High-Resolution Modeling Study of the Kuroshio Path Variations South of Japan

The Kuroshio Current, characterized by high eddy activity, is the western boundary current of the North Pacific subtropical gyre. After coming through Tokara Strait (see Fig. 1) from the East China Sea (ECS), the Kuroshio joins the Ryukyu Current, which flows northward along the east side of the Ryukyu Islands, and heads eastward along the southern coast of Japan.

Figure 1.  Typical paths of the Kuroshio south of Japan (LM, large meander; nNLM, nearshore nonlarge meander; oNLM, offshore nonlarge meander). Solid and dashed lines denote the axes of Kuroshio pathways in AVISO altimetry data (with LM in Jan 2005, nNLM in Nov 2005, and oNLM in Sep 2007) and the model output (with LM in Sep of the model year 28, nNLM in Nov of the model year 34, and oNLM in Apr of the model year 34), respectively, which are both defined as the 110-cm sea SSH isoline. Line PN, OK and ASUKA denote the repeat hydrographic sections.

A unique feature of the Kuroshio south of Japan is its well-known bimodal path fluctuations, switching between a meandering and a relatively straight one (Fig. 1); namely, the large meander (LM) and non-large meander (NLM) paths (e.g., Taft, 1972). (Kawabe, 1986) further divided the NLM path into two categories——the nearshore non-large meander (nNLM) path and the offshore non-large meander (oNLM) path, with the latter looping southward over the Izu Ridge (near 140^°E——see Fig. 1). Both the LM and the NLM states can persist long-range from a few years to a decade, while the transition process between the two paths takes place in only a few months (Kawabe, 1986). Owing to its peculiar oceanographic features and interannual variability, the Kuroshio path variation south of Japan has been the subject of extensive interest for the past several decades.

Previous studies have suggested the bimodal feature of the Kuroshio south of Japan is a forced phenomenon that depends sensitively on the upstream transport of Kuroshio (e.g., Kawabe, 1980; Chao and McCreary, 1982; Yoon and Yasuda, 1987; Kawabe, 1995; Akitomo et al., 1996). According to a theoretical study (White and McCreary, 1976), the LM can be understood as a stationary Rossby lee wave, and the wavelength is scaled by , where U is the velocity of the Kuroshio and β is the meridional gradient of the Coriolis parameter. This result shows that the zonal scale of the LM could be related to the velocity (or transport) of the upstream Kuroshio. Using sea level data from tide gauges, (Kawabe, 1995) suggested that the LM would occur only when the upstream Kuroshio has large or medium transport, and the NLM path will dominate when the transport is less than 23.5 Sv (10⁶ m³ s^-1). However, in contrast, (Qiu and Miao, 2000) suggested that the path oscillation since 1975 could be explained by a self-sustained intrinsic oscillation system involving the Kuroshio and its southern recirculation, based on a two-layer primitive-equation model.

Recently, high-resolution ocean general circulation model (OGCM) simulations have enabled us to investigate the variability of the Kuroshio path in detail either through case studies (e.g., Endoh and Hibiya, 2001; Miyazawa et al., 2004; Tsujino et al., 2006; Miyazawa et al., 2008; Usui et al., 2008; Endoh and Hibiya, 2009; Usui et al., 2011) or long-term analysis (e.g., Douglass et al., 2012; (Tsujino et al., 2013)). The local dynamics of formation of the LM are discussed extensively in these studies, including the formation and propagation of a trigger meander (e.g., Endoh and Hibiya, 2001; Usui et al., 2008), the effect of bottom topography and deep anticyclone eddies (e.g., Tsujino et al., 2006; Douglass et al., 2012), the role of eddy-Kuroshio interaction (e.g., Miyazawa et al., 2004), or combinations of these factors. In addition to the local dynamics, (Tsujino et al., 2013) revealed the effects of large-scale wind forcing on the bimodality of the Kuroshio path by changing the Kuroshio Current transport, and highlighted the influences of eddies from the recirculation gyre.

Although both local and remote dynamics have been investigated in previous studies, the mechanisms governing the bimodality of the Kuroshio path still remain elusive due to limited observations and a lack of comparative high-resolution OGCM studies. The purpose of this study is to re-examine the possible mechanisms involved in the formation of the LM, using a (1/10)^° OGCM. The model reasonably reproduces the typical paths of the Kuroshio south of Japan, as observed in the altimeter, as well as the interannual variability from the NLM to the LM. Both the local dynamics and remote influence are examined.

