This study focuses on the uptake of a passive idealized tracer in the Northern Hemisphere oceans from two coupled ocean–atmosphere simulations: a standard horizontal diffusion case and the second case including the Gent and McWilliams (GM) eddy mixing parameterization. The results are compared with tracer uptake in stand-alone synchronous and asynchronous ocean simulations for the same cases. The GM set of integrations shows tracer penetration reduced from the standard set in all water mass formation regions. There is a strong similarity in the tracer distributions in the stand-alone ocean simulations in both the standard and GM cases. Changes in the velocity fields between the stand-alone ocean and coupled simulations explain many of the differences in the modeled tracer concentrations.
There is a particular focus in the study on the dynamics of modeled water mass formation for North Pacific Intermediate Water, Labrador Sea Water, northeast Atlantic mode water, and North Atlantic Deep Water. The model representation of these water masses is compared with observational data of passive tracers, and gives mixed results as the model water masses are on lighter density surfaces than the real ocean though the timing of the advance of the tracer plume within the water masses appears to be realistically modeled. For the coupled simulations, the North Atlantic and North Pacific Oceans have interdecadal signals that alter the circulation and hence tracer patterns. Some issues arising from the interdecadal signal are discussed in relation to tracer distribution on density surfaces within the ocean.
The last two decades have seen increasing measurements of chemical tracers in the ocean. The first group is “biologically active tracers” such as oxygen and nutrients (Fiaderio and Craig 1978; Wilson and Wallace 1990; Reid 1986); these have been routinely measured in water samples for decades, but have the drawback that they participate in chemical reactions and can be depleted through biological utilization. They can be replenished in the case of nutrients by upwelling from deeper waters, though the oxygen signal can differentiate old from freshly ventilated water. A second group of chemical tracers that is not biologically active is “passive tracers” such as CFCs, C14, tritium/helium (Warner and Weiss 1992; Fine 1993; Fiaderio 1982; Jenkins 1988); these are either globally conserved or have well-known chemical decay properties.
This latter group of passive tracers is often used by physical oceanographers to constrain some of the parameters in oceanographic models (England and Hirst 1997; Duffy et al. 1995, 1997) and to check if model processes such as mixing (horizontal, vertical, isopycnal, isopycnal thickness) have realistic magnitudes. Use of passive tracers can provide insights into whether the ocean model correctly represents the formation regions and volume of individual water masses because the tracer distribution is independent of the temperature/salinity (T/S) properties of the water masses. Inert CFCs are the closest match to the type of passive tracer used in this modeling study; in particular they are good tracers of regions of the ocean that have recently been ventilated in the eastern North Atlantic (Doney et al. 1997), deep water in the Greenland Sea (Bonisch et al. 1997), Antarctic Intermediate Water (Fine 1993; Warner and Weiss 1992), and Antarctic Bottom Water (England et al. 1994) since the exchange time between the ocean and atmosphere for these gases is relatively fast.
Another recent innovation in ocean models is the inclusion of the Gent and McWilliams (1990, hereafter GM) parameterization for subgrid-scale eddy mixing on isopycnal surfaces. This parameterization has had considerable impact on the density stratification in the global ocean, particularly in the more weakly stratified Southern Ocean; it considerably reduces the area and depth of convection and alters the rate of meridional overturning (Danabasoglu and McWilliams 1995; Hirst and McDougall 1996). The tracer experiments described in section 2 consider coupled ocean–atmosphere simulations with two mixing parameterizations: a standard configuration with mixing along isopycnals plus a background horizontal diffusivity and the GM eddy diffusivity with zero background horizontal diffusivity.
A further point to consider is that the rate of tracer uptake in coupled models may differ from ocean-only models since the higher frequency forcing under coupled conditions alters the location of convection compared with simulations with fixed boundary conditions (Power 1995). The coupled model used in these experiments uses monthly flux corrections that have an imprint of the patterns of convection obtained in the stand-alone ocean runs, and which adds to the complex boundary conditions seen under variable forcing in the coupled model with both a broader spectrum of frequencies and a wider range of amplitudes.
The purpose of this paper is to discuss the differences in ocean tracer uptake between the stand-alone ocean model using either asynchronous or synchronous time stepping (Bryan 1984) and the coupled model, for the cases with standard horizontal diffusion and GM eddy mixing parameterizations. The inclusion of the coupled model in this comparison gives an ocean that may be more representative of the real world through the inclusion of surface forcing across the full spectrum of frequencies. This allows us to evaluate the coupled models passive tracer uptake, which is a simple analogue for heat subduction in coupled climate change experiments, without accounting for the way the heat uptake will alter the density structure.
This paper builds on the existing work with the CSIRO ocean model discussed in two papers by Hirst and McDougall (1996) and Hirst et al. (2000). Hirst and McDougall described in detail how the addition of the GM eddy mixing parameterization alters the salinity, temperature profiles, and T/S relationships in the stand-alone ocean model. Hirst et al. (2000) discussed the meridional overturning, meridional heat transport, and convection patterns in the ocean stand-alone cases for the standard and GM simulations and discussed how these patterns were maintained in the coupled model simulations.
The present paper discusses tracers in the Northern Hemisphere ocean basins and focuses on water masses formed in the Northern Hemisphere. The uptake of tracers by water masses originating in the Southern Hemisphere and also in the Indian Ocean will be discussed in a second paper. Water masses such as North Atlantic Deep Water and Antarctic Intermediate Water, which are formed in one hemisphere but have a considerable role in the other, will be briefly mentioned in the paper focused on the other hemisphere. Section 2 briefly outlines the model description and the experiments undertaken. Section 3 describes the global zonal mean results and considers basinwide tracer distributions, while section 4 discusses the implications of the results and compares the results with observations in key regions. Section 5 presents the conclusions of this study.
