Manuscript : “ Oxygen and indicators of stress for marine life in multi-model global warming projections ”

CO2 and O2 replace the fugacities. RI not mentioned anymore. Pag. 10788, from line 2: The goals of this study are to provide multi-model estimates of decadal-to-century scale trends in O2, and CO2 (together with fO2, fCO2, and other biologically relevant variables), as well as to identify underlying mechanisms using global warming simulations from six fully coupled atmosphere-ocean general circulation models and one model of intermediate complexity. [...] Pag. 10789 line 2: The concentrations of dissolved inorganic carbon (DIC, i.e. the sum of CO2, HCO3^− and CO3^2concentrations) Pag. 10789 line 2: Higher dissolved CO2 concentrations (i.e. the sum of the CO2(aq) and H2CO3 concentrations) Pag 10796 lines 13-22: Similarly to the observation-based O2 frequency distribution, the modeled O2 frequencies range between 0 and 400 mmol m^-3, indicating the ability of the models to reproduce the correct range of variability of O2 in the upper pelagic layer. The highest peak in the observations is at ~300 mmol m^-3. Most models tend to overestimate the magnitude of this peak and/or exhibit it between 200 and 240 mmol m^-3. The observation-based and modeled CO2 values lie mostly between 0.01 and 0.06 mol m^-3, with the highest peak at ~0.02 mol m^-3. The magnitude of this peak is largely overestimated by CSM1.4, MPIM and, to a smaller extend, CCSM3 and GFDL. From Pag. 10797, line 9 to Pag. 10798 line 10: replaced fCO2 (fO2) with CO2 (O2). Pag 10814 line 20 – 10815 line 25: Section in the appendix replaced with “Projected changes in RI additional figure” The largest reductions are projected by BCM-C in the Southern Ocean, the North Atlantic and Pacific. Most of the decrease is related to the increase in fCO2 in many areas. RI is projected to increase in the Eastern South Atlantic in most models. This is consistent with the projected changes in fO2 in this region. The increase in fO2 dominates over the projected increase in fCO2 in this low oxygen region. The projected changes in RI in the eastern tropical and Southern Pacific and in the Bay of Bengal and the Arabian Sea are not consistent across the range of models, similarly as found for projected changes in fO2 in these low-O2 regions. (text previously at Pag. 10804 line 17-25) 2. Length: The paper is on the long side with a rather large number of figures. In my opinion, the paper has a substantial potential for a careful editorial pruning in order to make the manuscript easier to digest and follow. The introduction is a good example as it introduces many findings and concepts that are interesting but not relevant for this paper. We agree and deleted these paragraphs: Pag. 10789 lines 10-16 Pag. 10790 lines 13-19 Pag.10800: deleted line 23 (subsection: Attribution to mechanisms) Pag. 10790, line 2: deleted “and indicates increased stratification as the main driver of this decline.” Fig. 6 and 10 are not presented anymore. Fig. 14 is now in the appendix (as Figs. B3)


Introduction
The ocean is undergoing physical and chemical changes in response to climate change caused by anthropogenic emissions of carbon dioxide (CO 2 ) and other greenhouse gases. These changes include widespread ocean warming, alterations in the stratification, density structure, circulation and physical transport rates, and an increase in total carbon content by air-sea CO 2 uptake forcing the ocean towards more acidic conditions and altering acid-base relationships (Bindoff et al., 2007). Oxygen (O 2 ) and CO 2 are involved in aerobic respiration where organic compounds and O 2 are converted to CO 2 and water. Global warming,

