## 1. Introduction

The transport of heat by mesoscale eddies in the ocean is the subject of considerable interest and debate because it is suspected to be an important term in the time-mean ocean heat transport. The rectified eddy heat transport (hereafter referred to as the eddy heat transport) arises from the correlation of velocity and temperature variability about their time means. The fundamental dynamics that underlie the eddy heat transport in the ocean are not clear and have not been adequately explained. The role of the oceanic mesoscale eddy field in climate processes has been only marginally addressed observationally in the ocean (Bryden 1979; Bryden and Heath 1985; Bryden and Brady 1989; Stammer 1998; Wunsch 1999). Coarse-resolution ocean models, such as those used in climate models, do not resolve the transport processes associated with the oceanic mesoscale field. Therefore, it has been the subject of much speculation and theoretical consideration because of the need to include it as a subgrid-scale parameterization in coarse-resolution climate models (e.g., Gent and McWilliams 1990; Holloway 1992; Danabasoglu et al. 1994; Visbeck et al. 1997; Griffies 1998). Tests of these parameterizations have met with some success when applied to atmospheric data (Kushner and Held 1998) and ocean models (Böning et al. 1995; Rix and Willebrand 1996); however, little work has been done to date on investigating characteristics of the eddy heat transport in high-resolution, global ocean circulation models.

Unfortunately, the time-mean eddy heat transport is one of the most poorly observed quantities in the ocean. Wunsch (1999) recently compiled a collection of the available current-meter data in an attempt to assess its magnitude in the ocean. In general, he found it was significant in the western boundary current regions of the Pacific and Atlantic Oceans. It was also found to be very small in the interior of the ocean gyres. However, the sparseness of the current meter data precludes any separation into the rotational and divergent components, making the fluxes nearly impossible to interpret. Stammer (1998) recently used satellite altimetry data to compute an eddy diffusivity, which he then combined with a temperature climatology to estimate the eddy heat transport. His results were consistent with the picture that the eddy heat transport is large in the boundary currents and weak in the interior of the gyres. The time-mean eddy heat transport has also been addressed by several regional process studies, some of which will be discussed later (Bryden 1979; Bryden and Heath 1985; Bryden and Brady 1989; Bower and Hogg 1996; Cronin and Watts 1996).

In this paper we explore the characteristics of the eddy heat transport taken from a high-resolution global ocean general circulation model. The eddy heat transport is examined by basin, by depth interval, and by dynamical component (section 2). Then in section 3, the vector distribution of the eddy transport is considered. Following the suggestion of Marshall and Shutts (1981), it is separated into its rotational and divergent parts with the Gulf Stream as an example. The rotational component cannot transport heat across latitude circles and does not contribute to the poleward transport of heat by the oceans, whereas the divergent component of the heat transport does affect the local heat balance and does transport heat northward. It is shown that for a coherent meandering jet, the rotational eddy heat transport dominates and it is not necessarily downgradient, leading one to doubt the validity of claims that current meter data show downgradient temperature transport (section 4). A comparison of the model's northward eddy heat transport and a global estimate derived from TOPEX/Poseidon data (Stammer 1998) is made. The basic technique used by Stammer (1998) is tested and thereby the assumptions that were used in calculating it are evaluated. Last, the eddy heat transport at a few selected locations is examined by using its cospectrum for an indication of what part of the frequency spectrum is contributing to the heat transport (section 5).

## 2. General characteristics of the eddy flux

The numerical simulation output we use for this study is from run 4B of the Parallel Ocean Climate Model (POCM), archived at 3-day intervals. The POCM is a primitive-equation, level model configured for the global ocean between 75°S and 65°N, with realistic topography (see Jayne and Marotzke 2001 for a discussion of the model's time-varying heat transport). The model has an average resolution of about ¼° and 20 vertical levels. Previous studies (Stammer et al. 1996; McClean et al. 1997) have shown that the POCM model simulation's eddy kinetic energy is too weak by at least a factor of 4, and the sea surface variability is low by at least a factor of 2 when compared to the altimetry observations of TOPEX/Poseidon. However, it is hoped that while future higher-resolution model runs may change the quantitative details presented here, in general the overall qualitative picture would remain similar (as suggested by the analyses of Cox 1985; Beckmann et al. 1994).

Beckmann et al. (1994) found in their North Atlantic Ocean model that increasing the resolution did indeed change the eddy heat transport, that the heat transport by the mean flow was also affected, but that there was very little overall change in the total heat transport. Therefore, one might expect that increased resolution likely will increase the magnitude of the eddy heat transport, but that its overall structure would be qualitatively similar. However, it remains to be seen whether this will hold true. In part, this work on the eddy heat transport should be regarded as a demonstration of what analyses could be done on higher resolution, and hopefully more realistic, ocean model runs to understand the dynamics behind the eddy heat transport. Beyond this goal, we attempt to describe and understand the model's eddy heat transport and compare and contrast it to other estimates of the ocean's eddy heat transport.

