Simulations and NCAR-CCSM model configurations

Three model runs using different forcing fields and configurations have been analysed in the Greenwich Meridian section in the Weddell Sea (Fig. 1). This section was chosen because several hydrographical transects have been carried out in the region (e.g. Fahrbach et al. Reference Fahrbach, Hoppema, Rohardt, Schröder and Wisotzki2004, Klatt et al. Reference Klatt, Fahrbach, Hoppema and Rohardt2005), which allows for a direct comparison between observations and model results. The first experiment (hereafter referred to as POP-INT) consists of the ocean component of the NCAR-CCSM forced by 43 years of Large & Yeager (Reference Large and Yeager2004) fluxes. The second and third experiments (hereafter referred to as POP-NOICE and POP-CORE1, respectively) are forced by a repeat annual cycle obtained from the Large & Yeager (Reference Large and Yeager2004) forcing developed for coupled ocean-ice models. The POP-CORE1 simulation is the same run used in the Coordinated Ocean Ice Reference Experiments (COREs), discussed in Griffies et al. (Reference Griffies, Biastoch, Boning, Bryan, Danabasoglu, Chassignet, England, Gerdes, Haak, Hallberg, Hazeleger, Jungclaus, Large, Madec, Pirani, Samuels, Scheinert, Sen Gupta, Severijns, Simmons, Treguier, Winton, Yeager and Yin2009). It consists of the ocean component of the NCAR-CCSM coupled to the Community Sea-ice Model (CSIM, Briegleb et al. Reference Briegleb, Bitz, Hunke, Lipscomb, Holland, Schramm and Moritz2004). The CSIM serves as the sea-ice component of CCSM. It is the result of a community effort to develop a portable, efficient sea-ice model that can be run coupled to a global climate model or uncoupled as a stand-alone ice model. It is a dynamic-thermodynamic model that includes a sub-grid scale ice thickness distribution, energy conserving thermodynamics, and elastic viscous plastic (EVP) dynamics. POP-NOICE is not coupled to the CSIM. The runs are summarized in Table I.

Table I. Analysed Model Outputs.

The ocean component of the NCAR-CCSM solves the primitive equations in general orthogonal coordinates in the horizontal with the hydrostatic and Boussinesq approximations. Although this study focuses on the Weddell Sea, the model domain is global. The grid uses spherical coordinates in the Southern Hemisphere, but in the Northern Hemisphere the pole is displaced into Greenland at 80°N, 40°W. The horizontal grid has 320 zonal and 384 meridional grid points, and the resolution is uniform zonally but not meridionally. In the Southern Hemisphere, the meridional resolution is 0.27° at the equator, gradually increasing to 0.54° south of 33°S, and is constant at higher latitudes. There are 40 levels in the vertical, whose thickness increases monotonically from 10 m near the surface to 250 m in the deep ocean. The minimum depth is 30 m, and the maximum depth is 5.5 km. The surface fresh water flux is converted into an implied salt flux using constant reference salinity. Therefore, although the sea surface height varies locally, the ocean volume remains fixed. The time step used is one hour. The horizontal viscosity is a Laplacian operator that is anisotropic following the formulation of Smith & McWilliams (Reference Smith and McWilliams2003), and uses different coefficients in the east–west and north–south directions. Both coefficients vary in space and time, depending on the local rates of shear and strain, and the minimum background horizontal viscosity is 1000 m² s^-1. The vertical mixing scheme is the KPP scheme of Large et al. (Reference Large, McWilliams and Doney1994). Further details on the ocean component of the CCSM3 can be found in Smith & Gent (Reference Smith and Gent2004).

The model forcing follows Large & Yeager (Reference Large and Yeager2004) and is also described in Griffies et al. (Reference Griffies, Biastoch, Boning, Bryan, Danabasoglu, Chassignet, England, Gerdes, Haak, Hallberg, Hazeleger, Jungclaus, Large, Madec, Pirani, Samuels, Scheinert, Sen Gupta, Severijns, Simmons, Treguier, Winton, Yeager and Yin2009). The interannual forcing is composed of a 40 yr cycle (1958–1997) of 6-hourly winds near surface temperature and humidity from the National Center for Environmental Prediction (NCEP)–NCAR reanalysis. The model starts from a resting state of observed temperature and salinity distributions and is integrated until the solution has lost all memory of the initial conditions (the spin-up is 500 years). The model is then run for four 43 yr forcing cycles. The analysis is restricted to the fourth cycle. The repeat annual cycle (Normal Year Forcing, NYF) consists of a normal year constructed to retain synoptic variability (i.e. Yeager & Large Reference Yeager and Large2007). A detailed description of the NCAR-CCSM model configuration and biases can be found in the June 2006 special issue of the Journal of Climate.