2. Model experiments
The coupled model used in these experiments is the CSIRO Mark 2 coupled model (Gordon and O'Farrell 1997), which consists of a 9-level atmospheric model at R21 spectral resolution coupled to a dynamic ice model and ocean model at a horizontal resolution of 3.2° latitude and 5.6° longitude, which matches the atmospheric spectral grid. The GFDL ocean model is used (Cox 1984) for two different vertical resolutions. The standard model has 12 vertical levels with resolution of 25 m in the mixed layer and 900 m in the deep ocean. The second set of simulations includes the GM eddy mixing parameterization and no background horizontal diffusion; for numerical reasons it is discretized on 21 vertical levels (Hirst et al. 1996). There is a correspondence between the two vertical resolutions, in that the extra levels have been used to double the resolution at the deeper levels, so one can compare tracer distributions at a given depth by combining the values from the equivalent pairs in the 21-level model; the topography is identical in both cases (Hirst et al. 1996). The passive tracer input function is defined as 100% concentration in the surface layer and it is replenished to this value every time step. This configuration of tracer input is ideal for examining ventilation of the deep ocean, but is less useful as a tracer in the upper thermocline because the upper ocean (above 300 m) is nearly saturated in the mid–high latitudes after 10 years.
Six model cases will be discussed in this paper. In all the experiments the passive tracer has been tracked for 50 years with monthly averaged tracer, temperature, salinity, and velocity fields stored for each year. The tracer experiments use the CSIRO coupled model, with both standard horizontal diffusion and GM eddy mixing configurations. The additional four experiments were undertaken with stand-alone ocean simulations in both standard and GM configurations, but in one pair the ocean model has the same time step for tracer and velocity fields (synchronous run) and in the other pair the time step on the tracers is much longer than the velocity time step so as to speed the simulation of the ocean model to equilibrium (asynchronous run).
There are two reasons for making comparisons between the tracer distribution in synchronous and asynchronous ocean runs. The first compares the ventilation properties of the ocean between the two types of time steps. To spin up the ocean component of the CSIRO coupled model, the model was run for over 2000 years asynchronously and then for 700 years synchronously (Gordon and O'Farrell 1997). The long synchronous simulations were needed since it took a considerable time for the surface fluxes to reach a steady state in the regions of deep ocean ventilation in the high latitudes. A second reason is that tracer simulations such as those of England (1995) and Duffy et al. (1997), which calculate renewal rates of the ocean basins using age or radiocarbon tracers, need long equilibrium simulations. When testing such simulations for parameter sensitivity it is clear that only simulations with asynchronous time stepping can be considered. However, the synchronous simulation is more appropriate when comparing coupled model results where synchronous time steps were employed in the ocean component, as asynchronous time stepping distorts the speed of transient Rossby waves (Bryan 1984). The appendix provides details of the surface forcing fields used in the ocean-only experiments and choice of parameter settings for the ocean model.
3. Model results
a. Zonal mean tracer distributions
Figure 1 shows the zonal mean tracer distributions averaged across all ocean basins for all six cases after 50 years of simulation, while Fig. 2 shows the integrated volume of water affected by tracers for the same simulations. These figures show that, although there are small differences in the zonal uptake of tracer between the asynchronous and synchronous simulations, the major difference occurs with the uptake in the coupled simulations. This is caused by greater ventilation of the Southern Ocean under the high-frequency atmospheric forcing present in the coupled simulations.
The data are shown after 50 years of simulation, but the patterns are characteristic of the synchronous and asynchronous cases for the GM and standard simulations throughout the simulations, as the evolving patterns were checked at 10-yr intervals. The coupled model simulations, however, show that the relative height of tracer peaks in the zonal average volume integrated statistics do change as the simulation evolves, though the overall ratio of the volume of tracer uptake in the Southern Hemisphere, compared to the volume in the Northern Hemisphere (McDougall et al. 1996), is steady throughout the simulation. In the standard coupled simulation, the volume of water affected by tracer matches very closely the tracer distribution in the synchronous ocean model stand-alone run throughout the Northern Hemisphere and to the north of 40°S in the Southern Hemisphere (Fig. 2). The similarity in the Northern Hemisphere tracer uptake after 50 years is also clear. The exception is the region around 60°S where the synchronous stand-alone run has 8% greater volume of tracer at depth than the asynchronous run, and in the coupled run the volume of tracer taken up by the ocean is 30% greater than the asynchronous simulation. A time series of tracer volume in the North Atlantic region at depths 1250 to 1900 m shows that the volume of the water mass affected by the tracer in the asynchronous run is within 2% of the volume in the synchronous simulation but is 10% less than that affected in the coupled run after 50 years of simulation.
In contrast to the standard set of simulations (Fig. 1) the synchronous and asynchronous cases in the GM set have almost identical uptake of tracer by volume in both hemispheres (Fig. 2). The coupled model tracer data again follow a similar pattern with only minor differences in tracer uptake at 30°–35°N. In the Southern Hemisphere the synchronous and asynchronous runs again have very similar tracer uptake, while the coupled model shows about 75% greater volume of water affected by tracer at 60°S; this is a far greater increase than the 30% uptake seen in the standard set of simulations even though the overall volume of tracer uptake in the standard coupled run is double that of the GM coupled run.
b. Regional tracer evolution in the North Pacific
The volume of tracer uptake in the Northern Hemisphere has been subdivided into sections for the North Atlantic and North Pacific for the each of the six simulations. Time series for the tracer volume in the Pacific basin averaged between 20° and 60°N are shown in Fig. 3b. Almost all tracer is confined to the region above 800 m, with tracer below 800 m occurring only in the Sea of Okhotsk and the Sea of Japan. Intrusion of tracer into the North Pacific gyre at depth occurs only in the standard coupled run, but not in any of the simulations with the GM scheme. In the time series for the standard coupled run (Fig. 3b), it is seen that there is a rapid increase in the tracer volume after 20 years at this 800-m level; this is because a new convective source occurs to the east of Kamkatcha peninsula at 55°N, which is subsequently advected into the northern part of the North Pacific gyre. In the equatorial Pacific, there is an intrusion of tracer between 800 and 1250 m in all three standard simulations at around 160°W. The tracer represents mode water formed convectively in the Tasman Sea and south of Australia. Mixing of Southern Ocean Mode Water into the North Pacific gyre occurs at depths from 470 to 800 m (Fig. 4). This is below the main thermocline, which occurs at 270 to 470 m in both sets of simulations. Slightly greater volumes of tracer have been injected into this depth interval in the standard runs, but this is a reflection of the production levels of Southern Ocean Mode Water in the simulations.