The remainder of the paper is organized as follows. The model used in the study is described in section 2. Section 3 presents the results of the OGCM, and both the mean state and the bimodal variability of the Kuroshio system south of Japan are discussed. Possible mechanisms involved in the formation of the LM are examined in section 4. Finally, section 5 is devoted to a summary and conclusions.

The eddy-resolving model used for this study is a version of the Bryan-Cox-Semtner model (Bryan, 1969; Semtner, 1974; Cox, 1984) developed at the University of Wisconsin-Madison (Jacob, 1997; Tobis et al., 1997). The dynamics of this model are based on the Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model Version 2 (MOM2) (Pacanowski, 1995), with the split-explicit free surface formulation (Killworth et al., 1991). A model performance speed-up is achieved by implementing the gravity wave retardation method with little impact on the internal dynamics (Jensen, 1996; Tobis, 1996). The bathymetry is generated by interpolating the 2-Minute Gridded Global Relief Data (ETOPO2) database. We follow the protocol used to force ocean climate models proposed by (Griffies et al., 2009), and the model is initialized from Polar Science Center Hydrographic Climatology (PHC) with zero velocity. Within three degrees on the northern and southern boundaries of the model domain, the temperature and salinity are restored to the seasonal climatology of PHC 3.0, with the restoring timescale increasing inward linearly from 1 day to 720 days (Marchesiello et al., 2001).

The quasi-global configuration of the model covers a region from 75^°S to 75^°N, with a 0.1^° grid spacing able to reproduce oceanic mesoscale features such as eddies. The model has 50 vertical levels, and a water depth from 5 m to 5500 m.

The surface fluxes of wind stress, heat and freshwater used to drive the model are calculated from the Coordinated Ocean-ice Reference Experiment forcing version 2 (CORE.v2; Large and Yeager, 2009) according to the bulk formulae presented by (Large and Yeager, 2004). The model is run globally for 39 years due to a limitation of available computer time. Year 1-25 is considered as the spinup period of the model, and years 26-39 are forced by monthly historical forcing from 1950 to 1963. It should be noted that we focus on the physical mechanisms of the LM in the model rather than a thorough comparison with observations.

Figure 2a shows the mean sea surface height (SSH) field of the model during the simulation period. The model succeeds in reproducing the spatial pattern of the Kuroshio and Kuroshio Extension (KE) system, with a reasonable recirculation gyre in the Shikoku Basin and two quasi-stationary meanders [with ridges located at 144^° and 150^°E, as indicated by (Qiu and Chen, 2005)] after separating from the Japan coast at around 36^°N. The annual mean transport of the modeled Kuroshio (above 700 m) across the PN line (Fig. 1) and that of the Ryukyu Current (above 1000 m) across the OK line are about 23.5 and 16.0 Sv, respectively, which are comparable to previous observations (Kawabe, 1995; Andres et al., 2008). Across the ASUKA (Affiliated Surveys of the Kuroshio off Cape Ashizuri) line south of Japan (Fig. 1), the volume transport of the Kuroshio eastward current has been estimated as ranging from 29.0 to 40.2 Sv among three synoptic hydrographic surveys (Nagano et al., 2010). In the present high-resolution model, the annual mean transport across the ASUKA line is 38.5 Sv, which is also consistent with the observed value. In this sense, the present model can resolve the Kuroshio system by and large.

Figure 2.  Mean SSH (units: cm) for (a) the model, and (b) AVISO. Contour interval is 10 cm. Thick black lines denote the 110-cm SSH isoline, which is regarded as the Kuroshio axis. The white box in (a) indicates the averaged region for relative vorticity.