Optimum Multiparameter Analysis (OMP)

We have chosen to use the Optimum Multiparameter (OMP) technique to trace Weddell Sea deep water masses in this study. This method was introduced by Tomczak (Reference Tomczak1981) as an extension of classical water mass analysis through temperature-salinity (T/S) diagrams (Mamayev Reference Mamayev1975). Mackas et al. (Reference Mackas, Denman and Bennett1987) and Tomczak & Large (Reference Tomczak and Large1989) significantly improved the multiparameter analysis, showing the potential of the method to analyse mixing conditions and water mass distribution in both coastal and oceanic regions.

Some definitions are needed to fully understand the physical basis of the multiparameter analysis (Tomczak & Large Reference Tomczak and Large1989). First, a water mass is defined as a physical entity that occupies a finite volume in space (Tomczak & Large Reference Tomczak and Large1989) or as a body of water with a common formation history with its source in specific parts of the ocean (Tomczak Reference Tomczak1999). Second, the water masses are defined by water types, which refer to a set of hydrographical parameters that describe the properties of the water masses, or by source water types, which are a set of hydrographical parameters that describe the properties of newly formed water masses (Tomczak & Godfrey Reference Tomczak and Godfrey1994).

The OMP analysis quantifies mixing between selected water masses (i.e. water type percentage contribution to the total mixture) by solving an over-determined system of linear mixing equations that use several conservative and semi-conservative parameters. In matrix form, this can be written as:

(1)

where G is the matrix of water types, x is the OMP result (i.e. fractions of each water type), d is the vector of observed data (or model outputs as addressed in this study), and r is the vector of residuals. The mixing equations are normalized to make parameters with different units comparable, and the equations are weighted to account for environmental variability and/or measurement uncertainty (see below). This system of equations (Eq. 1) is solved by minimizing the residuals (r) in a least squares sense, which yields the optimum combination of water type contributions.

The standard OMP analysis package (OMP2 2005) is used here to estimate the distribution and fractions of mixture of the Weddell Sea deep water masses (based on local water mass definitions - i.e. water types) in the NCAR-CCSM simulations described above. The limited number of available hydrographical parameters restricts the number of water masses used to run the OMP. From the model simulations considered here, only potential temperature (θ) and salinity are considered, which means that the number of water types that can be used to separate the water masses with OMP is limited to three. Thus, the following water masses are considered in this study: WDW, WSDW, and WSBW. Only two physical restrictions are imposed in the OMP analysis: the first is mass conservation, which means that the contribution of all water types must add up to 100%, and second, this contribution cannot be negative.

It is necessary to gauge the system of mixing equations to account for environmental variability and/or measurement accuracy. Weights are obtained from parameters variance (see Tomczak & Large Reference Tomczak and Large1989 for more details). Input parameters with higher weights will have greater influence than parameters of lower weights to construct the final water mass distribution. In this study, high applied weights are correlated with parameters that are further realistic represented in the ocean simulations. Resolving the contribution of the deep water mass or those water masses which have only a very small gradient within the parameters used is directly associated with the established weights. Therefore, it is necessary to restrict the analysis to a depth range where homogeneous weights may be applied (J. Karstensen, personal communication 2007), e.g. disregarding the variable and well-mixed surface layer and considering only deep waters for the OMP application (this study). Although it is relatively easy to trace the mixing of only three water masses through the T/S diagram, applying the OMP technique leads to refined results (M. Tomczak, personal communication 2006). A detailed description, advantages, and other applications of the OMP method can be obtained from Tomczak & Large (Reference Tomczak and Large1989), Karstensen & Tomczak (Reference Karstensen and Tomczak1997, Reference Karstensen and Tomczak1998), and Leffanue & Tomczak (Reference Leffanue and Tomczak2004).