The difference in ocean stratification between the GM and standard simulations causes the difference in ocean response between the sets of simulations. Within the interval 470 to 800 m (Fig. 4) there is considerably more ventilation in the standard runs than in the GM simulations. Tracer at the center of the North Pacific gyre is at 25%–30% concentration in the GM ocean stand-alone runs, while in the coupled run the tracer is ventilated at this depth to 35% concentration due to extra ventilation from a source in the north of the region off Kamkatcha peninsula. In the standard case, the asynchronous run has 45% tracer concentration at the center of the gyre, while the synchronous run has a maximum concentration of 50%; the coupled simulation has higher tracer concentrations again of 55%–60% in the center of the North Pacific gyre. The synchronous and coupled model standard cases have higher velocities (2%–3%) in the North Pacific gyre circulation compared to the asynchronous simulation, which explains the slightly higher central tracer concentrations. Both ocean stand-alone cases have tracer concentrations of 50% near 55°N off the Kamkatcha peninsula, while the GM stand-alone cases had signals only barely above 10%, which represents the background signal derived from diffusion of the tracer. The standard coupled run has a major source of tracer in the polar gyre, which ventilates a region from 45°–60°N, 150°–170°E with values greater than 60% concentration.
The details of the tracer source off Kamkatcha Peninsula will be examined using tracer concentrations projected onto late-winter density surfaces. On density surfaces, the major differences between and within the sets of simulations is in the subduction of tracer in the region north of 40°N. The main gyre is ventilated on density surfaces <25 σt units, with water subducted along the axis of the return circulation; at intermediate densities there is a tendency for water to be subducted from the northeastern sector in the Gulf of Alaska, while surface densities greater than 26 σt units only occur in the Sea of Japan, Sea of Okhotsk, and in the Bering Sea off the Kamkatcha Peninsula. There are some differences in the tracer distributions on particular surfaces when comparing the coupled integrations to the ocean-only runs. The tracer distributions are similar in regions that allow ventilation of the main gyre but differ in the higher-latitude source regions. Figure 5 shows the distributions on 26 σt for all six simulations; this surface was chosen as it represents an upper bound on water that might be classified in the model as North Pacific Intermediate Water (NPIW). The GM coupled model is almost fully ventilated in the Bering Sea on this 26-σt surface, while the same surface outcrops farther south in the Gulf of Alaska in the GM stand-alone cases. This pattern in the GM coupled model, where the density surfaces outcrop at the ocean surface farther north than in the asynchronous/synchronous run, occurs across the density surfaces in the 25 to 26 σt range (not shown). However, in the standard coupled case (Fig. 5), the coupled model density surfaces outcrop farther south than in the standard ocean stand-alone cases. Unlike the model results in Fig. 4, on level surfaces it is clear that the coupled GM case is ventilated from the Kamkatcha region; however Fig. 5 indicates that ventilation is also taking place in the Sea of Okhotsk
The reason for the stronger concentrations on density surfaces in the stand-alone GM runs is that the surface densities are greater in these simulations. Hence the same surface is higher in the water column in the GM stand-alone cases and thus captures more of the tracer subducted into the main gyre.
The results from the stand-alone ocean and coupled model for the standard set of simulations on density surfaces show a number of differences. While there is more tracer present in the main gyre on any given density surface in the range from 24 to 26 σt units for the stand-alone cases compared to the coupled case, the reason is different from that seen in the GM simulation. The volume of tracer is greater in the stand-alone cases as the density surfaces intersect the ocean surface farther north than in the standard coupled model. However, the tracer maximum has advected farther in the coupled simulation than in the synchronous/asynchronous stand-alone cases with speeds 0.3 cm s−1 greater in the coupled simulation on density surfaces in the 25 to 26 σt unit range.
In the GM coupled model, the density of the ocean surface is 0.3–0.5 σt units lighter in the northern sector of the Pacific compared to the stand-alone case, while in the midlatitudes there is a smaller increase in density in the western sector where the only deep midwinter mixed layers in the North Pacific occur. The density changes are the result of both salinity and temperature changes across the region. In the standard coupled run the density generally increases in the northern and western sectors compared to the stand-alone case, while in the GM models the density increase is found only in the western sector of the coupled model. As a consequence of this surface density change, the tracer surfaces intersect the surface farther north in the standard ocean-only simulations compared to the coupled simulation, the reverse of the situation in the GM models.
It follows that the sharp contrast between the tracer results in the two coupled runs is based on the different location of the density surface. The synchronous and asynchronous simulations have a similar pattern in both GM and standard ocean stand-alone cases; the surface density field is set by the relaxation condition on temperature and salinity, so the density surfaces will outcrop at similar latitudes. Once the isopycnals are subsurface, the GM mixing scheme does not show larger differences in the tracer concentrations in the stand-alone cases. Concentration differences are less than 5% in most of the gyre region in the 25–26 σt density interval; differences between 5% and 10% occur along the Gulf of Alaska and in the equatorial region.
Figure 6 shows the tracer on the 26.5-σt surface for all six simulations. On this surface, the concentration plume near Kamkatcha is weak but still present in the GM coupled model, for although the last surface that was directly ventilated in the Bering Sea for the GM coupled model was 26.15 σt, vertical diffusion of tracer permits a weak signal of the tracer pattern to project onto denser isopycnal surfaces. In both sets of ocean stand-alone cases the plume is stronger, especially close to the Kamkatcha source with the most recently ventilated isopycnal from the surface at 26.4 σt. In the standard coupled model, Fig. 6 shows that in winter the 26.5-σt surface still outcrops in the Bering Sea away from the Kamkatcha peninsula. The Sea of Okhotsk is also not fully ventilated, as it was in the other five cases, with the last surface to outcrop in the standard coupled model occurring on the 26.75-σt isopycnal.
