We investigate the Pacific thermohaline circulation induced by the locally enhanced vertical mixing around the Kuril Straits by using a general circulation model. When the strong vertical mixing reaches the sea surface, localized upwelling is induced from the depth to the surface. On the contrary, in the cases where the strong vertical mixing does not reach the sea surface, downwelling of surface water is induced in the shallow layer while upwelling of deep water is induced below. The structure of the meridional overturning circulation in the Pacific Ocean significantly differs depending on whether the surface buoyancy flux is large enough to sustain upwelling of deep water or not. The mixing-induced downwelling of shallow water is not associated with the sea surface cooling, and the source of dense water is in the deep ocean. Among the non-surface-reaching mixing cases, the depth that separates the downwelling and upwelling is found to be relatively insensitive to the depth where the vertical diffusivity sharply changes.
 One of the most important mechanisms controlling the deep Pacific circulation is the vertical turbulent mixing [Munk and Wunsch, 1998]. Many previous studies concluded that the vertical mixing maintains the upwelling of deep water and affects the thermohaline circulation [e.g., Munk, 1966; Bryan, 1987]. The vertical profile of vertical diffusivity has significant effects on the thermohaline circulation [e.g., Gargett, 1984; Bryan and Lewis, 1979]. Tsujino et al.  demonstrated its importance in the deep Pacific circulation by using an ocean general circulation model (OGCM) with horizontally uniform vertical diffusivity. Many observational studies have further suggested that the intensity of vertical mixing is highly variable and larger near boundaries than in the open ocean [Polzin et al., 1997; Moum et al., 2002; Nash et al., 2004]. Several modeling studies have investigated the role of locally enhanced mixing near the rough bottom in deep ocean flows [Hasumi and Suginohara, 1999; Simmons et al., 2004]. These studies concluded that the local enhancement of vertical mixing has a considerable effect on the deep Pacific Ocean circulation.
 Several regions have been reported to exhibit localized strong vertical mixing in the Pacific Ocean. For instance, the Aleutian Archipelago, the Kuril Straits, the Luzon Straits, the Indonesian Archipelago, and the Hawaiian Ridge are known as “mixing hot spots” [e.g., Egbert and Ray, 2000; Niwa and Hibiya, 2001]. The enhanced mixing around the Kuril Straits is one of the best studied among them [Kitani, 1973; Talley, 1991; Nakamura et al., 2000a]. Since the K1 tide is subinertial around the latitude of the Kuril Straits (∼47°N) and topographically trapped waves are generated and amplified while circulating around the Kuril Islands [Tanaka et al., 2010], almost all of the tidal energy is locally dissipated to efficiently induce vertical mixing.
Nakamura et al. [2006a, 2006b] investigated the influence of the locally enhanced vertical mixing around the Kuril Straits on the formation of North Pacific Intermediate Water by using an OGCM. On the basis of the results of tide models [Nakamura et al., 2000a; Nakamura and Awaji, 2004] they assumed a value of 200 × 10−4 m2 s−1 for the localized high vertical diffusivity. Localized upwelling is formed around the Kuril Straits and it accompanies southward and northward transport of 2.5 Sv (1 Sv = 106 m3 s−1) in the North Pacific intermediate and deep layers, respectively. This magnitude of deep northward transport would account for a significant part of the deep-water inflow from the Southern Ocean toward the North Pacific Ocean, which is estimated to be ∼10 Sv by observations [Roemmich et al., 1996; Rudnick, 1997]. As the Kuril Straits are located at the northern end of the North Pacific Ocean, the locally intensified vertical mixing influences the whole of the Pacific Ocean circulation. The localized upwelling of deep water is a diapycnal flow induced by the vertical mixing, so it can be regarded as a part of the Pacific thermohaline circulation, together with the enhanced southward flow of the intermediate water and northward flow of the deep water.
 On the basis of the results of a tide model [Nakamura and Awaji, 2004] and the fact that very cold water is found at the sea surface in summer around the Kuril Straits [see Nakamura et al., 2000b, Figure 1a], Nakamura et al. [2006a, 2006b] applied the large value from the sea surface to the bottom. However, turbulence observations at many locations of strong tide-induced vertical mixing indicate that mixing becomes weaker as the distance from the bottom increases [St. Laurent et al., 2001; Moum et al., 2002]. A recent turbulence observation at the Bussol’ Strait, which is the deepest among the Kuril Straits, also suggested that the strong vertical mixing does not reach the sea surface [Yagi, 2008]. Moreover, the result of a tide model [Nakamura and Awaji, 2004] shows that the region where the strong mixing reaches the sea surface is confined within ∼10 km from the coast. When we prescribe strong vertical mixing up to the sea surface around the Kuril Straits in a coarse-resolution model, the areal extent of cold surface water is likely to be overestimated, which would lead to an unrealistically large surface buoyancy flux.
