MOMSO 1.0-a near-global, coupled biogeochemical ocean-circulation model configuration with realistic eddy kinetic energy in the Southern Ocean

We present a new near-global coupled biogeochemical ocean-circulation model configuration. The configuration features a horizontal discretization with a grid spacing of less than 11km in the Southern Ocean and gradually coarsens in meridional direction to more than 200 km at 64◦ N where the model is bounded by a solid wall. The underlying code framework is GFDL’s Modular Ocean Model coupled to the Biology Light Iron Nutrients and Gasses (BLING) ecosystem model of Galbraith et al. 5 (2010). The configuration is cutting-edge in that it features both a relatively equilibrated oceanic carbon inventory and a realistic representation of eddy kinetic energy a combination that has, to-date, been precluded by prohibitive computational cost. Results from a simulation with climatological forcing and a sensitivity experiment with increasing winds suggest that the configuration is suited to explore Southern Ocean Carbon uptake dynamics on decadal timescales. Further, the fidelity of simulated bottom water temperatures off and on the Antarctic Shelf suggest that the configuration may be used to provide 10 boundary conditions to ice-sheet models. The configuration is dubbed MOMSO a Modular Ocean Model Southern Ocean configuration.

and oceanic CO 2 concentrations which drive net-air sea fluxes. A comprehensive quantitative understanding is still work in progress but the consensus is that the variability in the extent, to which deep-water masses in the Southern Ocean are isolated from the atmosphere, is among the major drivers regulating atmospheric CO 2 -variability (e.g., Anderson et al., 2009;van 15 Heuven et al., 2014;Ritter et al., 2017). Consequently, the Southern Ocean has shifted into the limelight of climate research (DeVries et al., 2017;Tamsitt et al., 2017;Langlais et al., 2017, and many more).
As for now we know that the Southern Ocean accounts for almost half of the global oceanic CO 2 uptake from the atmosphere (Takahashi et al., 2012). But, there is concern that anticipated climate change (e.g., via changes of atmospheric circulation and sea-ice cover) may trigger substantial changes in the Southern Ocean carbon budget (e.g., Heinze et al., 2015;Abernathey 20 et al., 2016) such that the current rate of uptake may well decline in decades to come. Indications for the existence of such triggers have been revealed by observation-based atmospheric reanalysis products which show an ongoing strengthening and a poleward shift of the southern westerly winds since the 1970s ( Thompson and Solomon, 2002).
This observed trend is projected by climate scenarios to intensify (e.g., Simpkins and Karpechko, 2012) and it is straightforward to assume that the associated wind-driven circulation impinges on biogeochemical dynamics and, eventually, on the 25 oceanic carbon budget. A comprehensive understanding of the link between changing winds and oceanic upwelling of carbonrich deep waters (which, in turn, affects surface saturation and net air-sea CO 2 exchange) is, however, work in progress.
To this end, the role of mesoscale ocean eddies is especially uncertain: the current generation of coarse resolution (nonmesoscale-resolving) models suggests that a poleward shift and an intensification of the Southern Ocean westerlies results in a strengthening of the subpolar meridional overturning cell (e.g., Saenko et al., 2005;Hall and Visbeck, 2002; Getzlaff of the underlying eddy parameterization in coarse resolution models which can not afford to resolve mesoscale dynamics explicitly. To date we know that very different state-of-the-art approaches to parameterize eddies yield surprisingly similar sensitivities of oceanic carbon-inventories to changing winds Dietze et al. (2017). The question, however, as to how these results compare to high resolution coupled biogeochemical ocean circulation which actually resolve eddies explicitly, has not been answered yet. The main reason being the prohibitive computational cost that is associated to equilibrating simulated 5 dissolved inorganic carbon concentrations at depth.
In this study we present the model configuration MOMSO 1.0. The configuration features realistic levels and distributions of eddy kinetic energy. This suggests that the configuration explicitly resolves a substantial part of mesoscale-related variability rather than relying on parameterizing their effect. MOMSO is designed to explore the sensitivity Southern Ocean carbon uptake to atmospheric changes on decadal scales. The configuration is rendered feasible by recent advances in compute hardware 10 and, by chance, from its similarity with a spun-up coarse resolution model which delivered the initial conditions for the biogeochemical module. More specifically, we will showcase that the "level-of-equilibration" of simulated deep dissolved inorganic carbon allows to test the sensitivity of the Southern Ocean carbon budget to anticipated climate change patterns and include a preliminary result of a comparison to the results of a similar, but coarse resolution, model.
In addition, we showcase that MOMSO is potentially suited to explore the effects of changing eddying circulation patterns on 15 basal melting because it features a realistic representation of Antarctic Continental Shelf Bottom Water (ASBW) temperatures at the seabed off Antarctica. The realistic seabed temperatures, in combination with its biogeochemical module, empowers MOMSO to explore feedback loops such as: atmospherically-driven changes in ocean-circulation drive additional heat supply which fuel basal melting. The buoyant lens of meltwater may, for one, suppress the AABW-formation (Williams et al., 2016). This, in turn, effects oceanic carbon sequestration. Second, the meltwater carries bioavailable iron to the Southern Ocean -20 which affects oceanic primary productivity and the associated export of organic carbon to depth (Grotti et al., 2005;Lannuzel et al., 2008Lannuzel et al., , 2010Raiswell et al., 2008;Smith et al., 2007;Smith and Nelson., 1986;van der Merwe et al., 2009).
In summary, this study aims to (1) describe and present a new eddying coupled ocean-circulation biogeochemical model configuration, and (2) to depict potential applications and associated research questions. The project is dubbed MOMSO, a configuration of GFDL's Modular Ocean Model version 4p1 with enhanced resolution in the Southern Ocean. The naming is 25 an homage to the underlying framework, the MOM4p1 release of NOAAs Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model (Griffies, 2009). The framework was chosen because of its exceptional documentation, maturity, modularity and user community. The ocean circulation model is coupled to a sea-ice model and the biogeochemical module. The latter has been developed by Galbraith et al. (2010).
The configuration MOMSO is distributed, free of charge, in a joint effort of the biogeochemical modelling group at the This study is based on simulations with the Modular Ocean Model (MOM), version MOM4p1 (Griffies, 2009). The configuration is near-global, bounded by Antarctica and 64 • N. In the Southern Ocean the horizontal resolution is higher than 11 km up till 40 • S. The meridional grid resolution coarsens towards the North (Fig. 1). There are no open boundaries and there is no tidal forcing. 5 The biogeochmical model BLING, short for Biology Light Iron Nutrients and Gasses (Galbraith et al., 2010) is coupled online to the ocean-sea ice model. Atmospheric CO 2 concentrations are prescribed to a preindustrial level of 278 ppmv. The respective carbon inventories and fluxes are referred to as natural carbon.
The remainder of this section provides more details.