The OMP technique is applied to the model outputs in two different approaches. The first considers water type indices from historical observed data, while the second uses indices calculated from model climatology (Table II). The observed water types used in the first approach are calculated using in situ data obtained from the National Oceanographic Data Center (NODC) historical hydrographical dataset. In this case, the weights (Table II) were obtained using the variance equation described in Tomczak & Large (Reference Tomczak and Large1989). For the second OMP application, the water type indices were adjusted to better represent model water masses, for which θ and salinity may differ somewhat from observed water masses characteristics. This was accomplished using modelled θ and salinity values found in selected depth intervals (see Table II). In this case, the parameter weights were calculated based on the variances of the water types and of the model θ/S outputs.

Table II. Water types, parameters and weights used for OMP input.

Weddell deep ocean representation in the NCAR-CCSM model

The ocean component of the NCAR-CCSM model has been extensively studied and validated against observations (e.g. Large et al. Reference Large, Danabasoglu, Doney and McWilliams1997, Gent et al. Reference Gent, Bryan, Danabasoglu, Doney, Holland, Large and McWilliams1998, Danabasoglu Reference Danabasoglu2004, Large & Danabasoglu Reference Large and Danabasoglu2006). According to these studies, observations and model results are in good agreement with respect to gyres and inter-basin exchanges, despite the generally too strong near-surface winds. Drake Passage (Fig. 1) transport is too strong but has little effect on SST and sea surface salinity. Doney & Hecht (Reference Doney and Hecht2002) note that the surface boundary conditions in permanent sea-ice covered regions are leading to inadequate formation of dense, cold, and relatively saline shelf waters, which is an important precursor of the dense source water masses that form AABW. The origins of the Southern Ocean water masses have been the subject of much investigation but their rates of formation and contributions to global ocean ventilation are still under debate.

Some of the densest water masses in the Southern Ocean are dependent on salinity/freshwater fluxes from the marine cryosphere (i.e. sea-ice, icebergs, and ice shelves). Thus, the inclusion of sea-ice related processes, as well as coupling a sea-ice model to general global circulation models is a desirable model development goal (e.g. Fichefet & Maqueda Reference Fichefet and Maqueda1997, Lipscomb & Hunke Reference Lipscomb and Hunke2004, Holland et al. Reference Holland, Bitz, Hunke, Lipscomb and Schramm2006, Huntley et al. Reference Huntley, Tabak and Suh2007). Climate models do not normally represent the fluxes from icebergs and ice shelves. Recently, advances in parameterizing ice shelf processes and representing the flow within ice shelf cavities have been discussed and included in both regional and global models (e.g. Timmermann et al. Reference Timmermann, Hellmer and Beckmann2002, Beckmann & Goosse Reference Beckmann and Goosse2003, Thoma et al. Reference Thoma, Grosfeld and Lange2006). For example, Thoma et al. (Reference Thoma, Grosfeld and Lange2006) showed that changes in ice shelf configurations have significant impacts on the water mass properties of the eastern Weddell Sea.

The integration period for the simulations present here may be shorter than the equilibrium time for the middle to deep oceans. However, this work focuses at model-model intercomparison so the integration time is sufficient to reveal differences between models and drift with respect to observations. Examinations of the integrated mean kinetic energy (KE), and the KE per layer for each of the simulations (not shown) reveal that the runs are dynamically stable. With respect to temperature and salinity, there is a fast initial adjustment with small drifts in the deeper oceans, but it should be noted that the present analysis relies on mean properties, not variability or changes in the thermohaline properties of the Southern Ocean.

As can be seen in the θ/S diagram for the Greenwich Meridian region (Fig. 2), neither simulation exactly represents the observed absolute θ/S values for the entire water column. Ocean only simulations (i.e. POP-INT and POP-NOICE runs) represent the WDW layer more realistically than do simulations with the interactive sea-ice model (i.e. POP-CORE1 run). The latter show the highest temperature values for the WDW layer. The layer that encloses the WSDW and WSBW is badly represented in all simulations. In general, while for POP-INT and POP-NOICE these waters are saltier and colder, including the interactive sea-ice model renders the WSDW and WSBW fresher and warmer. Biases with respect to the CORE-I simulation results are discussed in Griffies et al. (Reference Griffies, Biastoch, Boning, Bryan, Danabasoglu, Chassignet, England, Gerdes, Haak, Hallberg, Hazeleger, Jungclaus, Large, Madec, Pirani, Samuels, Scheinert, Sen Gupta, Severijns, Simmons, Treguier, Winton, Yeager and Yin2009), where the most notable are found exactly in the Southern Ocean, suggesting the importance of examining ice-ocean interactions and feedbacks. The sections of θ and salinity and their differences with respect to observations can be seen in Figs 3 & 4.

