OCTOBER 2000 HOGAN AND HURLBURT 2535

Impact of Upper Ocean±Topographical Coupling and Isopycnal Outcropping in 1 1 *Japan/East Sea Models with ⁄8؇ to ⁄64؇ Resolution

PATRICK J. HOGAN AND HARLEY E. HURLBURT Naval Research Laboratory, Stennis Space Center, Mississippi

(Manuscript received 3 November 1998, in ®nal form 23 November 1999)

ABSTRACT A regional primitive equation ocean model is used to investigate the impact of grid resolution, baroclinic instability, bottom topography, and isopycnal outcropping on the dynamics of the wind and through¯ow-forced 1 surface circulation in the Japan/East Sea. The results demonstrate that at least ⁄32Њ (3.5 km) horizontal grid resolution is necessary to generate suf®cient baroclinic instability to produce eddy-driven cyclonic deep mean ¯ows. These abyssal currents follow the f/h contours of the bottom topography and allow the bottom topography to strongly in¯uence mean pathways of the upper-ocean currents in the Japan/East Sea. This upper ocean± topographical coupling via baroclinic instability (actually a mixed baroclinic±barotropic instability) requires that 1 mesoscale variability be very well resolved to obtain suf®cient coupling. For example, ⁄32Њ resolution is required to obtain a realistic separation latitude of the East Korean Warm Current (EKWC) from the Korean coast when Hellerman±Rosenstein monthly climatological wind stress forcing is used. Separation of the EKWC is more 1 realistic at ⁄8Њ resolution when the model is forced with climatological winds formed from the ECMWF 10-m 1 reanalysis due to strong positive wind stress curl north of the separation latitude, but at ⁄8Њ the level of baroclinic instability is insuf®cient to initiate upper ocean±topographical coupling. Hence, this major topographical effect is largely missed at coarser resolution and leads to erroneous conclusions about the role of bottom topography 1 and unexplained errors in the pathways of current systems. Results from a ⁄64Њ simulation are similar to those 1 at ⁄32Њ, particularly where the EKWC separates from the Korean coast, suggesting statistical simulation con- 1 vergence for mesoscale variability has been nearly achieved at ⁄32Њ resolution. Isopycnal outcropping and as- sociated vertical mixing provide an alternate mechanism to topographical control in developing and maintaining a boundary current along the west coast of Japan, but are less important than baroclinic instability in driving deep mean ¯ows.

1. Introduction ubiquitous mesoscale eddy ®eld. Because the JES contains processes similar to those occurring in large- Located between Japan, Korea, and Russia, the Ja- scale ocean basins, but is smaller and may adjust more pan/East Sea (hereafter referred to as the JES) is a rapidly, and hence be easier to study, the JES offers part of the chain of adjacent marginal seas in the ocean scientists a unique natural laboratory for in- northwest Paci®c. With a maximum depth of about vestigating oceanic processes. For the same reasons, 3700 m, the JES has been characterized as a micro- the JES is an ideal basin for studying dynamics via cosm of the global ocean (Ichiye 1984). Indeed, large- numerical simulations. The region is characterized by scale circulation features similar to those found in the global ocean include cyclonic and anticyclonic gyre complex circulation patterns and mesoscale variabil- systems, a western boundary current that separates ity, yet is small enough for multiple simulations to be from the coast, and a deep homogeneous water mass. economically feasible, including some with very high Other interesting phenomena include wind and ther- horizontal grid resolution. This feasibility is further mohaline driven circulation, deep convection, and a enhanced by using the Naval Research Laboratory's Layered Ocean Model (NLOM), a modular design that provides an effective means to dissect ocean dynamics (Hurlburt et al. 1996) and one that is extremely ef- * Naval Research Laboratory Contribution Number NRL/JA 7323- ®cient and portable to several different computer ar- 98-0063. chitectures (Wallcraft and Moore 1997). In¯ow through Tsushima Strait and out¯ow through the Tsugaru and Soya Straits (Fig. 1) couples the JES Corresponding author address: Dr. Patrick J. Hogan, Naval Re- search Laboratory, Stennis Space Center, MS 39529-5004. to the Paci®c Ocean. These straits have shallow sill E-mail: [email protected] depths; the deepest sill depth (about 200 m) is in the

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2536 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