 Upwelling of thermohaline circulation is maintained by transport of the buoyancy obtained at the sea surface toward the deep ocean. This means that the thermohaline circulation induced by locally enhanced vertical mixing would be different depending on whether the mixing reaches the sea surface or not. We investigate the structure and formation mechanism of the thermohaline circulation induced by strong mixing localized around the Kuril Straits which does not reach the sea surface.
 The paper is organized as follows. The model and experimental design are described in section 2. In section 3, the results of realistically configured experiments are presented. On the basis of the results of box ocean experiments, the mechanism of vertical flows formed by the mixing are described in section 4. Section 5 provides a discussion on the proper way of incorporating the strong mixing around the Kuril Straits into models. Finally, the conclusion is described in section 6.
2. Model Description and Experimental Design
 The OGCM employed in this study is CCSR Ocean Component Model (COCO) version 4 [Hasumi, 2006]. The model incorporates a uniformly third-order tracer transport algorithm [Leonard et al., 1993], isopycnal diffusion [Cox, 1987], and isopycnal layer thickness diffusion [Gent et al., 1995]. The turbulence closure scheme of Noh and Kim  is applied for diagnosing vertical viscosity and diffusivity. The horizontal viscosity and vertical viscosity are 2.5 × 104 and 1.0 × 10−4 m2 s−1, respectively. The isopycnal diffusivity and isopycnal layer thickness diffusivity are 1.0 × 103 and 7.0 × 102 m2 s−1, respectively. Background vertical diffusivity is described later.
2.1. World Ocean Model
 The model domain is global excluding the Arctic Ocean. The bathymetry is constructed from a 5 min topography data set (ETOPO5 [National Geophysical Data Center, 1998]). The horizontal resolution is 1°. There are 45 vertical levels. The grid spacing varies from 5 m (top) to 200 m (bottom).
 The wind stress is monthly climatology based on reanalysis by the European Centre for Medium-Range Weather Forecasts [Roske, 2001]. The sea surface heat and freshwater fluxes are imposed by restoring the temperature and salinity of the uppermost level to climatological monthly sea surface temperature and salinity [Levitus et al., 1998] with a damping time scale of 3 days. In one of the conducted experiments, restoring of sea surface salinity is not applied but climatology monthly surface freshwater flux [Roske, 2001] is imposed around the Kuril Straits. Because the target of this study is the Pacific thermohaline circulation, temperature and salinity are weakly (with a damping time scale of 100 days) restored to observed monthly climatology below 1000 m depth in basins other than the Pacific. The background vertical diffusivity is specified by the global mapping of Hibiya et al. , which is based on expendable current profiler observations and whose Pacific average is 0.2 × 10−4 m2 s−1. An almost steady state is obtained by integrating the model for 3000 years, and the results presented herein are the average of the last 10 years.
 We conduct several experiments with different vertical profiles of vertical diffusivity around the Kuril Straits (Figure 1). The vertical diffusivity around the Kuril Straits is vertically constant in the case CONST, as in a previous study [Nakamura et al., 2006b]. The vertical diffusivity around the Kuril Straits is small (0.2 × 10−4 m2 s−1) at the sea surface and large at depth with a sharp change around 200 m depth in the case 200M. Because of the uncertainty in vertical profile of the vertical diffusivity, further additional experiments are conducted. The depth where the vertical diffusivity sharply changes is set to 300 and 500 m in the cases 300M and 500M, respectively. The value of large vertical diffusivity around the Kuril Straits is set to 200 × 10−4 m2 s−1. The basal depth to which the large vertical diffusivity is applied is 3000 m, based on the result of a tide model [Nakamura and Awaji, 2004]. For comparison, the case BG is conducted where the vertical mixing is not enhanced around the Kuril Straits.