Published by Copernicus Publications on behalf of the European Geosciences Union.
expanding hypoxia and higher CO 2 levels represent physiological stresses for marine aerobic organisms that may act synergistically with ocean acidification (Pörtner and Farrell, 2008).
The fugacity of carbon dioxide (f CO 2 ) and oxygen (f O 2 ) are two key thermodynamical variables for reporting the physiological state of aerobic marine animals (Hofmann et al., 2011;Seibel et al., 2012). The combined mechanisms of global CO 2 (and f CO 2 ) increase and more local deoxygenation (and f O 2 decrease) have just begun to be addressed (e.g., Paulmier et al., 2011;Pörtner et al., 2011;Gruber, 2011;Mayol et al., 2012;Brewer and Peltzer, 2009) and need to be further investigated. Surprisingly, there is a lack of studies that project the combined evolution of these variables under global warming and anthropogenic carbon emissions using fully coupled Earth System Models.
The goals of this study are to provide multi-model estimates of decadal-to-century scale trends in O 2 and CO 2 (together with f O 2 , f CO 2 , and other biologically relevant variables), as well as to identify underlying mechanisms using global warming simulations from six fully coupled atmosphere-ocean general circulation models and one model of intermediate complexity. The analysis focuses on how anthropogenic carbon emissions and climate change affect decadal-to-century scale changes in CO 2 and O 2 within the upper mesopelagic layer (UML, from 100 to 600 m of depth) under the SRES A2 high-emission scenario. The performance of current Earth System Models in representing observation-based variables is assessed. Using the range of models, uncertainties in the projected changes in different environmental variables that are potentially relevant for marine life are explored. Thereby, the present analysis complements available studies on temperature, pH or calcium carbonate saturation state (i.e., the product of the concentrations of calcium ions and carbonate ion, divided by the apparent stoichiometric solubility product). Here we focus our analysis on the UML because within this layer O 2 is depleted the most as a consequence of the O 2 consumption associated with the remineralization of organic matter.
The concentrations of dissolved inorganic carbon (DIC, i.e., the sum of CO 2 , HCO − 3 and CO 2− 3 concentrations) as well as f CO 2 are increasing in the surface ocean (0-100 m) and in the UML because a substantial fraction of the anthropogenic carbon emissions from fossil fuel burning and land use change is taken up by the ocean. Higher dissolved CO 2 concentrations (i.e., the sum of CO 2(aq) and H 2 CO 3 concentrations) induce, through acid-base equilibria, a decrease in pH and carbonate ion concentration, in turn reducing calcification in many calcareous organisms (Orr et al., 2005;Kroeker et al., 2010). Many physiological processes are sensitive to CO 2 levels, and elevated CO 2 concentrations may impose physiological stress on marine organisms more generally , even altering sensory responses and behavior of marine fishes (Nilsson et al., 2012). High CO 2 may have detrimental effects on the survival, growth, and physiology of marine animals and impair oxygen transport by lowering blood pH (Pörtner et al., 2011).
Many observational (e.g., Pfeil et al., 2012;Watson et al., 2009;Lüger et al., 2006) and modeling (e.g., Keller et al., 2012;Tjiputra et al., 2012;Thomas et al., 2008) studies are directed towards understanding the air-sea CO 2 fluxes and uptake of anthropogenic carbon by the ocean. Recently, observational studies focused on upwelling regions (Mayol et al., 2012;Paulmier et al., 2011), typical low-O 2 -high-CO 2 areas of particular vulnerability, where elevated f CO 2 levels represent a key threat connecting aerobic stress and calcification challenges. However, f CO 2 and CO 2 are not state variables in ocean biogeochemical models and coupled Earth System Models, and their distribution below the surface layer has been studied poorly.
Observational studies indicate a mostly negative trend in the oxygen content over recent decades in different basins of the world's ocean (Stramma et al., 2008;Chan et al., 2008;Whitney et al., 2007;Mecking et al., 2006;Emerson et al., 2004;Joos et al., 2003;Keeling et al., 2010;Takatani et al., 2012). A recent global-scale observational study (Helm et al., 2011) supports the evidence of a widespread ocean O 2 decrease between the 1970s and the 1990s. These authors, as well as Stramma et al. (2012), however, report also large areas where O 2 has increased during recent decades.
The observed O 2 variations are relatively small and trends can therefore be difficult to detect, although they can result in changes up to 50 % in the low-O 2 regions. There is no consensus in considering anthropogenic global warming as the main driver of the observed O 2 changes because of the relatively short and sparse observational records. Natural variability may tend to mask the anthropogenically induced trends in dissolved O 2 , as suggested from both modeling ) and observational (Mecking et al., 2008;Deutsch et al., 2011) studies. Despite the uncertainties about the last-decade trends, a long-term decrease in the oceanic O 2 inventory is expected under global warming and consistently simulated across a range of models (Sarmiento et al., 1998;Frölicher et al., 2009;Plattner et al., 2002;Bopp et al., 2002). Surface warming, lower sea surface O 2 concentration, enhanced stratification, reduced ventilation of the thermocline, and slowed thermohaline circulation tend to decrease the resupply of O 2 to the ocean interior, and to increase the residence time of water at depth, enhancing biological O 2 utilization.
Higher CO 2 levels may enhance O 2 consumption at depth because of the higher C/N ratio of the organic matter produced at high CO 2 concentration, as shown in mesocosms experiments (Riebesell et al., 2007). As long as the O 2 /C ratio is not adjusted, this effect would lead to an expansion of the suboxic water volume in particular in the tropical oceans . On the other hand, reduced upwelling of nutrient-rich waters may decrease export of biologically produced organic carbon and the respiratory oxygen consumption in subsurface water. Many modeling studies identify Biogeosciences, 10, 1849-1868, 2013 www.biogeosciences.net/10/1849/2013/ circulation and mixing changes as main drivers of the ongoing and future O 2 decline Schmittner et al., 2008;Duteil and Oschlies, 2011;Bopp et al., 2002;Plattner et al., 2002). The observations reported by Helm et al. (2011) as well as earlier observation-based studies (e.g., Emerson et al., 2001) also support this explanation.
In the next section, simulations and experimental details are described. In the Appendix the model descriptions and additional plots are provided.

Methods
In this study we consider global warming simulations from seven Earth System Models including representations of terrestrial and marine biogeochemistry of different complexities. The models are forced with prescribed CO 2 emissions from reconstructions (1870-2000 AD) and a high-emission scenario, SRES A2 (2000-2100 AD). The Earth System Models represent the interactions between the physical climate system, biogeochemical cycles, and marine ecosystems under global warming. Such interactions are included in global Earth System Models, but not in global ocean-only or high-resolution, eddy-resolving regional models. The strategy applied here is to analyze results from a broad suite of Earth System Models; the model spread in results provides a first indication of uncertainty. The same simulations with IPSL, MPIM, CCSM3 and CSM1.4 models considered in this study are analyzed in Steinacher et al. (2010) to investigate the ability of the models to represent the present spatio-temporal pattern of net primary production and to project its future changes under the high-emission scenario SRES A2. Further details about these four models and the experimental setup of the simulations considered here are provided in Steinacher et al. (2010). Schneider et al. (2008) present results for three (IPSL,MPIM,and CSM1.4) of the seven models used in this study. They provide detailed information on the performance of these three models under current climate conditions and compare modeled physical (temperature, salinity, mixed layer depth, meridional overturning, ENSO variability) and biological (primary and export production, chlorophyll concentration) results with observation-based estimates. Roy et al. (2011) investigate the impacts of climate change and rising CO 2 on ocean carbon uptake with the simulations from IPSL, MPIM, CSM1.4 and BCM-C.
The models are forced by anthropogenic CO 2 emissions from fossil fuel burning and land use changes as reconstructed for the industrial period (from 1870 to 1999 AD) and following the SRES A2 emission scenario after 2000 AD. The CSM1.4, CCSM3 and the MPIM model also partly include non-GHG forcings. Volcanic eruptions and changes in solar radiation over the historical period are also taken into account by CSM1.4 and CCSM3. The variables used in the analysis have been interpolated onto a common 1 • × 1 • grid using a Gaussian weighted average of the data points within a radius of 4 • with a mapping scale of 2 • . Control simulations in which atmospheric CO 2 and other forcings are set to constant preindustrial levels are used to remove possible century scale model drifts for each grid point and for each month ). This procedure is applied to the three-dimensional field of temperature, salinity, oxygen, DIC, alkalinity and nutrients used in the following analysis.
As a point of reference we use annual GLODAP  and World Ocean Atlas 2009 (WOA 09) Antonov et al., 2010;Garcia et al., 2010a,b) gridded data sets for data-model comparisons. The gridded data sets provide structured information, extremely valuable in model-data comparisons and to estimate global properties.