The Kuroshio path fluctuates significantly downstream of the Tokara Strait. A quasi-periodic meandering state has been described using sea level derived from tide gauges since the 1960s (Kawabe, 1985). As shown in Fig. 1, three typical paths of the Kuroshio south of Japan in the model (dashed line) are presented in comparison with that observed by altimeter data (solid line). The altimeter products are collected by the Data Unification and Altimeter Combination System (DUACS) and distributed by Archiving Validation and Interpretation of Satellite Data in Oceanography (AVISO)/the Centre National d'Etudes Spatiales of France. For both the altimeter data and the model output, the Kuroshio axis south of Japan is defined as the 110-cm SSH isoline. As indicated by the thick line in Fig. 2a, the 110-cm SSH contour is consistently located at the maxima of the meridional gradient of the SSH, and can be recognized as a good indicator for the Kuroshio axis (Qiu and Chen, 2005).

It should be noted that for the past two decades during which the altimeter data are available, only one stable LM case lasting for about 1 year has been observed, with its first appearance in July 2004 [for details, readers are referred to (Miyazawa et al., 2008) and Usui et al. (2008, 2011)], which seems to be much shorter and weaker than historical LM cases (e.g., Kawabe, 1995). In the present model, like most reproductions of the LM in other eddy-resolving models (e.g., Maltrud and McClean, 2005; Douglass et al., 2012; Tsujino et al., 2013), the spatial scale of the LM is much larger than the 2004-05 LM case (see Fig. 1).

Figure 3.  Yearly paths of the Kuroshio defined by the 110-cm contours in the monthly SSH fields during the simulation period.

It is noted that the shape of the LM evolves progressively over time along the Kuroshio axis. For a better view of the simulated LM, we plot the yearly paths of the Kuroshio south of Japan in Fig. 3. At the stage of formation (year 28), the initial shape of the LM looks like that observed in the altimeter data, which would not last for the whole lifetime of the LM. After that (year 30), the modeled LM separates from the coastline of Japan at the Tokara Strait south of Kyushu. This process has also been reported by many eddy-resolving models (e.g., Maltrud and McClean, 2005; Douglass et al., 2012; Tsujino et al., 2013), and (Tsujino et al., 2006) stated that this behavior is unrealistic and should be attributed to model bias. In their recent work, (Tsujino et al., 2013) suggested that this can be solved by considering the wind speed as relative to the surface current in evaluating surface stress, which will reduce the model eddy kinetic energy (EKE). Since the observed 2004-05 LM case only lasted for about 1 year, it is still unknown whether this phenomenon would happen in a realistic longer-lived LM.

Regarding the time scale of the Kuroshio path variations, it is worth mentioning that most previous eddy-resolving models are unable to simulate the Kuroshio path variations with realistic time scale even under realistic historical forcing (e.g., Miyazawa et al., 2004). Using sea level data from tide gauges, (Kawabe, 1995) reported five LM cases with irregular interannual variations during the 1975-91 LM-dominant period. (Qiu and Joyce, 1992) stated that the persistence of the LM varies from 1 year to 5 or 6 years. In subsequent work, (Qiu and Miao, 2000) modeled the interannual variations of the path oscillation with a preferred timescale of 4 years. In the present model, as shown in Fig. 3, during the 14-yr simulated period, the model reproduces the transition from the NLM to the LM three times with a primary LM case developing in year 28 and lasting for about 6 years (LM1). This time scale is comparable to one significant historical LM case which occurred during the period 1975-80 (Kawabe, 1995). This 6-year-long LM ceases at the beginning of year 34 and the Kuroshio stays in a straight state for nearly 3 years before the formation of the second LM. Therefore, the modeled Kuroshio path variability shows evident interannual changes and it would be meaningful to look into the lifecycle of the simulated 6-year-long LM case.

A general criterion for evaluating the Kuroshio path variations south of Japan is the path length. When the Kuroshio is in its LM state, the path length is much longer than that of the NLM path. Figure 4 shows the path length between 131^°E and 141^°E during the modeled time, with three LM periods indicated by the dashed black lines. We can see that the simulated 14 years are nearly dominated by the LM state, which explains why the mean Kuroshio shown in Fig. 2a takes the meandering path. To show how the modeled LM evolves, the lifecycle of the meander event beginning in year 27 is shown by a sequence of SSH fields (Fig. 5). Figure 5a shows the Kuroshio in its oNLM path state, with the main stream looping southward over the Izu Ridge. Figure 5b shows the Kuroshio becomes meridionally elongated prior to the formation of the LM. Then, the northward stream steadily switches to the western flank of the Izu Ridge and the LM emerges (Figs. 5c-e). Unlike the observed 2004-05 LM, which developed from a small meander south of Kyushu (e.g., Miyazawa et al., 2008; Usui et al., 2008), the present model presents a similar formation process to that simulated by (Qiu and Miao, 2000). The 6-year-long LM decays together with the southwestward movement of the southern anticyclonic eddy (Figs. 5f and g) and then the Kuroshio path returns to its typical NLM state (Fig. 5h).