2.2. Box Ocean Model
 A box ocean model is utilized to qualitatively interpret the results obtained by the World Ocean model. The model domain spans from 0°E to 60°E in the zonal direction and from 84°S to 60°N in the meridional direction. Although the width of the box ocean is smaller than the Pacific Ocean, it does not affect the qualitative discussion made herein. The horizontal resolution of the box ocean model is set to 3°, which is coarser than that of the World Ocean model. In terms of the governing physics of the model, however, both resolutions are categorized as noneddying and are expected to represent the same level of complexity. Therefore, this resolution is adequate in interpreting what happens in the World Ocean model. The bottom is flat (4200 m depth). There are 25 vertical levels, and the layer thickness varies from 5 m (top) to 250 m (bottom).
 To simplify the model, salinity is kept constant. The sea surface heat flux is imposed by restoring the temperature of the uppermost level to 10.0°C with a damping time scale of 3 days. To form the deep water in the southern part of the basin, temperature to the south of 40°S is restored to annual mean climatology [Levitus et al., 1998] averaged zonally over the Pacific sector. In the case MIX, large vertical diffusivity (200 × 10−4 m2 s−1) is specified at the northwestern corner of the basin (0°E–12°E, 51°N–60°N), and a small value (0.2 × 10−4 m2 s−1) is specified in the remaining. The vertical distribution of the enhanced vertical diffusivity is the same as in case 200M of the World Ocean model. For comparison, the small (0.2 × 10−4 m2 s−1) vertical diffusivity is uniformly applied in the case NOMIX.
3. Results of World Ocean Model
 In BG, vertical motion is not significant at the latitudes of the Kuril Straits (45°N–50°N) because there is not locally enhanced vertical mixing (Figure 2a). Since the vertical flow formed by the strong vertical mixing around the Kuril Straits is not captured in the zonally integrated view because of the subarctic Ekman upwelling in the latitudes of the Kuril Straits, differences of the Pacific meridional overturning stream function from BG are shown for CONST, 200M, 300M, and 500M in Figures 2b–2e. Deep water upwells at the latitudes of the Kuril Straits as the result of enhanced vertical mixing, and the northward transport in the deep layer is increased compared with BG (Figures 2b–2e). The upwelling of deep water reaches the shallow layer and a southward transport is induced in the shallow and intermediate layers by vertical mixing in CONST (Figure 2b). This result is similar to that of a previous study [see Nakamura et al., 2006b, Figure 11]. In cases 200M, 300M, and 500M, on the other hand, there is an anomalous northward transport at shallow and deep layers and a southward transport at an intermediate layer (Figures 2c–2e). Note that the difference of the meridional overturning stream function for CONST is not completely the same as that in the previous study [Nakamura et al., 2006b], which has a little upper cell corresponding to the strengthening of the dense shelf water formation in the Okhotsk Sea. Because our model does not include sea ice, the formation of dense shelf water and the associated upper cell do not appear in CONST.
 In CONST, unrealistically large downward freshwater flux (equivalent to precipitation of ∼0.03 m d−1) is induced around the Kuril Straits. Sea surface salinity (SSS) becomes very high where sea surface-reaching strong mixing is specified compared with the surrounding regions. Such a feature is actually observed in reality, but its area is much narrower. Furthermore, the observed climatology [Levitus et al., 1998], to which the model SSS is restored, does not resolve such a fine structure. As a result, restoring boundary condition for SSS induces the unrealistic freshwater flux in CONST. To exclude the influence of such an unrealistic behavior of SSS restoring, an experiment (CONST_wflx) is conducted where the monthly mean climatology of surface freshwater flux [Roske, 2001] is specified around the Kuril Straits (140°E–160°E, 40°N–60°N) without restoring the sea surface salinity. Localized downwelling of surface water is induced around the Kuril Straits in this case (Figure 2f). When the freshwater flux around the Kuril Straits is realistic, the ocean cannot receive sufficient buoyancy at the sea surface to sustain the surface-reaching upwelling even under the surface-reaching strong mixing. It should also be noted that a high salinity bias is found at the sea surface in the northwestern Pacific region centered around the Kuril Straits in CONST_wflx as a result of surface-reaching strong vertical mixing.
 The horizontally integrated vertical velocity near the Kuril Straits (140°E–155°E, 40°N–50°N) is upward at all depths in CONST, but it is downward above about 1000 m depth and upward below in cases 200M, 300M, 500M, and CONST_wflx (Figure 3). The direction of vertical flow differs depending on whether large buoyancy is locally provided through the sea surface around the Kuril Straits. It is notable that the depth of the boundary between the upwelling and downwelling is almost invariant at ∼1000 m in cases 200M, 300M, and 500M. We discuss its reason in subsection 4.2.