Fig. 2. a. Full range potential temperature-salinity diagram for POP-INT (circles), POP-NOICE (squares) and POP-CORE1 (triangles) simulations and the NODC/NOAA dataset (dots) at the Greenwich Meridian for the years 1984, 1986, 1992, 1996, and 1998, which were used to obtain the observed mean water mass distribution in Fig. 7. The rectangles limit water masses based on Robertson et al. (Reference Robertson, Visbeck, Gordon and Fahrbach2002). b. Same as (a) but zoomed in on the salinity of deep water masses. Solid lines indicate σ₀ and dashed lines σ₂.

Fig. 3. a–c. Potential temperature (θ) distribution, and d–f. θ-difference (simulated minus observation) for POP-INT, POP-NOICE, and POP-CORE1 simulations, respectively, in the Greenwich Meridian section for depths >500 m.

Fig. 4. Same as Fig. 3, but for salinity field.

The θ section of each simulation (Fig. 3a–c) clearly shows the WDW core in the intermediate layer up to 1500 m despite small differences in the absolute values of θ. However, significant differences exist in the deep ocean representations. Both the POP-INT and POP-NOICE simulation results are colder (~0.3°C) than those with the interactive sea-ice model. The θ-difference (Fig. 3d–f) shows the simulated WDW layer (i.e. 500–1500 m) with temperature cores in excess of 0.5, 0.3, and 0.5°C warmer for the POP-INT, POP-NOICE, and POP-CORE1 runs, respectively. Considering the layer between 1500 and 2000 m, which is the mixing zone between WDW and WSDW, the simulated temperature is still warmer with respect to observations. For this depth interval, the ocean-only runs shows that θ varies between 0.1 and 0.4°C for POP-INT (Fig. 3a), and -0.1 and 0.2°C for POP-NOICE (Fig. 3b), which is 0.2 to 0.5°C higher than the observed values for this layer (Fig. 3d & e). POP-CORE1 simulations show a pattern similar to that described for POP-INT. Between 2000 and 4000 m (i.e. the WSDW layer), distinct patterns are evident for each simulation. For POP-INT (Fig. 3d), this layer displays temperatures approximately 0.1 to 0.3°C warmer than the observations for the upper levels, while POP-NOICE simulations show temperatures in general 0.1 to 0.2°C lower (Fig. 3e). On the other hand, for the POP-CORE1 results, the WSDW layer is approximately 0.2°C warmer (Fig. 3f). Below 4000 m, the ocean-only runs yield temperatures 0.1 to 0.2°C colder (Fig. 3d & e) than POP-CORE1 (Fig. 3f).

The POP-CORE1 run simulates salinity better for the Greenwich section (Fig. 4c) than the ocean-only runs (Fig. 4a & b). The WDW core with salinity 34.7 is displaced upward and salinities around 34.65 appear near the bottom, better characterizing this region (Fig. 4c), but there are still differences. The highest reported salinity difference (i.e. 0.015 to 0.03 saltier; Fig. 4d–f) up to 1500 m shows that the WDW layer is saltier for all simulations than the observed data. Conversely, the deep ocean (between 2000 and 4000 m) shows salinity values slightly saltier than the observations, but without differentiating the WSDW and WSBW layers for the POP-INT and POP-NOICE simulations (Fig. 4d & e). The POP-CORE1 simulation produces a relatively fresh core around 3000 m (Fig. 4f), with salinity values near observed values, which may indicate the first step to better representing the WSBW layer. The salinity values that differentiate the water masses considered in this study are very similar, making it difficult to characterize the water masses in the water column. Further discussion of this matter is presented in the next sections.