Figure 4.  Time series of the simulated Kuroshio path length south of Japan. The path length is defined as the distance between 131^°E and 141^°E along the Kuroshio axis. The thick line is smoothed over 1 year (12-month running mean) and the dashed lines below denote the LM periods.

Figure 5.  Lifecycle of the meander event that developed in year 27 indicated by the sequence of the modeled SSH field. The contour interval is 10 cm and regions with SSH <110 cm are shaded.

Although the overall performance of the simulation seems to be good, especially in the region south of Japan, discrepancies between the model and the observations nevertheless exist. As shown in Fig. 2a, after separating from the Japan coast at around 36^°N, the mean upstream KE jet develops a larger-amplitude meander than that observed in the altimeter data (Fig. 2b). With a closer look at the yearly path of the Kuroshio shown in Fig. 3, we find that when the Kuroshio is in its LM state, the upstream KE jet tends to shift northward more often than that of NLM state. Using SSH data from multiple satellite altimeters, (Qiu and Chen, 2005) pointed out that the meridional path change of the KE jet is independent of the bimodal Kuroshio path variability south of Japan. Therefore, it seems that the present model may not show good skill in simulating the amplitude of the upstream KE jet migration, but this aspect is beyond the scope of the following analysis. Another shortcoming is that the decaying process of the 6-year-long LM is not an observed feature of the 2004-05 LM case (e.g., Usui et al., 2011). Although a significant impact of anticyclonic eddies on the formation of the LM has been clarified by previous studies (e.g., Mitsudera et al., 2001; Waseda et al., 2003), the present model indicates that movement of anticyclonic eddies might also play a role in weakening the meander. In this case, the focus of our study is on identifying possible conditions for the development of the LM, more than on its recession. Despite the problems with our model results, we believe that the present model captures the essential characteristics of the Kuroshio system and that the Kuroshio path variability is adequately simulated in the region south of Japan.

In the self-sustained oscillation system proposed by (Qiu and Miao, 2000), they concluded that the accumulation of low-PV (potential vorticity) anomalies intensifies the offshore recirculation gyre south of Japan and forces the Kuroshio to flow along the Japan coast when the Kuroshio is in its straight path state. And this process could increase the vertical velocity shear and further lead to baroclinic instability of the straight-path Kuroshio, which causes the development of the LM.

In the present study, following (Qiu and Miao, 2000), the relative vorticity, ζ=(? v/? x-? u/? y), is averaged in the Kuroshio region of (25^°-35^°N, 132^°-140^°E) (boxed region in Fig. 2a), where the horizontal and meridional velocity for the calculation is averaged vertically over the upper 500 m. As shown in Fig. 6a, the relative vorticity is calculated over the above area and a decreasing trend is found when the Kuroshio is in its straight path state, indicating the accumulation of low-PV anomalies south of Japan. While in the LM state, the area-averaged ζ is relatively stable due to the presence of a cyclonic eddy north of the Kuroshio axis.

Following the idea of the self-sustained oscillation mechanism, we then calculate the strength of the Shikoku recirculation gyre (SRG) and the vertical shear of the alongshore Kuroshio, the results of which are shown in Figs. 6b and c, respectively. Here, we should distinguish the SRG from the anticyclonic eddy south of the Kuroshio. The SRG is associated with the basin-scale circulation and is much larger than the anticyclonic eddy (≈200 km in diameter for the velocity core) (Mitsudera et al., 2001). To evaluate the strength of the SRG, we follow (Qiu and Chen, 2005) to define:

where A denotes the area of SSH >150 cm in the boxed region, where the relative vorticity is calculated (Fig. 2a). As shown in Fig. 2a, the 150-cm isoline matches the outer boundaries of the SRG well. Besides, the vertical shear of the alongshore Kuroshio is defined as U_0-200-U_200-600, where U_0-200 and U_200-600 denote the vertically-averaged alongshore Kuroshio velocity over the upper 200 m and 200-600 m, respectively. The SRG strength increases continuously when the Kuroshio is in its straight path state, corresponding to the decreasing trend of the averaged relative vorticity. While the increasing trend of the vertical shear of mean velocity is not obtained as in (Qiu and Miao, 2000) when the Kuroshio is in NLM state, the occurrences of the LMs do accompany a sharp decrease in the velocity shear, which agrees with their results.