 The horizontal distribution of the differences of density and horizontal and vertical velocities between cases 200M and BG are shown in Figure 4. Dense water is formed by the locally intensified vertical mixing around the Kuril Straits (Figure 4a); northward and southward currents are formed around the dense water at sea surface and 1000 m depth, respectively (Figures 4b and 4c); and the strong downwelling induced by the vertical mixing is confined around the Kuril Straits (Figure 4d). Note that the dense water is not confined around the Kuril Straits but extends eastward by the wind-driven eastward flow. These currents and the density distribution are also formed in cases 300M, 500M, and CONST_wflx (figure not shown).
 It is well known that the downwelling of shallow water in the northern North Atlantic and the Antarctic marginal seas, which drives the global thermohaline circulation, is associated with cooling at the sea surface and consequent deep convection. However, the downwelling of surface water in cases 200M, 300M, 500M, and CONST_wflx does not accompany deep convection. Such a type of thermohaline circulation has not previously been discussed. It is difficult to extract the thermohaline component in the realistic model because of the existence of wind-driven circulation. Thus, the box ocean model without wind stress forcing is utilized to investigate the mechanism of the downwelling of surface water induced by the locally enhanced vertical mixing. To utilize the simple sea surface restoring method while avoiding large buoyancy flux at the sea surface, we employ the non-surface-reaching strong vertical mixing (as in case 200M).
4. Results of Box Ocean Model
4.1. Structure of Downwelling and Upwelling Induced by Strong Vertical Mixing
 The zonally integrated meridional overturning stream function for NOMIX and MIX is shown in Figure 5. The deep water from the south of the basin upwells over the whole basin and returns to the south in NOMIX (Figure 5a). This characteristic of MOC is the same as that in BG (Figure 2a). In MIX, upwelling of deep water and downwelling of surface water are induced at the latitudes of localized mixing (Figure 5b). Northward flows are induced in the shallow and deep layers and a southward flow is induced in the intermediate layer as in cases 200M, 300M, 500M, and CONST_wflx of the World Ocean model. The result of the box ocean model with wind stress also shows upwelling and downwelling induced by the non-surface-reaching locally enhanced vertical mixing (figure not shown).
 Strong downwelling of surface water is found at the western part of the northern end (Figure 6a). This downwelling does not accompany deep convection. The sea surface flow toward the downwelling consists of a northward western boundary current and a cyclonic current around the region of strong mixing (Figure 6b). The flow at 1075 m depth from the downwelling consists of a southward western boundary current and an anticyclonic current around the northwestern corner (Figure 6c). The direction of flow is opposite between the sea surface and 1075 m depth. A relatively cold column of water is formed at the northwestern corner by the vertical mixing (Figure 6d).
 The meridional section of potential temperature along the western boundary for NOMIX, MIX, and their difference (MIX minus NOMIX) is shown in Figure 7. Isothermal layers become thick around 1000 m depth, and a cold water column is formed above 800 m depth. The horizontal currents at the sea surface and 1075 m depth flow around this cold water column. The relatively cold water does extends not eastward along the northern boundary but southward along the western boundary (Figure 6d), because the displacement of isothermal layers at the northwestern corner propagate not eastward but southward by Kelvin waves. As a result, the horizontal current around the cold water runs almost parallel with the western boundary and hits the northern boundary (Figure 6b). Note that there is relatively weak upwelling along the western boundary at the shallow layer (Figure 6a) because the isothermal line is not completely parallel to the western boundary.
 This kind of circulation structure is also seen in the case of deepwater formation induced by the sea surface cooling, which has been well documented by many previous studies [e.g., Suginohara and Aoki, 1991; Marotzke, 1997]. The downwelling of surface water induced by cooling at the sea surface and that induced by locally enhanced vertical mixing are commonly accompanied by the formation of a cold water mass, a cyclonic current in an upper layer, and an anticyclonic current in a lower layer along the margin of cold water. The difference between the two is in the sink of buoyancy (source of cooling): it is the atmosphere in the downwelling formed by the sea surface cooling whereas it is the deep ocean in that formed by the vertical mixing. The aforementioned characteristics are also found in the realistically configured model (Figure 4).