Representation of the Weddell Sea Deep Water masses from observations

The mean water mass distribution from observations obtained by Kerr (Reference Kerr2006) is shown in Fig. 5. This mean distribution was obtained from the OMP analysis for each repeat section sampled on the Greenwich line, which considered in the years 1984, 1986, 1992, 1996, and 1998 (for more details see Kerr Reference Kerr2006). Klatt et al. (Reference Klatt, Fahrbach, Hoppema and Rohardt2005) and Fahrbach et al. (Reference Fahrbach, Hoppema, Rohardt, Schröder and Wisotzki2004) describe the characteristic properties and the variability of this area. The water mass structure of the Greenwich Meridian section is well captured, showing the WDW occurring in the intermediate layer down to ~2000 m representing more than 30% of the contribution to the water column (Fig. 5a). The WSDW core contribution is above 50%. It is distinct between 1000–3500 m (Fig. 5b), and the WSBW is present below 3000 m (> 50%) with a greater contribution (> 70%) below 4000 m in the region north of the Maud Rise (Fig. 5c).

Fig. 5. Observed (OBS) mean water mass distribution (% of mixing) from the NODC/NOAA dataset at the Greenwich Meridian for the years 1984, 1986, 1992, 1996, and 1998, using the water types based on NODC data (see Table II). a. Warm Deep Water, b. Weddell Sea Deep Water, and c. Weddell Sea Bottom Water. The seamount is the Maud Rise.

Representation of the Weddell Sea Deep Water masses in the NCAR-CCSM

There is a growing need to investigate the performance of the Southern Ocean water mass properties in ocean general circulation models (OGCMs). Recently, Connolley & Bracegirdle (Reference Connolley and Bracegirdle2007) carried out an Antarctic assessment of several IPCC AR4 coupled models and showed that the NCAR-CCSM did not score high when compared with other models. In fact, climate models in general do a poor job of representing water masses, in particular with respect to the Southern Ocean. One reason may be that climate models do not accurately represent the ice-ocean interactions necessary for proper water mass formation, ventilation, and spreading. The lack of vertical resolution could also be a problem in correctly representing the Antarctic water masses.

The present study addresses this issue by performing an OMP analysis on Weddell Deep Water Masses represented in the ocean component of the NCAR-CCSM model. It should be noted that the concept of the water type used to fingerprint water masses is an artificial quantitative analysis construction that does not occupy any volume in space (Tomczak & Large Reference Tomczak and Large1989). The OMP analysis simply allocates the water mass that is closer in parameter space to the water type definition used regarding the hydrographical properties.

Two approaches are considered when using the OMP analysis with the NCAR-CCSM model outputs. First, water types are obtained from observed data (i.e. calculated from the NODC historical dataset). Next, taking into account the overall differences between model-simulated and observed water mass characteristics, the OMP technique is applied using redefined water types. The redefined water types are adjusted for each specific simulation. This approach embraces more precisely the deep water masses in each different run. The root-mean-square (RMS) differences between the water mass representations obtained from model results and NODC data are calculated to assess the OMP outputs.

Water mass representation with observed water types

OMP results obtained from the model simulations show that the ocean-only runs (i.e. POP-INT and POP-NOICE) do not adequately represent the water mass distribution in the Weddell Sea (Fig. 6). The layer thickness of WDW and WSBW is overestimated in both simulations. As can be seen in the POP-INT simulation results (Fig. 6a), the WDW contribution reaches high values in the mixture (between 50–70%) at levels below 2000 m, which is an unrealistic depth for this water mass (see Fig. 5a). The same is true for the WDW in the POP-NOICE simulation results (Fig. 6b); i.e. a high water mass contribution (between 50–70%) at deeper levels. Using observed water type indices yields an unrealistic WDW distribution and incorrect mixture fractions at the intermediate water column for both ocean-only runs.

Fig. 6. Dense water mass distribution (% of mixing) at the Greenwich Meridian section obtained from POP-INT, POP-NOICE, and POP-CORE1 simulations, as indicated in the first box, using the water type index based on NODC data (see Table II). Each column indicates specifics water masses: a–c. Warm Deep Water, d–f. Weddell Sea Deep Water, and g–i. Weddell Sea Bottom Water. The seamount is the Maud Rise.