10
The underlying bathymetry is ETOPO5 (c.f. Data Announcement 88-MGG-02, Digital relief of the Surface of the Earth. NOAA, National Geophysical Data Center, Boulder, Colorado, 1988). Using a bilinear scheme the bathymetry is interpolated onto an Arakawa B-grid (Arakawa and Lamp, 1977) with 2400 × 482 tracer grid boxes in the horizontal. The ocean-circulation and the sea-ice model share the same horizontal grid.
The vertical discretisation comprises a total of 55 levels. Fig. 1 shows the nominal depth and thickness of each level. The model 15 bathymetry is smoothed with a filter similar to the Shapiro filter (Shapiro, 1970). The filter weights are 0.25, 0.5 and 0.25. The filtering procedure can only decrease the bottom depth, i.e. essentially, it fills rough holes. The filter is applied three times consecutively. The resulting bathymetry contained lakes which we filled after visual inspection. In addition, we filled narrow inlets which had a width of less than three grid boxes. In total, MOMSO has 42.429.759 wet tracer grid boxes.
We use the zstar coordinate (Stacey et al., 1995;Adcroft and Campin, 2004) in the vertical, i.e., the depth and thickness of 20 each level varies with time. zstar (z ) is calculated as a function of nominal depth (z), water depth (H) and the free sea surface height (η) which varies with time: (equation 6.6 in Griffies, 2009). The approach overcomes the problem with vanishing surface grid boxes which appears in generic z-level discretisation when sea surface height variations are of similar magnitude than the thickness of the uppermost 25 grid box.