CO 2 , f CO 2 and f O 2
For this study, CO 2 and f CO 2 are computed using the standard OCMIP carbonate chemistry routines (Orr et al., 2005;Steinacher et al., 2009) as a function of DIC, temperature, alkalinity, salinity, phosphate, and silicate (from observed or modeled quantities). f O 2 is computed following the Garcia and Gordon equation (Garcia and Gordon, 1992), in the version corrected by Sarmiento and Gruber (2006), as a function of dissolved O 2 , temperature and salinity (from observed or modeled fields). Changes in f CO 2 are attributed to changes in individual variables, i.e., to changes in DIC, temperature, alkalinity or salinity. This is done by keeping all input variables except one at the 1870-1879 mean concentration in the carbonate chemistry computation. Similarly, changes in f O 2 are attributed to changes in salinity, temperature, and O 2 .
www.biogeosciences.net/10/1849/2013/ Biogeosciences, 10, 1849-1868, 2013 The mechanisms responsible for the 21st century O 2 and DIC variations are investigated in more detail using CSM1.4 results and taking advantage of its relatively simple formulations for export production and remineralization. In CSM1.4 the stoichiometric O 2 to PO 4 ratio for production and remineralization is fixed at −170. This allows splitting of the changes in O 2 ( O 2 ) in individual components representing different drivers (see Frölicher et al., 2009). O 2gas reflects the contribution by air-sea gas exchange on the O 2 concentration; this component reflects the influence of warming on the solubility of O 2 . O 2bio is computed by multiplying the change in phosphate concentration with the O 2 to PO 4 Redfield ratio; this reflects the influence from changes in the cycling of organic matter.
Similarly, the CSM1.4 changes in DIC due to marine production and remineralization of organic matter and calcite are linearly linked to variations in phosphate and alkalinity according to the fixed Redfield ratios (Plattner et al., 2001;Gruber and Sarmiento, 2002). This allows assignment of DIC changes ( DIC) to variations in the marine biological cycle ( DIC bio ) and in air-sea gas exchange ( DIC gas ) . DIC bio is given by the sum of two terms, representing the organic matter cycle and the calcium carbonate cycle contributions. The difference between DIC and DIC bio can be assigned to DIC gas . Brewer and Peltzer (2009) proposed the "respiration index" (RI) as an indicator for estimating the physiological limits for deep-sea aerobic life. RI is a quantity linearly related to the available energy involved in the basic oxic respiration.

Respiration index
Generally, the limits to aerobic life in the ocean are defined in terms of a minimum dissolved O 2 concentration, which is typically set to 5 mmol m −3 . Below this concentration microbes turn to other electron acceptors as it becomes inefficient to consume O 2 . However, by defining the limit in terms of oxygen only, the potential influence of CO 2 on respiration is neglected. Brewer and Peltzer (2009), considering the basic oxic respiration equation C org + O 2 → CO 2 and the related Gibbs free energy (proportional to log 10 (f O 2 /f CO 2 )), proposed the RI. Brewer and Peltzer (2009) proposed the RI with the following thresholds, which can be used to map changing conditions: RI ≤ 0 corresponds to the thermodynamic aerobic limit (formal dead zone); the range 0 ≤ RI ≤ 0.4 has been proposed as "practical dead zone"; RI between 0.4 and 0.7 represents the practical limit of aerobic respiration; the interval from 0.7 to 1 may define an aerobic stress regime. The actual limits will be, however, strongly species dependent and still need to be experimentally validated, as noted by Brewer and Peltzer. Criticisms of the concept were raised and its practical usefulness has been questioned (Seibel et al., 2012;Pörtner et al., 2011). The mechanisms of transport of O 2 and CO 2 between the environment and tissues are important controls influencing aerobic performance of large animals; a factor not included in the definition of RI. Acknowledging this limitation, we project changes in RI as another possible indicator of aerobic stress.

Model evaluation
We start comparing model results with observation-based estimates of the present-day annual mean distributions of f O 2 , O 2 , f CO 2 , DIC, and RI. First, the skill of the different models in representing the observation-based fields for the world ocean between 100 and 600 m of depth is assessed by Taylor diagrams (Taylor, 2001) (Fig. 1). The Taylor diagrams allow us to visualize the correspondence between model results and observation-based variables. The polar coordinates represent the correlation coefficient R (polar angle) and the normalized standard deviation σ model /σ obs (radius). In such a diagram the points corresponding to the observation-based variables would lie all at (R =1, σ model /σ obs = 1). The Taylor diagrams of temperature, salinity, phosphate and mixed layer depth can be found in Roy et al. (2011) for the MPIM, IPSL, BCM-C, and CSM1.4 models. O 2 mean value and standard deviation (σ ) for WOA 09 and the models are listed in the table in Fig. 1a. BCM-C oxygen mean value and σ match the observations, GFDL has a slightly low mean value and the other models present in general too high mean values and too low σ . The correlation coefficient (R) for O 2 and f O 2 ranges for all models between 0.7 and 0.9, with a normalized standard deviation (σ model /σ obs ) between 0.7 and 1.1 for both variables (Fig. 1). DIC shows very similar R values, and slightly higher normalized standard deviations (between 0.8 and 1.25).
The results for f CO 2 and RI are less satisfying. The correlation with the observation-based fields is lower than 0.6 for CCSM3 and CSM1.4 (for both variables) and for BCM-C (for f CO 2 ). The normalized standard deviation for f CO 2 lies between 1.1 (MPIM) and 1.9 (BCM-C), while for RI it ranges from 0.9 (IPSL) to 2.75 (CSM1.4 and CCSM3). The simplest model, UVIC2-8, produces the highest R values for all variables and normalized standard deviations between 0.9 and 1.2 for O 2 , f O 2 , DIC and RI. These results are in good agreement with the observation-based annual mean fields, compared with other models. This is true for the annual mean distribution; however, the situation for UVIC2-8 would look considerably worse considering the seasonal cycle.
Next, we compare the simulated and observation-based frequency distributions for O 2 and CO 2 within the UML. The frequency distributions allow us to assess whether the models are able to represent the volume associated with a given concentration interval. Unlike the information in a Taylor diagram, the frequency distribution does not depend on the  geographical location and is as such insensitive to a geographical mismatch between observed and modeled water masses. Modeled and observation-based O 2 and CO 2 frequency distributions are shown in Figs the observations is at ∼ 300 mmol m −3 . Most models tend to overestimate the magnitude of this peak and/or exhibit it between 200 and 240 mmol m −3 . The observation-based and modeled CO 2 values lie mostly between 0.01 and 0.06 mol m −3 , with the highest peak at ∼ 0.02 mol m −3 (Fig. 3). The magnitude of this peak is largely overestimated by CSM1.4, MPIM and, to a smaller extent, CCSM3 and GFDL. Modeled and observation-based volumes related to three O 2 concentration intervals in the low-O 2 range (Table 1) are analyzed next. For O 2 < 80 mmol m −3 IPSL, UVIC2-8, CSM1.4, and CCSM3 underestimate the associated volume by a factor between 1.8 (UVIC2-8) and 3 (CSM1.4), compared to the data-based estimates. The other models overestimate it slightly (by a factor of 1.3 for the MPIM and 1.1 for the BCM-C). In the hypoxic range (O 2 < 50 mmol m −3 ) the situation is analogous.
On the other hand, all models overestimate the water volume in the suboxic regime (O 2 < 5 mmol m −3 ). The IPSL results are close to WOA 09 O 2 (respectively, 0.08 % and 0.05 % of the total volume of each data set). The other models simulate large discrepancies from the observations, in particular GFDL and MPIM. As pointed out by Bianchi et al. (2012), these discrepancies in the suboxic regions could be only partially attributed to deficiencies of the gridded WOA O 2 data set in representing the O 2 minimum zones (OMZs), whose extent is considered by Table 1. Observed and modeled present-day and 2090s (decadal average) volume of three water classes identified by different O 2 regimes, expressed in m 3 and in percent of the total oceanic volume. In the first line, the revised WOA 05 values from Bianchi et al. (2012) are listed.
Volume [10 15 m 3 ] (% of the total) 1990s 2090s The models underestimate O 2 and overestimate CO 2 in the Atlantic eastern boundary upwelling regions, and, to some extent, in the Bay of Bengal, while they tend to overestimate O 2 and underestimate CO 2 in the Arabian Sea and in large areas of the Pacific -in particular in the North Pacific. These mismatches are linked to too low export production in the North Pacific and the Arabian Sea and too high export in the Bay of Bengal in the models (Steinacher et al., 2010). The relatively pronounced O 2 and CO 2 biases simulated by BCM-C in the Southern Ocean are related to the strong mixing process. Despite these deficiencies, the models represent many features of the observation-based CO 2 and O 2 fields between 100 and 600 meter depth, such as low O 2 in the equatorial and eastern boundary upwelling systems, high O 2 in the gyres and in the North Atlantic, and lower O 2 in the North Pacific (Fig. B1 in Appendix B). Other reproduced features are the low CO 2 in the gyres and higher values in the equatorial and eastern boundary upwelling regions (Fig. B2 in Appendix B).
In conclusion, the comparison between model results and data-based estimates reveals that, while representing major large-scale features and global trends, the models suffer from considerable shortcomings in the representation of O 2 and CO 2 . Regionally, large deviations in the CO 2 in the UML compared to observations are not a surprise. Deviations in modeled DIC, alkalinity, temperature, and salinity all contribute to deviations in CO 2 and resulting errors can be large even if deviations in the four state variables (DIC, ALK, TEMP, and SALT) are modest as the carbonate chemistry is non-linear. In any case, the top-down forcing by rising Biogeosciences, 10, 1849-1868, 2013 www.biogeosciences.net/10/1849/2013/ atmospheric CO 2 concentrations leads to rising DIC concentrations (and f CO 2 ), as found in the models and in observations. Perhaps more problematic is the misrepresentation of the O 2 fields in the tropical UML. As the interest is to a large extent on the future evolution of low-oxygen regions, the challenge is in the correct representation of subtle changes in the balance between O 2 supply to the thermocline by physical mixing and advection, and O 2 consumption by remineralization of organic material. Thus, we expect that projections of the evolution of low O 2 regions will vary among the models and be affected by large uncertainties.