Our calculation of the eddy heat transport includes variability on all timescales, that is, everything that is time varying. Therefore we include the mesoscale band as well shorter period variability, such as tropical instability waves in the equatorial regions, and Ekman layer variability. Here we will consider the northward eddy heat transport in a few different manners: separated by basin, by depth, by dynamical components, and finally as a function of longitude and latitude. As will be discussed later, the eddy heat transport consists of a rotational part and a divergent part, which need to be teased apart, in order to better understand the dynamics at work.

To begin, the zonal integral of the northward eddy heat transport for both the World Ocean and the individual ocean basins are compared to the heat transport by the time-mean circulation in Fig. 1, and then plotted together in Fig. 2. The ocean basins defined separately here extend from 37°S (the latitude of the Cape of Good Hope on Africa, south of which the ocean basins become ill-defined) to the northern end of the model domain at 65°N for the Atlantic and Pacific Oceans, and the natural boundary at 26°N for the Indian Ocean. This is similar to the analysis done by Semtner and Chervin (1992) and McCann et al. (1994) on the ½° POCM run. The results are remarkably similar to theirs. Notably the eddy transport in the Antarctic Circumpolar Current (ACC) is higher in the ¼° run, suggesting that increasing model resolution enhances the modeled eddy heat transport while leaving its structure generally unchanged, as was suggested by Beckmann et al. (1994). Over the World Ocean, the eddy heat transport is a significant contributor to the total time mean, particularly on either side of the equator, where there is a southward eddy heat transport of 0.9 PW at 5°N and 0.5 PW northward eddy heat transport at 5°S. The large convergent eddy heat transport at the equator is consistent with observations from current meter arrays by Bryden and Brady (1989) and Baturin and Niiler (1997) and previous modeling work by Philander and Pacanowski (1986). Farther from the equator there is a second peak in eddy heat transport, which is especially noticeable in the Indian Ocean south of the equator at around 15°S; it will be shown later that this is associated with an area of intense activity in the western half of the Indian Ocean, likely from the intense monsoonal activity there. Another peak is in the Pacific Ocean at 15°N, also concentrated in the western part of the basin. Elsewhere, the eddy transport is large in the southern Indian Ocean along the Antarctic Circumpolar Current, where approximately 0.6 PW of heat is carried southward by the eddies, which is the same order of magnitude as the observed time-mean heat transport there (Macdonald and Wunsch 1996). Overall, in the northern midlatitudes, there is a small eddy heat transport of peak amplitude of 0.2 PW for the World Ocean.

Figure 3 shows how the eddy heat transport is distributed over four depth bins: the model surface layer (0–25 m), the near surface (25–985 m), the mid depths (985–2750 m), and the deep ocean (2750–5200 m). Several features stand out. First, most of the eddy heat transport is contained in the near surface (0–985 m), suggesting that the dynamics that lead to the eddy heat transport is confined to the upper ocean, in line with previous model results from Böning and Cox (1988). Second, there is a broad-scale eddy heat transport over the midlatitudes of both hemispheres in the surface (0–25 m) layer of order 0.1–0.2 PW, which is associated with a rectification of the Ekman layer velocity and temperature variability. Finally, the two deepest depth bins account for very little of the eddy heat transport despite their covering about 75% of the total ocean depth. An exception is the middepth range (985–2750 m), which has a small but significant contribution near the equator. Wunsch (1999) in his quasi-global estimate of eddy heat transports from current meter data also found that the majority of the eddy heat transport, as well as the eddy kinetic energy, was confined to the upper 1000 m.

Finally, we can consider the eddy heat transports due to the different dynamical overturning components. Figure 4 shows the zonal basin integrals of the eddy heat transport for the four components of the overturning decomposition in Lee and Marotzke (1998) and Hall and Bryden (1982) (see also Jayne and Marotzke 2001). They are as follows:

- The contribution to the meridional velocity due to the external mode (or barotropic gyre circulation) flowing over varying topography. Essentially, it is the flow governed by the Sverdrup relation taking into account time dependence, bottom topography and frictional effects.
- The surface Ekman flow minus its vertical average to represent its barotropic compensation. The Ekman component of velocity, is taken here to be the shear velocity in the four surface levels referenced to velocity at the fifth model level (117.5 m); however, nearly all the Ekman transport takes place in the top level (uppermost 25 m).
- The zonal average over the basin of vertical shear flow, which is generally associated with thermal wind shear balanced by zonal density gradients, as well as smaller contributions from the ageostrophic shear from frictional and nonlinear effects.
- Deviations from the zonally averaged vertical shear flow, which is generally associated with baroclinic eddies.

These heat transport associated with the dynamical components are respectively equivalent to 1) “barotropic,” 2) “Ekman,” 3) “baroclinic,” and 4) “eddy” components of the classic Hall and Bryden (1982) analysis, and are additive parts of the total heat transport.

