1. Introduction

A ubiquitous feature in the Northern Hemisphere subtropical ocean circulation is the existence of an anticyclonic recirculation gyre (RG) on the southern flank of the wind-driven western boundary current outflow. The southern RGs in the Kuroshio Extension (KE) and the Gulf Stream have been documented in many pioneering studies of these two current systems (e.g., Worthington 1976; Richardson 1985; Kawai 1972; Mizuno and White 1983). The RGs significantly enhance the eastward volume and heat transport of the western boundary currents (e.g., Wijffels et al. 1998) and their variability has been recognized in recent years to be crucial in understanding the low-frequency midlatitude oceanic changes (Spall 1996; Qiu and Miao 2000; Schmeits and Dijkstra 2001; Dewar 2003; Hogg et al. 2005; Pierini 2006; Berloff et al. 2007; among others).

In the North Atlantic Ocean, various observations also reveal the presence of a RG north of the Gulf Stream jet between Cape Hatteras and the Grand Banks over a sloping bottom topography (Hogg et al. 1986; Hogg 1992). In contrast, evidence for a RG north of the Kuroshio Extension jet is fragmentary. Based on two years of moored current meter data along 152°E, Schmitz et al. (1987) observed steady westward flows of 2∼5 cm s⁻¹ at the 1200-m and 4000-m depths flanking the narrow, eastward-flowing KE jet. Similar westward-recirculating abyssal flows were found along 165°E by Schmitz (1987) and Joyce and Schmitz (1988). During the World Ocean Circulation Experiment (WOCE) era, scattered moored current meter measurements at 4000 m have further captured westward mean flows at 36.4°N, 150.2°E and 38.4°N, 148.9°E by Owens and Warren (2001), and at 38.0°N, 146.5°E and 38.0°N, 145.5°E by Fujio and Yanagimoto (2005). In addition to the moored current meter measurements, westward abyssal flows north of the KE jet were also evident in a lowered acoustic Doppler current profiler (ADCP) transect along 152.5°E reported by Yoshikawa et al. (2004).

While our knowledge of the deep flows in the western North Pacific is increasing, an overall picture regarding the time-mean circulation of the KE’s northern RG remains lacking. Unlike the RG south of the KE, there exists no surface expression revealing the presence of a northern RG in the KE system. For example, Fig. 1 shows the absolute mean surface dynamic height field determined by combining concurrent surface drifter, surface wind stress, and satellite altimeter data (Niiler et al. 2003). While a surface dynamic height plateau is clearly visible to the south of the KE jet centered along ∼32°E, the surface dynamic height values decrease monotonically poleward north of the KE jet. In other words, clarifying the existence of a mean northern RG requires long-term, subsurface velocity measurements over the broad transition region between the KE and the subarctic boundary.¹

A collaborative research program named the Kuroshio Extension System Study (KESS) was initiated in 2003. One of the objectives of KESS is to clarify the RG patterns as an integrated part of the Kuroshio Extension system. (Complete scientific objectives of KESS can be found online at http://www.uskess.org.) To achieve this objective, an array of moored profilers (MPs), current meters, pressure recording inverted echo sounders (PIES), and upward-looking ADCPs was deployed across the Kuroshio Extension jet from 32.5° to 37.0°N near 146°E over the period from June 2004 to May 2006. In addition, a total of 48 APEX profiling floats were deployed during the KESS deployment and turnaround cruises in May/June 2004 and 2005 surrounding the MP/CM/PIES/ADCP array. The profiling floats have a parking depth at 1500 m and measure temperature–salinity profiles every 5 days. In the present study, we describe the mean circulation pattern at the 1500-m depth inferred from the profiling-float drift trajectories. Results based on the moored MP/CM/PIES/ADCP data will be reported in a separate paper (S. Jayne and N. Hogg 2007, personal communication).