1 FIG.1.THE ⁄32Њ bottom topography for the Japan/East Sea. Locations of the in¯ow and out¯ow ports are marked as black arrows. Geographical locations used in text are marked. Contour interval is 250 m. western channel of Tsushima Strait. However, ¯ow meandering current that episodically sheds eddies. The through these straits has a profound impact on the cir- highly variable spatial and temporal characteristics of culation features within the JES. In particular, a fun- the TWC make it dif®cult to support one theory over damental circulation feature in the JES is the Tsushima the other. Indeed, hydrographic surveys (Kawabe 1982; Warm Current (TWC), which transports warm salty wa- Kim and Chung 1984), analyses of satellite infrared data ter into the basin through Tsushima Strait. The historical (Kim and Legeckis 1986; Cho and Kim 1996), and anal- interpretation of Suda and Hidaka (1932) and Uda ysis of satellite-tracked drifters (Beardsley et al. 1992) (1934) that the TWC splits into three distinct branches demonstrate the complex and chaotic behavior of this after entering the basin is still widely accepted, but oth- current system. ers (Moriyasu 1972) have regarded the TWC as a single In contemporary literature, the three branches of the

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2537

FIG. 2. Observationally based schematic patterns of surface circulation in the Japan/East Sea (a) in summer by Naganuma (1977) and (b) annual mean by Yarichin (1980). The following currents are labeled: East Korea Warm Current (EKWC), Nearshore Branch (NB), Liman Cold Current (LCC), and North Korea Cold Current (NKCC). Warm (cold) rings are denoted by a ``W'' (``C''). From Preller and Hogan (1998).

TWC (from east to west) are named the Nearshore instance, what mechanism determines the separation lat- Branch (NB), the Offshore (or Middle) Branch, and the itude of the EKWC from the coast of Korea? What East Korea Warm Current (EKWC), respectively. In the causes the formation and persistence of the NB of the three-branch view, the NB ¯ows eastward along the TWC? What role does the bottom topography play in coast of Honshu and exits into the Paci®c Ocean through in¯uencing the surface circulation? Of course the ability Tsugaru Strait (Fig. 2). The Offshore Branch, which is to address questions such as these depends on the re- more variable in space and time, is situated to the west- alism of the numerical model being used to investigate northwest of the Nearshore Branch and ¯ows along the dynamics. The realism of any given simulation is Honshu farther offshore. The EKWC ¯ows northward dependent on many criteria, including choice of model along the Korean continental slope to about 37Њ±38ЊN, physics, numerical implementation of the equations of where it meets the southward ¯owing North Korean motion, the choice of forcing functions and parameter Cold Current (NKCC). At the con¯uence, the currents space, and model grid resolution. For the simulations separate from the coast and ¯ow east-northeast toward used in this study, the numerical implementation of the Tsugaru Strait along the polar front. Much of that ¯ow equations of motion is identical for all simulations (see exits through Tsugaru Strait, while the remainder con- Hurlburt and Thompson 1980; Wallcraft 1991; Hurlburt tinues northward along the coast of Hokkaido as the et al. 1996; Moore and Wallcraft 1998). The model Tsugaru Warm Current. Some of this ¯ow exits into the physics are varied only to the extent that they are used Sea of Okhotsk through Soya Strait, while the remainder to explore the parameter space for the simulations. The eventually returns southward along the Siberian coast primary forcing functions are in¯ow and out¯ow as the Liman (or Primoriye) Cold Current (LCC) and through the straits and monthly climatological wind the NKCC (south of Vladivostok), forming a cyclonic stress. Nearly all of the simulations are forced with sea- gyre in the northern part of the basin (Fig. 2). sonally varying in¯ow and out¯ow through the straits A primary objective of this study is to identify the and the Hellerman and Rosenstein (1983, hereafter HR) critical dynamical processes responsible for the ob- monthly wind stress climatology, although that wind served and modeled circulation patterns in the JES. For climatology may not give the best results in the JES.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2538 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