There is an eddy heat transport associated with the Ekman component that is about 0.1 PW in the midlatitudes and looks remarkably similar to the eddy heat transport in the upper 25 m of the ocean. This distribution suggests two important conclusions: The first is that there is a small, order 0.1–0.2 PW, eddy heat transport due to covarying Ekman layer transports and temperature fluctuations not being fully in quadrature. The second is that the majority of the eddy heat transport is associated with the deviations from the zonal mean in the baroclinic shear term. The confinement of the eddy heat transport to the upper 1000 m and its association with the deviations in the baroclinic shear intimate that baroclinic processes are the dominant process. The barotropic gyre component also contributes strongly to the eddy heat transport in the Antarctic Circumpolar Current, implying that barotropic eddies interacting with topography may play a role there.

## 3. A meandering jet

So far only the zonal integral of the northward component of the eddy heat transport has been considered; however, the depth-integrated eddy heat transport is a two-dimensional vector quantity in space. Figure 5 shows the depth integrated vector eddy transport for the region of the Gulf Stream off the eastern coast of the United States. This particular region was chosen for its general interest as well as its historical current-meter coverage. While there are obvious model deficiencies in this area, notably that the Gulf Stream tends to follow the coast too tightly and it separates too far north of Cape Hatteras, there are some qualitative conclusions that can be drawn by comparing the eddy heat-transport field to observations. The SYNOP arrays that were located in the Gulf Stream at 55°W, discussed by Bower and Hogg (1996), and at 68°W, analyzed by Cronin and Watts (1996), provide guidance here. Both these studies found significant eastward (downstream) eddy heat transport with small meridional (cross-stream) eddy heat transports, consistent with the overall behavior of the model.

As was discussed by Marshall and Shutts (1981), the eddy heat transport is composed of two different dynamical components: the rotational and the divergent. The rotational component does not contribute to the globally integrated poleward transport of heat by the oceans, nor does it contribute to the local energy balance, as it transports as much energy into any given region as it does out of the region. The divergent component, on the other hand, does affect the local heat budget, and therefore is dynamically active. To examine the eddy heat transport in more detail, it must be separated into its two parts and the divergent part examined on its own. However, the distribution of the rotational flux and divergent flux must be determined globally, as they cannot be locally separated.

The separation of the divergent and rotational fluxes is done numerically by taking the divergence of the eddy heat transport, then inverting the divergence with a Laplacian inverter with Neumann boundary conditions (no heat transport through the lateral boundaries) to find a potential function, and then taking the gradient to recover the vector quantities for the divergent part of the eddy heat transport. A note on the boundary conditions is appropriate: since the model uses the condition that *u,* *υ* = 0 on the boundaries, it automatically implies that **v**′*θ*′ = 0 on the boundaries as well. This does not necessarily mean that the rotational and divergent components must be zero on the boundaries, only their sum need be. However, for their sum to be zero would imply that the rotational part of the eddy heat transport interacts with the divergent part, which should not be so since the rotational part of the flux does not affect the dynamics. Thus, both the divergent and rotational fluxes should be zero on the boundary to satisfy the boundary condition.

**v**′

*θ*′

**v**′

*θ*′

_{D}is the divergent component of the eddy heat transport; and (

**v**′

*θ*′

_{R}is the rotational component:where

*ϕ*is like a velocity potential, andsince by definition, the divergence of the rotational flux is zero:

**k**is the unit vector in the vertical direction, orwhere

*ψ*is like a velocity streamfunction, andgiven that by definition, the curl of the divergent flux is zero:

Returning to Fig. 5, we see that the rotational component dominates the total in the Gulf Stream region. In the divergent component, there appears to be a very weak southward eddy heat transport to the south of the jet axis, which is consistent with a downgradient transport as the meridional temperature gradient reverses sign south of the Gulf Stream in the model and is consistent with the eddy driving mechanism in the southern recirculation gyre of the Gulf Stream proposed by Bryden (1982). The estimated rotational and divergent contributions summed to the total eddy heat transport to within numerical precision.

*θ*and the velocity (

*u,*

*υ*) have distributions as they would in a quasigeostrophic jet, and are set by the following relations:where

*ψ*is the streamfunction,

*ξ*is the spatial and temporal distribution function for the axis of the jet. For example, the “Bickley” jet with a standing wave in it would have a streamfunction with the formwherewith the variables

*x,*

*y,*

*t*describing space and time;

*L*being the half-width of the jet,

*λ*the meander amplitude,

*l*the meander wavelength, and

*τ*the meander period.

## 4. The global distribution of eddy transport

*θ*

*κ*was estimated from

*κ*

*x,*

*y*

*αK*

_{E}

*x,*

*y*

*T*

_{alt}

*x,*

*y*

*K*

_{E}and the eddy mixing timescale

*T*

_{alt}were calculated from TOPEX/Poseidon observations; and

*α,*the mixing efficiency, was deduced to be 0.005 from a combination of TOPEX/Poseidon and current meter observations (Stammer 1998).