In addition to the description of the inferred northern RG, we further explore in this study its formation mechanism with the aid of a global eddy-resolving ocean general circulation model (OGCM). Regarding the formation of the RG north of the Gulf Stream, two competing mechanisms have been put forth. The first regards the unstable Gulf Stream jet as an eddy energy source, with the time-mean northern RG being forced by radiating eddy fluxes (e.g., Holland and Rhines 1980; Hogg 1988; Jayne et al. 1996; Nakamura and Chao 2000). The second proposed mechanism emphasizes the importance of the sloping bottom topography and the intruding deep western boundary current (DWBC); in this scenario, the northern RG is generated by the downslope DWBC south of the Great Banks through bottom vortex stretching (e.g., Greatbatch et al. 1991; Zhang and Vallis 2007).

It is worth emphasizing that the oceanic conditions north of the KE are different from those north of the Gulf Stream in several important ways. First, the bottom topography north of the KE jet is largely flat with a depth of ∼6000 m (see Fig. 1). Prominent topographic features in the region include the deep Japan Trench parallel to the east coast of Japan and the Shatsky Rise located along ∼158°E. These topographic features, as we will find in this study, do have a localized impact upon the middepth circulation. Second, no equivalent deep western boundary current, such as that driven by thermohaline processes in the subpolar North Atlantic, exists along the western boundary of the North Pacific Ocean. Third, rather than being confined to the north by a shallow continental shelf as in the North Atlantic, the region of the KE system is open to the north and is in direct contact with the western boundary current of the wind-driven subpolar gyre, that is, the Oyashio and its continuation, the subarctic boundary, and the Subarctic Front. Given these differences between the two current systems, a better description and exploration of the KE’s northern RG can also help improve our understanding of the formation mechanism of the Gulf Stream’s northern RG.

This paper is organized as follows. In section 2, we provide a detailed analysis of available profiling float data and describe the middepth mean circulation pattern around the KE jet. In section 3, we compare the observed mean circulation pattern with that obtained in an eddy-resolving OGCM output. The formation mechanism for the northern RG of the KE is examined in section 4 through momentum and potential vorticity budget analyses of the OGCM output. Section 5 summarizes our results from the present study.

2. Float-derived middepth mean circulation

During the KESS deployment cruise in May–June 2004, 20 APEX profiling floats were deployed in the KE region in order to capture the time-varying upper-ocean thermal structures and the flow pattern at the float parking depth of 1500 m. A second set of 28 APEX floats was deployed during the turnaround cruise in May–June 2005. Maps of the float release locations from the two cruises can be found in Fig. 1 of Qiu et al. (2006, 2007, respectively. A profiling KESS float measures temperature–salinity (T–S) at 72 prespecified pressure levels during its rise from the parking depth to the surface; the rise takes about 4.5 h. While at the sea surface, the float transmits the T–S data to the Advanced Research and Global Observation Satellite (ARGOS) system, and its position is “fixed” by ARGOS overpasses on an approximately hourly basis. On average, the float archives about seven position fixes during its 8.5 h on the surface. After diving to its parking depth, the float follows the 1500-m depth Lagrangian flow and it makes its next ascent to surface with a 5-day repeat cycle.

To derive the flow velocity at the parking depth, we follow Park et al. (2005) and utilize the float’s ARGOS position fixes. Specifically, we estimate the float’s diving and resurfacing locations by extrapolating the satellite position fixes by assuming a linear and inertial velocity model; the parking depth velocity is then obtained by dividing the distance between the diving and resurfacing locations by the time elapsed from the diving to resurfacing. There are two sources of error in this method of velocity estimation. The first is related to estimating the diving and resurfacing locations by extrapolating the float’s position fixes. By analyzing a subset of the KESS floats and by withholding part of the ARGOS position fix information, we find that the error of the extrapolation has a median value of 881 m (Chen et al. 2007). The second source of error comes from the upper-ocean velocity shear that causes the float to drift during its ascending and descending periods. The size of this velocity shear can be evaluated from the float drifts at the sea surface and the parking depth. The median value for the drift due to the velocity shear is estimated at 2096 m (Chen et al. 2007). Assuming the above two errors are uncorrelated, we find that the uncertainty in the parking depth velocity estimated during a single float cycle is on the order of 0.8 cm⁻¹.

