day as a consequence of ocean stratification and reduced nutrient https://doi.org/10.5194/gmd-2019-344 Preprint. Discussion started: 27 February 2020 c © Author(s) 2020. CC BY 4.0 License.

It has been suggested that both increased grazing pressure (respiration process) by zooplankton, and primary production by phytoplankton (photosynthesis process), will have an impact on export production in a warmer world (Paul et al., 2015;Turner, 2015). However, in the simplified biologically induced export flux (Maier-Reimer, 1993) scheme (Fig. 2) we apply in cGENIE, we cannot explicitly consider the impact of increased grazing pressure in the surface waters. Rather, we are interested in the wider 100 question of the interaction of (any) temperature-dependent community production (as export production), with temperaturedependent microbial remineralisation in the ocean interior, and its impact on the global ocean carbon cycle.
We apply only the direct effect of temperature on large scale metabolic processes (plankton photosynthesis (growth) and microbial respiration). Other factors such as involving particle size distributions, particle density (Cram et al., 2018) and 'ballasting' (e.g. Wilson et al., 2012), and sinking speed (determined by particle characteristics) (McDonnell et al., 2015), are generally 105 determined within the food web and may be considered to be secondary impacts of temperature dependence as a controller on community structure. In past-climates some of these community structures may have been significantly different. Recently Boyd et al. (2019) defined additional particle pumps in the ocean, involving eddy-subduction, diel vertical migration, mesopelagic migration and seasonal lipid pumps. Including these processes is outside the scope of this study where here we focus on the large-scale effect of temperature on metabolism in the ocean biological carbon pump. 110

Standard, non-temperature-dependent model formulation
In the original version of the biological uptake scheme, -'BIOGEM' (Ridgwell et al., 2007) nutrients are taken out of the surface ocean layer according to several factors including light incidence, ice fraction, nutrient uptake limitation (Michaelis-Menten type), and a prescribed maximum uptake rate (Eq. 1). Here, the maximum uptake rate, (maximum rate of conversion of dissolved PO4, phosphate, into organic matter by phytoplankton), 125 has a calibrated value of 9.0 x10 -6 mol kg -1 yr -1 (Cao et al., 2009), while the calibrated Michaelis Menten half saturation value is 9.0 ×10 -7 mol kg -1 (Cao et al., 2009). Nutrient uptake is instantaneously converted into organic matter export (both particulate organic matter (POM) and a fraction as dissolved organic matter (DOM)), and this represents community production (see Fig. 2). This production encompasses the entire surface food web, including the action of primary producers (phytoplankton) and the effect of consumers (e.g. grazers). 130 In this export production model, increases in nutrient uptake are directly transferred to increased production of organic matter that sinks out of the surface layer (at 80m depth). In this single (PO4) nutrient scheme, dissolved inorganic carbon (DIC) is taken out of solution in the surface layer at a molar ratio of 106:1 to PO4and O2 at a ratio of -138:1 with PO4 (Redfield et al., 1963). POM is partitioned into two fractions, which conceptually are: labile (fraction 1, 'POM1'), and recalcitrant POM (fraction 2, 'POM2') (Ridgwell et al., 2007). POM sinks vertically out of the surface layer and settles with a given velocity (here: 125 m day -1 ). POM is 135 remineralised throughout the water column using a prescribed remineralisation 'curve' reflecting the decay of POM as it sinks. The prescribed remineralisation 'curve' of relative sinking flux vs. depth (e.g. see: Hülse et al., 2017) is always adhered to (Eq. 2a for POM1, Eq. 2b for POM2). In the sinking curve, the relative flux at each layer (z) is calculated according to an exponential decay function (Ridgwell et al., 2007). Where: =ℎ POM exported out of the surface layer (at 80m) length-scale (556m for POM1; 1x10 6 m for POM2effectively infinite and hence no water column decay) 145 initial proportion of POM into fraction 2 (0.055) Any POM not remineralised within the water column is instead remineralised at the ocean floora 'reflective' boundary condition assumption (see Hulse et al. (2017) for discussion).