Indeed, a simulation forced with a different wind set is Finally, the summary and conclusions are presented in offered to accentuate this point. However, because we section 7. are interested in the dynamics of the circulation, a less than optimal wind set can actually facilitate this task 2. The NRL Layered Ocean Model (NLOM) because systematic changes between different solutions can be easier to identify. The NLOM is a semi-implicit primitive equation The overall strategy employed in this study is to ex- ocean model with vertically integrated model equations ploit the modularity of the NLOM to investigate the for each layer. In the NLOM, the layered model equa- circulation dynamics in the JES. In this context we ex- tions are cast in transport form and (in the version used amine the solutions from an ensemble of simulations here) the interfaces between the layers are isopycnal with increasing (dynamical) complexity by including or surfaces. The model retains the free surface and can excluding certain features and dynamical processes. The include realistic bottom topography as long as it is con- simplest end-member in this ensemble is a linear 1.5- ®ned to the lowest layer. The fundamental model design layer reduced-gravity design, while the most complex is described in detail in Hurlburt and Thompson (1980), is a fully nonlinear 4-layer design with realistic bottom although Wallcraft (1991) has signi®cantly enhanced the topography. Information on the model design is dis- current version. The mathematical and numerical for- cussed in section 2 and speci®c details of the model mulation of the equations in spherical coordinates are implementation for the JES are presented in section 3. discussed in Moore and Wallcraft (1998), with particular In section 4 the lowest order dynamics of the JES are attention given to the proper formulation of the isopyc- 1 investigated with a series of linear ⁄8Њ, 1.5-layer reduced- nal diffusion. The simulations for this study use a por- gravity simulations. In section 5 we examine the sen- table, scalable version of the NLOM (Wallcraft and sitivity to external forcing, as well as the impact of Moore 1997), which runs ef®ciently and interchange- higher order dynamics, such as nonlinearity and iso- ably on a wide variety of computer architectures using pycnal outcropping. The impacts of horizontal grid res- a tiled data parallel programming paradigm. olution, mesoscale ¯ow instabilities, and realistic bot- The equations for the n-layer ®nite depth, hydrody- tom topography are discussed in section 6. In these sim- namic model are given below for layers k ϭ 1,´´´,n 1 ulations, horizontal grid resolution varies from ⁄8Њ (14 with k ϭ 1 for the top layer. In places where k is used 1 km) to ⁄64Њ (1.7 km), which allows the role of upper to index model interfaces, k ϭ 0 is the surface and k ocean±topographical coupling to be examined in detail. ϭ n is the bottom:

(Vu cos␪)ץ (Uu)ץ U 1ץ kkkkkϩϩϪV (u sin␪ ϩ a⍀ sin2␪) ␪ kkץ ␾ץ[] tacos␪ץ

ϭϩmax(0, Ϫ␻kϪ1)ukϪ1 ϩ max(0, ␻kk)u ϩ1 Ϫ (max(0, Ϫ␻kk) ϩ max(0, ␻ Ϫ1))uk ( h Ϫ H)ץ h n ϩ max(0, ϪC ␻ )(u Ϫ u ) ϩ max(0, C ␻ )(u Ϫ u ) Ϫ k G jj MkϪ1 kϪ1 kMkkϩ1 kkj͸ ␾ץ a cos␪ jϭ1 (he cos2␪)ץ (he cos␪)ץ A ϩ (␶ Ϫ ␶ )/␳ ϩϩk ␾␾kkk ␾␪ (1) ␪ץ ␾ץ[] ␾␾kϪ1 k o a22cos ␪

(V ␷ cos␪)ץ ( U ␷)ץ V 1ץ kkkkkϩϩϩU (u sin␪ ϩ a⍀ sin2␪) ␪ kkץ ␾ץ[] tacos␪ץ

ϭϩmax(0, Ϫ␻kϪ1)␷ kϪ1 ϩ max(0, ␻kk)␷ ϩ1 Ϫ (max(0, Ϫ␻kk) ϩ max(0, ␻ Ϫ1))␷ k ( h Ϫ H)ץ h n ϩ max(0, ϪC ␻ )(␷ Ϫ ␷ ) ϩ max(0, C ␻ )(␷ Ϫ ␷ ) Ϫ k G jj MkϪ1 kϪ1 kMkkϩ1 kkj͸ ␪ץ a jϭ1 (he cos2␪)ץ (he cos␪)ץ A ϩ (␶ Ϫ ␶ )/␳ ϩϩk ␾␪kkk ␪␪ (2) ␪ץ ␾ץ[] ␪␪kϪ1 k o a22cos ␪ hץ (V ϭ ␻ Ϫ ␻ Ϫ KÃ ٌ22[ٌ (h Ϫ H )], (3 ´ ١ k ϩ t kkkϪ14 kkץ