The analysis of the level model results suggested that the volume of the intermediate water mass was greater in the standard coupled simulation than the GM coupled case and reached a deeper ocean model level. This is because the tracer plume is at a different density in the standard coupled simulation, though the volumes are not necessarily greater than in the GM model. However, any volume calculation will be biased due to the larger grid size at depth in the level model. As the model in density space shows less difference than in level space between stand-alone and coupled simulations, it is unlikely that the addition of sea ice production in the peripheral seas of the North Pacific is driving additional NPIW in the coupled runs. Ice does not form in the western Bering Sea directly in the region of downwelling of the densest NPIW, but there is still considerable interannual variability in surface heat and freshwater fluxes in this location. The fluxes alter the large-scale density regime of the North Pacific, and hence the depths of individual isopycnals on longer timescales, so the pathways of NPIW into the interior of the gyre may be altered by interdecadal variability.
The model displays considerable decadal variability in the North Pacific with a strong node of variability in the region near the Kamkatcha Peninsula where the model NPIW originates (Vimont and Hirst 1999, personal communication). Analysis of the surface density after 30 years of the coupled simulations shows less difference between the GM and standard cases and this is reflected in the tracer concentrations projected onto density surfaces. Hence the difference documented above for the GM and standard cases is aliased by taking a time slice after 50 years.
c. Description of tracer in the North Atlantic in the standard simulations
In this section the tracer distribution between 20° and 70°N in the North Atlantic below a depth of 270 m will be considered in more detail. Because the tracer at depth in the North Atlantic is much greater than that of the North Pacific, it dominates the zonal values for the Northern Hemisphere (Fig. 1). Tracer sections from the standard coupled simulation show a slightly higher uptake than the synchronous ocean stand-alone case, except for a zonal maximum at 40°N, which was about 2% higher in the zonal average in the asynchronous case (Fig. 2). In the GM simulations the standard ocean stand-alone cases are very similar, with the coupled GM model having a 10% greater volume uptake at 40°N and 2%–3% greater at other latitudes (Fig. 2). The analysis next focuses on the tracer distribution and its interpretation in the North Atlantic for the standard set of simulations.
Above 470 m, the ocean is well ventilated with high concentrations across the domain except beneath the Gulf Stream. Below 470 m, the waters with the highest concentrations of tracer after 50 years of simulation are confined to a limited part of the North Atlantic north of 50°N and east of 15°W, as shown in Fig. 7 for the layer 470–800 m. The tracer in the eastern part of the section has slightly higher concentration with some influence from the Mediterranean outflow, and a more northerly source in the Bay of Biscay, which is the result of localized winter convection. This eastern water mass had higher tracer concentrations in the stand-alone simulations than in the coupled simulations, so it is unlikely that differences in tracer values are due to increased convective mixed layers in the coupled model. Both the asynchronous and synchronous stand-alone cases have higher levels (3%–5%) of the “Bay of Biscay Water” tracer recirculating in the gyre than the coupled case, despite the same concentration in the source region. In contrast, the other main feature of note is the increase in ventilation in the Labrador Sea in the coupled simulation compared to the stand-alone simulations.
Between 800 and 1250 m (Fig. 8) in the model, high concentrations (>60%) only occur north of 50°N. There is also evidence of a counterflow along the Gulf Stream axis occurring with the 60% concentration isoline trending toward the southwest at (45°N, 45°W) in all three simulations, representing the upper level of the deep western boundary current. Here, however, the concentrations are 10%–15% higher in the coupled simulation compared to the stand-alone cases.
Two possible causes for the difference in tracer concentration in the three simulations were investigated: the depth of winter convection and the velocities on individual levels. The depth of convection in the North Atlantic was measured from monthly averages of a density criterion of 0.3 σt units calculated each time step. This slightly overestimates the true convection, but shows the depth of penetration of wintertime mixing; there was minimal change for any season between the synchronous and asynchronous runs in this mixed layer measure. The coupled model, with its high-frequency atmospheric forcing, had mixed layers deeper by up to 100 m in winter, between 20° and 50°N in the main gyre and increases of 400 m in the mixed layer depth in the northeast Atlantic east of 40°W and north of 40°N.
The velocities in the main gyre in the North Atlantic are of similar magnitude in all three simulations (Fig. 9a). The difference in velocities in the coupled model compared to the asynchronous run (Fig. 9b) for the depth interval 470–800 m shown is greater than for the synchronous case (Fig. 9c; note the difference in scale from Fig. 9b).
The velocity difference between the synchronous and asynchronous simulations showed a small increase of flow in the center of the gyre (25°N) (Fig. 9c) of about 1% of the magnitude of the total flow. In comparison the coupled model showed a slackening of the flow of up to 15% along the lower axis of the gyre at 30°N where the concentrations are reduced in Fig. 7 compared to the stand-alone simulations. This change in the flow of the gyre in the synchronous and coupled cases, relative to the asynchronous case, is seen at levels down to 1250 m, so presumably it relates to the barotropic signal driven by changes in the wind stress in the coupled simulation.
At depths from 800 to 1250 m (Fig. 8), considerably less tracer is circulating in the main gyre than higher in the water column. The velocities in the main gyre are an order of magnitude smaller than observed in the depth interval 270–800 m, with maximum flow of 0.6 cm s−1. While there is still some evidence of a gyrelike flow in this depth interval in the asynchronous/synchronous simulation in the region of west to east flow at 40°N, in the coupled simulation there is a strong flow reversal up to 60°W in the gyre region. We conclude that this change in flow pattern explains the stronger tracer concentration in the western side of the basin in the coupled model at this depth.