 Vertical velocity at 2950 m, horizontal velocity fields at 2075 and 3825 m, and potential temperature at 3075 m depth are shown in Figure 8 for MIX. Localized upwelling occurs at the northern end of the strong mixing region (Figure 8a). The flow at the bottom toward the upwelling consists of a northward western boundary current and a cyclonic current surrounding a warm water column formed by the locally enhanced vertical mixing (Figures 8b and 8d). At 2075 m depth, the flow from the upwelling consists of a southward western boundary current and an anticyclonic current around the water column (Figure 8c). These upwelling and horizontal currents at depth construct a mirror image of the downwelling of surface water and associated horizontal currents.
4.2. Depth of Boundary Between the Downwelling and Upwelling
 In the realistically configured model, the depth of boundary between the downwelling and upwelling induced by the vertical mixing is insensitive to the top depth of the strong mixing (cases 200M, 300M, and 500M; Figures 2c–2e). On the basis of the result of the box ocean model, we consider how the depth of the boundary is determined.
 The strong vertical flow confined to the northern end [>O(10−5) m s−1] is not a consequence of convergence or divergence of geostrophic currents due to the β effect, because the downwelling velocity due to the β effect accompanying the northward flow in the shallow layers is estimated to be O(10−6) m s−1. The strong vertical motion is formed by the existence of lateral boundaries. The theoretical background is given in Appendix A, and only the result is described here. We assume that the horizontal currents are geostrophic and the vertical flows are determined by the divergence or convergence formed by the boundary-reaching geostrophic currents. The water in the vertically mixed region becomes denser in the upper layer and lighter in the lower layer than its surrounding water (Figure 9a). Pressure in the mixed water column becomes lower in the shallow and deep layers and higher in between compared with the surrounding region (Figure 9b). The vertical integral of this horizontal pressure difference becomes as in Figure 9c, to which the vertical velocity at the lateral boundary is proportional under the present assumption. The actual vertical velocity at the boundary is shown in Figure 9d, and its profile is well captured by Figure 9c. The depth that separates the downwelling and upwelling is determined by the divergence and convergence of geostrophic currents, which are induced by the locally enhanced vertical mixing, on the lateral boundary.
 We also carried out the additional experiments where the region of enhanced vertical mixing is the same as MIX but the upper bound of strong vertical mixing is taken to be 500 and 800 m (hereafter MIX500M and MIX800M) to explain how the depth separating the upwelling and downwelling is determined. The aforementioned consideration well accounts for the vertical profile of vertical velocity in MIX500M and MIX800M as in MIX (Figure 9). The depth which separates the downwelling and upwelling is insensitive to the top depth of strong mixing among cases MIX, MIX500M, and MIX800M.
 When we assume an idealized situation where the strong vertical mixing locally modifies the water column without exchanging water with the surrounding region or affecting the surface buoyancy flux, the depth where vertical velocity becomes zero (except for the sea surface and the bottom) can be calculated (Figure 10). The depth is not invariant against the top depth of strong mixing under this theoretical consideration. Because the present assumption is well satisfied in MIX800M, the depth where the vertical velocity at the boundary vanishes coincides well with that indicated in Figure 10. However, it is not the case for MIX and MIX500M. In these cases, the positive density anomaly in the upper layer is much larger than what is expected under the preceding assumption. The large positive density anomaly is confined along the western boundary (Figures 11a–11c), and it is induced by the upwelling at the western boundary to the south of the strong downwelling region (Figures 11d–11f). In MIX and MIX500M, stronger upwelling and a larger positive density anomaly than in MIX800M are formed. The upwelling at the western boundary is formed by the fact that the isothermal line is not completely parallel to the western boundary (Figures 11a–11c). Thus, the strength of this upwelling is also determined by the density anomaly similarly to the downwelling at the northern boundary. In MIX and MIX500M, the larger positive density anomaly induces the stronger upwelling along the western boundary, and the stronger upwelling intensifies the positive density anomaly at the shallow layer further than in MIX800M. Because of this interaction between the upwelling at the western boundary and the positive density anomaly in the shallow layer, the depth of the boundary between downwelling and upwelling becomes less sensitive to the top depth of the strong vertical mixing (Figure 10). Since the relation between the strength of the upwelling at the western boundary and the top depth of strong vertical mixing cannot be derived by a simple consideration, further quantitative discussion, which gives an account for the almost perfect insensitivity, is not feasible.
