Proxy records indicate that export production was enhanced at the onset of the last glaciation. We examine how glacial cooling affects marine export production through temperature-dependent phytoplankton productivity and organic carbon remineralization using glacial climate simulation. Compared with the standard assumption of no temperature dependence as applied in existing paleoclimate models, including temperature-dependence enhances export production globally under glacial climate conditions. The near freezing temperatures of Antarctic seawater significantly reduce organic carbon remineralization. Less remineralization than production results in increasing carbon export to the deep ocean. Nutrients remaining near the surface at high latitudes are advected to lower latitudes via Antarctic Intermediate Water, enhancing primary production, and hence export, in temperate and tropical regions as well. Including temperature-dependence improves the model's agreement with the glacial proxy records of export production and stable carbon isotopes even under stably stratified conditions with a weakening of North Atlantic Deep Water.
 During glacial periods, export production (EP) of particulate organic carbon (POC) was enhanced globally in proxy records [Bopp et al., 2003]. EP is the amount of the phytoplankton produced POC exported to the deep ocean, which is determined by how much primary production (PP) occurs and/or how much carbon produced at the surface remineralizes within the euphotic layer. Various studies of sediment core samples and models have found that dust-induced iron supply can increase PP and EP in the Southern Ocean at the last glacial maximum (LGM) [e.g.,Martin, 1990; Bopp et al., 2003]. Enhanced EP was furthermore evident at the onset of glaciation [Kohfeld et al., 2005] when the observed dust deposition was relatively low [Petit et al., 1999].
 EP may be sensitive to climate variability, because phytoplankton production and organic matter remineralization depend on water temperature. According to Eppley , marine phytoplankton (algae) grows about twice as rapidly when the ambient water temperature increases by 10°C (Q10 = 2). Especially in the winter euphotic environment, planktonic autotrophic growth rate is limited by temperature rather than by nutrients that are amply supplied via strong mixing [Felip et al., 1996]. At low temperatures where growth rates are relatively low, organic matter also remineralizes more slowly [Laws et al., 2000], whereas both remineralization and production proceed faster at high temperatures. The temperature dependencies of both growth and remineralization together strongly affect the rate of recycling of organic matter within the euphotic zone [Matsumoto, 2007].
 Sea surface temperature (SST) reconstructions of the last glacial maximum (LGM) suggest that tropical SST was 2–3°C cooler than for the preindustrial period [Intergovernmental Panel on Climate Change, 2007]. The temperature pattern was a spatially non-uniform and the cooling was greater at higher latitudes [MARGO Project Members, 2009]. Such a distinct cooling would affect phytoplankton growth and remineralization processes in the ocean.
 In association with changes in phytoplankton production and remineralization, EP has the potential to affect the carbon partitioning between atmosphere and ocean. In surface waters, biological production incorporates dissolved inorganic carbon (DIC) into organic matter, which removes carbon dioxide (CO2) from the ocean surface and the atmosphere. Remineralization by organisms in turn produces CO2 from organic matter, which releases carbon to the atmosphere. Thus, the relative rates of PP and remineralization within the euphotic zone [Volk and Hoffert, 1985] control the efficiency of oceanic carbon uptake by the biological pump. Indeed, an idealized −5°C forcing of water temperatures reduces atmospheric pCO2 by 33 ppmv through slow remineralization [Matsumoto, 2007]. Glacial cooling may reduce atmospheric pCO2 by changes in both CO2 gas solubility and EP.
 In this study, we examine how temperature-dependent phytoplankton growth and remineralization affect marine EP and the associated ocean carbon cycle during glacial periods. Here we use the glacial climate simulation obtained from an Atmosphere-Ocean coupled climate model, MIROC (Model for Interdisciplinary Research on Climate [K-1 Model Developers, 2004]), and additionally calculate the ocean carbon cycle using an offline ocean carbon cycle model. By comparing the experiments considering temperature dependence of growth and remineralization with those using the standard assumption of no temperature dependence, we evaluate special characteristics of the temperature-dependent production and remineralization responses to glacial cooling.