Between 1250 and 1900 m the tracer distribution is governed by the recirculation beneath the main gyre. This means that the highest concentrations south of 45°N are along the western boundary of the domain rather than in the eastern section, as was seen in the levels above 1250 m. The concentration distributions for the asynchronous and synchronous simulations are very similar, with only 2% higher values in the synchronous simulation (Fig. 2a). The coupled simulation has concentrations that are 7% higher than the synchronous simulation, which is the result of additional tracer being injected at depth by winter convection in the source regions in the Labrador Sea and overflowing Denmark Strait. The concentrations in all the simulations in this depth interval are higher than those seen in the 800–1250 m interval. Examination of the meridional overturning streamfunction for the North Atlantic indicates that the outflow region of North Atlantic Deep Water (NADW) is beneath 1250 m, which is marked by the tracer.
Between 1900 and 2800 m the tracer in the standard simulations is almost entirely confined to the western Atlantic sector with only 20%–40% in the eastern sector (Fig. 10). The Mid-Atlantic Ridge does not intersect this depth interval, but in the north of the region the flow between 35° and 50°N is blocked by the model's topographic representation of the seamounts on the ridge; this steers the circulation to the west of 40°W and prevents basinwide tracer circulation. The concentrations are higher in the western section than in the previous depth interval, which is a consequence of the NADW being transported by the faster, deeper currents.
Interestingly, at this depth it is the synchronous simulation that has highest tracer values rather than the coupled simulation (Fig. 10). The velocity fields in the deep western boundary current are slightly greater in the synchronous and asynchronous simulations compared to the coupled model simulations, which partly explains the difference seen in the concentrations. The concentrations in the source region in the Labrador Sea were the same in each of the three simulations at over 95%, but the region to the south of the Labrador Sea has stronger concentrations in both the synchronous and asynchronous runs.
A further explanation for the reversal of tracer differences between runs at this depth is the strength of the overturning streamfunction in the coupled and synchronous ocean cases. Gordon and O'Farrell (1997) showed that the mean overturning streamfunction in the coupled simulation is less than in the ocean stand-alone synchronous simulation used here. As well as the lower mean strength of the overturning, the tracer input over the 50-yr time period will be influenced by the peaks and troughs in the multidecadal cycle in the overturning streamfunction. During the 50-yr period, less tracer has reached this depth, so the meridional circulation in the coupled model is weaker than in the synchronous simulations.
d. Comparison of North Atlantic tracer distribution between standard and GM coupled models
Most of the dynamic features that control the concentrations in the North Atlantic Ocean in the standard model simulation also hold for the GM set of simulations. This discussion will point out noticeable differences between the coupled model results. For ease of comparison, the 21-level model tracers have been converted to 12 levels by averaging.
The concentrations in the 470–800 m interval in the GM model runs were considerably less than in the standard runs (Fig. 7d). For tracer being subducted into the subtropical gyre in the GM coupled model, the source region is from the Mediterranean outflow rather than transport from the high-latitude convection region. The tracer in the standard coupled model appears to be sourced from farther north in the Bay of Biscay. On closer examination, both Mediterranean and Bay of Biscay sources exist in each set of the models, but in the standard coupled run, the Bay of Biscay tracer introduced in late winter convection (Mar–Apr) is so strong that it overwhelms the signal from the more southerly (40°N) source at the Mediterranean outflow. In the GM simulation the Bay of Biscay convection is weaker and does not occur every winter season, so the signals from both regions are distinct out to 50 years of simulation.
The concentration in the plume in the subtropical gyre at 30°N in the 470–800 m interval is less in the GM coupled model than the standard coupled model. In both cases, tracer values for the coupled model runs are only slightly lower than the respective ocean stand-alone cases (5%). However, there is a strong contrast in the strength of the velocity pattern between the two coupled simulations in the 470–800 m interval, with speeds in the subtropical gyre in the GM model less than 50% the amplitude in the standard run. In the thermocline this contrast was less strong, with the standard coupled model being only 20% faster than the GM case. This weaker circulation contributes to the reduced concentrations seen in the subtropical gyre of the GM model and is accompanied by a strengthening of the stratification below the thermocline in the GM model of 0.3 σt units.
In the next depth interval, 800–1250 m, the differences between the tracer distribution in the GM coupled simulation and standard coupled model are less marked (Fig. 8d). The subtropical gyre circulation is weak in the standard run, though still present, while in the GM coupled model a deep western boundary current of just under 1 m s−1 has developed. The similarity of the tracer patterns, despite the different flow regimes, suggests that, with slacker currents at this depth due to the transition between the subtropical gyre and deep western boundary current regimes, diffusion of tracer may be a controlling factor particularly along isopycnals.
In the standard coupled model between depths of 800 and 1250 m, the majority of the North Atlantic region lies on the 27.2-σt density surface (Fig. 11). Hence, the tracer distribution at 800 to 1250 m south of 50°N is almost entirely governed by advection and diffusion along the density surface. In the region of the main gyre, where we have seen that the currents are weaker, the tracer distribution is zonal south of 40°N on the surface. This 27.2-σt isopycnal surface outcrops at 52°N, so the high tracer distribution in the northern sector of the level distribution (Fig. 8c) is related to the downward arm of the North Atlantic overturning cell.
While the tracer simulation for the GM model showed some similarities to the standard simulation in the 800–1250 m interval, it is due to a different mechanism. In the GM simulation the transport of tracer is governed by the reverse flow beneath the main gyre and increasing strength in the deep western boundary current. In isopycnal space, the 27.4-σt surface occurs between 800 and 1000 m across most of the North Atlantic south of 50°N; the distribution of tracer on this surface is very similar to that seen between 800 and 1250 m (Fig. 8d).
The broadscale tracer patterns for 1250 to 1900 m are also very similar for all the GM and standard models, but here the similarity is due to similar ocean current regimes at these depths. All are dominated by the deep western boundary current transporting tracer in the NADW equatorward (Fig. 12).
The reduced plume strength in the midlatitudes in the GM coupled simulations compared to the standard coupled simulation is due to the deep western boundary current in the GM coupled simulation being up to 40% weaker than in the standard coupled simulation, a reduction that is replicated throughout the deeper part of the water column. In the lowest model levels the differences in tracer between the coupled simulations is due to a combination of the reduced strength of the deep western boundary current and to reductions in tracer input from the Labrador Sea region modulated by the coupled models interdecadal signal.