As of November 2007, a total of 5831 dive–resurface position pairs are available from the KESS profiling floats. To improve the spatial and temporal coverage for the mean flow estimation, we further include in this study the profiling float data from the international Argo program in the region of our interest: 28°–45°N, 135°–165°E. As shown in Fig. 2, inclusion of the Argo profiling floats increases the overall number of float dive–resurface position pairs to 8813 over the 9-yr period since 1999. Notice that the parking depths of the Argo profiling floats vary from 1000 to 1500 to 2000 m. With the majority of floats drifting at the 1500-m depth, velocity values estimated from the floats drifting at other depths are adjusted to the 1500-m level by applying a correction based on climatological geostrophic shears calculated from the World Ocean Atlas 2001 (Conkright et al. 2002).

Figure 3 shows the individual velocity vectors at the 1500-m depth estimated from all available dive–resurface position pairs. In the figure, velocity vectors with a positive (negative) zonal component are indicated in red (blue). Despite the high level of eddy variability in the region, the presence of the eastward-flowing KE jet along ∼34°N, the generally westward-moving flows on its two flanks, and the eastward flow associated with the subarctic boundary along ∼40°N, can clearly be gleaned from the figure.

Using these available velocity estimates, we derive the time-mean flow pattern at the 1500-m depth by adopting the objective mapping technique of Bretherton et al. (1976). In mapping both u and υ fields, we assume that the covariance function is in the form of a distance-weighted Gaussian function: exp [−(Δx² + Δy²)/R²], where Δx and Δy are spatial distances in the x and y directions, and R is the spatial decorrelation scale. Based on the inferred velocity vectors shown in Fig. 3, the value for R is estimated to be ∼100 km. Figure 4 shows the resultant mean velocity field objectively mapped from the velocity vectors presented in Fig. 3. Although it represents a nearly 9-yr mean flow field, Fig. 4 is biased to the period of 2005–06 when float measurements were most abundant (recall Fig. 2).

The position of the KE jet at the 1500-m depth tracks the mean surface KE jet very well (cf. Fig. 1). In Fig. 4, the crest of the second quasi-stationary meander of the KE jet, which exists nominally near 148°E, appears relatively obscure. This obscurity is largely due to the fact that several intense, westward-propagating, cold-core eddies stalled south of the second quasi-stationary meander in 2005–06 (see Figs. 2 and 3 in Qiu et al. 2007), causing the local, float-derived KE path to have a southerly bias. Figure 4 shows that the KE path veers northward after encountering the Shatsky Rise near 158°E and that there is a broadening in the KE width downstream of the Shatsky Rise. Both of these features are discernible in the pattern of the surface KE jet shown in Fig. 1.

It is interesting to note that the subarctic boundary and the Subarctic Front have their respective middepth expressions in Fig. 4: the former corresponds to the eastward flow along ∼40°N and the latter, the northeastward flow along ∼43°N. While both the subarctic boundary and the Subarctic Front have been considered in the past to be shallow upper-ocean features, the result of Fig. 4 indicates that their associated mean flows can reach at least to the 1500-m depth. Notice that a narrow southward current exists against the 1500-m isobath east of Japan, which corresponds to the lower portion of the Oyashio (see Kono and Kawasaki 1997; Mitsudera et al. 2004). Offshore of this southward current exists a continuous northward flow that overrides the eastern flank of the sloping Japan Trench. While exhibiting little surface expression, this mean northward flow has been detected by moored current meter measurements and its existence has been regarded as a rectified along-slope current driven by the underlying Japan Trench slope (Owens and Warren 2001; Fujio and Yanagimoto 2005).

Persistent westward flows are seen in Fig. 4 to exist south of the KE jet. These westward flows constitute part of the KE’s southern RG and their existence is evident in the surface dynamic height field shown in Fig. 1 as well. In between the KE jet and the subarctic boundary, broadscale westward flows, corresponding to the KE’s northern recirculation, are clearly discernible in Fig. 4. Although abyssal westward flows have been observed at several locations in this region (see references cited in the introduction), Fig. 4 provides the first, time-mean flow pattern that delineates the horizontal extent of the northern RG in the KE system. Unlike its southern counterpart, the northern RG shows no expression in the mean surface dynamic height field.