Temperature dependent processes 150
In the temperature-dependent version of biological export production, a temperature-dependent growth rate limiter is applied to a characteristic time-scale of ambient nutrient depletion (Eq. 3). A similar scheme but without temperature-dependent remineralisation has previously been applied by Meyer et al. (2016) for PO4-only uptake, and for 2 nutrients (PO4 and NO3) by Monteiro et al. (2012).
The "Eppley curve" is often applied to model metabolic response to temperature change (Table 1). An improved fitted curve was proposed by Bissinger et al. (2008), with both being based on fitting the model to data from empirical studies. The largest difference between the Bissinger curve and the Eppley curve is the value of A (Eq.4). It makes little difference which curve we use because we calibrate Vmax (Eq.3) which is also a multiplier for the temperature growth limitation term (in Eq. 4). We use the original Eppley et al. (1972) values for A and b (in Eq. 4) as per Monteiro et al. (2012). Both the Eppley and Bissinger curves gives a Q10 170 value (where Q10 is the increase in the rate of the metabolic process with a 10°C increase in temperature) for nutrient uptake as 1.88 (Bissinger et al. 2008).
To calculate the remineralisation rate of POM, an Arrhenius-type equation is applied (as in John et al. 2014) (Eq. 5).
Where: 175 Activation energy (J mol -1 ) gas constant (J K -1 mol -1 ) absolute temperature (K) rate constant as T approaches infinity 180 This rate is calculated for local temperature and applied to the local POM flux for each POM fraction individually. For both fractions, sinking rate is 125 m day -1 , so for cGENIE's non-uniform ocean depths, the fractional loss of POM due to remineralisation in each layer (z) is as Eq. 6. https://doi.org/10.5194/gmd-2019-344 Preprint. Discussion started: 27 February 2020 c Author(s) 2020. CC BY 4.0 License. 185 Where n denotes POM fraction (either labile (1) or recalcitrant (2) -distinguished as these have different k(T) values), ∆ ( ) is the time that sinking particles on average spend in layer z.

Model tuning methodology
In previous published applications of the cGENIE model, either a temperature dependence in export (e.g. Meyer et al., 2016) OR a temperature dependence in remineralisation (John et al., 2014) have been explored in addressing varying paleo questions. Or more 190 commonly, neither (e.g. Ridgwell and Schmidt, 2010), have been used. Here, we now explore both temperature dependence in export and remineralisation scheme together (Table 2) and as a consequence, it is necessary to jointly re-tune the respective scaling factor in each scheme.
We identify three parameters needing joint re-tuning: 1, The maximum nutrient uptake rate Vmax (Eq. 3) important for export production. 2, The activation energy, Ea(1) (Eq.5) (the minimum energy required for the transformation of organic carbon 195 into inorganic carbon through respiration processes for the remineralisation of labile POC1, Particulate Organic Carbon type 1) where the labile POC1 dominates that exported from the surface. 3, The fraction of recalcitrant POC2 (denoted as rPOM Eq.2a and Eq.2b, note that in this paragraph values are described for carbon, but apply for nutrient as well) formed at the surface that plays a role in the down-column total POC flux and especially how much of the total POC reaches the very deep ocean.