As both WDW and WSBW are overestimated in the POP-INT and POP-NOICE runs, it is not surprising that the OMP analysis yields an unrealistic distribution for the WSDW (Fig. 6d & e). In these two simulations, the WSDW contribution to the total mixture was below 30% (Fig. 6d & e). Tracing WSDW in the water column becomes difficult because of the occurrence of this water mass throughout a wide depth range (i.e. approximately 3000 m) and a relatively large temperature range (for example, between 0 and -0.7°C for the WSDW). These facts may indicate that more than one water type should be used to mark that particular water mass in the ocean as suggested in the OMP technical report for some specific cases (OMP2 2005). However, the unrealistic separation of the Weddell Sea water masses presented here is probably associated with: i) the choice of the water type salinity index (Table II), considering the closer salinity range in the deep layers of the Weddell Sea (Fig. 4), and ii) the use of equal weights applied for θ and salinity in this approach, which does not account for the fact that the temperature field is better represented than the salinity field (with respect to the observations). WSBW is also overestimated in POP-INT and POP-NOICE (Fig. 6g & h). The expected outcome would be WSBW predominant (> 70%) below 4000 m, but in the simulation WSBW is displaced upwards. The OMP results show contributions > 70% below 3500 m (Fig. 6g) and below 3000 m (Fig. 6h) for POP-INT and POP-NOICE, respectively.

The POP-CORE1 simulation yields a somewhat more realistic water mass distribution than other runs when compared with observations (e.g. Fahrbach et al. Reference Fahrbach, Hoppema, Rohardt, Schröder and Wisotzki2004, Kerr et al. Reference Kerr, Mata and Garcia2005), as illustrated by the smallest RMS difference (Fig. 7). This is a significant result that indicates the importance of including a sea-ice model to improve regional representation of dense water masses in ocean models. The improvement is probably due to the fact that the simulated salinity values are closer to the observed. WDW occurs down to a depth of 2000 m with contributions above 50% in the model (Fig. 6c). The WSDW occupies the layer between 2000–4200 m with contributions higher than 50% (Fig. 6f), and the WSBW is confined to the deep oceanic basin north of Maud Rise (Fig. 6i). However, differences still occur when the simulated water mass distribution is qualitatively compared to the observed (Fig. 5).

Fig. 7. Root-mean-square difference (RMS) for dense water masses analysed with observed water types in the Greenwich Meridian section. a, b. Warm Deep Water (WDW), c, d. Weddell Sea Deep Water (WSDW), and e, f. Weddell Sea Bottom Water (WSBW). For all graphics, each simulation is represented as indicated in (b).

In summary, the interactive sea-ice model yields an overall vertical displacement of the simulated deep water masses relative to observations. Changes related to the contribution of each water mass are also noted. This reflects the lack of coastal processes representation in the model (such as coastal polynyas and local sea-ice formation). These govern the formation of Shelf Waters in the model configuration, which could lead to inaccurate deep water mass properties. As reported by Large & Danabasoglu (Reference Large and Danabasoglu2006), near-surface regions has not been well captured by simulations using the ocean component of the NCAR-CCSM model, and small systematic displacements will lead to large SST biases due to strong horizontal gradients.

We calculate RMS differences in order to quantify disparities among water mass representations. The greatest agreement between the numerical representations of WDW, WSDW, and WSBW is for the smallest implied normalized RMS differences (or observational uncertainty) between the OMP results. The run in which the sea-ice model is interactive has the smallest RMS (Fig. 7). The salinity results from POP-INT are not sensitive enough to allow separation between the WSDW and WSBW (they yield one water mass only) in this OMP analysis, which explains why its associated RMS difference with respect to the observations is less for the WSBW (Fig. 7e & f).

Salinity has a fundamental role in determining the density field in the polar regions. Thus, any salinity change will significantly impact water mass distribution. NCAR-CCSM ocean simulations do not reproduce the observed salinity patterns exactly, as demonstrated by the salinity differences shown in Fig. 4d–f. Salinity patterns are better simulated in the POP-CORE1 run, suggesting that it is necessary to couple with a sea-ice model to correctly simulate deep salinity values, and consequently, deep water mass characteristics. As it stands, salinity corrections are needed to identify specific deep water masses. However, this is not the ultimate solution for correctly representing the dense water mass characteristics (see discussion below). Including other cryospheric processes in the model is also necessary, as are improvements in the sea-ice parameterizations.