*κ*were constant, there would be only a divergent component. However, the rotational part of Stammer's (1998) estimate is still minor since the curl of a diffusive flux is small according to the following:The first term of the right-hand side of (27) is zero by definition, and the second term of (27) is small since ∇

*κ*and ∇

*θ*

*κ*∇

^{2}

*θ*

*κ*and ∇

^{2}

*θ*

A first glance examination of the POCM's eddy heat transport compared to Stammer (1998) shows some distinct similarities and differences between the two. First, the POCM shows a convergent eddy heat transport along the equator, which is completely lacking in the Stammer (1998) calculation. Stammer's (1998) methodology does not include an estimate of the eddy heat transports arising from tropical instability waves that are important in the low latitudes (Fig. 4). Second, the Antarctic Circumpolar Current shows a significant southward eddy heat transport in both estimates, though there is some discrepancy in its magnitude. Third, Stammer (1998) found a strong and widespread northward eddy heat transport across both the Gulf Stream and the Kuroshio, which is somewhat weaker in the model. The POCM's heat transports in the ACC and the western boundary currents may suffer from the model's weak mesoscale variability (Stammer et al. 1996). Despite all the differences there are some general similarities: in particular, the weak eddy heat transports in the centers of the ocean gyres seem to be robust and are supported by the current meter compilation by Wunsch (1999). On a final note, Stammer (1998) does not include the eddy heat transport due to rectification of Ekman layer variability, but it is a small contribution (order 0.1 PW) to the total and could be computed separately from the wind stress and Ekman layer temperature from climatologies.

To understand the differences between the Stammer estimate of the eddy heat transport and the POCM's eddy heat transport, his method is applied, following the steps described in his paper, directly to the model output using the time-varying sea surface height field and time-mean temperature field from the model. If the result of this calculation were to agree well with the model's directly estimated eddy heat transport, then it would indicate that the Stammer (1998) method is a robust way to calculate the eddy heat transport and would support applying it to the general ocean. If it does not, then it suggests that his method may not correctly predict the eddy transports. Indeed, the resulting estimate (Fig. 8e) is very similar in character to the Stammer estimate and is very unlike the model's actual eddy heat transport in many parts of the ocean, which implies that the method used by Stammer to calculate the eddy heat transport is not appropriate. Additionally, this can be considered a test of the model's simulation of the ocean's eddy field. The model, at least qualitatively, appears to be reproducing the large-scale eddy patterns observed by TOPEX/Poseidon (Stammer et al. 1996), as it gives nearly the same result for the eddy diffusivity derived using the method of Stammer (1998). Some possible reasons for the differences in the two eddy heat transport estimates will be discussed below.

There are three regimes in which the model's eddy heat transport will be compared to the Stammer (1998) estimate in detail. The first is in the Southern Ocean. The two estimates of the eddy heat transport in the Antarctic Circumpolar Current region are generally consistent, though the model estimate is larger there, 0.6 PW in the POCM versus what Stammer (1998) considers to be a lower bound for the ACC of 0.3 PW. The amplitude of the eddy heat transport across the ACC is consistent with the estimate of 0.45 ± 0.3 PW from de Szoeke and Levine (1981), but is somewhat larger than the values of 0.3 PW from Gordon and Owens (1987) and 0.2 PW from analysis of the Fine Resolution Antarctic Model by Thompson (1993).

The current meter data from the ACC are inconclusive. Bryden (1979) found a significant southward eddy heat transport in Drake Passage, and, by extrapolation to the rest of the ACC, thought that it was sufficient to balance the atmospheric heat loss to the south of 60°. However, the Bryden (1979) results were not corrected for mooring blow-over, which could have overestimated the eddy heat transport by as much as 20% (Nowlin et al. 1985). Bryden and Heath (1985) measured the ACC southeast of New Zealand and found that the eddy heat transport there was too weak to account for the expected loss of heat to the atmosphere, but their measured transports were not statistically significant. Meanwhile, Phillips and Rintoul (2000) using data from a current meter array south of Australia found a large poleward heat transport. If their observations were extrapolated to the rest of the ACC, the eddy heat transport would amount to 0.9 PW southward, more than enough to account for the heat loss to the atmosphere south of the Polar Front. It appears from analysis of the POCM that the extrapolation from either of these three locations is not appropriate, as the strongest southward heat transports in the model are in the southwestern Indian Ocean sector of the ACC, an area of no current meter coverage. The eddy heat transport is very inhomogeneous, increasing the difficulty of using scattered current meters to extrapolate the eddy heat transport for the rest of the ACC.

The second regime is the area around the western boundary currents. In the Stammer (1998) estimate the Gulf Stream and Kuroshio both have a large northward eddy heat transport that is somewhat smaller in the model. There are, of course, numerous model deficiencies that could play a role in the discrepancies, chief among them being the general weakness of the model temporal and spatial variability, and the failure of the model to reproduce the extension of eddy kinetic energy east of the separation points in the Kuroshio and Gulf Stream as was pointed out by Stammer et al. (1996). In fact, comparing the model's alongstream eddy heat transport to current meter data from the Gulf Stream at 55°W (Bower and Hogg 1996) and 64°W (Cronin and Watts 1996) shows that the model underestimates the eddy heat transport by a factor of about 4. At 55°W, Bower and Hogg (1996) found that the alongstream eddy heat transport was about 150°C m^{2} s^{−1}, while the model only produces about 40°C m^{2} s^{−1} in the axis of the stream at 55°W. Upstream at 68°W, Cronin and Watts (1996) found a stronger eddy heat transport of 900°C m^{2} s^{−1} as compared with the model's 200°C m^{2} s^{−1} there.