The Vmax range was chosen by testing the model while retaining the remineralisation version used in John et al. (2014), and 200 selecting a range that gave a reasonable agreement with PO4 and O2 concentrations, at 4, 7 and 10. For the initial fraction of POC2, we took the John et al. (2014) version, and applied a testing range that encompassed a range 25-400% around this value (×0.25, ×1.0, ×4.0). For the Ea(1) (Eq.5) setting, John et al. (2014) used the mid of a range of 50 to 60 kJ/mol for labile POC (range identified in Arndt et al. 2013). We use a lower and higher value of that range, and a selection of values nearer the mean (Table 3). Our ensemble hence consisted of 3 different choices for Vmax, 3 different choices for initial fraction of POC2, and 6 different choices for 205 Ea (1), for a total of 3×3×5 = 45 different parameter combinations and hence model ensemble member experiments. Values for the two rate constants, A (Eq 5, for POC1 at 9×10 11 yr -1 , for POC2 at 1×10 14 yr -1 ) that were calibrated for the modern ocean and the sinking speed of 125 m day -1 in John et al. (2014) are retained.
Each of the 45 experiments in the ensemble are spun-up for ten thousand years with pre-industrial boundary conditions: the atmosphere restored to 280 ppm CO2 and -6.5 ‰ δ 13 CO2. Following on from each respective spin-up, each model ensemble 210 member is then forced from year 1700 to 2010 in a transient simulation with atmospheric composition conforming to observed (rising) mean annual trends in CO2 and (falling) δ 13 CO2. This is because global datasets are based on modern (i.e. the past few decades) oceanographic observations, where (especially shallow) distributions of nutrients and oxygen may already have been impacted by historical warming, so model-data comparison with the model pre-industrial steady-state is arguably inappropriate.
Direct atmospheric measurements and ice core data has shown that atmospheric δ 13 CO2 has dropped with increasing CO2 due to fossil fuel emissions (that have a characteristic low δ 13 C) known as the Suess effect (Keeling, 1979;Rubino et al., 2013). This affects ocean δ 13 C in a non-uniform manneraffecting (in general) nearer-surface waters more strongly due to ocean physics and circulation patterns.

Model-data comparison method
For the model-data comparison, World Ocean Atlas 2009 (WOA 2009, Levitus et al. 2010 5° data for phosphate (PO4) and dissolved 220 oxygen (O2) was rescaled to a 10°x10° grid (with a simple linear upscaling) and to the cGENIE model depth scale by averaging over the data depth points that most closely correspond to the cGENIE ocean model depth layer distribution. This depth rescaling produces a global mean depth-uncertainty of 2.2% (of the targeted cGENIE depth). As a result the depth rescaling results in additional small uncertainties of ± 0.01µmols l -1 (at 1 standard deviation) for PO4, and ± 0.02 µmols l -1 for O2. The cGENIE model output from year 2010 of the transient experiments was also rescaled to 10°×10°, and converted to µmols l -1 (from mol kg -1 using 225 modelled water density) for modelled O2 and PO4, for a direct comparison with the data. In all cases, latitudes higher than 80° were neglected. For PO4, we statistically compare the surface concentration, important for constraining nutrient uptake rates, and the global ocean distribution, which strongly reflects remineralisation and hence the strength and efficiency of the biological pump (plus ocean circulation). For dissolved O2, we statistically compare model and data between 283m to 411m (cGENIE ocean level 4 centred at 346m) as an indication of the dissolved oxygen depletion caused by remineralisation near the bottom of the mixed layer and how 230 well the model can represent this. We also compare the global ocean dissolved oxygen distribution. Given that we are utilizing model-derived temperature distributions in the ocean to project nutrient and oxygen concentrations (plus δ 13 C distributions) which we then contrast with the respective observed data, we additionally re-grid temperature data (producing an uncertainty of ±0.1°C) in the same way as O2 and PO4 so as to enable us to elucidate biogeochemical biases arising from model-data temperature mismatch.
For assessing water column profiles, we defined a set of ocean regions, shown in Figure 3. These regions are similar to 235 those used by Weber et al. (2016), but with some regions reduced in size or separated (Subtropical Pacific limited to South Pacific and North Indian Ocean added). This was done so that within each region ocean water characteristics are broadly similar (including temperature, nutrients, oxygen, salinity) as well as particle fluxes being similar (as Weber et al. 2016). We compare the model distribution of δ 13 C of DIC with data from Schmittner et al. (2013) by grouping this data into regions (Fig. 3) and creating representative (mean with standard deviation) down-column profiles for δ 13 C for visual comparison with model outputs in the 240 matching region.