Projections
Figure 4 presents an overview of the current and projected distribution in the UML for a set of variables of potential biological relevance: temperature, O 2 , CO 2 , RI, pH, and saturation state of water with respect to aragonite ( A ). A widespread warming of the UML is projected under the SRES A2 scenario with typical temperature changes over the 21st century of 1 to 2 • C. The 21st century changes in O 2 are less conspicuous on a global scale and no large expansion or contraction of thermocline regions with low O 2 is projected (variations are in general around ±10 mmol m −3 ). On the other hand, a large increase in CO 2 is projected everywhere in the UML. Concomitantly, a decrease in RI, pH, and A is evident in Fig. 4. The different models simulate a consistent increase in sea surface temperature, CO 2 uptake and O 2 outgassing (Fig. 5). Therefore, a widespread change in physical and chemical ocean properties is projected. Changes in O 2 , CO 2 , f O 2 , f CO 2 , and RI for the individual models are analyzed in more detail in the following sections.

Oxygen and f O 2
A global decline in the oceanic dissolved oxygen inventory is projected by all models, with a 1870-2100 decrease between 2 % and 4 % (for details about CSM1.4 see Frölicher et al., 2009). The 21st century changes in the frequency distribution of O 2 in the UML are modest (Fig. 2), with higher values becoming somewhat less abundant (in particular for MPIM, GFDL,UVIC2-8, and CCSM3). O 2 changes in the UML (Fig. 6) show a complex pattern that reflects the influence of different competing factors such as circulation, production, remineralization and temperature changes. The models simulate consistently, i.e., at least five of the seven models agree in sign, a decrease in O 2 in the UML of the northern Pacific, the tropical and subtropical South Pacific, the Southern Ocean, the eastern part of the Indian Ocean and the subpolar North Atlantic. A relatively small increase in O 2 is consistently found in the multi-model average in the UML of the tropical oceans -including the eastern boundary upwelling system, the Caribbean Sea, western parts of the Indian Ocean, and the California upwelling region. The decrease is particularly pronounced (up to −60 mmol m −3 ) in the northwestern Pacific in UVIC2-8, BCM-C, IPSL and GFDL. Modest changes are found in the UML around 40 • S, north of the equatorial Pacific, and in the subtropical Atlantic.
We next turn our attention to the low-O 2 regions and analyze whether they are projected to expand or to contract until 2100 under the SRES A2 forcing. Specifically, we address changes in the volume occupied with waters that hold less than 80 mmol m −3 , less than 50 mmol m −3 , and less than 5 mmol m −3 (Table 1 and Fig. 7). In Table 1 the projected 2090s (2090-2099 average) volume for the three low-O 2 regimes and the different models is listed. While most models project a contraction of the suboxic waters (O 2 < 5 mmol m −3 ), the model projections are not in agreement about the more oxygenated waters. Figure 7 illustrates the temporal evolution of the water volumes with low-O 2 , corresponding relatively to the 1870s (1870-1879 average) volume for each model and each low-O 2 class. By year 2100, four models project a modest expansion of the regions with O 2 concentration below 80 mmol m −3 (Fig. 7a) exhibit relevant changes, and GFDL and MPIM show, respectively, a 2 % and 26 % reduction. Projected changes in the volume with O 2 concentrations below 50 mmol m −3 (Fig. 7b) are smaller: two models (CCSM3, CSM1.4) project a 4.5 % volume increase, while the others project a decrease between 0.4 % (UVIC2-8) and 4.5 % (MPIM). For suboxic conditions (O 2 concentration below 5 mmol m −3 , Fig. 7c), the volume is projected by most models to contract slightly (up to −10 %). However, IPSL projects a 25 % contraction of the suboxic volume in the 2090s with respect to the condition in the 1870s.
Internal variability may mask part of the model-model differences, since we are analyzing single-mode simulations. Some models feature large interannual and decadal-scale variability in the volume evolution (especially IPSL). If such  Next, we quantify the different mechanisms responsible for the changes in the O 2 concentration following the procedure described in the experimental design section and adopted from Frölicher et al. (2009). Rising temperatures affect O 2 concentrations directly through the temperature dependence of gas solubility -estimated in previous studies being responsible for 25 % to 50 % of the total O 2 changes (Bopp et al., 2002;Plattner et al., 2002;Frölicher et al., 2009). But warming induces also indirect effects, via stratification, circulation and soft-tissue pump alterations. As an illustration, we discuss average O 2 changes in the UML of the entire Atlantic Basin and in the Atlantic OMZs (i.e., O 2 smaller than 50 mmol m −3 ) as simulated by CSM1.4 over the period from 1870 to 2100 (Fig. 8b). The focus is on the Atlantic Basin because CSM1.4, compared to other models, simulates OMZs of limited extension in the Pacific.
In the Atlantic as a whole, O 2 concentrations decrease on average by around 7 mmol m −3 in the upper 800 m. This decrease is driven by ( O 2gas ), i.e., by changes in solubility and air-sea fluxes not directly associated with changes in saturation concentration, whereas the reorganization of the biological cycle leads to a decrease in phosphate and a related increase in O 2 ( O 2bio ) that partly offsets the decrease.