By comparing the SRG strength with velocity shear (Fig. 6), at the formation of the LMs, it is readily concluded that the sudden release of velocity shear corresponds well with the weakening of the SRG. It seems that the weakening of the recirculation gyre could help the fully-charged velocity shear to break down; that is, through baroclinic instability.

Figure 6.  (a) Time series of relative vorticity, ζ, averaged in the Kuroshio region of (25^°-35^°N, 132^°-140^°E), where horizontal and meridional velocity for the calculation is vertically averaged in the upper 500 m. (b) Time series of the southern recirculation strength in the region same as ζ. (c) Time series of the vertical shear of the alongshore Kuroshio, U_0-200-U_200-600, where U_0-200 and U_200-600 denote the vertically averaged alongshore Kuroshio velocity of the upper 200 m and 200-600 m, respectively; denotes the spatial average in the region (30^°-35^°N, 132^°-140^°E). For all three time series, the thick black lines are smoothed over 12 months. LM periods are shaded in the three panels.

Previous studies have emphasized the importance of baroclinic instability on the formation of the LM (e.g., Yoon and Yasuda, 1987; Hurlburt et al., 1996; Endoh and Hibiya, 2001; Tsujino et al., 2006). A useful tool to evaluate the flow instabilities is eddy energetics analysis. However, due to the great fluctuations of the Kuroshio path south of Japan, it is not easy to define the background flow field in a satisfactory way. (Tsujino et al., 2006) proposed decomposing the instantaneous field into the field at the previous time interval (mean field) and the deviation from it (eddy field). Here, we adopt the same method in the monthly model outputs and analyze the local energetics.

Following (Tsujino et al., 2006), the barotropic (BT) conversion rate from the mean kinetic energy (MKE) to eddy kinetic energy (EKE), indicating the barotropic instability, is defined as

where ρ₀ is the typical value of the seawater density under the Boussinesq approximation, is the eastward (northward) velocity in the mean field, and u'(v') is the eastward (northward) perturbed velocity in the eddy field.

The baroclinic (BC) conversion rate from mean potential energy (MPE) to eddy potential energy (EPE), indicating the baroclinic instability, is defined as:

where g is the gravity acceleration, u'=(u',v',0) is the perturbed field of a geostrophic flow, ∇ is a horizontal gradient operator, is the depth-dependent background density field, and δρ(x,y,z,t) is the weakly varying density field, with the overbar and prime denoting the mean and eddy component, respectively.

Besides, considering that the rotational component of the eddy buoyancy flux () plays no role in converting energy between mean and eddy fields (Tsujino et al., 2006), the dynamically baroclinic (dynBC) conversion rate, which only takes the divergent component of the eddy buoyancy flux in the BC conversion rate, is also calculated due to its dynamical importance:

For a detailed discussion on the calculation of energy conversion rates, readers are referred to Tsujino et al. (2006, Appendix B).

The three conversion rates (200 m) at the formation of the modeled 6-year-long LM are compared in Fig. 7. It is seen that both the BT and the BC conversion rates show high positive values when the northward Kuroshio stream switches to the western flank of Izu Ridge and the LM emerges, indicating the important impact of flow instabilities on the formation process. As expected, both the BC and the dynBC show a much more significant signal than the BT, suggesting that the baroclinic instability bears greater responsibility for the growth of the LM than the barotropic instability. Another feature is that the large positive values of the BC and the dynBC are located between the Japan coast and the Kuroshio axis, where a cyclonic eddy accompanying the LM forms.

Figure 7.  Energy conversion rates (shading; units: kg m^-1 s^-3) at 200 m on the formation of the 6-year-long LM for (a) BT, (b) BC, and (c) dynBC. Contours denote the potential density at 200 m (contour interval = 0.15 kg m^-3).