The model's weak northward eddy heat transport in the Kuroshio extension and the Gulf Stream recirculation region compared to that strong northward heat transports seen by Stammer (1998) may result from the overall weakness of the model's variability. On the other hand, neither Bower and Hogg (1996) nor Cronin and Watts (1996) observed the widespread strong cross-stream eddy heat transport that Stammer (1998) estimates for the Gulf Stream (Fig. 5). Also, while Stammer (1998) shows a uniformly northward eddy heat transport across the Gulf Stream, the model shows a more complex structure with a convergent eddy heat transport along the stream axis. Wunsch (1999) found strong southward eddy heat transports in a few locations in both the Gulf Stream and the Kuroshio. In the Kuroshio, he found one location with a southward eddy heat transport of 190°C m^{2} s^{−1}, compared to a weaker, but still southward eddy heat transport of 70°C m^{2} s^{−1} in POCM. These southward heat transports are simply not seen in the Stammer (1998) estimate, and might simply have been averaged out by his 5° smoothing. Moreover, as was argued in the previous section, it is difficult to know exactly what the current-meter observations are showing in the meandering jets, given the dominance of the rotational eddy heat transports. However, the eddy fields in the western boundary current systems are not robust features of the numerical model because of its inadequate spatial resolution. Smith et al. (2000) demonstrated in a model of the North Atlantic Ocean that increasing the spatial resolution to ° significantly improved both the path and eddy statistics for the Gulf Stream. However, even higher resolution may be required as Hogan and Hurlburt (2000) found in their model of the Japan/East Sea that it was necessary to reduce their grid spacing to ° before they found statistical convergence for the mesoscale variability. Resolution to these discrepancies will to await analysis of high-resolution models and more observational evidence.

The third area is the equatorial region where the model shows a convergent eddy heat transport along the equator in all three of the ocean basins, which is different from the southward eddy transport along the equator in the Indian Ocean and the northward transport in the Pacific Ocean estimated in Stammer (1998). In the equatorial region the POCM is thought to simulate the ocean well, and the convergent eddy heat transport along the equator is supported by previous modeling studies (i.e., Semtner and Holland 1980; Cox 1980; Philander and Pacanowski 1986).

The POCM's eddy heat transport in the equatorial Pacific is consistent with the observational evidence from the region. The POCM finds a convergent eddy heat transport along the equator of 190 W m^{−2} over the region between 110° and 140°W. Using data from current meters, Bryden and Brady (1989) found a convergent eddy heat transport along the equator between 110° and 152°W of about 245 W m^{−2}. Two studies using drifters also found a convergent eddy heat transport along the equator of 180 W m^{−2} between 105° and 120°W (Hansen and Paul 1984), and 100 W m^{−2} between 110° and 140°W (Baturin and Niiler 1997). So the model appears to be within the range of eddy heat convergence suggested from the observational work. All three of the observational studies, as well as the modeling studies mentioned above, pointed to the importance of tropical instability waves as the energetic fluctuations responsible for the convergent heat transport. These waves have a wavelength of about 1000 km (well resolved by the model's resolution) and periods of 3–4 weeks (Qiao and Weisberg 1995). They occur within about 5° of the equator and derive their energy from the barotropic velocity shear between the South Equatorial Current and the North Equatorial Countercurrent, and can be easily observed by satellite sea surface temperature observations (Legeckis 1977; Legeckis et al. 1983; Legeckis and Reverdin 1987; Chelton et al. 2000). The development of a parameterization for the eddy transports by tropical instability waves is needed to augment the Stammer (1998) methodology.

*κ*

*θ*

^{2}

*κ*and |∇

*θ*

^{2}are positive, so there would always be a net conversion from the mean-flow energy to eddy potential energy. However, Hansen and Paul (1984) found positive values for the conversion of eddy potential energy due to temperature south of the equator in the Pacific Ocean, indicating a conversion of energy back from the eddy field to the mean flow there. In the POCM the values of

**v**′

*θ*′

*θ*

The conversion of eddy potential energy may then provide a diagnostic to test when and where eddy parameterizations based on downgradient fluxes will fail in the ocean; they will fail where there is a baroclinic energy conversion from the eddy field to the mean flow. In essence, the eddy parameterization by Stammer (1998) fails to work in the model for the same reason it probably fails to work in the ocean; there are regions in the ocean where the eddy heat transport is upgradient associated with a net conversion of energy from the eddy field to the mean flow. While it would be gratifying to find a unified argument why the mixing length scalings fail in some locations and work elsewhere, it appears as of now that they may fail for different reasons. In the western boundary currents, the meandering jet gives a large apparent eddy diffusivity, even though there is not necessarily any mixing taking place. In the equatorial regions the apparent upgradient transport of temperature by the tropical instability waves plays a significant dynamical role. It should be noted that these two regimes are both places where there are narrow, barotropically unstable jets, and the solution to the problem may ultimately come from including those dynamics in the parameterization. In the Antarctic Circumpolar Current, in contrast to these other two regimes, baroclinic instability is dominant and the parameterization gives a consistent estimate; however, it certainly remains possible that they agree for reasons of pure chance.