3. Middepth circulation in the OFES simulation

To clarify the nature and formation mechanism of the northern RG in this study, we examine model results from a multidecadal hindcast run of the OGCM for the Earth Simulator (OFES). In this section, we explore the model’s realism in comparison with the float-inferred middepth circulation pattern. The representativeness of Fig. 4 as the long-term mean flow pattern at the 1500-m depth and the three-dimensional structure of the KE’s northern RG are investigated using the model’s output.

The OFES model covers the global domain from 75°S to 75°N; it has an eddy-resolving horizontal resolution of 0.1° and 54 vertical levels. The model code is based on Modular Ocean Model version 3 (MOM3), modified for optimal performance by the Earth Simulator of Japan (Sasaki et al. 2008). The model was spun up for 50 years with monthly climatological atmospheric forcing from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kistler et al. 2001). This climatological run was followed by a 55-yr hindcast integration for the period 1950–2004 using the NCEP–NCAR reanalysis daily-mean forcing data. In our present study, the model output from the 55-yr hindcast run was analyzed. For more details on the OFES model, such as its initial/boundary conditions and subgrid-scale eddy parameterizations, readers are referred to Nonaka et al. (2006).

Figure 5a shows the flow field at the 1500-m depth averaged from 2000 to 2004 using the OFES hindcast run. This 5-yr duration overlaps with the period of the profiling float measurements although, as we noted in section 2, the float-inferred flow field of Fig. 4 is likely biased to 2005–06. Despite this possible bias, the modeled flow pattern compares favorably to the observed flow field. All major zonal flows identified in Fig. 4, that is, the KE jet, its northern and southern westward recirculating flows, and the eastward flows associated with the subarctic boundary and the Subarctic Front) can be located in Fig. 5a. Along the 1500-m isobath east of Japan, the southward intruding Oyashio appears less clear in the model than in Fig. 4, although the OFES hindcast captures well the northward flow overriding the eastern flank of the sloping Japan Trench. Notice that the flow pattern appears smoother in Fig. 5a than Fig. 4; this is because the model result was based on the full 5-yr averaging, whereas Fig. 4 was based on the temporally and spatially irregular float measurements.

It is worth emphasizing that the mean flow pattern shown in Fig. 5a does not depend sensitively on the period over which the averaging is taken. Averaging the flow field over the entire 55-yr period from 1950 to 2004, for example, results in a flow pattern at the 1500-m depth (Fig. 5b) quite similar to that of Fig. 5a.² To illustrate this point further, we plot in Fig. 6 the 1500-m depth zonal velocity as a function of latitude and time along 150°E over the 55-yr hindcast period. While the instantaneous flows are clearly influenced by transient mesoscale eddy signals, all of the major zonal flows consistently appear once an interannual or longer time scale, averaging is taken. This model result is important as it suggests that, while the accumulation of the profiling float data in the coming years will undoubtedly improve the inferred middepth flow pattern, the overall picture presented in Fig. 4 is likely to remain unchanged.

To obtain a three-dimensional picture of the northern RG, we plot in Fig. 7 meridional sections of the mean zonal velocity along 145°E, 150°E, and 155°E, respectively, from the OFES hindcast run. The time averaging, as in Fig. 5a, is taken from 2000 to 2004. The 145°E section is located just east of the northward flow overriding the Japan Trench and is close to the western edge of the northern RG. Abyssal flows (below 2000 m) underneath the KE jet and the northern RG along this section are weak and fragmented. The volume transport associated with the northern RG across 145°E is 9.2 Sv (Sv ≡ 10⁶ m³ s ⁻¹; see the dashed line in Fig. 8). The northern RG, with the maximum westward flow centered on 38.5°N, becomes fully developed along 150°E. Compared to the upstream 145°E section, Fig. 7b shows that both the KE jet and the northern RG are more vertically coherent, or equivalent barotropic. The volume transport of the northern RG increases to 26.4 Sv across the 150°E section. The 155°E section is located west of the Shatsky Rise where the path of the mean KE jet veers northeastward (recall Figs. 4 and 5). Accompanying the broadening of the KE jet along this section, the volume transport of the northern RG drops to 20.9 Sv (see the gray line in Fig. 8). Downstream of the Shatsky Rise, the width of the KE jet broadens further and the northern RG becomes no longer well defined; the westward flow north of the KE along 160°E, for example, has a volume transport less than 5 Sv (figure not shown).