The Ocean Component
MOM4p1 is a z-coordinate, free surface ocean general circulation model which discretizes the ocean's hydrostatic primitive equations on a fixed Eulerian grid. The vertical mixing of momentum and scalars is parameterized with the K-Profile-Parameterization approach of Large et al. (1994) with the same parameters applied in eddy-permitting global configurations of The background values apply also below the surface mixed layer throughout the water column. Both parameterizations of the nonlocal and the double diffusive (vertical) scalar tracer fluxes are applied.
We apply a state-dependent horizontal Smagorinsky viscosity scheme (Griffies and Hallberg, 2000) to keep friction at the minimal level demanded by numerical stability. We use the PPM advection scheme (Colella and Woodward, 1984) for active 5 tracers and a flux-limited scheme following Sweby (1984) for biogeochemical tracers. We do not apply an explicit horizontal background diffusivity other than the contribution that is implicit to the advection scheme.
Several decades into the spin-up the configuration became unstable in coarsely-resolved places where strong currents met rough topography. Setting an additional horizontal isotropic Laplacian Viscosity of 600 m 2 /s from 10 • S to 50 • N, of 1200 m 2 /s above 50 • N and 1800 m 2 /s and above 60 • N until the northern boundary of the model domain kept the respective oscillations 10 in check. In addition we added Laplacian Viscosity at the exit of Drake Passage (Fig. 2).

The Sea Ice Component
The ocean component is coupled to a dynamical sea ice module, the GFDL Sea Ice Simulator (SIS). SIS uses elatic-viscousplastic rheology adapted from Hunke and Dukowicz (1997). In the standard version, the simulated sea ice impacts sea surface height. This led to a viscous cycle at some places where sea ice attracts ever more sea ice resulting in unrealistic anomalies in 15 sea surface height. We solved this problem by switching to levitating sea ice (by applying a small change to the code).

The Biogeochemical Component
In our setup, the ocean component is coupled to the BLING ecosystem model of Galbraith et al. (2010). BLING is a prognostic model that, in the basic version, explicitly resolves four biogeochemical tracers: dissolved inorganic phosphorous, dissolved organic phosphorous, dissolved iron and dissolved oxygen. In this study we use BLING in conjunction with a carbon module 20 that explicitly resolves dissolved inorganic carbon and alkalinity as described, e.g., in Bernadello et al. (2014).
The design idea behind the "reduced-tracer" model BLING is a low computational cost and yet, complex-enough, framework to be utilized in high-resolution configurations. Our choice of BLING is motivated by it's fidelity which is comparable to much more complex (and computationally expensive) models (Galbreith et al., 2015). Further, the choice allows for a comparison with the coarse-resolution setup described in Dietze et al. (2017) which uses the exact same BLING configuration.

Initial Conditions and Spin-up Procedure
The circulation model starts from rest (i.e. initial velocities are nil) with initial values for temperature and salinity taken from WOA2009 Antonov et al., 2009, respectively). After 20 years of physics-only spin-up, the biogeochemical model is hooked on. The initial conditions for the biogeochemical tracers are interpolated from the fully spun-up coarse resolution configuration used by Dietze et al. (2017) (their "FMCD" simulation) which is, apart from the spacial discretization dynamics substantially compared to using observational products. After a subsequent 60 year-long spin-up with on-line biogeochemistry the model allows already (as we will put forward in Section 3.2) for an investigation of circulation-driven decadal changes of the Southern Ocean Carbon Budget.