 Such a tendency is found in the result of the realistically configured model. There is upwelling confined to the eastern coast of Honshu Island (Figure 4d). In cases 200M and 300M, the upwelling is further intensified, and a larger dense anomaly is formed in the shallow layer than that in 500M. Thus, the depth separating downwelling and upwelling depends less on the top depth of strong vertical mixing than the analytical result, as in the case of the box ocean model.
5.1. Vertical Profile of Vertical Diffusivity Around the Kuril Straits
 In the World Ocean model, the direction of subsurface vertical flow around the Kuril Straits is determined by whether sufficient buoyancy is supplied at the sea surface to sustain surface-reaching upwelling. What is essential for the thermohaline circulation is gain or loss of buoyancy rather than the value of vertical diffusivity. In discussing the volume transport of thermohaline circulation, it is necessary to focus on the amount of potential energy generated by vertical mixing. The dissipation rate E is represented by
where ρ0 = 103 kg m−3 is the density of seawater, γ = 0.2 is the mixing efficiency, κv is the vertical diffusivity, and N is the Brunt-Väisälä frequency. We estimate the dissipation rate by integrating within the region of the locally enhanced mixing. In CONST, this value amounts to 293 GW. In cases 200M, 300M, 500M, and CONST_wflx, the dissipation rates are 125, 108, 80, and 170 GW, respectively, which are considerably smaller than that in CONST. The dissipation rate is significantly different depending on whether the ocean receives a large amount of buoyancy through the sea surface, because the stratification of the shallow layer becomes strong and the vertical mixing efficiently works in the shallow layer under large surface buoyancy flux.
 The result of a tide model [Nakamura and Awaji, 2004] shows that the strong vertical mixing at the sea surface is confined within ∼10 km from the Kuril Islands. When such strong vertical mixing reaches the sea surface in a coarse-resolution model, the region of surface-reaching strong vertical mixing becomes too large (more than 100 km wide). As a result, cold, salty surface water covers a much wider region of the sea surface than observed. It is possible to prevent such a bias from inducing unrealistic surface freshwater flux, as in the case of CONST_wflx. However, a large SSS bias appears around the Kuril Straits in that case, and it is difficult to prevent erroneous surface heat flux. As a pragmatic approach to represent the effect of tidal mixing around the Kuril Straits in coarse-resolution modeling, therefore, it seems reasonable to specify non-surface-reaching strong vertical mixing. In high-resolution modeling, with a horizontal grid size of much less than 10 km, it would be possible to directly specify the surface-reaching strong mixing in narrow regions, as tide models and observations suggest. For that purpose, we need to know the actual regions where the strong mixing reaches the sea surface, which have not yet been identified.
5.2. Strength of Vertical Mixing Around the Kuril Straits
 In this study, the value of 200 × 10−4 m2 s−1 is employed as the vertical diffusivity around the Kuril Straits based on estimates by tide models [Nakamura et al., 2000b; Nakamura and Awaji, 2004] and observation [Yagi, 2008]. However, Tanaka et al.  reported that the value of vertical diffusivity averaged around the Kuril Straits is about 8 × 10−4 m2 s−1, which is smaller by more than an order of magnitude than that of previous estimations [Nakamura et al., 2000b; Nakamura and Awaji, 2004b], y using a horizontally two-dimensional tide model and altimeter data. This discrepancy may be caused by the difference of averaging area and parameterization used for determining the energy dissipation rate and its vertical profile. A robust value has not been obtained for the vertical diffusivity around the Kuril Straits because of the uncertainty in estimation methods. So, in this study, additional experiments (Kv20, Kv50, and Kv100) are conducted where the strong vertical diffusivity is chosen to be 20, 50, and 100 × 10−4 m2 s−1 with the same vertical profile as in case 200M. The larger the vertical diffusivity is, the stronger the volume transport of the Pacific thermohaline circulation becomes. On the other hand, the qualitative characteristics of the Pacific thermohaline circulation induced by the localized mixing around the Kuril Straits, such as the depth to which surface water downwells, do not depend on the value of vertical diffusivity.
 The dissipation rate calculated according to equation (1) is 22, 46, 78, and 125 GW for cases Kv20, Kv50, Kv100, and 200M, respectively. The reason why the dissipation rate does not increase linearly with the vertical diffusivity is that stronger vertical mixing leads to weaker stratification and smaller Brunt-Väisälä frequency. When the vertical profile of vertical diffusivity is given by that of case 200M, the vertical diffusivity of 50–100 × 10−4 m2 s−1 is comparable (46–78 GW) to the estimations of global tide models [Le Provost and Lyard, 1997; Egbert and Ray, 2000] and the Okhotsk regional one [Tanaka et al., 2007] (30–100 GW). When the vertical profile is taken to be that of cases 300M or 500M, the dissipation rate is expected to be comparable with the above estimates, even if the vertical diffusivity is larger than 100 × 10−4 m2 s−1. It should be noted that what is physically important is not vertical diffusivity but energy dissipation rate. Vertical diffusivity is a widely used measure, but one should not be restrained to only that quantity without paying attention to the aspect of energetics.