In concluding this section, we note that the net volume transport values across the 28°–45°N segment shown in Fig. 8 (e.g., 33 Sv across 150°E) agrees well with the Sverdrup transport values calculated from the wind stress curl data of the NCEP–NCAR reanalysis. This agreement indicates that the wind-driven circulation dominates the large-scale flows in the western North Pacific and that the enhanced eastward transport of the mean KE jet (e.g., 93 Sv across 150°E) is due to the presence of the neighboring southern and northern RGs.

4. Formation mechanism of the northern RG

With the OFES hindcast run simulating well the observed middepth circulation around the KE jet, we use the high-resolution model output in this section to explore the formation mechanism of the detected northern RG. Given the essentially flat regional bottom topography beneath the northern RG, we hypothesize that the northern RG is driven by eddy fluxes related to the unstable KE jet. The fact that the KE jet is a highly unstable system in the region of 140°–160°E has been investigated quite extensively. Readers interested in this topic are referred to Qiu and Chen (2005) and Taguchi et al. (2007) for references and recent analyses.

To keep this paper concise, we limit our presentation of the model analysis results to the isopycnal surface 27.575 σ_θ. Instead of an isobaric surface, an isopycnal surface is used in this section because the isopycnal coordinates annihilate the vorticity-generating solenoidal term and are a natural choice to conduct budget analysis of potential vorticity (Vallis 2006). In the region of our interest, the 27.575 σ_θ isopycnal surface has a depth ranging from 1300 to 1700 m and a mean value at 1500 m (see Fig. 9a). As we noted in section 3, the modeled mean flow pattern at this depth agrees well with the profiling float result and is representative of the flows below the main thermocline (Fig. 7). Because of the weak vertical shear, the mean velocity u on the 27.575 σ_θ surface from the 2000–04 OFES output looks very similar to that shown in Fig. 5a. Instead of showing this u field, we plot in Fig. 9b its associated geostrophic streamfunction field on the 27.575 σ_θ surface for the purposes of following discussions. In Fig. 9b, the northern RG corresponds well to the region north of the KE jet where < −1000 m² s⁻¹. Notice that due to the equivalent barotropic nature of the mean flow field, the spatial pattern of the field somewhat resembles the isopycnal depth field of Fig. 9a.

Figure 10 shows the mean potential vorticity (PV) field Q on the 27.575 σ_θ surface. Here instantaneous Q is defined by

View Expanded

where ζ denotes the relative vorticity, f_o the Coriolis parameter at the reference latitude (35°N here), β the latitudinal gradient of f, σ_θ the potential density, and 〈σ_θ〉 the potential density averaged in time and space (i.e., the domain of Fig. 10), respectively. Thus defined, 〈σ_θ〉 is a function of z only. The PV definition of Eq. (1) has been used by Kagimoto and Nakamura (2007, manuscript submitted to Dyn. Atmos. Oceans) in their study of quasigeostrophic transient wave activity and fluxes in the North Pacific Ocean basin. On the broad scale, Fig. 10 reveals that Q increases with the latitude due to the planetary vorticity βy. In the region surrounding the KE jet, the stretching vorticity term in Eq. (1) works to reduce the meridional PV gradient ∂Q/∂y and its effect is clearly identifiable in Fig. 10. The fact that ∂Q/∂y is reduced in the vicinity of the KE jet has been noted in the previous study of the PV distribution in the North Pacific by Talley (1988). Superimposed on Fig. 10 by the white lines are the geostrophic streamfunction contours shown in Fig. 9b. It is important to emphasize that the and Q contours are not fully parallel to each other, implying that the mean circulation on this isopycnal surface is not exactly inertial and that other physical processes, such as eddy forcing and dissipation, must play a role in the determination of the u field.