Boundary Conditions and Sponges
The boundaries towards the Arctic (i.e. the northern end of the model domain shown in Fig. 1) are closed (i.e. they are 5 represented by solid and flat walls). Temperature and salinity are restored to climatological estimates Antonov et al., 2009) in so-called sponge zones located in the coarse-resolution domain. The sponge zones along with restoring timescales are shown in Fig. 3. The purpose of these sponges is to ensure realistic deep-water characteristics even though northern-hemisphere deep-water formation processes are handicapped by the combination of coarse resolution with the absence of eddy-parameterizations. 10 At the air-sea boundary we apply climatological atmospheric conditions taken from the Corrected Normal Year Forcing (COREv2 Large and Yeager, 2004). In addition we apply a surface salinity restoring to climatological values ) with a timescale of 1/2 year throughout the model domain.
Atmospheric CO 2 concentrations are prescribed to a preindustrial level of 278 ppmv. Thus the simulated oceanic carbon is also referred to as natural carbon. Biogeochemical air-sea fluxes (of iron) are identical to the ones applied in Galbraith et al.

Results
In the following we evaluate our model (simulation REF) by comparing our climatological results from the nominal years 1980 -2024 to observational data (Sec. 3.1). One problem is the tradeoff between data density and the length of the period the data is representative for. For any given year data densities are typically insufficient to compile a comprehensive 3-dimensional gridded 25 data product. Binning data of several years into one product closes spacial data gaps, but then, this blurs the referencing to an ever (anthropogenically-driven) changing system state. This problem is especially pronounced in the Southern Ocean where in-situ data acquisition is complicated by hostile environmental conditions.
The climatological atmospheric boundary conditions which drive our ocean model are representative for the period 1958-2000. Climatological data products are typically biased in that they contain more recent data being the result of recent techno- 30 logical advances (such as the development of autonomous platforms). Hence, a model evaluation is not straightforward and it is difficult to define meaningful model-data misfit metrics. Our pragmatic approach to this problem is to put plots of observed and modeled properties which are typically found in respective publications side to side. Further, we make an effort to put our climatological model results into the context of observed trends/interannual variability.
In addition to the observations being moving targets, unfortunately, the climatological simulation REF does also exhibit a certain amount of drift. This drift is associated to a spin-up procedure that is of finite duration (i.e. starting from rest, it has only been integrated for a finite number of model years). Subsection 3.2 puts this drift into perspective by comparing the 5 drift in simulation REF with the temporal evolution of the simulation WIND which is driven by idealized (but realistic) wind anomalies over the Southern Ocean.
Hence, by putting the configuration's sensitivity in relation to its persistent drift Subsection 3.2 provides a measure of the signal-to-noise ratio.
We close this Section in 3.3 with technical issues.

Evaluation of the Climatological Simulation
The major aim of our model setup is to explore the role of mesoscale features, or, eddies in determining the CO 2 -uptake of the Southern Ocean. One hypothesis this model is set-up to test is whether spatially-unresolved dynamics in IPCC-type coarse resolution models biases their carbon uptake sensitivity. In order to come to a meaningful conclusion on this, our high-resolution model has to perform with a fidelity similar or superior to that of IPCC-type coarse resolution models. In the 15 following we list and explain our choice of model assessments which we deem relevant in this respect. Please note, that a comprehensive coarse versus high resolution comparison is beyond the scope of this high-resolution model description.
-Ocean circulation (Sec. 3.1.1) which, e.g., effects the transport of carbon-rich deepwater to the surface, shapes the locations of fronts and constitutes a major pathway for nutrients that are essential for phytoplankton growth. For the evaluation of surface currents, we use exemplary snapshots showcasing main circulation paths and spacial variability -Eddy kinetic energy (EKE, Sec. 3.1.2), which is an important measure for the mesoscale activity and thus a key proxy for realistically reproducing eddy-dynamics. At the surface the EKE can be derived from the variability of the sea surface height (SSH), a measure that can be directly observed from space by satellite altimetry. and causes convection. Spatial temperature gradients are associated to geostrophic circulation (if they are not salinitycompensated). Further, sea surface temperature (SST) serves as a proxy for the realism of the surface mixed layer depth whose dynamics is a major process involved in the supply of nutrients from depth to the sun-lit surface ocean. The sea surface temperature (SST) can be directly observed from space and, therefore, is available in an unrivaled (compared to in-situ measured properties) spacial and temporal resolution. Temperatures at depth are important for basal melting 5 and, thus, are related to the formation rate of Antarctic Bottom Water (AABW). AABW formation, in turn, affects the solubility and biotic pump of carbon.
-Salinity (Sec. 3.1.4) is related to the density of sea water. Saltening by brine rejection can cause convection, meltwater on the other hand can build lenses thus increasing the local stability of the water column which prevents vertical mixing.
Spatial salinity gradients are associated to geostrophic circulation (if they are not temperature-compensated).