This section discusses some of the issues raised by the study and briefly compares some of the model results with observational data of intermediate and deep water masses formed in the Northern Hemisphere to determine which of our six simulations provides the most realistic representation. It also considers the results of other modeling studies of ocean tracers.
In the North Pacific, the most significant ventilated water mass in the NPIW. The water mass is characterized by a salinity minimum between 400 and 800 m centered on the 26.8-σt isopycnal surface. There have been several papers discussing the origins of this water mass and its low salinity, as there is no major outcrop of this isopycnal in the observational data in the North Pacific basin. There are theories that the NPIW is driven by water mass modification in the Alaskan gyre and in the Oyashio off Kamkatcha in the mixed water region between the Oyashio and Kuroshio (Van Scoy et al. 1991; Yasuda et al. 1996; Yasuda 1997; Talley 1993). The only place where surface waters are observed to be as dense as 26.8 σt is the Sea of Okhotsk in winter, when brine rejection from sea ice formation causes dense water to occur locally.
There are a number of modeling studies to explain the properties and formation of NPIW and why it carries a low salinity signal compared to the ocean layers above and below it (Yasuda 1997). The consensus regarding the lower salinity is that the local precipitation exceeds the evaporation in the north of the region, while in the Kuroshio area the evaporation is as large as the precipitation. The most detailed model study that we can compare directly with our results is that of Yamanaka et al. (1998a), using the Meteorological Research Institute model. Their study focuses on the role that the two regions could have in the intermediate water formation, for waters with winter densities greater than 26.2 σt in the western Bering Sea and Sea of Okhotsk. Other studies have also focused on how the denser waters of the Sea of Okhotsk overflow into the North Pacific through gaps in the Kuril island chain (Watanabe and Wakatsuchi 1998).
The CSIRO model's representation of the salinity minimum is seen in a section along 180° in all simulations in Fig. 13. This figure shows that all simulations have some signal of a low salinity tongue, but underestimate its strength and extent compared to the climatology of Levitus et al. (1994) (Fig. 13f). As in the Yamanaka et al. (1998a) study, the simulation with the GM mixing scheme gives an improved result for the strength and extent of the salinity minimum compared to the standard case, which is equivalent to their isopycnal mixing case. Interestingly, the two coupled cases do not show a major improvement in the strength of the salinity minimum compared to the stand-alone cases, with only some broadening of the low salinity region occurring in the GM coupled case. In density space, all cases have the salinity minimum centered on the 26.5-σt surface; however this less dense surface for the salinity minimum is similar to the results of Yamanaka et al. (1998a), who were only able to generate a salinity minimum on the 26.8-σt density surface by using perpetual artificial wintertime forcing over the entire depth of the water column in the Bering Sea and Sea of Okhotsk.
In both the standard and GM stand-alone cases the wintertime surface density is at a maximum of 26.3 σt units in the western Bering Sea off Kamkatcha and in the Sea of Okhotsk. In contrast, in the coupled models the surface winter density in this area can be as high as 26.5 σt in the model over a wide area and locally as high as 26.7 σt in the Sea of Okhotsk in most winters. There are occasional winters when the surface densities are even higher at 26.9 σt in the standard coupled model and other winters in the GM coupled model when the density is not above 26.0 σt.
Despite the increased winter surface density in the coupled models in most winter seasons, there has been no major shift in the position of the salinity minimum related to the density structure. This is presumably due to the fact that the surface forcing is transient and only present at the higher densities for 2–3 months of the year. Also, these higher densities only overflow shallow sills in the Kuril island chain into the Pacific Ocean in the winter months in the model simulation. The lower surface densities that occur in the Sea of Okhotsk on an interdecadal timescale arise from a combination of both higher temperatures (by 1 K) and fresher surface waters (by 1‰), which mean less sea ice growth has occurred over the winter season. Unlike Yamanaka et al. (1998b), we have not undertaken experiments to distinguish the relative source of NPIW from the Sea of Okhotsk and Kamkatcha, but the concentrations in Figs. 5 and 6 suggest in our model that the Kamkatcha source is of an equal magnitude to the Sea of Okhotsk source where the pre-NPIW source waters are trapped beneath the sill depth for about eight months of the year. This may help account for the salinity minimum being on a lighter isopycnal surface compared to the Yamanaka et al. (1998a) perpetual winter experiments.
In the North Atlantic sector, a direct comparison of water masses in the model with observations is hampered by the deepest water masses in the model being lighter than observations, with maximum densities of 27.5 σt in the standard simulation and 27.68 σt in the GM simulation. England and Holloway (1998) compared stand-alone GFDL ocean model simulations, focusing on CFC tracer concentration patterns in the middepth to deep North Atlantic Ocean. They also had incorrect densities for their deep water, but attempted to overcome this shortcoming by examining tracer patterns on the model level where the deep western boundary current is at a maximum. Of the five cases considered by them, only models with a topographic stress term, or artificial forcing of salinity and temperature in the northern North Atlantic convective regions, allowed sufficient tracer into the deep western boundary current region.
Although the CSIRO ocean model is also based on the GFDL model code, there are differences in resolution and topography, which suggest the CSIRO model results have more realistic meridional overturning and deep circulations. However, as noted in section 3.4, the velocity field in the GM version of the model is very sluggish in the deep western boundary current below 1900 m. The densest waters are formed in the Greenland Sea in the GM model, with highest surface densities of 28.2 σt in the northern Greenland Sea and Barents Sea in winter. However, very little of this dense water overflows Denmark Strait each year, with the majority remaining in the Greenland Sea Basin, with the densest waters in the bottom of the North Atlantic basin having a density of 27.68 σt. As this water was not formed locally, or in the Labrador Sea, it must have originated in the Greenland Sea and overflowed Denmark Strait. However, there is no evidence in the model of water overflowing the sills between Iceland, Faeroes, and Shetland to form the Icelandic–Scotland overflow water. This water was identified by Smethie et al. (2000) as contributing overflow water of low CFCs and slightly lower density than the dense water overflowing Denmark Strait. The densest waters carried south by the deep western boundary current are on the 26.67-σt surface. Figure 14a shows the concentration on the 27.55-σt surface, which is the densest that is ventilated in year 50 of the GM simulation from the Greenland Sea through Denmark Strait. The model ocean at this density is also ventilated from the Labrador Sea where open ocean convection occurs and the subsurface waters are influenced by a local recirculation of Denmark Strait overflow water. This means that, unlike the England and Holloway (1998) study, the CSIRO model does have a weak representation of the Denmark Strait overflow water in the deep western boundary current at midlatitudes, though it is almost indistinguishable from the waters coming from the Labrador Sea, which is modified Denmark Strait overflow water.