2. Model and Experiments
 To simulate glacial ocean carbon cycle, we use the glacial simulation in the MIROC model [Chikamoto et al., 2012; A. Abe-Ouchi et al., Role of Southern Ocean in simulating the Atlantic Meridional Overturning Circulation (AMOC) with climate models at the Last Glacial Maximum, manuscript in preparation, 2012]. Climatology for the application to an offline carbon cycle model is obtained from the last 30-year averages of MIROC outputs: zonal and meridional ocean velocities, vertical diffusivity, sea surface height, sea surface wind speed, sea ice fraction, and sea-surface solar radiation. The MIROC-based offline ocean biogeochemical model is based on a simplified ecosystem of nutrient-phytoplankton-zooplankton-detritus (NPZD) ecosystem model, which is a three dimensional ocean model coupled with a one-layer atmosphere [Chikamoto et al., 2012]. The biogeochemical components are two plankton function groups of phytoplankton and zooplankton, suspended and sinking particulate detritus, four dissolved inorganic components (alkalinity, DIC, nitrate,and oxygen), and two carbon isotopes (13C and 14C).
 The temperature-independent control experiment uses a maximum growth rate of 0.8 day−1 and a remineralization rate of 0.1 day−1 above 100 m and 0.02 day−1 below 100 m depth [Plattner et al., 2001]. These parameters are impartial throughout the model domain, independent of temperature.
 In the temperature-dependent run, we incorporate the temperature dependent parameterization for the maximum growth rate (Pmax), following Oschlies and Garçon .
Here T is water temperature (°C) and T0 is the upper critical temperature for the maximum growth rate (set to 20°C). For water temperature greater than the optimum temperature of 20°C, the maximum growth rate is constrained to its maximum value [Eppley, 1972]. The temperature-dependent remineralization rate (r) is parameterized according toMatsumoto  (r = 0.02 × eln2/10×T). In all experiments, the detritus-sinking rate is set to 8 m day−1 [Chikamoto et al., 2012]. The production rate of calcium carbonate (CaCO3) is assumed to be proportional to that of POC (CaCO3:POC = 0.08).
 The ocean carbon cycle simulations with temperature-independent or dependent parameterization are each respectively spun up for 10,000 years with the prescribed atmosphericpCO2 of 280 ppmv. After equilibrium is attained between the atmosphere and the ocean carbon cycle, we calculate atmospheric CO2concentration for 1000 years (PI and PI-temp). In glacial experiments (LG and LG-temp), the marine carbon cycle is run for 5000 years starting from each equilibrated state of the preindustrial carbon cycle. Atmospheric stable carbon isotopes (δ13C) and radiocarbon (Δ14C) are set to −6.5, and 0 per mil, respectively. For comparisons of simulated δ13C pattern with proxy records, we additionally simulate the glacial ocean carbon cycle forcing the prescribed atmospheric pCO2 of 199 ppmv and δ13C of −6.6 per mil. In all experiments we analyzed the steady-state biogeochemical fields. The preindustrial biogeochemical fields in PI-temp are similar to those in PI (Figure S1 in theauxiliary material).
Chikamoto et al.  described preindustrial and glacial climate states (corresponding to experiment LGb in Chikamoto et al. ). In the glacial simulation, the global mean temperature is 2.5°C cooler than in the preindustrial simulation. The maximum NADW formation is reduced from 16 Sv in the preindustrial simulation to 8 Sv in the glacial simulation. The pattern of shallower NADW and enhanced AABW produces the surface-deep gradient ofδ13C in the Atlantic Ocean [Chikamoto et al., 2012]. The Atlantic distribution of δ13C difference between LG and PI is consistent with that in proxy records (Figure S3). We performed further sensitivity studies of LG-temp, including temperature dependence of growth rate and remineralization.