Water mass representation using water types adjusted to each individual simulation

Here, the θ/S structure for each experiment is examined separately (see Fig. 2). Water types and weights used to perform the OMP analysis are adjusted for each analysed simulation (see Table II). Unlike the previous results, an observation-based water type index is used for each water mass and the same weight was applied for both θ and salinity. In this approach, the OMP analysis is able to distinguish all the deep water masses at the Greenwich section for all simulations. Based on the results of the previous analysis, the POP-CORE1 simulation represents an improvement because, as discussed earlier, the salinity values are closer to the observations and differ for all water masses used (see Table II). This allowed a refined separation of WSDW and WSBW in deeper layers. Compared to the observed water mass distribution, the representation of all water masses is improved when the interannual forcing is used (POP-INT). Note that this simulation does not incorporate the interactive sea-ice model, but accounts for the intra-seasonal forcing corrected for sea-ice.

Visually, the WDW base limit with a 30% contribution (Fig. 8a–c) fits well with the distribution obtained from observed data (Fig. 5a). However, the WDW contribution to the total mixture varies slightly (~20–30%) between the simulations and the observational results, as indicated by low RMS differences (~20%). Although the RMS differences are slightly higher for the POP-NOICE run, they also show a realistic representation for WDW. In the water column, the RMS differences (not shown) show that the layer near the base limit of the WDW (~1000–2000 m) is better represented than that closer to the surface (~500–1000 m).

It becomes possible to distinguish deep waters when water mass indices based on the simulation results for all model runs are used (Fig. 8d–f). However, the thickness of the WSDW layer is underestimated compared to observations (Fig. 5b). The upper level of the WSDW layer is not adequately represented, probably because only one water type was used, as described earlier. Nevertheless, the RMS differences show the same behaviour for the WSDW as for the WDW along the section. In this case, the results of the POP-INT and POP-NOICE runs are much closer to the observed pattern (differing ~15%) than that of the POP-CORE1 run (~25%). At depth for the POP-INT and POP-NOICE runs, the WSDW layer is better represented between 1500 and 3000 m than the top layer (above 1500 m). This can be seen relative to the distribution with contributions below 30% in this depth interval (Fig. 8d & e). Conversely, for the POP-CORE1 simulation, the WSDW core is displaced downward (Fig. 8f), implying that the depth interval between 1000 and 2000 m is weakly represented (Fig. 8f). Considering the bottom layer, the WSBW is well represented in all simulations at the Greenwich Meridian (Fig. 8g–i) despite the increasing RMS differences (~20% to 40%) along the entire section and depth. The pattern in RMS differences can be associated with the weak representation of the near-bottom layer properties in the model simulations. Along the latitude regions where the WSBW is expected to fill the water column (i.e. region north of Maud Rise), the RMS differences are constant, around 20%.

Fig. 8. Dense water mass distribution (% of mixing) at the Greenwich Meridian section obtained from POP-INT, POP-NOICE, and POP-CORE1 simulations, as indicated in the first box, using the water type index based on each model output (see Table II). Each column indicates specifics water masses: a–c. Warm Deep Water, d–f. Weddell Sea Deep Water, and g–i. Weddell Sea Bottom Water. The seamount is the Maud Rise.

The upward shift of the RMS difference curves at high, near-coastal latitudes (between 66°S and 70°S; Fig. 7) is an indication that near-coastal waters are underrepresented, relative to open ocean waters in the OMP analysis. Considering the bottom layer (WSBW), the missing or incorrect representation of the near-bottom layers suggests that the physical processes that govern the formation and spreading of dense water also requires attention.

It is important to note that the water mass distributions discussed in this section were obtained using different water type indices for each model simulation (Table II). Thus, water mass distributions should not be compared across models. Consequently, the POP-CORE1 water mass distribution cannot be considered the worst simulation to represent the deep water masses. Nevertheless, the previous analyses and the model observation property comparisons clearly show that the inclusion of a sea-ice model in the NCAR-CCSM ocean model leads to a better simulation of the deep ocean.

The work presented here is somewhat sensitive to the choices of water masses and weights, which implies that results may vary depending on the set of OMP parameters used (i.e. water types) and the property weights applied. This demonstrates the challenges inherent not only in correct simulation of water masses in the Southern Ocean but also in evaluation of model results. These models lack the resolution and the representation of some smaller-scale physical processes needed to correctly represent the temperature and salinity range found over the continental shelves of Antarctica. The coarse representation likely also contributes to erroneous representation of several deep ocean characteristics in those models.