Pickart and Smethie (1998) have undertaken several repeat sections over a 10-yr period of temperature salinity, and CFCs in the (42°–44°N, 54°–56°W) region, and were able to distinguish different rates of formation of Labrador Sea Water in different years, with localized maxima and minima in the CFCs downstream from the Labrador Sea in both upper (above 1500 m) and lower water masses. Deeper waters from the Denmark Strait overflow are found on the CFC sections below 2500 m. Tritium and 3He measurements (Doney and Jenkins 1994) provide further passive tracers in the region with different source functions and time histories.
In the CSIRO model the contributions of Upper and Lower Labrador Sea Water are less distinct in the density field. However, a section through the GM coupled model at 49°N (Fig. 15a) shows a separate Labrador Sea core at about 800-m depth (concentration above 98%) with a deeper tracer core between 1500 and 2000 m centered at 40°W. At a model grid point farther north at 52°N, there is only a larger maximum in the Labrador Sea, though the area greater than 90% concentration is much larger than seen at 49°N (Fig. 15b). It is proposed that the lower part of this core is the direct contribution from the Denmark Strait overflow to the tracer concentration field at a density greater than 27.5 σt, while the upper core is slightly lighter at 27.4 σt and involves water from Denmark Strait that has recirculated in the Labrador Sea below the pycnocline at 270 m. In the standard coupled model, the deep western boundary current core is much deeper due to the faster currents, which were discussed in earlier sections, with maximum concentrations now occurring between 2000 and 4000 m depth. In this case, on the section at 49°N (Fig. 15c) there is no core of Upper Labrador Sea Water at 800 m, while in contrast at 52°N there is a region of Labrador Sea Water having maximum concentrations between 1000 and 1500 m, with a deeper core of overflow water at slightly lower concentration.
Figure 16 shows two meridional sections from the two coupled models at 57°W and 34°W. At 57°W in both cases we see the pool of well-mixed water in the mouth of the Labrador Sea and a weaker southerly core. The depth of maximum convection for the standard coupled models in the Labrador Sea, and the depth of the deep western boundary current tracer core, again drive the differences between the pairs of sections. These contrasts are also true at 34°W with the tongue of high concentration water reaching 3500 m depth in the standard coupled simulation and only 2000 to 2500 m in the coupled GM case. Most of the high tracer concentration waters in the deeper part of this 34°W section have originated from the Denmark Strait region. Smethie et al. (2000) and Doney and Jenkins (1994) have estimated the age of the Denmark Strait overflow water mass downstream in the deep western boundary current using CFC-11: CFC-12 ratios and the Tritium:He3 ratio. While our simulations have not included an “age” tracer, which would allow for dilution of the newly ventilated water with older recirculating waters, the timing of the advancing edge of the tracer plume in the model is in broad agreement with the observed tracer data, which means that the velocities in the deep western boundary current are not unrealistic.
In the northeast Atlantic and in the Bay of Biscay the observations document the formation of Subpolar Mode Water on the 27.2–27.5 σt surface (McCartney and Talley 1982). Again, in the model the densities for this water mass are less than in observations. These waters are remnants of late winter convection, which varies considerably from year to year in this region, particularly in the less convective GM model. In the modeled ocean these water masses do not inject tracer below 800 m (Fig. 7) and there is little clear separation from other water masses in the tracer field on meridional sections. In density space the concentrations on the 26.8 and 27.2 σt density surfaces (Figs. 17 and 11) show the influence of the pathways of the mode waters that are ventilated in the NE Atlantic sector; the 27.2-σt surface in the GM coupled model is better ventilated than the standard model in the mode water region to the west of the British Isles.
Both the time series data for the different water masses given in Fig. 3 and the plan sections (e.g., Fig. 4) document small differences between the asynchronous and synchronous standard simulations in the Northern Hemisphere ocean basins. The differences in the midlatitude tracer distributions (Figs. 4 and 7) occur where different velocities are generated by the distorted physics of the asynchronous time stepping; this causes the tracer plumes to be driven slightly further in the 50-yr period in the synchronous simulation. In the coupled model simulation, which forms the most realistic representation of the ocean in terms of the variable forcing on daily to interdecadal timescales, some of the differences at middepth in the tracer distributions can be attributed to real changes in water masses, while the changing velocity regime in the coupled model also plays a considerable role. This causes the tracer plume to be driven farther in some depth intervals in the coupled model, most notably in the upper levels of the North Atlantic Deep Water layer (Fig. 3a), while in other cases the speed of the tracer plume in the coupled runs is modulated by multidecadal signals of the coupled system. The other large differences in the standard coupled model tracer concentration fields occur in the high latitudes, due to the high-frequency forcing causing much greater ventilation and deeper mixed layers.
In contrast to the standard set of simulations, in the GM set of simulations there were even smaller differences between the asynchronous and synchronous simulations, with the greatest differences showing up in the areas of high ventilation in the Southern Ocean (Fig. 2). There are still small differences in the actual tracer amounts after 50 years of simulation between the synchronous and asynchronous cases, but differences overall were less than 5% locally with zonal averages of less than 1%. In the GM ocean-only cases, the difference in the depth of winter convection between the asynchronous and synchronous runs is smaller than in the standard runs. The GM coupled model also shows reduced ventilation at depth, compared to the ocean-only cases in the North Atlantic, but marginal increases in ventilation in the North Pacific; however, these changes are less marked than in the standard model set of results due to the increased stratification and denser deep waters with the GM scheme.