Figure 1illustrates PP and EP changes due to temperature-dependent phytoplankton growth and remineralization. In areas where water temperature is greater than 4.5°C (T > ln(0.8/0.6)/ln(1.066)), the phytoplankton growth rate and PP are higher in LG-temp than in LG (Figure 1a). Remineralization of organic carbon below the euphotic layer tends to be faster in LG-temp for waters having temperature greater than 0°C (T > ln(0.02/0.02) × 10/ln2) (Figure 1b). Faster growth and faster remineralization translate into more intense organic carbon recycling, and therefore DIC overall increases in near-surface waters (not shown). The net result is greater EP (Figure 1c).
 The near freezing temperatures of Antarctic water reduce both production and remineralization. Less remineralization than production leads to a net increase in POC export to depth in polar regions (Figure 1c). The fraction of carbon export at 100 m depth to PP (pe ratio) significantly increases at high latitudes in LG-temp compared to in LG (Figure 1d). The temperature-dependent biological processes and the accompanying nutrient redistributions implicate organic carbon preservation, especially at high latitudes.
 The changes in temperature-dependent biological processes contribute to nutrient redistribution in the upper ocean. In LG-temp, phytoplankton growth weakens at polar region, which causes unutilized nutrients to transport northward along upper-ocean isopycnal surface of Antarctic Intermediate Water (Figure 2a). On the other hand, the remineralization change redistributes nitrate between upper and deeper 100 m (Figure 2b). Within the euphotic layer remineralization reduces in temperatures lower than 23°C (T < ln(0.1/0.02) × 10/ln2) whereas that below the euphotic layer reduces in temperatures lower than 0°C. The sensitivity difference at higher latitudes attributes to the nutrient decrease (increase) at depths upper (deeper) than 200 m. At lower latitudes, the temperature-dependent remineralization enhances nitrate recycling of the aphotic zone.
 Temperature dependent change in maximum growth and remineralization rates increases global EP at 100 m depth to 8.9 GtC yr−1by 39 % in LG-temp compared with that of 6.4 GtC yr−1in LG. In LG glacial cooling enhances ocean stratification in association with the weakening of NADW formation and the northward penetration of southern source intermediate-bottom water [Chikamoto et al., 2012]. This feature reduces the nitrate supply from the deep ocean and hence weakens EP (Figure 3a). In addition, the polar sea ice reduces (increases) EP south of 60°S (north of the modern Antarctic Polar Front (APF)) due to the limited insolation and the nutrient transport from polar region.
 Compared with the experiment in LG, the temperature-dependent PP reduction at polar region in LG-temp causes nitrate to furthermore transport northward. This results in amplifying the EP increase north of the modern APF (Figure 3b). The change in nitrate recycling through temperature-dependent processes enhances EP in the upwelling region of the eastern equatorial Pacific and western North Pacific as well (Figure 3b). Net primary production increases by 5.2 GtC yr−1in LG-temp from 32.7 GtC yr−1 in LG and EP increases by 2.5 GtC yr−1in LG-temp from that in LG, suggesting the enhancement of productivity. The EP response agrees with the proxy records of EP change between glacial and modern [Bopp et al., 2003].
 Along the margin of the Southern Ocean during the glacial period, enhanced EP would be partly supported by iron supply [e.g., Martin, 1990; Bopp et al., 2003]. An added glacial iron deposition in LG increases EP in the Atlantic and Indian sector of the Southern Ocean and in the entire Pacific Ocean [Oka et al., 2011]. On the other hand, we find temperature-induced enhancement of EP at low latitudes in the Atlantic and Indian Ocean as well as in the Pacific Ocean. The record of EP starts to increase even during Isotope Stage 5a-d [Kohfeld et al., 2005], when significant dust deposition was not found [Petit et al., 1999]. The EP variability may be partly explained by the fast response of temperature-mediated biological processes through the large temperature drops during early glaciation.