In the OMZs of the Atlantic, the O 2 concentration is projected to slightly increase between 200 and 500 m depth (i.e., positive O 2OMZ , Fig. 8b). At 250 m depth O 2OMZ shows the maximum increase of +3 mmol m −3 . This increase results from the slightly positive balance between a reduced O 2 consumption by remineralization and the reduced physical flux of O 2 related to air-sea exchange.
A comparison of the spatial distribution of the changes in O 2 and phosphate ( Fig. 6) points to linkages between O 2 changes in the thermocline and changes in the marine biological cycle. In particular, a decrease in primary production, export of organic material and remineralization (see Steinacher et al., 2010) is linked to the projected O 2 (f O 2 ) increase in the tropical Pacific and Atlantic, and in the Pacific eastern boundary upwelling systems. Regional changes in O 2 and phosphate in the thermocline differ among models similarly as for f O 2 . This reflects differences in how stratification and circulation is changing in the different models and how the balance between remineralization and physical tracer transport is altered in the thermocline. Such changes can reflect complex feedbacks and interactions between the physical and biogeochemical system, but their identification in the seven models is beyond the scope of this study. The seven models simulate consistently a warming UML under the SRES A2 scenario (Fig. 6), except in a few grid cells. Hence, in the UML the pattern of changes resulting from alteration in the biological cycle is shifted to smaller O 2 concentration because of the decrease in O 2 solubility due to rising temperatures.
In conclusion, the net changes in O 2 depend on changes in the subtle balance between O 2 supply by physical transport and O 2 consumption by remineralization. Both processes are linked as remineralization is primarily driven by the rate of export of organic matter, which itself depends sensitively on the physical transport of nutrients into the euphotic zone (Steinacher et al., 2010).

CO 2 and DIC
As expected under a high CO 2 emission scenario, a widespread increase in CO 2 in the UML (Fig. 9), as well as a shift towards higher values in the CO 2 frequency distribution (Fig. 3), is projected by the end of the 21st century as a consequence of the atmospheric CO 2 invasion. The highest peak in the frequency distribution, situated at 0.1-0.2 mol m −3 in the 1990s, is found between 0.2-0.4 mol m −3 by the end of this century. CO 2 is projected to increase by up to 0.045 mol m −3 and large changes are simulated in the Atlantic and Pacific eastern boundary upwelling systems, in the Bay of Bengal (CSM1.4 and CCSM3) and in the North Pacific (GFDL, IPSL, BCM-C).
The simulated increase in DIC in the UML (Fig. 10) reveals the familiar pattern of high changes in the intermedi- ate water masses around 40 • N and 40 • S and in the wellventilated North Atlantic, and relatively small changes in the tropical upwelling regions and around Antarctica. This pattern in DIC change is consistent among the different models and also consistent with observation-based reconstructions for the historical period . This increase in DIC is predominantly driven by the invasion of anthropogenic carbon from the atmosphere. Changes in DIC due to the reorganization of the marine biological pump can be approximately quantified by multiplying the simulated changes in phosphate (Fig. 6) with the typical C : P Redfield ratio of 117 after Anderson and Sarmiento (1994) (a value in the middle of the range given by Takahashi et al., 1985;Redfield et al., 1963). These changes are typically much smaller than the simulated DIC increase.
We attribute the f CO 2 change to changes in DIC, temperature, ALK, and salinity, following the procedure described in the experimental design section (Fig. 11a, b). The f CO 2 increase is predominantly driven by the uptake of anthropogenic CO 2 from the atmosphere. On global average, changes in f CO 2 in the UML as well as in the deep ocean are predominantly driven by the increase in DIC, with additional contributions resulting from the warming thermocline and minor contributions from changes in alkalinity on the global scale. For example, for CSM1.4 (Fig. 11b) the f CO 2 increase at the surface is ∼ 470 µatm of which 400 µatm are attributed to the change in DIC. At 230 m depth the total increase is ∼ 560 µatm; the effect of temperature changes (+30 µatm) is one order of magnitude smaller than the DIC contribution (+510 µatm). The contributions from changes in salinity, phosphate and silicate to changes in f CO 2 are negligible on global average.
In addition to anthropogenic carbon uptake, the change in DIC is also partly attributed to the reorganization of the marine biological cycle. The total DIC for the period 1870-2100 is about +130 mmol m −3 at the surface (Fig. 11b). A slightly more nuanced picture emerges when considering the contribution by individual drivers to the f CO 2 changes in the thermocline on a regional level. In Fig. 12 the ratio between the f CO 2 change due to DIC changes ( f CO 2DIC ) and the total f CO 2 change ( f CO 2tot ) for the different models is plotted. The f CO 2DIC / f CO 2tot ratio is generally below one, with the exceptions of the Atlantic and Arctic oceans for IPSL, the Atlantic and Indian basins for GFDL and the North Atlantic for the two NCAR models and BCM-C. This results from the compensating effect of alkalinity changes (negative f CO 2ALK values) in these regions.  In conclusion, all seven models consistently simulate an increase in f CO 2 in the UML. Some regional differences remain among the models and are likely related to differences in the thermocline ventilation and its change with time. The f CO 2 increase is predominately driven by the uptake of anthropogenic CO 2 from the atmosphere.