## 5. Frequency distribution

*υ̂*

*f*)is the Fourier transform of the velocity time series

*υ*′(

*t*), and

*θ̂*

*f*) is the complex conjugate of the Fourier transform of the temperature time series

*θ*′(

*t*), and

*τ*is the averaging period.

The cospectra for four locations are shown in Fig. 11. The cospectra have been multiplied by the frequency to emphasize the higher frequency range while making sure that the area under the curve is proportional to the heat transport at that frequency. The frequency distribution of the eddy heat transport is widely variable around the global ocean. In the equatorial Pacific, most of the covariance is in the frequency band of 20–50 days, consistent with the hypothesis that the tropical instability waves by Hansen and Paul (1984) and Baturin and Niiler (1997) are responsible for the eddy heat transport there. This is to be contrasted with the tropical Indian Ocean location where the covariance is spread over a much broader range of frequencies from 50 to 500 days, and the Kuroshio where periods around the annual cycle appear to dominate. The ACC location, which is south of Madagascar, is perhaps the most difficult to understand. The cospectra there are noisy, which could be due to the variability from the strong meandering current there, and it appears that even very long fluctuations with periods of 1000 days contribute significantly. One of the weaknesses of this analysis is its inability to distinguish the rotational eddy heat transport from the divergent eddy heat transport that is strong in the ACC. Surely the two must have different frequency distribution, but it is not obvious how to compute the separated cospectra.

## 6. Discussion and conclusions

Globally, internal oceanic instabilities (eddies) play only a minor role in the time-dependent heat transport (Jayne and Marotzke 2001). They do, however, contribute to the time-mean heat transport in a number of locations. The eddy heat transport was examined in a number of ways. The interiors of the ocean gyres have little eddy heat transport in agreement with analyses of current meter data (Wunsch 1999) and an estimate derived from mixing length arguments (Stammer 1998). Despite some differences, both the Stammer estimate and the model come to the same conclusion: that eddy heat transports are important only in a few places in the ocean. The most significant eddy heat transport activity was found in western boundary currents, equatorial regions, and the Antarctic Circumpolar Current. The zonally averaged eddy heat transport makes a significant contribution to the total time-mean heat transport in the tropics and in the Antarctic Circumpolar Current, and the majority of the eddy heat transport is concentrated in the upper 1000 m of ocean depth. Finally, there is a global zonally integrated eddy heat transport of about 0.2 PW at its maximum due to the correlation of the velocity and temperature variability in the Ekman layer.

Further tuning of the Stammer (1998) estimate would likely bring it into closer agreement with the model. However, this model has significant deficiencies in its eddy field statistics, and that may be the source of the differences between the two estimates. On the other hand, the POCM's eddy heat transport differs from the estimate constructed by Stammer from a mixing length argument using altimetry data and a temperature climatology. That method was tested by applying it to the model fields, and it was shown that it failed to reproduce the model's directly computed heat transport, bringing into question the validity of the eddy parameterization. In the western boundary currents it was found that there is a large rotational component to the eddy heat transport which results from the meandering of the jets and that it obscures the dynamically important divergent component. An analytical argument shows that for a coherent meandering jet there can be a large rotational eddy heat transport, which is not necessarily downgradient. This rotational eddy heat transport may make analyses of scattered current-meter records difficult to interpret. Furthermore, the meandering jet has associated with it high levels of eddy kinetic energy, which using the Stammer (1998) method would imply high levels of eddy diffusion where none necessarily exist.

Along the equator the rotational component is weak, but there is a convergent eddy heat transport that comes from tropical instability waves with periods from 20–50 days, in agreement with results from current meter observations (Bryden and Brady 1989), mixed layer drifters (Hansen and Paul 1984; Baturin and Niiler 1997), satellite observations (Chelton et al. 2000), and previous modeling work (Semtner and Holland 1980; Cox 1980; Philander and Pacanowski 1986). In some locations the eddy heat transport is upgradient and is associated with areas of conversion of eddy potential energy to the mean flow energy. In the Antarctic Circumpolar Current region, the model's eddy heat transport agrees well with the estimate by Stammer, possibly because baroclinic instability dominates there. Finally, in the interiors of the gyres the eddy heat transport is very weak and the mixing length argument (Stammer 1998), the evidence from current meters (Wunsch 1999), and the model are all in unanimous agreement that the eddy transports are small there.

Resolution of these discrepancies between these model results and those of Stammer will need to await analysis of high-resolution models and more observational evidence. The eddy fields in the western boundary current systems are not robust features of the numerical model because of its inadequate spatial resolution. Smith et al. (2000) and Hogan and Hurlburt (2000) have demonstrated that, in order to realistically simulate the ocean's eddy field in numerical models, higher spatial resolution is required, at least

## Acknowledgments

Funding for this research came from the Department of Defense under a National Defense Science and Engineering Graduate Fellowship (SRJ), from the National Science Foundation, the American Automobile Manufacturers Association, and the Tokyo Electric Power Company through the TEPCO/MIT Environmental Research Program. The computational resources for this work were provided by the National Center for Atmospheric Research, which is supported by the National Science Foundation. Special thanks are owed to Robin Tokmakian and Bert Semtner from the Naval Postgraduate School, who generously provided their numerical model output. We are also grateful to Peter Gent, Nelson Hogg, Mike Spall, Detlef Stammer, and Carl Wunsch for helpful discussions. We thank the anonymous reviewers whose comments led to considerable improvement of the manuscript.