 A more efficient biological pump would be expected to result in more carbon sequestration to the deep ocean compared with the temperature-independent simulation. The vertical gradient of DIC between surface and deep ocean increases by 13 mmol m−3in LG-temp compared with that in LG. However, the atmosphericpCO2difference between LG-temp and LG is only 2 ppmv. The change in phytoplankton growth rate raises atmosphericpCO2 by 2 ppmv, mainly because the significant weakening of EP in the Southern Ocean reduces oceanic carbon uptake via biota. In contrast, less remineralization and accompanying carbon export lower atmospheric CO2 by 4 ppmv. The POC deposition at the ocean floor below 100 m depth is 0.02 GtC yr−1larger (+5 %) in LG-temp than that in LG, whereas that below 1500 m depth is 0.006 GtC yr−1 lower (−30 %). This implies fast organic carbon recycling at depths upper 1500 m and less organic carbon export to the deep ocean. Furthermore, low pCO2sensitivity to the temperature-mediated biological processes is partly due to the coupling of carbon export with CaCO3 export within the euphotic zone. We assume that the CaCO3 production is proportional to the POC production with a ratio of 0.08. Therefore, the export rain ratio of CaCO3 to POC is constant within the euphotic zone, and the CaCO3export increases by 39 % as well as the POC export in LG-temp. Between 100 and 200 m, CaCO3 dissolves by 4 % in both LG and LG-temp. On the other hand, the settling POC remineralizes by 14 % in LG and by 29 % in LG-temp at 200 m depth. The different remineralization/dissolution rates for POC and CaCO3 increase the rain ratio of CaCO3to POC at 200 m depth to 0.11 from 0.08 (LG-temp vs. LG). This strengthens the alkalinity pump, thereby leading to a rise of atmospheric CO2.
 In order to compare the atmospheric CO2sensitivity to temperature-dependent processes with the previous study [Matsumoto, 2007], we additionally ran an idealized simulation of global uniform temperature perturbation from the preindustrial climate state. In a 5°C cooling experiment (the same as in Matsumoto ), atmospheric CO2lowers by 9 ppmv. Because temperature cannot decrease below the freezing point in polar regions, the global-mean temperature drops 4.5°C. This corresponds to a sensitivity of atmosphericpCO2to cooling of −2 ppmv/(°C cooling) in our −5°C temperature experiment, which is larger than the response of −0.8 ppmv/(°C cooling) in LG-temp. The simplistic imposition of uniform temperature change may overestimate the atmosphericpCO2 sensitivity. The atmospheric CO2 response in our model is about one third of the estimate of −6.6 ppmv/(°C cooling) in the previous 5°C cooling experiment [Matsumoto, 2007]. In our model, CaCO3 export is directly proportional to POC export at low latitudes, which strengths the alkalinity pump and thereby increases atmospheric pCO2. In Matsumoto , particulate CaCO3 is assumed to dissolve in waters highly undersaturated with respect to calcite, which would decouple CaCO3export from POC export. Deep-water nutrient concentrations also increase especially in the South Atlantic Ocean associated with the glacial change in Atlantic meridional overturning, whereas nutrient concentrations increase in the North Pacific Ocean under modern ocean circulation [Matsumoto, 2007]. Improvements of the representation of the glacial ocean circulation would reduce uncertainty regarding precisely how temperature affects the organic carbon and nutrient pools. Ocean warming of 5°C increases atmospheric pCO2 by 6 ppmv, which is a stronger effect than the reduction found in the 5°C cooling experiment.