Tuning the temperature-dependent versionmodel vs. data
We first assess model skill in simulating the temperature distribution in the ocean, given its critical importance in the temperaturedependent calculations of metabolic processes (Fig. 4). We find a generally good model fit to ocean temperature data in mid and 245 low latitudes near-surface waters, and in capturing the first order patterns in benthic temperatures. At high latitudes, cGENIE shows https://doi.org/10.5194/gmd-2019-344 Preprint. Discussion started: 27 February 2020 c Author(s) 2020. CC BY 4.0 License.
larger differences as compared with observations, due to deficiencies in modelled ocean circulation and/or surface climate. For instance, the temperature discrepancy throughout the water column in the North Atlantic may be due to an overly-strong AMOC has the strongest control on PO4 and O2 distribution of the three variables. The overall best-fit to the data for the final CBRU setting was selected as Vmax = 10, Ea(1) = 54 kJ mol -1 , initial fraction POC2 = 0.008, where the best-fit is determined as the setting with the 255 combined overall lowest RMSE for the PO4 and O2 distributions.
Surface nutrients are important in constraining export production. Fig. 6 shows cross-plots for surface PO4 concentration for the best fit CBRU and the standard model for the selected ocean regions. The addition of temperature dependence in CBRU generally increases the surface nutrient concentration and is in better agreement with data than the standard model. The high surface nutrient regions, Antarctic and the North Pacific, are lower than data in all model cases and this is likely due to the lack of iron 260 limitation in this version of the model, as biological activity removes too much nutrient from the surface waters. In these regions, the temperature-dependent version shows slightly better fit to data than the standard model, as the colder surface water reduces nutrient uptake rates. However, the very lowest nutrient regions (e.g. some south Pacific and some west tropical Atlantic) are slightly higher compared to data for CBRU the temperature dependent version.

Performance of the temperature-dependent model compared to the standard model 265
We contrast the performance of the existing tuned, but non temperature-dependent BIOGEM scheme (CB), with the new tuned temperature dependent scheme (CBRU). Regional water-column profile model outputs for CBRU and CB are plotted against PO4 from WOA 2009 (Levitus et al. 2010) in Fig. 7. In both schemes, the model was tuned according to its fit to PO4 and both CBRU and CB show a good fit to data. Some differences can be seen between CBRU and CB in surface mid and low latitude waters, e.g.
in the North Indian Ocean and Eastern Tropical Pacific, where nutrients are higher in CBRU in better agreement with data. In higher 270 latitude waters model-data differences may be more related to ocean circulation, as for temperature (Fig. 4), with CB and CBRU very similar in Southern Ocean and North Pacific and CBRU slightly better in the North Atlantic.
We find that the addition of temperature dependent microbial processes generally increases surface nutrient concentrations (as shown in Fig. 8) compared to the standard model (except for the very high southern latitudes). This is especially the case in the low-nutrient gyres with up to 4-times higher PO4 concentration in CBRU than in CB. In the deeper ocean, nutrient concentration is 275 reduced in the temperature-dependent version except for the North Atlantic (where higher surface nutrients are delivered to the deep via the AMOC) and the high Southern latitudes (with slightly higher PO4 than the standard model).