10
-Sea Ice (Sec. 3.1.5) caps the direct exchange between atmosphere and ocean and thus controls the air-sea gas exchange of CO 2 . It also modulates the air-sea buoyancy forcing by, e.g., insulating the surface from heat loss or by brine rejection during ice formation. Further, it shields the surface water from solar irradiance and hampers the assimilation of CO 2 by autotrophic plankton.
-Nutrients (Sec. 3.1.6) which are essential to the growth of autotrophic plankton and whose availability exert major 15 control on the biological pump. The most important macronutrient is bioavailable phosphorous such as phosphate (PO 4 ) because its availability is essential to all phytoplankton (and cyanobacteria). The distribution of PO 4 is determined by the interplay of ocean circulation transporting PO 4 dissolved in sea water and marine biota which utilize phosphorous to build biomass (typically at the surface) and release PO 4 in the course of degradation of organic material (typically at depth). In addition we assess simulated iron concentrations since, the Southern Ocean is well-known for being a site 20 where this is limiting the growth of autotrophs (Boyd and Ellwood, 2010). day and the highly non-linear characteristics of eddy-dynamics renders an "eddy-to-eddy" similarity without data assimilation impossible. So the purpose of Fig.5 is to demonstrate the similarity of patterns and the major transport pathways which coincide remarkably. The closeup into the Agulhas retroflection zone (Fig. 6) highlights that this remarkable similarity is sustained right 30 up into the transition to coarser, non-eddy-resolving resolution. been reported in other high-resolution configurations and may, according to Dufour et al. (2015), be related to a deficient representation of the overflow of dense waters, formed along the Antarctic coasts. If so, the ACC bias may well be endemic to z-level models which struggle to represent complex topography (in comparison to more elaborate numerical approaches such 5 as, e.g., finite elements). A comprehensive investigation is beyond the scope of this manuscript. But, still, the problem is an intriguing one -especially since, historically, (coarse resolution) models started out from an opposing bias dubbed Hidaka's Dilemma (Hidaka and Tsuchiya, 1953), where an excessive ACC transport could, only by application of unrealistically high friction, be fenced into realistic bounds.

Ocean Circulation
In terms of meridional overturning in the Southern Ocean our model values are consistent with the Southern Ocean State and an abyssal cell of 13 ± 6 Sv. We find a climatological mean value for the upper cell of 12 ± 4 Sv and 8 ± 3 Sv for the lower cell.
In terms of transports of the Weddell and Ross gyre our simulation is slightly biased high. In the Weddell gyre we simulate conclude that the energetics of simulated mesoscale has realistic levels and we see no evidence of a general low bias. This suggests that our spacial resolution in the Southern Ocean is high enough to allow for a meaningful investigation of eddydriven processes. A caveat here is that the observed EKE may be biased low as suggested by, e.g., Fratantoni (2001).  On average the simulated Southern Ocean SSS is biased low (compare reference in Fig. 14) compared to recent observational estimates during 2005 -2017 (Fig. 15). The underlying reason is subject to current investigation. For know, Fig. 14  is weak enough to allow for substantial SSS dynamics. Figure 16 shows a comparison of the simulated number of sea ice-covered months per year with an observation estimate (HadISST Rayner et al., 2003). Overall the agreement is good with the following exception: (1) The Weddell and Ross Sea the ice coverage is underestimated by two months. We speculate that this triggers elevated air-sea momentum fluxes and, 15 eventually, biases the respective gyre strengths high (c.f. Section 3.1.1).