## REFERENCES

Baturin, N. G., , and P. P. Niiler, 1997: Effects of instability waves in the mixed layer of the equatorial Pacific.

,*J. Geophys. Res.***102****,**27771–27793.Beckmann, A., , C. W. Böning, , C. Köberle, , and J. Willebrand, 1994: Effects of increased horizontal resolution in a simulation of the North Atlantic Ocean.

,*J. Phys. Oceanogr.***24****,**326–344.Böning, C. W., , and M. D. Cox, 1988: Particle dispersion and mixing of conservative properties in an eddy-resolving model.

,*J. Phys. Oceanogr.***18****,**320–338.Böning, C. W., , W. R. Holland, , F. O. Bryan, , G. Danabasoglu, , and J. C. McWilliams, 1995: An overlooked problem in model simulations of the thermohaline circulation and heat transport in the Atlantic Ocean.

,*J. Climate***8****,**515–523.Bower, A. S., , and N. G. Hogg, 1996: Structure of the Gulf Stream and its recirculations at 55°W.

,*J. Phys. Oceanogr.***26****,**1002–1022.Bryden, H. L., 1979: Poleward heat flux and conversion of available potential energy in Drake Passage.

,*J. Mar. Res.***37****,**1–22.Bryden, H. L., 1982: Sources of eddy energy in the Gulf-Stream recirculation region.

,*J. Mar. Res.***40****,**1047–1068.Bryden, H. L., , and R. A. Heath, 1985: Energetic eddies at the northern edge of the Antarctic Circumpolar Current in the southwest Pacific.

*Progress in Oceanography,*Vol. 14, Pergamon, 65–87.Bryden, H. L., , and E. C. Brady, 1989: Eddy momentum and heat fluxes and their effects on the circulation of the equatorial Pacific Ocean.

,*J. Mar. Res.***47****,**55–79.Chelton, D. B., , F. J. Wentz, , C. L. Gentemann, , R. A. de Szoeke, , and M. G. Schlax, 2000: Satellite microwave SST observations of transequatorial tropical instability waves.

,*Geophys. Res. Lett.***27****,**1239–1242.Cox, M. D., 1980: Generation and propagation of 30-day waves in a numerical model of the Pacific.

,*J. Phys. Oceanogr.***10****,**1168–1186.Cox, M. D., 1985: An eddy-resolving numerical model of the ventilated thermocline.

,*J. Phys. Oceanogr.***15****,**1312–1324.Cronin, M., , and D. R. Watts, 1996: Eddy-mean flow interaction in the Gulf Stream at 68°W. Part I: Eddy energetics.

,*J. Phys. Oceanogr.***26****,**2107–2131.Danabasoglu, G., , J. C. McWilliams, , and P. R. Gent, 1994: The role of mesoscale tracer transports in the global ocean circulation.

,*Science***264****,**1123–1126.de Szoeke, R. A., , and M. D. Levine, 1981: The advective flux of heat by mean geostrophic motions in the Southern Ocean.

,*Deep-Sea Res.***28A****,**1057–1085.Gent, P. R., , and J. C. McWilliams, 1990: Isopycnal mixing in ocean circulation models.

,*J. Phys. Oceanogr.***20****,**150–155.Gordon, A. L., , and W. B. Owens, 1987: Polar oceans.

,*Rev. Geophys.***25****,**227–233.Green, J. S. A., 1970: Transfer properties of the large-scale eddies and the general circulation of the atmosphere.

,*Quart. J. Roy. Meteor. Soc.***96****,**157–185.Griffies, S. M., 1998: The Gent–McWilliams skew flux.

,*J. Phys. Oceanogr.***28****,**831–841.Hall, M. M., , and H. L. Bryden, 1982: Direct estimates and mechanisms of ocean heat transport.

,*Deep-Sea Res.***29****,**339–359.Hansen, D. V., , and C. A. Paul, 1984: Genesis and effects of long waves in the equatorial Pacific.

,*J. Geophys. Res.***89****,**10431–10440.Hogan, P. J., , and H. E. Hurlburt, 2000: Impact of upper ocean–topographical coupling and isopycnal outcropping in Japan/East Sea models with ⅛° to ° resolution.

,*J. Phys. Oceanogr.***30****,**2535–2561.Hogg, N. G., 1994: Observations of Gulf Stream meander-induced disturbances.

,*J. Phys. Oceanogr.***24****,**2534–2545.Holloway, G., 1992: Representing topographic stress for large-scale ocean models.