The main difference between the GM and standard coupled models is the reduced penetration of tracer to depth in all the water mass formation regions. This is due to the higher densities at depth and stronger stratification in the GM simulations. This implies that the tracers are carried on denser surfaces in the GM models, while not penetrating as deeply. One other feature is the weaker and shallower deep western boundary current in the GM coupled model compared to the standard simulation, which presumably is related to the density structure in the region.
The analysis of the North Pacific tracer distributions on density surfaces showed that a direct comparison of tracer distribution between matching years in the coupled runs can be deceptive, as the results can be aliased by longer term climate signals on interdecadal timescales. There is also an interdecadal signal in the North Atlantic in the coupled models that modulates the meridional overturning streamfunction, surface temperatures and salinity, and heat and freshwater fluxes. However, this signal appears to have had less influence on our interpretation of the tracer concentration fields, which have pre-dominantly been considered in level space. The interdecadal signal does modulate the horizontal velocity fields so that 50-yr averages of the coupled run are required for comparison with stand-alone cases. The interdecadal signal also modulates the injection of tracer and newly formed NADW in the deepest model levels of the North Atlantic basin and explains why the densities at the deepest levels close to Denmark Strait are not seen overflowing the strait in year 50 of the simulation.
Comparison with observations of tracers (e.g., CFCs or tritium) on density surfaces is limited by the fact that the ventilated water masses are not being formed in either the North Pacific or North Atlantic on the correct density surface. The North Pacific Intermediate Water low salinity anomaly does not penetrate far enough south in the model and is found on the 26.5-σt surface, despite the coupled model including sufficiently dense water after brine rejection in the Sea of Okhotsk. In our model results, this water is trapped most of the year below the sill depth and the source off Kamkatcha Peninsula is of equal magnitude. In the North Atlantic the densities of all the main water types are too low by up to 0.3 σt in the standard case and 0.1 σt in the GM case. However, the dynamics in the formation region of North Atlantic Deep Water, Upper and Lower Labrador Sea Water and Northeast Atlantic mode water are realistic. The Mediterranean outflow water dense plume (not discussed) is poorly represented and the dense water plume overflowing Denmark Strait is too weak, so transects in the deep western boundary current do not provide a good match with CFC and tritium observations. The tracers however, do not provide a clear criterion to prefer the GM over the standard coupled model distributions in the Northern Hemisphere, despite the GM case having more realistic deep water densities. Many of the errors in the water masses are common in both simulations. However, the tracer plume of the Denmark Strait overflow water is more realistic in the standard coupled case.
In conclusion, the CSIRO coupled climate model has represented the main ventilated water masses formed in the Northern Hemisphere. Our studies with a single passive tracer cannot be calibrated against particular tracer sections (e.g., CFCs) to determine whether the volume and timing of water mass formation in the coupled model are realistic. However, studies by England and Hirst (1997) and England and Rahmstorf (1999) suggest that the models have yet to achieve this degree of sophistication and reliability in the coarse-resolution models used for climate studies. Only when we are more confident that the models represent well all the physical processes that are driving the water mass formation, and its subsequent circulation in the ocean, can we expect coupled models to pass this more stringent test.
I acknowledge the contribution of members of the climate modeling group who contributed to the development of the CSIRO coupled climate model, in particular Dr. Hal Gordon. I also would like to thank Drs. Tony Hirst and Matthew England for discussions and comments on a draft of this paper, and the anonymous reviewers whose comments have improved the manuscript. This work was completed under funding from Environment Australia for climate change research. The simulations were performed on the CRAY-YMP and CRAY-J90 of the CSIRO High Performance Computing Centre.
Model Parameters and Forcing Fields
The experiments with the asynchronous/synchronous ocean model were forced with Hellerman and Rosenstein (1983) monthly wind stresses, with the surface fluxes derived by relaxing the model surface temperatures and salinities to Levitus (1982) monthly climatologies with a 20-day timescale. The coupled simulations are flux corrected, so in the monthly mean the wind stress climate is close to Hellerman and Rosenstein values and the monthly mean surface temperatures and salinities of the ocean are close to those in the stand-alone ocean case (Gordon and O'Farrell 1997). The coupled simulation is initiated using the equilibrated solution of the synchronous ocean stand-alone simulation. The synchronous simulation used hourly time steps for tracer and velocity fields, while the asynchronous simulations used daily time steps for tracer and 20-minute time steps for velocity fields.
The internal parameters of the model include mixing of tracers along the sloping neutral density surfaces of the model; this is performed using the Cox (1987) implementation of the Redi (1982) scheme using an isopycnal tracer diffusion of 107 cm2 s−1. The vertical diffusivity is the stability-dependent function of Gargett (1984) as outlined in Hirst et al. (2000). A lower limit of 0.3 cm2 s−1 is imposed on the vertical diffusivity except between the uppermost layers where a larger lower limit is set to represent wind-forced mixing. Oceanic convective mixing is simulated by applying an enhanced vertical diffusivity of 106 cm2 s−1, in regions of static instability. The vertical viscosity is set to 20 cm2 s−1 throughout the water column and the horizontal viscosity at 9 × 109 cm2 s−1.
The standard version of the model has a background horizontal diffusion that tapers from 1.2 × 107 cm2 s−1 at the surface to 0.5 × 107 cm2 s−1 in the deep ocean with an e-folding depth of 500 m. In the GM simulation there is no background horizontal diffusion and the eddy diffusion in the eddy-induced advection term (Hirst and McDougall 1996) is set to 107 cm2 s−1 below 270 m and is tapered linearly to zero at the surface.
Corresponding author address: Dr. Siobhan O'Farrell, CSIRO Atmospheric Research, Private Bag No. 1, Aspendale, Victoria, 3195 Australia. Email: Siobhan.O'Farrell@dar.csiro.au