To quantify the effect of the eddy forcing, we first evaluate the convergence of the Reynolds stresses in the x-momentum equation:

View Expanded

that is, the rhs of Eq. (2) (see Fig. 11). Here an overbar denotes the mean value averaged from 2000 to 2004 and a prime denotes the deviation from the mean. As indicated in Eq. (2), a positive (negative) Reynolds stress convergence tends to enhance (weaken) the mean zonal flow. Although the result in Fig. 11 is somewhat patchy due to the meandering KE jet, a persistent picture is, nevertheless, detectable. Specifically, along the path of the KE jet (i.e., along the continuous = 0 contour around 35°N in Fig. 11), the Reynolds stress convergence is, by and large, positive, indicating the eastward-flowing KE jet is being accelerated by the lateral eddy momentum flux forcing. One exception occurs over the Shatsky Rise near 156°–159°E where the Reynolds stress convergence has a large negative value. Northwest of this negative Reynolds stress convergence region, a large positive Reynolds stress convergence area exists with its center at 37.5°N, 156°E. This area, as we noted in Fig. 5a, corresponds to where the KE jet bifurcates and forms its northern branch. The result of Fig. 11, thus, suggests that the eddy momentum forcing around the Shatsky Rise works to accelerate the northern branch of the KE jet and weakens, at the same time, the southern branch. This result regarding the importance of eddy forcing in generating the northern KE branch is consistent with the high-resolution model study by Hurlburt and Metzger (1998). On the two sides of the meandering KE jet, Fig. 11 reveals that the Reynolds stress convergence has largely negative values, indicating the laterally radiating eddy momentum fluxes work to accelerate the recirculating westward mean flows. This result, based on the model’s Reynolds stress analysis, thus supports the hypothesis that the northern RG is eddy driven.

It is also instructive to examine the eddy’s impact on the northern RG from the PV point of view. Specifically, we are interested in whether the mean northern RG could be understood as resulting from the turbulent Sverdrup balance theorized by Rhines and Holland (1979):

View Expanded

where the perturbation PV Q′, following Eq. (1), is given by

View Expanded

Using the geostrophic streamfunction , the lhs of Eq. (3) can also be written as J(, Q), where J is the Jacobian operator. Physically, Eq. (3) relates the convergence of eddy PV fluxes to the mean flow generation across the mean PV gradient. In Fig. 12, we compare the distributions of J(, Q), and −∇ · () on the 27.575 σ_θ surface. Despite the dominance by patchy small-scale features, a favorable correspondence between these two terms is visually discernible. To examine this correspondence in a more quantitative way, we present in Fig. 13 a scatterplot of J(, Q), versus −∇ · () inside the northern RG region of our interest (defined as the area where < 1000 m² s⁻¹ in Fig. 9b). The linear correlation coefficient between the two terms is 0.45. While confirming the general validity of the turbulent Sverdrup balance, the result of Fig. 13 also underlines the importance of the eddy dissipation term neglected in Eq. (3). Note that the linear correlation coefficient value at 0.45 between the J(, Q), and −∇ · () terms is typical for the other areas as well.

Like for the wind-driven Sverdrup transport streamfunction, the convergent eddy flux-driven mean flow streamfunction can be solved by integrating the rhs of Eq. (3) pseudowestward along constant Q contours:

View Expanded

where p is the arc length along the constant Q contour, rescaled according to ∂|x|/∂p = |∇Q| (Rhines and Holland 1979). Here the integration is taken pseudowestward because the large-scale ∇Q, as shown in Fig. 10, is directed northward. Figure 14 shows the field derived from Eq. (5) using the −∇ · () values shown in Fig. 12b and p = 0 along 165°E. Notice that by integrating along the constant Q contours, many of the small-scale features in the eddy PV flux convergence field are smoothed out and the resultant pattern exhibits broader spatial scales. Compared to the “true” field of Fig. 9b, the inferred field shown in Fig. 14 generally has the correct sign and this is particularly true in regions of the northern RG and the subarctic boundary. Overall, the inferred values have larger amplitudes than those shown in Fig. 9b, suggesting that the eddy-induced mean circulation is likely weakened by the dissipation processes neglected in the turbulent Sverdrup balance, Eq. (3). This result is consistent with the findings in Figs. 12 and 13 that J(, Q) and −∇ · () are not exactly balanced.