Sea Ice
(2) Overall the simulated ice extent is biased high (compare Fig. 17, black line to Fig. 18). This may (or may not) be associated with a mismatch between climatological forcing and observation period. In any case Fig. 17 suggests that increasing the wind speeds to levels being more representative for the time period of observations shown in Fig. 18 alleviates the model bias.

20
Simulated Southern Ocean PO 4 surface concentrations are biased low by down to 0.6 mmol P/m 2 locally (Fig. 19 a, c). The reason is not straightforward to identify because it could be associated to a deficient physical module, a deficient biogeochemical module, or both. In the following we will present an indication that the problem is associated to a deficient formulation of iron limitation, argue why the formulation of light-limitation is unlikely to be the main problem, voice a caveat, and, finally, put the model-data misfit into perspective.

25
The uptake of PO 4 at the sun-lit surface by autotrophic phytoplankton is known to be limited by the availability of light and the availability of the micronutrient iron (in the Southern Ocean). Fig. 20 features a comparison of simulated iron concentrations with observations. Even though the spacial and temporal coverage of iron measurements is still sparse (because iron is expensive to measure) the emerging pattern is one where the simulated biotic iron-drawdown at the surface appears to be too strong. Surface iron concentrations are biased low, just like the PO 4 concentrations and they appear so throughout an annual 30 cycle. Such deficient model behavior can be caused by insufficient throttling of phytoplankton growth by both iron and light limitation. Looking closer in seasonal model-data misfits, however, suggests that a deficient iron limitation is more likely to be the cause: The dependency of growth (and associated micro-and macronutrient drawdown at the surface) is known to be a highly nonlinear function of environmental drivers. We find that the bias in surface PO 4 concentrations is almost constant over the course of a seasonal cycle (Fig. 19 d), even though the photosynthetically available radiation varies dramatically from season to season in the respective latitudes (also because radiation experienced by phytoplankton cells dispersed in surface waters, is a function of the seasonally varying surface mixed layer depth). By chance, this could be the result of nonlinear forcing 5 modulating a deficient nonlinear formulation of PO 4 -limitation such that the model bias stays constant over a wide range of environmental conditions (here seasons). But this is unlikely. Looking into the seasonal bias of simulated iron concentrations ( Fig. 20 d) we find that it varies substantially from season to season compared to the respective PO 4 variability (Fig. 19 d) just as is expected when a deficient nonlinear model formulation is exposed to substantial variations in driving environmental conditions.

10
A caveat remains. Simulated phosphate concentrations at depth (Fig. 19) are also biased low. This may be associated to a biological pump that is controlled by a deficient iron module. Equally likely, however, is that the biases at depth are caused by a deficient representation of ocean physics and associated abyssal circulation pathways.
In summary, the model's fidelity in reproducing observed patterns of biogeochemical properties is state-of-the-art (being nonjudgmental here). It is comparable to the coarse resolution twin (with twin referring to the biogeochemical model) configuration 15 described in Dietze et al. (2017).

Sensitivity Experiment (WIND) versus (Reference) model drift
In the following we analyze the results of a sensitivity experiment. purpose there being to relate model-data mismatches to potentially mismatching (climatological) model forcing.
In the following the focus differs. We aim to showcase that the sensitivity of our model-configuration to a typical decadal change in boundary conditions is high relative to persistent model drift.
As the winds increase, the Southern Ocean overturning increases Fig. 21. The respective trend in the maximum overturning is clearly distinguishable from the very weak drift of the reference simulation (Fig. 22). Locally, the wind-induced increase 25 in overturning exceeds 0.3 Sv yr −1 (Fig. 23). Averaged over the Southern Ocean, the overturning results in a surface cooling and saltening (Fig. 10). Both, the cooling and the saltening signals are clearly distinguishable from the very weak trend of the reference simulation.
Close to Antarctica the winds transport warmer waters to the surface where they melt (prevent) ice (formation). Fig. 17 shows that this trend is clearly distinguishable from the variability in the reference simulation. 30 Surprisingly, this does also apply to simulated trends in the temperature of the Antarctic Continental Shelf Bottom Water (ASBW) which, naturally, take much longer to equilibrate than surface processes. Fig. 24  can not only be attributed to the recent increases in wind-strength. In any case, Fig. 24 suggests that the equilibration of the model has set in sufficiently so that model responses to decadal forcing variability are detectable.
A caveat here is the ACC transport through Drake passage. Both simulations WIND and the reference feature a comparable and relatively strong drift (Fig.7).
The above mentioned trends in ocean (circulation) physics and changes in the sensitivity experiment WIND have their 5 counterparts in the dynamics of biogeochemical processes. Some of these equilibrate quickly, such as surface nutrients or biomass while others are associated with long timescales. Carbon dynamics is notorious in that respect and this, for decades, has hindered the interpretation (and development) of eddy-resolving coupled ocean-circulation biogeochemical carbon models.
To this end the results summarized in Fig. 25  yields a drift-corrected 5 P gC/(1000 yr 2 − 1 P gC/(1000 yr 2 ) = 4 P gC/(1000 yr 2 )). A comprehensive comparison is a natural application of MOMSO but beyond the scope of this manuscript.