 We investigated the Pacific thermohaline circulation induced by the locally enhanced vertical mixing around the Kuril Straits by using a general circulation model. When strong vertical mixing reaches the sea surface, localized upwelling is induced from depth to the surface. On the contrary, when strong vertical mixing does not reach the sea surface, downwelling of surface water is induced above ∼1000 m and upwelling of deep water is induced below. The structure of thermohaline circulation significantly differs depending on whether the ocean receives a large amount of buoyancy at the sea surface around the Kuril Straits. The downwelling of shallow water is not accompanied by cooling through the sea surface and deep convection.
 The mechanism of sinking of surface water induced by vertical mixing which does not reach the sea surface is examined by using a box ocean model. It is shown that sinking is associated with the formation of dense water and geostrophic current around it. Although this downwelling is similar to that induced by sea surface cooling in terms of the above processes, the cold water source is in the deep ocean in the former whereas it is in the atmosphere in the latter.
 The depth separating the upwelling and downwelling is insensitive to the top depth of the strong vertical mixing. The vertical flow is confined to the boundary and governed by the convergence or divergence formed by the boundary-reaching geostrophic currents. The locally enhanced vertical mixing, which does not reach the sea surface, forms denser and lighter water than in its surrounding region in the upper and lower layers, respectively. As a result, a low-pressure anomaly is formed in the upper and lower layers and a high-pressure anomaly is formed in between, and the downwelling and upwelling is induced in the upper and lower layers, respectively. Assuming that the difference of density between the strong vertical mixing region and its surrounding region is formed only by the vertical exchange of density, the depth separating the upwelling and downwelling depends on the upper limit depth of the strong mixing. However, a larger density anomaly is induced by the upwelling in the western boundary at the south of strong downwelling. As a result, the depth that separates the downwelling and upwelling becomes insensitive to the top depth of strong vertical mixing.
 Although the strong mixing around the Kuril Straits is known to reach the sea surface, it does not necessarily mean that we should directly apply the surface-reaching strong mixing to models. It seems more appropriate to specify non-surface-reaching large vertical diffusivity in coarse-resolution models (typical horizontal grid size is ∼1°), because the region of cold, salty sea surface influenced by the mixing becomes too broad under the surface-reaching case. It would be possible and reasonable to directly specify the surface-reaching strong mixing in high-resolution models which can resolve its narrowness (less than 10 km).
 The Pacific meridional overturning circulation induced by the vertical mixing around the Kuril Straits transports a significant amount of heat. For instance, the northward heat transport at 24°N increases by ∼10% (from 0.44 to 0.48 PW) by the mixing-induced northward flow at the shallow layer in case 200M. Schematic diagram of the Pacific thermohaline circulation and heat transport induced by the strong vertical mixing around the Kuril Straits is shown in Figure 12. A larger amount of heat is received at the sea surface in low latitudes and transported to the latitudes of the Kuril Straits by the intensified Kuroshio and subarctic gyre in the shallow layer. It is then transported to the deep ocean first by the localized downwelling to ∼1000 m and then by the localized strong vertical mixing. The transported heat maintains the localized upwelling of deep water around the Kuril Straits. It means that the shallow meridional overturn (localized downwelling) is required for the maintenance of the deep meridional overturn (the localized upwelling of deep water), since the deep ocean cannot receive the necessary heat from the sea surface by the non-surface-reaching mixing. Many previous studies investigated the effect of the vertical mixing on the thermohaline circulation and reported that the vertical mixing induces the upwelling of deep water [e.g., Bryan, 1987; Simmons et al., 2004; Jayne, 2009]. This study is first to suggest the possibility of downwelling formation by the vertical mixing.