,*J. Phys. Oceanogr.***22****,**1033–1046.Jayne, S. R., , and J. Marotzke, 2001: The dynamics of ocean heat transport variability.

,*Rev. Geophys.***39****,**385–411.Kushner, P. J., , and I. M. Held, 1998: A test, using atmospheric data, of a method for estimating oceanic eddy diffusivity.

,*Geophys. Res. Lett.***25****,**4213–4216.Lee, T., , and J. Marotzke, 1998: Seasonal cycles of meridional overturning and heat transport of the Indian Ocean.

,*J. Phys. Oceanogr.***28****,**923–943.Legeckis, R., 1977: Long waves in the eastern equatorial Pacific Ocean: A view from a geostationary satellite.

,*Science***197****,**1179–1181.Legeckis, R., , and G. Reverdin, 1987: Long waves in the equatorial Atlantic Ocean during 1983.

,*J. Geophys. Res.***92****,**2835–2842.Legeckis, R., , W. Pichel, , and G. Nesterczuk, 1983: Equatorial long waves in geostationary satellite observations and in a multichannel sea surface temperature analysis.

,*Bull. Amer. Meteor. Soc.***64****,**133–139.Levitus, S., , and T. P. Boyer, 1994:

*Temperature*. Vol. 4,*World Ocean Atlas 1994,*NOAA Atlas NESDIS 4, 117 pp.Macdonald, A. M., , and C. Wunsch, 1996: An estimate of global ocean circulation and heat fluxes.

,*Nature***382****,**436–439.Marshall, J., , and G. Shutts, 1981: A note on rotational and divergent eddy fluxes.

,*J. Phys. Oceanogr.***11****,**1677–1680.McCann, M. P., , A. J. Semtner, , and R. M. Chervin, 1994: Transports and budgets of volume, heat, and salt from a global eddy-resolving ocean model.

,*Climate Dyn.***10****,**59–80.McClean, J. L., , A. J. Semtner, , and V. Zlotnicki, 1997: Comparisons of mesoscale variability in the Semtner–Chervin ¼° model, the Los Alamos Parallel Ocean Program ⅙° model, and TOPEX/POSEIDON data.

,*J. Geophys. Res.***102****,**25203–25226.Nowlin, W. D., , S. J. Worley, , and T. Whitworth, 1985: Methods for making point estimates of eddy heat flux as applied to the Antarctic Circumpolar Current.

,*J. Geophys. Res.***90****,**3305–3324.Philander, S. G. H., , and R. C. Pacanowski, 1986: The mass and heat budget in a model of the tropical Atlantic Ocean.

,*J. Geophys. Res.***91****,**14212–14220.Phillips, H. E., , and S. E. Rintoul, 2000: Eddy variability and energetics from direct current measurements in the Antarctic Circumpolar Current south of Australia.

,*J. Phys. Oceanogr.***30****,**3050–3076.Qiao, L., , and R. H. Weisberg, 1995: Tropical instability wave kinematics: Observations from the Tropical Instability Wave Experiment.

,*J. Geophys. Res.***100****,**8677–8693.Rix, N. H., , and J. Willebrand, 1996: Parameterization of mesoscale eddies as inferred from a high-resolution circulation model.

,*J. Phys. Oceanogr.***26****,**2281–2285.Rossby, T., 1987: On the energetics of the Gulf Stream at 73W.

,*J. Mar. Res.***45****,**59–82.Semtner, A. J., , and W. R. Holland, 1980: Numerical simulation of equatorial ocean circulation. Part I: A basic case in turbulent equilibrium.

,*J. Phys. Oceanogr.***10****,**667–693.Semtner, A. J., , and R. M. Chervin, 1992: Ocean general circulation from a global eddy-resolving model.

,*J. Geophys. Res.***97****,**5493–5550.Smith, R. D., , M. E. Maltrud, , F. O. Bryan, , and M. W. Hecht, 2000: Numerical simulation of the North Atlantic Ocean at °.

,*J. Phys. Oceanogr.***30****,**1532–1561.Stammer, D., 1998: On eddy characteristics, eddy transports, and mean flow properties.

,*J. Phys. Oceanogr.***28****,**727–739.Stammer, D., , R. Tokmakian, , A. J. Semtner, , and C. Wunsch, 1996: How well does a ¼° global circulation model simulate large-scale oceanic observations?

,*J. Geophys. Res.***101****,**25779–25811.Thompson, S. R., 1993: Estimation of the transport of heat in the Southern Ocean using a fine-resolution numerical model.

,*J. Phys. Oceanogr.***23****,**2493–2497.Visbeck, M., , J. Marshall, , T. Haine, , and M. A. Spall, 1997: On the specification of eddy transfer coefficients in coarse resolution ocean circulation models.

,*J. Phys. Oceanogr.***27****,**381–402.Warren, B. A., 1999: Approximating the energy transport across oceanic sections.

,*J. Geophys. Res.***104****,**7915–7919.Wunsch, C., 1999: Where do ocean eddy heat fluxes matter?

,*J. Geophys. Res.***104****,**13235–13249.

^{*}

Woods Hole Oceanographic Institution Contribution Number 10549.