The distribution of dissolved oxygen also provides information about the biological pump. Photosynthesis removes CO2 from ocean waters and adds O2, where respiration does the opposite. In surface waters and the mixed layer, ventilation with the https://doi.org/10.5194/gmd-2019-344 Preprint. Discussion started: 27 February 2020 c Author(s) 2020. CC BY 4.0 License.
atmosphere also results in higher oxygen concentration. As a general pattern, respiration progressively reduces dissolved oxygen 280 concentrations down the water column until a minimum is reached. Below that depth -the 'oxygen minimum zone' (OMZ)ocean circulation reintroduces more oxygenated water masses (ventilated from higher latitude and colder surface waters), slightly increasing dissolved oxygen concentrations again with further depth. In the Antarctic zone (Fig. 9), circulation patterns appear to dominate oxygen content (as CB and CBRU are very similar, but fairly dissimilar to data indicators). The North Pacific region also shows offsets between model and data for O2, PO4 and temperature, also suggesting a circulation difference. In low latitude waters 285 CBRU shows a better fit to data between the surface and 500m than CB, suggesting that oxygen depletion rates due to respiration are better described here. Overall, the intensity of the OMZ in both CB and CBRU are in reasonable agreement with data, although in low and mid latitudes warmer waters the OMZ occurs higher in the water column in CBRU than CB.

Tracing Carbon-13
Carbon-13 data are widely used as indicators of ocean circulation and of changes in that circulation over time (Lynch-Stieglitz 290 2003). Water masses have characteristic carbon-13 signatures, so changes in δ 13 C measured at any one location may be at least partially attributed to changes in the water sources. The biological pump also affects the δ 13 C of ocean waters. The process of photosynthesis fractionates the carbon that is exchanged (from the dissolved inorganic to the organic form); carbon-12 is preferentially taken up, leaving more carbon-13 in the surface waters (Schmittner et al. 2013). As summarized by Kirtland Turner and Ridgwell (2016), fractionation between POC (and DOC) and δ 13 C of CO2(aq) in cGENIE is a function of the CO2(aq) 295 concentration and based on an approximation of the model of Rau et al. (1996) (Ridgwell, 2001). This gives rise to a spatial distribution in the δ 13 C of exported organic carbon, with lower (more negative values) at higher latitudes, and higher (less negative) value towards the equator, primarily reflecting the temperature control on the concentration of CO2(aq) in surface waters. The mean flux-weighted δ 13 C of POC is around -23‰ for the pre-industrial period, and around -26‰ by the year 2010 due to the Suess effect.
As POC is remineralised in the water column, low δ 13 C carbon is released into the water, modifying the ambient δ 13 C of DIC. 300 The regional mean and standard deviation of data δ 13 C, and model CBRU and CB are shown in Fig. 10. For all regions, general patterns are similar to those seen in O2 concentration, except the Antarctic zone. Benthic and deep water absolute δ 13 C values are generally similar to data for both model settings. The model-data offset in mid-depth waters (~800m) in the sub-Antarctic zone may be due to a reduced Antarctic intermediate waters contribution in the model. This may explain similar model-offsets at this depth in the South and East-Tropical Pacific regions. In warm surface waters, δ 13 C reduces more quickly with depth in CBRU 305 than CB, as nutrient recycling is occurring faster.

POC export and implications for biological carbon pump efficiency
The inclusion of a temperature dependence term in remineralisation strongly affects export production via the rate of nutrient recycling, and fundamentally affects the efficiency of the biological carbon pump. To demonstrate the impact of each varied parameter, the export flux of POC (measured at 80m) for every simulation (not only the best-fit CBRU) is shown in Fig. 11. With 310 a lower activation energy requirement (low Ea(1) value), less energy is needed for the remineralisation process to occur, this means https://doi.org/10.5194/gmd-2019-344 Preprint. Discussion started: 27 February 2020 c Author(s) 2020. CC BY 4.0 License.