Respiration index
The observation-based RI distribution in the UML (Fig. 4) reveals minima in the eastern tropical Pacific, in the Arabian Sea and in the Bay of Bengal as well as in the eastern upwelling region of the Atlantic and Pacific, while maximum values are found in the Arctic.
Typically, simulated RI in the UML is on average reduced by 0.2 to 0.6 units between 1870 and 2100 ( Fig. B3 in Appendix B).
The RI depth profile in the eastern tropical Pacific (105-115 • W and 0-10 • N) and for CCSM3 is analyzed as an illustrative example to demonstrate the influence of changes in f O 2 and f CO 2 on RI (Fig. 13). This region is projected to become part of a large low-RI area in CCSM3. At preindustrial time, simulated RI is well above unity, taken as threshold for aerobic stress, at all depths. RI is projected to decline in the upper 1000 m and RI falls on average below unity between 300 and 600 m until 2100 under the SRES A2 scenario. This decline is the result of both a decline in f O 2 and an increase in f CO 2 .
In Fig. 14  project modest changes in the volume of water with RI < 0 (below 1 %). BCM-C represents an exception, projecting a 3 % expansion, mainly caused by a contraction in the volume with RI > 1 (−3.5 %). UVIC2-8 exhibits a 0.6 % increase for water with RI < 0, while MPIM and IPSL project a 0.6 % and 0.3 % decrease, respectively. The other three models exhibit smaller changes. Diverging changes are found in the volume with RI > 1, which is projected to shrink by four of the seven models: CCSM3 (−0.4 %), CSM1.4 (−0.8 %), GFDL (−0.4 %) and BCM-C (−6 %). The other three models project an increase: +0.4 % for the MPIM, +0.3 % for the IPSL and +0.2 % for the UVIC2-8, mainly driven by the hypoxic and suboxic waters contraction.
In conclusion, the models do not suggest a widespread expansion of waters with RI below unity, consistent with small changes in the volume of hypoxic and suboxic waters. However, RI declines in the UML in most regions due to the combined influence of rising f CO 2 and decreasing f O 2 levels.

Summary and discussion
In this study we have mapped the changes in a few potential stressors for aerobic organisms in the UML for a businessas-usual greenhouse gas emission scenario using a range of Earth System Models. The focus of this study is on the evolution of CO 2 , f CO 2 , O 2 , and f O 2 . Changes in these variables come in combination with changes in other environmental parameters, including a rise in temperature and hydrogen ion concentrations, and implied changes in the speciation of trace metals and other acid-base relationships. These collective changes likely influence the performance and functioning of marine organisms in systematic and synergistic ways. There exists a range of studies discussing projections of pH and calcium carbonate saturation state under future anthropogenic carbon emission and global warming, and we refer the reader to the literature for further details (e.g., Kleypas et al., 1999;Orr et al., 2005;Steinacher et al., 2009;Frölicher and Joos, 2010;Joos et al., 2011).
The Earth System Models represent the interactions between the physical climate system, biogeochemical cycles, and marine ecosystems under global warming. The limitations in projecting O 2 changes arise from the limited understanding of the marine biological processes and the uncertainties about the future evolution of production/remineralization and marine ecosystem functions under rising CO 2 Steinacher et al., 2010). Changes in nutrient fluxes from land to coastal zones and increasing atmospheric nitrogen deposition (Duce et al., 2008) -which can result in local eutrophication and O 2 depletionare not included. An intrinsic difficulty is the complexity and multitude of interactions and teleconnections. Gnanadesikan et al. (2012) analyze the O 2 changes in GFDL in detail. They identify an additional supply of O 2 along isopycnals that is fed by changes in convection and vertical mixing associated with salinification of the surface ocean off Chile. Their results suggest that the simulated changes in the volume of suboxic water in climate models can depend sensitively on changes in the heat and salt balance in nearby convective regions and thus on the patterns of global warming and precipitation minus evaporation. These authors examine also the pattern of the change in advective supply in GFDL and find that low-latitude O 2 changes are driven by a decline in net upwelling across 1500 m with more influence from high-O 2 , low-nutrient, young surface waters and less from the older low-O 2 , high-nutrient deep waters.
For CSM1.4, Frölicher et al. (2009) identify a significant warming of the ocean and a decreasing trend of ocean ventilation, which both tend to lower oceanic O 2 , as well as a decreasing trend in organic matter production, export and remineralization, which tend to increase subsurface O 2 and to decrease O 2 production at surface. These authors also highlight the role of variability related to climate modes such as the North Atlantic Oscillation and the Pacific Decadal Oscillation and to volcanic forcing. The models simulate consistently an increase in upper ocean temperature, carbon dioxide, and a decrease in pH and in the global O 2 inventory. , 10, 1849-1868, 2013 www.biogeosciences.net/10/1849/2013/ As expected under a high-emission scenario, a widespread increase of CO 2 in the UML is projected. The increase in CO 2 is primarily the direct result of the invasion of anthropogenic carbon from the atmosphere, and it is mainly responsible for the widespread decrease in RI outside low O 2 regions. RI in the UML is on depth average reduced by 0.2 to 0.6 units. Changes in dissolved O 2 (and f O 2 ) are overall relatively small compared to the changes in CO 2 (and f CO 2 ). The frequency distribution of O 2 in the UML is projected to change little over this century even for the high emission A2 scenario, a result that is consistent across the seven models. But projected changes in O 2 and dissolved O 2 in the UML are regionally of opposite sign. At least five out of seven models simulate in the UML a decrease in O 2 in the Southern Ocean, in the northern North Atlantic and North Pacific and in most of the subtropical Pacific and subtropical Indian Ocean, whereas O 2 in the UML is projected to increase in the equatorial and subtropical Southeast Atlantic, off California, in the Caribbean and tropical Indian Ocean in the multi-model average. Projected changes in the volume of hypoxic waters (< 50 mmol m −3 ) are small and within ±7 % in the seven models. Similarly, projected changes in suboxic waters are small. Analogously to the evolution of the volume occupied by low-O 2 waters, in general only small changes are projected for the volume of waters with a low RI.