In this case, we calculate the volume-integrated energy conversion rates over the area of SSH <40 cm in the boxed Kuroshio region of (25^°-35^°N, 132^°-140^°E) (Fig. 8). A sharp increase in all these three conversion rates can be seen before the onset of the LM, indicating energy transfer from the mean to the eddy field. Comparing with the vertical shear of the alongshore Kuroshio shown in Fig. 6c, we find that such an increase in conversion rates corresponds well to the sudden breakdown of the velocity shear. By averaging the volume-integrated conversion rates over five months before the onset of each LM event (two months for LM3), it is shown that both the BC and the dynBC are at least twice as much as the BT (Table 1). Thus, the above analyses confirm that baroclinic instability is mainly responsible for the formation of the LM, which agrees with previous studies (e.g., Endoh and Hibiya, 2001; Miyazawa et al., 2004).

Figure 8.  Time series of three conversion rates volume-integrated (upper 1000 m) over the area where SSH <40 cm in the boxed Kuroshio region of (25^°-35^°N, 132^°-140^°E). LM periods are shaded.

Relations between the upstream transport and the bimodality south of Japan have been discussed extensively, based on both observational data (e.g., Kawabe, 1980, 1995; Qiu and Miao, 2000) and modeling studies (e.g., Chao and McCreary, 1982; Yoon and Yasuda, 1987; Akitomo et al., 1996; Douglass et al., 2012). Here, we also re-examine the possible interaction between the upstream transport and the path variability.

Figure 9 shows the volume transport across the Tokara Strait during the 14-yr period. Although each of the three LM cases accompanies a transport higher than 23.5 Sv (Kawabe, 1995), the transport decreases several months before the LM forms when the meander gradually develops, indicating that the generation of the LM may not be explained by the increase in upstream transport. However, by comparing the SRG strength to the transport through Tokara Strait, we find an interesting correlation between them (1-yr smoothed, lagged correlation = 0.65), with the recirculation strength leading the transport by 7 months. Such correspondence is much more significant when the Kuroshio takes the meandering path and the SRG stays on the southwest of the main body of the LM, suggesting an impact of downstream (south of Japan) path variability on the temporal change of upstream (ECS) volume transport. Recently, using the Parallel Ocean Program (POP), (Douglass et al., 2012) also showed a similar phenomenon and suggested that the increase of transport through the Tokara Strait might be a response to the shift of the LM state. They stated that during the LM, the SRG could force the transport through the Kerama Gap (shown in Fig. 1) and further increase the Kuroshio transport in ECS, owing to its southwestward position compared with that in the NLM state.

Figure 9.  Time series of the SRG strength and volume transport through the Tokara Strait.Thick gray lines denote the LM periods.

Considering that the SRG also plays an essential role in modulating the Kuroshio path variability south of Japan, it seems that the present model indicates a regional oscillation system that includes the SRG, volume transport of the Kuroshio in ECS and the Kuroshio path bimodality south of Japan. The potential interactions among the three elements involved in the regional system are worth considering for future studies.

The above analysis points to the important role of the SRG strength in modulating the regional transport and path variations of the Kuroshio. But what causes the temporal change of the SRG strength? As an inseparable part of the Kuroshio system, the SRG arises from the accumulation of low-PV anomalies carried northward by the western boundary current, indicated by both theoretical and modeling studies (e.g., Cessi et al., 1987; Ierley and Young, 1988; Liu, 1997).