nutrients are returned to surface waters more quickly, production is higher, and so POC flux at 80m is higher. Conversely, the higher the Ea(1) value, the more energy is required to remineralise organic carbon. So, at higher Ea(1), proportionally more organic carbon reaches depth making surface processes less important. The fraction of the POC exported that is recalcitrant and the maximum nutrient uptake rate at the surface becomes less important as Ea(1) increases. 315 The remineralisation curves for each ocean region is shown in Fig. 12 for the best fit CBRU and CB model for POC (in gC m -2 yr -1 ). CBRU and CB have differing initial POC export fluxes with lower latitude warmer waters showing higher export in CBRU due to the increased nutrient recycling there. A dataset of POC flux (Mouw et al. 2016a) is overlaid on the remineralisation curves ( Fig. 12). In both model configurations, the measured Antarctic zone POC flux at shallow and intermediate depths (< 1500m) is significantly lower than in the model. We do not apply iron limitation in the Southern Ocean (or elsewhere), which would limit 320 productivity and POC export, and could explain some of the mismatch at shallower depths. However, the measured flux at depth appears well represented. In deeper waters elsewhere, measured POC flux in warmer regions (e.g. East tropical Pacific, North Indian, East Tropical Atlantic) is generally higher than in the model. This likely reflects additional processes that may increase POC fluxes to depth such as ballasting by minerals (Klaas and Archer, 2002;Wilson et al 2012) and the lower reactivity of POC associated with increased recycling in low latitude plankton ecosystems (Aumont et al., 2017). 325 Overall, the pattern of the efficiency of the transfer of particles from 80m to 1040m (Fig. 13) in CBRU is similar to that found in Weber et al. (2016), where efficiency of transfer is essentially a measure of the rate of remineralistion; what fraction of the POC exported at 80m reaches 1040m. Colder waters show higher transfer efficiency, with the lowest transfer efficiency seen in the sub-tropical gyres. The CB model has a fixed decay rate for all locations, so the transfer efficiency at any particular depth has a global uniform value. 330 It should be noted that here we have included all available data from Mouw et al. (2016a) without any attempt to ensure these data are representative of the annual mean (where the model output represents the annual mean). POC flux measurements can be highly dependent on time of year and number of data measurement points available. Some of the model-data mismatch may then be due to a mis-match between the interval in time represented by the data, and the annual mean of the model. For instance, blooms, which are not well represented in the model, may explain some of the very high POC flux values (for example 0.2 gC m -2 yr -1 ) in 335 the North Atlantic and hence why the model annual mean appears to underestimate the flux.

Implications of including temperature-dependent microbial processes
This temperature-dependent version of the cGENIE model treats two large scale processes of the biological carbon pump. Firstly, nutrient uptake rates due to the metabolic temperature dependence of photosynthesising marine biota; secondly remineralisation rates of sinking particulate organic matter due to the metabolic temperature dependence of respiring marine biota feeding on that 340 sinking organic matter. The calibrated temperature response of the respiration-based mechanism of remineralisation in the water column is more sensitive to temperature change (a mean Q10 of 2.28 over 0°C to 26°C, from Eq.5 using 54 kJ mol -1 for Ea(1)) than the photosynthesis-based one (the Eppley curve has a Q10 of 1.88, in Eq. 3 and 4, Bissinger et al. 2008), in agreement with https://doi.org/10.5194/gmd-2019-344 Preprint. Discussion started: 27 February 2020 c Author(s) 2020. CC BY 4.0 License.
fundamental studies (Brown 2004). Historical temperature rise, which we induced in the cGENIE.muffin Earth system model by prescribing the observed CO2 transient in the atmosphere, provides an illustrative example of the role and importance of including 345 sufficient temperature-dependent processes in models. In this section we therefore discuss in more detail the transient differences between CB and CBRU model configurations.