Biogeosciences
The dominating mechanisms leading to changes in temperature and CO 2 versus those that affect O 2 in the UML are fundamentally different under rising anthropogenic greenhouse gas emissions. The decadal-to-century scale uptake of CO 2 and heat by the ocean is driven "top down" by the increase in emissions of anthropogenic CO 2 and greenhouse gases that cause atmospheric CO 2 and radiative forcing to rise. The resulting perturbations in the air-sea temperature and CO 2 gradients cause a net flux of heat and CO 2 from the atmosphere to the ocean. The temperature and CO 2 content of the surface ocean increase and these perturbations are further penetrating the ocean by physical transport (advection, convection, diffusion). However, due to different uptake of CO 2 by the land, individual models are forced by different atmospheric pCO 2 , even though emissions are identical.
The partial pressure or fugacity of O 2 in the surface ocean is, except for areas under sea ice, very close to the almost constant atmospheric O 2 partial pressure. The typical time scale to equilibrate the well-mixed surface layer with atmospheric O 2 is roughly one month -much shorter than the multi-decadal ventilation time scales of the thermocline. Thus, in contrast to CO 2 or pH and temperature, f O 2 in the surface layer does hardly vary over multi-decadal timescales. The long-term changes in f O 2 and O 2 within the thermocline reflect changes in the subtle balance between O 2 supply from the surface layer to the thermocline, O 2 consumption by the remineralization of organic matter within the thermocline, and advection and diffusion of low-O 2 water masses from depth. In general, the models project an increase in stratification and a reduction in surface-to-deep tracer exchange under global warming in mid-and low latitudes (Steinacher et al., 2010). In turn there is a decrease in O 2 advection to the thermocline, as well as nutrient supply to the surface ocean, production, export, remineralization, and O 2 consumption within the thermocline. Remineralization rates are largely driven by the vertical flux of organic particles that settle due to gravitation, whereas tracer transport has typically a large horizontal component, resulting from advection and diffusion along the sloping layers of equal density. Not surprisingly then, the net balance of these opposing processes in terms of f O 2 and O 2 changes are regionally distinct.
We conclude that temperature, DIC and CO 2 will continue to rise, and pH and calcium carbonate saturation will continue to decrease in most regions in the thermocline. The link between these large-scale and substantial environmental changes and anthropogenic carbon emissions is firmly established and well understood. In contrast, changes in dissolved O 2 resulting from global warming appear relatively small, regionally different, and uncertain. This calls for a distinction of virtually certain ocean warming and acidification versus the postulated deoxygenation of thermocline waters in response to anthropogenic climate change. The challenge is in the observational detection as well as in correct representation of the subtle changes in the balance between O 2 supply to the thermocline by physical mixing and advection, and O 2 consumption by remineralization of organic material.

A1 IPSL
The IPSL-CM4-LOOP (IPSL) model consists of the Laboratoire de Météorologie Dynamique atmospheric model (LMDZ-4) with a horizontal resolution about 3 • × 3 • and 19 vertical levels (Hourdin et al., 2006), coupled to the OPA-8 ocean model with a horizontal resolution of 2 • × 2 • · cos φ and 31 vertical levels and the LIM sea ice model (Madec et al., 1998). The terrestrial biosphere is represented by the global vegetation model ORCHIDEE (Krinner et al., 2005) and the marine carbon cycle is simulated by the PISCES model (Aumont et al., 2003). PISCES simulates the cycling of carbon, oxygen, and the major nutrients determining phytoplankton growth (PO 3− 4 , NO − 3 , NH + 4 , Si, Fe). The model has two phytoplankton size classes (small and large), representing nanophytoplankton and diatoms, as well as two zooplankton size classes (small and large), representing microzooplankton and mesozooplankton. For all species the C : N : P ratios are assumed constant (122 : 16 : 1, Takahashi et al., 1985), while the internal ratios of Fe : C, Chl : C, and Si : C of phytoplankton are predicted by the model. Phytoplankton growth is limited by the availability of nutrients and www.biogeosciences.net/10/1849/2013/ Biogeosciences, 10, 1849-1868, 2013 light. For a more detailed description of the PISCES model see Aumont and Bopp (2006) and Gehlen et al. (2006). Further details and results from the fully coupled model simulation of the IPSL-CM4-LOOP model are given in .

A2 MPIM
The Earth System Model employed for this study at the Max Planck Institute for Meteorology (MPIM) consists of the ECHAM5  atmospheric model with 31 levels with the embedded JSBACH terrestrial biosphere model and the MPIOM physical ocean model, which includes a sea ice model (Marsland et al., 2003) and the Hamburg Ocean Carbon Cycle model HAMOCC5.1 marine biogeochemistry model (Maier-Reimer, 1993;Six and Maier-Reimer, 1996;Maier-Reimer et al., 2005). The coupling of the marine and atmospheric model components, and in particular the carbon cycles, is achieved by using the OA-SIS coupler. HAMOCC5.1 is implemented into the MPIOM physical ocean model configuration using a curvilinear coordinate system with a 1.5 • nominal resolution where the North Pole is placed over Greenland, thus providing relatively high horizontal resolution in the Nordic seas. The vertical resolution is 40 layers, with higher resolution in the upper part of the water column (10 m at the surface to 13 m at 90 m). A detailed description of HAMOCC5.1 can be found in Maier-Reimer et al. (2005). The marine biogeochemical model HAMOCC5.1 is designed to address largescale, long-term features of the marine carbon cycle, rather than to give a complete description of the marine ecosystem. Consequently, HAMOCC5.1 is a NPZD model with one phytoplankton group (implicitly divided into calcite (coccolithophorids) and opal (diatoms) producers and flagellates), one zooplankton species, and particulate and dissolved dead organic carbon pools. The model contains over 30 biogeochemical tracers, which include dissolved inorganic carbon, total alkalinity, oxygen, nitrate, phosphate, silicate, iron, phytoplankton and zooplankton.

A3 CSM1.4
The physical core of the NCAR CSM1.4 carbon climate model Fung et al., 2005) is a modified version of the NCAR CSM1.4 coupled physical model, consisting of ocean, atmosphere, land and sea ice components integrated via a flux coupler without flux adjustments (Boville et al., 2001;Boville and Gent, 1998). The atmospheric model CCM3 is run with a horizontal resolution of 3.75 • and 18 levels in the vertical (Kiehl et al., 1998). The ocean model is the NCAR CSM Ocean Model (NCOM) with 25 levels in the vertical and a resolution of 3.6 • in longitude and 0.8 • to 1.8 • in latitude . The sea ice component model runs at the same resolution as the ocean model, and the land surface model runs at the same resolu-tion as the atmospheric model. The CSM1.4-carbon model includes a derivate of the OCMIP-2 (Ocean Carbon Cycle Model Intercomparison Project Phase 2) ocean biogeochemistry model (Najjar et al., 2007). In the ocean model, the biological source-sink term has been changed from a nutrient restoring formulation to a prognostic formulation inspired by Maier-Reimer (1993). Biological production is modulated by temperature (T ), surface solar irradiance (I ), mixed layer depth (MLD), and macro-and micro-nutrients (PO 3− 4 and iron). The C : P ratio is constant and set to 117 (Anderson and Sarmiento, 1994).