To determine what controls the temporal variation of SRG, we show the sea surface height anomaly (SSHA) along 29^°N in Fig. 10b, with the dashed black line denoting the westward propagating speed of 0.048 m s^-1, which is comparable to the phase speed of first-mode baroclinic Rossby waves estimated by the T/P SSH data (Qiu, 2003). Large anticyclonic disturbances with positive SSHA can be found before the formation of the LMs (green lines in Fig. 10b). These positive SSHAs can be traced back to the eastern North Pacific basin (160^°-140^°W), and were presumably induced by negative wind stress curl anomalies in the central to eastern North Pacific basin, propagating westward as baroclinic Rossby waves. By comparing with the SRG strength shown in Fig. 10a, we find that these westward-propagating disturbances correlate well with the variability of the SRG strength. During the NLM state, the anticyclonic disturbances with low-PV water accumulate west of the Izu Ridge (near 140^°E, see Fig. 1), strengthening the offshore SRG. Onsets of the LMs are accompanied by a shift from anticyclonic disturbances to cyclonic ones. As discussed in section 4.1, weakening of the recirculation gyre would help the velocity shear to break down and cause the LM. In this case, the westward-propagating negative SSHAs might account for the weakening of the SRG before the formation of the LMs, and act as a remote trigger for the baroclinic instability of the Kuroshio south of Japan. It is worth mentioning that for the observed 2004-05 LM event, (Usui et al., 2008) also reported remarkable mid-latitude westward-propagating negative SSHAs that contribute substantially to the formation of the LM.

Figure 10.  (a) Time series of the 1-yr smoothed SRG strength (RS); for comparison with (b), it is plotted as the deviation from the mean RS value. (b) Hovmöller diagram of SSH anomaly along 29^°N. The dashed black line denotes the westward-propagating speed of 0.048 m s^-1, and green lines denote the LM periods.

(Tsujino et al., 2013) also mentioned the remote forcing conditions as potential factors that can affect the formation of the LM. They suggested that the wind-induced low-latitude positive SSHA could be advected by the Kuroshio in the ECS, causing an increase of the Kuroshio transport, which promotes the formation of the LM. In comparison, our analysis shows that, being independent of the upstream Kuroshio transport, the remote disturbances caused by large-scale wind in the mid-latitudes may directly influence the SRG strength, stimulating the variability of the Kuroshio path south of Japan.

In this study, a high-resolution OGCM with realistic bathymetry was used to model the Kuroshio path variations south of Japan. It succeeds in reproducing many important aspects of the Kuroshio system with its interannual bimodal variability south of Japan. During the 14-yr simulation, the model reproduces the transition from the NLM to LM three times with a primary LM that lasts for about 6 years, which is comparable to an historical LM case (1975-80). Possible mechanisms for the development of the LM have been examined in this paper.

Following the idea of the self-sustained oscillation mechanism, we first examined the Kuroshio path variability associated with the local relative vorticity, the SRG strength, and the vertical velocity shear of the alongshore Kuroshio. The spatial averaged relative vorticity shows a decreasing trend when the Kuroshio takes the NLM path, indicating the accumulation of low-PV anomalies south of Japan. During the transition period from the NLM to LM, a sudden release of velocity shear corresponds well to the weakening of the SRG, which plays a key role in modulating the Kuroshio path variations. Analysis of eddy energetics indicated that both the BC and the dynBC show a much more significant signal than the BT, suggesting that baroclinic instability bears greater responsibility for the growth of the LM than the barotropic instability. A sharp increase in the volume-integrated conversion rates can be seen before the onset of the LM, which also corresponds well to the sudden breakdown of the alongshore velocity shear. This study confirms the previous conclusion that baroclinic instability is mainly responsible for the formation of the LM.

Relations between the upstream transport and the Kuroshio path variability south of Japan have also been examined. Despite the temporal limit of the present study, an interesting connection was found between the SRG strength and the transport through the Tokara Strait, with the recirculation strength leading the transport by 7 months.

The simulation also indicated a possible mid-latitude remote forcing that could directly influence the strength of the SRG. The large anticyclonic (cyclonic) disturbances with positive (negative) SSHAs, propagating westward as baroclinic Rossby waves adjustment to the wind stress curl anomalies in the North Pacific basin, correlate well with the variability of the SRG strength. In addition, onsets of the LMs accompany the switch from anticyclonic disturbances to cyclonic ones. Thus, we propose that the negative SSHAs might account for the weakening of the SRG before the formation of the LMs, and act as a remote trigger for the baroclinic instability of the Kuroshio.

Overall, the present model under realistic historical wind forcing reproduces many important features of the Kuroshio system. In the region south of Japan, our analysis emphasizes both local and remote roles in modulating the flow instability and regional oceanic variability. Although disparities exist between the modeled LM and the 2004-05 LM case, the present study might provide some insight into the intrinsic oceanographic mechanisms and the effect of large-scale wind forcing to better understand the Kuroshio variability south of Japan.