Finally, we note that the mean KE jet itself cannot be entirely explained by horizontal convergence of the eddy PV fluxes; the inferred streamfunction values along the mean path of the KE jet appear not well organized in Fig. 14. One possibility for this is that the mean KE jet at the 27.575 σ_θ level is not purely eddy driven but is also influenced by the large-scale wind forcing.

5. Summary

Over the past several years, a large number of profiling floats have been deployed in the western North Pacific Ocean as part of the Kuroshio Extension System Study (KESS) and the International Argo program. Using the drift information from these floats at their parking depths, we examined the regional mean circulation at the 1500-m depth. The float-inferred mean flow field confirmed several middepth circulation features that have been noted in previous observations: for example, the existence of the deep, meandering KE jet, its northeastward branching upstream of the Shatsky Rise, and the intense RG south of the KE jet. The float measurements also revealed new circulation features at the 1500-m depth. In particular, a well-defined RG is found to exist north of the KE jet: spatially, it spans from east of the Japan Trench at ∼145°E to west of the Shatsky Rise at ∼156°E and is confined to the north by the subarctic boundary along 40°N. Although westward-recirculating flows have been observed north of the KE jet in the past, the float results presented in this study provide the first observational picture of the spatial pattern of the mean northern RG.

The presence of the deep-reaching, eastward-flowing subarctic boundary and northeastward-flowing Subarctic Front is another new middepth circulation feature revealed by the float measurements. Both of these currents have been regarded heretofore as “shallow” circulation features in the western subarctic gyre of the North Pacific Ocean. In addition, a continuous northward flow is found to exist offshore of the southward-flowing Oyashio; it overrides the eastern flank of the Japan Trench and delineates the western end of the northern RG.

Notice that the linear Sverdrup dynamics predicts that the wind-driven, subtropical western boundary current (i.e., the Kuroshio) should separate from the coast of Japan at ∼42°N. That the actual separation of the KE jet occurs at ∼35°N is a direct consequence of the presence of the northern RG. By causing the KE jet to separate at a southerly latitude, the northern RG creates an intergyre area, commonly known as the “Mixed Water Region,” within which the subtropical- and subarctic-origin waters are mixed and transformed (e.g., Talley 1993; Yasuda et al. 1996).

With the bottom topography underlying the KE’s northern RG essentially flat, we hypothesized that the northern RG is driven by eddy fluxes associated with the time-varying KE jet. To test this forcing mechanism, we analyzed the output from the OGCM for the Earth Simulator (OFES) hindcast run. The OFES hindcast model is eddy resolving and it simulates well the middepth circulation features inferred from the float measurements. An examination of the time-mean x-momentum balance in the OFES hindcast run revealed that the Reynolds stress convergence worked to accelerate westward zonal mean flows on the two flanks of the meandering KE jet, consistent with our hypothesis that the northern RG is driven by radiating eddy fluxes. While the calculation of the Reynolds stress convergence helps determine the direction of the accelerating mean flow, it provides little information on the strength of the resulting mean circulation. To evaluate this latter quantity, we evaluated the horizontal convergence of eddy PV fluxes based on the OFES hindcast run. By applying the turbulent Sverdrup balance equation, we found that the convergent eddy PV fluxes alone tended to overestimate the resultant northern RG and that the eddy dissipation processes are likely to play a role in controlling the strength of the spunup northern RG.

In the OFES hindcast run, the KE’s northern RG has a maximum volume transport of 26.4 Sv near 150°E and its presence as an equivalent-barotropic gyre appears to persist on the interannual and longer time scales. As the dynamic state of the KE system undergoes clear stable and unstable cycles (e.g., Qiu and Chen 2005; Taguchi et al. 2007), it will be interesting in future analyses of the profiling float data to clarify the time-varying nature of the northern RG.