A4 CCSM3
The NCAR CCSM3 carbon is a coupled climate model that uses no flux correction (Collins et al., 2006b). The model components are the Community Atmosphere Model version 3 (CAM3, Collins et al., 2006a), the Community Land Model version 3 (CLM3, Dickson et al., 2006), the Parallel Ocean Program version 1.4 (POP1.4, Danabasoglu et al., 2006), and the Community Sea Ice Model (CSIM, Holland et al., 2006). In CCSM3 carbon, the land model is on the same horizontal grid as CAM3 (T31) and the sea ice model shares the same horizontal grid as the ocean model (gx3v5). Overall, CCSM3 carbon includes improvements in the parametrization of the physics as well as the biogeochemical cycles compared to earlier NCAR model versions (e.g., NCAR CSM1.4-carbon). New treatments of cloud processes, aerosol radiative forcing, land-atmosphere fluxes, ocean mixed layer processes, and sea ice dynamics are included. The CCSM3 Biogeochemical Elemental Cycling (BEC) model includes several phytoplankton functional groups, one zooplankton group, semilabile dissolved organic matter, and sinking particles (Moore et al., 2004). Model-data skill metrics for the simulated marine ecosystem in uncoupled ocean experiments are reported in Doney et al. (2009). The BEC includes explicit cycling of C, N, P, Fe, Si, O and alkalinity. Phytoplankton functional groups include diatoms, diazotrophs, picoplankton, and coccolithophores. Phytoplankton Fe/C, Chl/C, and Si/C ratios adjust dynamically to ambient nutrient and light, while the C/N/P ratios are fixed within each group (Moore et al., 2004). The CCSM3 ocean circulation model is a coarse resolution version of the Parallel Ocean Program (POP) model with longitudinal resolution of 3.6 degrees and a variable latitudinal resolution from 1-2 • . There are 25 vertical levels with eight levels in the upper 103 m (Smith and Gent, 2004;Yeager et al., 2006).

A5 BCM-C
The Bergen Earth System Model coupled with terrestrial and oceanic carbon cycle models (BCM-C) is described in detail in Tjiputra et al. (2010), where regional climate-carbon cycle feedbacks are assessed. The physical ocean part of the BCM-C model is based on the Miami Isopycnic Coordinate Ocean   Biogeosciences, 10, 1849-1868, 2013 www.biogeosciences.net/10/1849/2013/ Model (MICOM) and is documented in Bleck and Smith (1990) and Bleck et al. (1992). Updates to the model code are described in Bentsen et al. (2004) and Assmann et al. (2010). With the exception of the equatorial region, the grid configuration employed here is almost regular with a horizontal grid spacing of approximately 2.4 • × 2.4 • . In order to better resolve the dynamics near the equator, the horizontal spacing in the meridional direction is gradually decreased to 0.8 along the equator. The marine carbon cycle is represented by the HAMOCC5.1 model . However, the biogeochemical was refined and adapted for isopycnic coordinates . The current version of the model includes an NPZD-type (nutrient, phytoplankton, zooplankton and detritus) ecosystem model following Six and Maier-Reimer (1996). The inorganic carbon chemistry in the HAMOCC5.1 model is based on Maier-Reimer and Hasselmann (1987) and is updated with the OCMIP (Ocean Carbon Cycle Model Intercomparison Project) carbon chemistry protocols.

A6 GFDL
The ESM2.1 Earth System Model was developed by the Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, USA, the physical fundament of the model being the CM2.1 coupled climate model (Delworth et al., 2006). In a fully coupled state, the model includes components for ocean (MOM-4 Griffies et al., 2004), atmosphere (AM2.1) and terrestrial biosphere (LM3 Anderson et al., 2004). LM3 is based on potential vegetation, so no land use is included. MOM-4 has 50 vertical levels, its spatial resolution is nominally 1 • globally, with a higher resolution of 1/3 • near the equator. Ocean biogeochemistry is represented by TOPAZ (Dunne et al., 2005) and includes a number of major nutrients (N, P, Si and Fe), both labile and semi-labile dissolved organic pools and parameterizations to represent the microbial loop (Henson et al., 2010). The ocean ecosystem is based on three classes of phytoplankton, biological production is modeled as a function of Chl : C ratios and limited by nutrients and light. Also, a simplified version of the ocean iron cycle is included, comprising biological uptake and remineralization, particle sinking and scavenging and adsorption/desorption.

A7 UVIC2-8
The UVIC2-8 model is the University of Victoria (UVic) Earth System Climate Model (Weaver et al., 2001) in the Redfield stoichiometry configuration described by Oschlies et al. (2008). The oceanic component is a fully threedimensional primitive-equation model with nineteen levels in the vertical, ranging from 50 m thickness near the surface to 500 m in the deep ocean. It contains a simple marine ecosystem model with the two major nutrients nitrate and phosphate and two phytoplankton classes, nitrogen fixers and other phytoplankton, with the former being limited only by phosphate. The trace nutrient iron is not explicitly included in the model. By tuning the biological parameters, in particular the relatively low maximum phytoplankton growth rate (0.13 day −1 at 0 • C), the model nevertheless achieves a reasonable fit to observed biogeochemical tracer distributions . Organic matter is produced, processed, and remineralized according to a fixed elemental stoichiometry of C : N : P = 112 : 16 : 1. The ocean component is coupled to a single-level energy-moisture balance model of the atmosphere and a dynamic-thermodynamic sea ice component. The terrestrial vegetation and carbon cycle component is based on the Hadley Centre's TRIFFID model (Cox et al., 2000). All model components use a common horizontal resolution of 1.8 • latitude × 3.6 • longitude.

Appendix B
Projected changes in RI -additional figure Figure B3 illustrates the projected RI changes in the UML by year 2100. The largest reductions are projected by BCM-C in the Southern Ocean, the North Atlantic and Pacific. Most of the decrease is related to the increase in f CO 2 in many areas. RI is projected to increase in the eastern South Atlantic in most models. This is consistent with the projected changes in f O 2 in this region. The increase in f O 2 dominates over the projected increase in f CO 2 in this low-oxygen region.