 In this study, we use different temperature dependency of phytoplankton growth and remineralization: For growth the formulation assumes Q10 = 1.89 (1.066T = eln(1.89)/10×T) and for remineralization, Q10 = 2.0. Thus remineralization is more sensitive to temperature than is growth. This difference, however, does not substantially affect the atmospheric pCO2 change. The setting both sensitivities to the same Q10 = 1.89 results in atmospheric pCO2reduction of 3.5 ppmv due to remineralization, which differs by only 1-ppmv from the original case with LG-temp. Rather, the lower sensitivity of phytoplankton growth to temperature, compared with remineralization, reflects that growth tends to be limited more by the availability of nitrate or light.
Kwon et al.  suggest that the shoaling of the remineralization depth reduces atmospheric pCO2using surface nutrient restoring. On the other hand, in our 5°C cooling experiment, the surface nitrate concentration reduces by 9 % from in the preindustrial control run, when the model does not impose nutrient restoring. Since climate dynamics was the same between two experiments, the nutrient difference must have been the result of temperature-dependent processes. Our result suggests that the surface nutrient distribution is sensitive to the temperature-mediate biological process, which would be also important for the transient response of atmospheric CO2 to climate change.
 In our model, all POC particles are not burial but dissolve at the sea floor. If considering burial processes in sediments, enhanced POC transport to the deep ocean results in increasing the POC burial as well. This would change the total inventory of carbon and nutrient in the ocean, thereby changing atmospheric CO2.
 Our model with temperature-dependent parameterization represents glacial EP well. However, we still have the biogeochemical biases in PI or PI-temp, especially in the upwelling region where the model overestimates the observed EP (Figure S1). EP is sensitive not only to biogeochemical model structure but also to ocean circulation and the parameter values. Correction for this biogeochemical bias, due to broad parameter values or ocean circulation patterns, would reduce the uncertainty of the EP response to glacial cooling. Further simulations using multi models will be also helpful for better understanding of the temperature-mediate biological processes behind the glacial climate state.
 We evaluate the impact of glacial cooling on EP through temperature-dependent growth and remineralization using an atmosphere-ocean climate model (MIROC) and an offline biogeochemical model. At high latitudes, colder water reduces remineralization significantly, which enhances the carbon-partitioning gradient between surface and deep waters especially in the Southern Ocean. This enhances EP even when ocean stratification reduces the nitrate supply to the surface in association with the slowdown of NADW formation. At low latitudes, fast nitrate recycling maintains EP. The enhancement of EP is found in the Southern Ocean and upwelling region, such as the eastern equatorial Pacific and western North Pacific, consistent with proxy records of EP during the glaciation [Bopp et al., 2003]. This supports the case for adding temperature-dependence in these models. The atmospheric CO2 reduction due to the temperature dependence is approximately 2 ppmv. The sensitivity of atmospheric CO2to temperature-dependent growth and remineralization is reduced somewhat by the counteracting effect of increased CaCO3:POC rain ratio in near surface waters.
 In our glacial simulation, the expanded sea ice cover in the Southern Ocean and enhancement of AABW formation contribute to much less biological production and nutrient utilization in the polar Southern Ocean. This in turn increases biological production and nutrient utilization in the subantarctic. The strength of NADW formation, on the other hand, affects the nutrient supply to the surface and associated biological production at lower latitudes. Improvements of the representation of glacial NADW formation and Southern Ocean climate dynamics would reduce the uncertainty of biogeochemical response at both lower and higher latitudes.
 In the glacial simulation, the magnitude of the cooling effect on the ocean carbon cycle is limited by the fact that present polar waters are already near the freezing points under the modern condition. Temperature effects of marine biology on atmospheric pCO2 can be amplified under global warming because there is no such limit on the extent of warming [Taucher and Oschlies, 2011].
 We are grateful to two anonymous reviewers of the manuscript for numerous constructive suggestions. We would like to thank K. Matsumoto and A. Timmermann for valuable comments in the fields of ocean carbon cycle and stable carbon isotopes. The numerical simulations were preformed on the super computer system at JAMSTEC.
 The Editor thanks 1 anonymous reviewer for his/her assistance in evaluating this paper.