 In this study, we investigated the role of strong vertical mixing localized around the basin boundary in the thermohaline circulation. In the Pacific Ocean, there are several other boundary regions where locally enhanced mixing occurs (e.g., the Aleutian Archipelago, the Luzon Straits, and the Indonesian Archipelago). Their effects should also be examined to clarify the structure of the Pacific thermohaline circulation. There is another well-known “mixing hot spot” which is not located around the boundary, the Hawaiian Ridge. The observation of the turbulent mixing shows that the enhancement of the vertical mixing does not reach the sea surface around the Hawaiian Ridge [Finnigan et al., 2002]. Because the lateral boundary where a strong vertical flow can occur is far apart, its role in the Pacific thermohaline circulation would be significantly different from what is found in this study and is left to be investigated.
Appendix A:: Relation Between Geostrophic Currents in Ocean Interior and Vertical Flows on Boundary
 Let us consider the situation where a water column of density ρ1, which is formed by locally enhanced vertical mixing, is surrounded by water of density ρ2 (Figure A1). Using the hydrostatic relation, the pressure difference between the water column and the ambient water, p′, is represented by
where ρ′ is the density difference between the water column and the ambient water (ρ1 − ρ2), g is the gravitational acceleration, ρ0 is a reference density, η1 and η2 are the sea surface height of the strong mixing region and its surrounding region, respectively, and p′s represents the difference of pressure caused by the difference of sea surface height. The geostrophic horizontal velocity, ug, is proportional to the pressure difference. At the boundary, which the geostrophic currents hit, the continuity equation yields
where u and w are horizontal and vertical components of velocity, respectively, and ∇H represents the horizontal nabla operator. Since w = 0 at the sea surface, the relation between the vertical velocity at the boundary and the pressure difference is obtained from equation (A2) as
where η is the sea surface height at the boundary. Because w = 0 at the bottom (hereafter defined as z = −H),
This constraint determines ps′ in equation (A1), and w is solved when p′ is given. The results of the box ocean model (the difference of density between the strong mixing region and its surrounding region and the vertical flows on the boundary) satisfy equations (A1)–(A4) (see section 4.2).
 Let us consider a simple situation to clarify why downwelling of surface water is induced by the enhanced vertical mixing which does not reach the sea surface. When a column of water is vertically mixed over a certain range of depth (−hB ≤ z ≤ − hA), the water in the column becomes denser in the upper layer and lighter in the lower layer than the surrounding water (Figure A2a). Using equations (A1) and (A4), the pressure difference becomes as shown in Figure A2b. It should be noted that ps′ becomes negative, which means that the sea surface height of the vertically mixed water column is lower than that of the surrounding region. The vertical velocity at the boundary, which is calculated by equation (A3), is shown in Figure A2c. Downwelling and upwelling are formed in the upper and lower layers, respectively. The depth of the boundary separating the upwelling and downwelling comes between the top and bottom of the vertically mixed depth range (−hB ≤ z ≤ −hA).
 Let us consider an idealized situation where the strong vertical mixing locally modifies the water column without exchanging water with the surrounding region. Since the enhanced vertical mixing does not reach the sea surface, there is not significant heat exchange at the sea surface. In this case the density difference (ρ′) is constrained by
The vertical distribution of density of the ambient water (ρ2) below the mixed layer is assumed as
where ρs, Δρ, and h are the density at the bottom of the mixed layer, difference of density between the shallow and deep layers, and the scale height for density change, respectively. This exponential function well approximates the vertical distribution of the density in the region surrounding the strong mixing in the box ocean model (not shown). Since the density is vertically linear in the intensified mixing region (also not shown), the vertical distribution of density in the strong mixing region (ρ1) is defined as
where a can be calculated from equations (A5), (A6), and (A7). A schematic view for the vertical density profiles is presented in Figure A3. A typical vertical profile of potential density around 60°N results in Δρ = 0.8 kg m−3, ρs = 26.8 kg m−3, and h = 700 m. The depth where vertical velocity becomes zero (except for the sea surface and the bottom) obtained from the density difference defined earlier is shown in Figure 10.
 This work is supported by JST/CREST. Numerical calculations were done by SR11000 at the Information Technology Center, University of Tokyo. We are deeply grateful to M. Endoh and M. Kawabe for their helpful comments and continuous encouragement. We feel grateful to A. Oka, H. Tatebe, Y. Komuro, K. Kusahara, Y. Sasajima, M. Watanabe, E. Watanabe, Y. Matsumura, and S. Urakawa for their helpful comments and discussions. Almost all figures are drawn using the Grid Analysis and Display System (GrADS) developed by M. Doty and M. Fiorino.