Between the years 1700 and 2010, global mean air temperature in cGENIE increases by 0.94°C. In turn, warming at the ocean surface induces stratification in the water column, reducing nutrient re-supply to the surface from subsurface waters. In the CB model, this results in a pronounced drop in POC export at 80m (Fig. 14). In CBRU this effect is largely offset by the intensified 350 recycling of nutrients in warmer surface and near-surface waters. However, the transfer efficiency is affected in CBRU, with a drop of over 5% in the proportion of POC exported at 80m that reaches 1040m. The largest transfer efficiency drops are seen in low and mid latitude waters (Fig. 15). This reduction in biological pump transfer efficiency is a result of increased rates of remineralisation in the warming water column, principally in surface and near-surface waters (while whole (volumetrically-weighted) ocean warming is 0.12°C over this period, 0.6°C occurs on a global mean basis in surface waters, and 0.02°C in deepest waters). 355 Between simulation pre-industrial and present-day model states, we found a significantly smaller drop in POC flux at 80m when temperature dependence was included (CBRU) compared to the standard model (CB). Global POC flux at 80m reduces by 0.3% between pre-industrial and present-day in CBRU, but with increases in the Southern Ocean of around 10% and in the tropics Southern Oscillation) which is on face value inconsistent with our simulation of a possible increase in NPP in the tropics. Their estimate utilised satellite based NPP products from data on chlorophyll and light incidence, and found that in no data-constraint did NPP increase in the tropics (although the data constraint varied according to the NPP product used). However, Behrenfeld et al. (2015) noted that a reduction in chlorophyll does not necessarily represent a reduction in productivity, due to photoacclimation. The satellite based NPP products do not account or correct for this effect, so may well underestimate NPP in warming conditions. In an 365 earlier study, Taucher and Oschlies (2011) found an increased NPP when temperature dependence was included in modelled future projections. Laufkotter et al. (2017) found that when including a temperature-dependence and oxygen content-dependent remineralisation, NPP increased on warming due to intensified nutrient recycling in near-surface water. They suggested this was largely due to an initial positive bias in surface ocean nutrients. In a second set-up they reduced nutrient recycling in surface waters and find little impact on NPP between the temperature sensitive and temperature independent model in a future projection to 2100 370 CE.
Most of the regional variability in the flux of POC in the deep sea was explained via particle size and the effect of temperature on remineralisation in a study by Cram et al. (2018), with oxygen concentration providing only a small improvement (by reducing nutrient recycling in the Eastern Tropical Pacific). Cavan et al (2017) concluded that the large oxygen minimum zone in the Eastern Tropical North Pacific reduces the rate of remineralisation due to the almost complete absence of zooplankton particle 375 disaggregation within, and provides a negative feedback to warming. Particle size plays a role in sinking speeds, as larger particles sink faster (generally), and particle size is a factor in export and transfer efficiency (Mouw et al 2016b). The version of cGENIE we https://doi.org/10.5194/gmd-2019-344 Preprint. Discussion started: 27 February 2020 c Author(s) 2020. CC BY 4.0 License. used does not account for particle size and has a fixed sinking speed globally. The lack of particle size variability and oxygen concentration's role in remineralisation may explain some of the increased POC flux at 80m that the model shows since the preindustrial period in tropical waters. This tropics POC export increase may also be partly due to initial higher nutrient concentrations 380 compared to data, or to the increased remineralisation rates re-supplying nutrients to the surface. In summary, projected and predicted changes in NPP in low-nutrient warm waters are still subject to large uncertainties (Turner et al. 2015, Cross et al. 2015. There is also still uncertainty as to the causes and even patterns in POC flux differences in different ocean regions (Henson et al., 2012;Marsay et al., 2015;Weber et al., 2016;Cram et al., 2018). We find the patterns of transfer efficiency (Fig. 13) for CBRU are in broad agreement with Marsay et al. (2015) and Weber et al. (2016). This transfer efficiency is not dependent on surface 385 waters NPP patterns or on how much POC is exported at 80m in cGENIE, however, the absolute amount of carbon reaching the deep ocean does depend on NPP and export. On warming since the pre-industrial period we found a reduced POC flux at 80m as well as a reduction in transfer efficiency, combining to produce a reduction in the strength of the biological carbon pump with warming. This further implies an increased carbon pump strength in cooler climates, as per Heinze et al. (2016).