Computational cost
Initial development, testing and spin-up of the physical configuration was carried out on two 32-core workstations, based on a 6320 AMD Opteron (Abu Dhabi) (8-core CPU, 2.80 GHz, 16MB L3 Cache, DDR3 1600) interconnected with a QDR

Summary and conclusions
We set out to develop a near-global coupled ocean-circulation biogeochemical model which explicitly resolves -in contrast version MOM4p1) and SO referring to the Southern Ocean. We use the biogeochemical/carbon module BLING developed by Galbraith et al. (2010).
Overall, we find in a climatological simulation (REF) an impressively eddying surface circulation that is in good agreement with observations from space. Further, simulated temperatures and sea surface salinity show a close agreement with observations. Sea-ice cover is biased high but is still in the range of observed values during particularly cold winters. A remaining 5 caveat is a low bias in the Drake Passage (99 Sv compared to observational estimates ranging from 110 to 170 Sv).
The simulated biogeochemistry is also biased. Surface phosphate concentrations are too low in the Southern Ocean which is indicative of a deficient formulation of the limitation of phytoplankton growth. Seasonally varying biases in simulated surface iron concentrations suggest that the problem is associated to an, up-to-date, uncomprehensive quantitative understanding of iron-dynamics. The model performance with respect to biogeochemistry is similar to what is state-of-the-art in coarse resolution 10 models (e.g., Dietze et al., 2017).
MOMSO is a quantum leap in the field of eddying coupled ocean-circulation biogeochemical carbon modeling in that it allows to investigate the effects of decadal changes in atmospheric boundary conditions on the oceanic carbon uptake. Previous attempts have been hindered by the computational cost that is associated to run simulations into a semi-equilibrated state which features trends considerably lower than the climate signals (as effected by prescribed anomalies in boundary conditions) under 15 investigation. To this end MOMSO benefitted from: (1) an ever increasing ease of access to computing power that is associated to Moore's law and (2) the fortunate coincidence that the spun-up coarse resolution restart from Dietze et al. (2017) was close enough to the equilibrated high-resolution state of MOMSO.
In this study we showcased that the remaining drift in MOMSO's Southern Ocean carbon uptake is substantially lower than changes driven by typical decadal variability of atmospheric variations. We illustrate that the respective sensitivity experiment 20 is suitable for a comparison to coarser-resolution model versions and present a first impression in Fig. 25. Similarly we find that the drift in simulated temperatures of the Antarctic Continental Shelf Bottom Water is small enough to allow to link atmospheric decadal variability to oceanic temperature variations at the boundary to ice shelves around Antarctica.
We illustrate that the respective sensitivity experiment is suitable for a comparison to coarser resolution model versions and present a first impression in Fig. 25. Code and data availability. The circulation model code MOM4p1 is distributed by NOAA's Geophysical Fluid Dynamics Laboratory (http: //www.gfdl.noaa.gov/fms). We use the original code without applying any changes to it apart from very minor changes (≈ 10 lines of code)