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2539

where Wk(␾, ␪) ϭ kth interface weighting factor for global F cos␪) diapycnal mixing designed to conserve)ץ F 1ץ 1 F ϭϩ␾␪ ´ ١ ␪ mass within a layer in compensation forץ ␾ a cos␪ץ a cos␪ ϩ explicit diapycnal mixing due to hkkϽ h ⌽ (i.e., ␻ϩϪϪ ␻ ), and net transport throughץץ ⌽2 1ץ 1 ⌽ϭ2 ϩ cos␪ kkٌ ␪ the lateral boundaries of layer kץ␪ ΂΃ץ ␾ 2a 2cos␪ץ a22cos ␪

hk ϭ kth layer thickness CM ϭ coefficient of additional interfacial friction associated with entrainment vk ϭ kth layer velocity ϭ eជ␾uk ϩ eជ␪␷ k A ϭ coefficient of isopycnal eddy viscosity Vkkkϭ h v ϭ eជ␾Uk ϩ eជ␪Vk X(␾, ␪) ϭ region wide area average of X eជk ϭ angular deformation tensor à K4 ϭ coefficient of biharmonic horizontal layer ␷ thickness diffusivity ץukk ץ e␾␾k ϭϪcos␪ ϭϪe ␪␪k (␪ ΂΃cos␪ a ϭ radius of the Earth (6371 kmץ ␾΂΃cos␪ץ

Ϫ2 ( u g ϭ acceleration due to gravity (9.8 m s ץ␷ kk ץ e␾␪k ϭϩcos␪ ϭ e ␪␾k ␪ ΂΃cos␪ t ϭ timeץ ␾΂΃cos␪ץ

Hk ϭ kth layer thickness at rest ⍀ϭangular rotation rate of the earth Ϫ5 Ϫ1 nϪ1 (7.292 205 ϫ 10 s ) H ϭ D(␾, ␪) Ϫ H nj͸ ␪ ϭ latitude jϭ1 D(␾, ␪) ϭ total depth of the ocean at rest ␾ ϭ longitude.

␳0 ϭ constant reference density A hydrodynamic reduced-gravity model with n active layers has the lowest layer in®nitely deep and at rest, ␳k ϭ kth layer density, constant in space and time ١hnϩ1 ϭ 0. The model that is, vnϩ1 ϭ 0, hnϩ1 ϭϱ, and equations for the active layers are identical to those for gjՆ k n-layer hydrodynamic ®nite depth model, except that Gkj ϭ Άg Ϫ g(␳kjoϪ ␳ )/␳ j Ͻ k Hn ϭ const

␶w ϭ wind stress Gkl ϭ g(␳nϩ1 Ϫ ␳k)/␳ 0, l Յ k

Ck ϭ coefficient of interfacial friction Gkl ϭ g(␳nϩ1 Ϫ ␳l)/␳ 0, l Ͼ k

Cb ϭ coefficient of bottom friction ␶ k ϭ ␶ w, k ϭ 0  ␶w, for k ϭ 0 ␶ k ϭ Ck␳ 0|vk Ϫ vkϩ1|vk Ϫ vkϩ1), k ϭ 1,´´´,n C ␳ |v Ϫ v |(v Ϫ v ), ␶ ϭ  ko k kϩ1 kkϩ1 ␻ ϭ 0, k ϭ 0 k  for k ϭ 1, ´´´, n Ϫ 1 k max(0,ϩϪ ) max(0, ) h à , C ␳ |v |v , for k ϭ n ␻k ϭ ␻␻␻kkkϪ Ϫ k bo n n k ϭ 1,´´´,n 0 for k ϭ 0, n The model equations were integrated on a C-grid ␻ ϭ k ϩϪ (Mesinger and Arakawa 1976) using second-order, cen- Ά␻kkϪ ␻ Ϫ W kk␻à for k ϭ 1,´´´, n Ϫ 1 tered ®nite differences in space. The integrations in time ϩϩϩ2 ␻kkϭ ␻Ä [max(0, h kkkϪ h )/h ] used an explicit numerical scheme for the reduced-grav- ity simulations and a semi-implicit scheme for the ®nite ␻ϪϪϩϭ ␻Ä [max(0, h Ϫ h )/h ]2 kk kkk depth simulations. For the ®nite depth simulations, the ϩϪ ␻à kkkkϭ (␻ Ϫ ␻ )/W external and internal gravity waves are treated semi- implicitly in a manner similar to that described by Kwi- ␻Ä ϭ kth interface reference diapycnal mixing k zak and Robert (1971). This treatment allows a longer velocity time step than required for primitive equation models ϩ hk ϭ kth layer thickness at which entrainment that use an explicit free surface or a rigid lid. Overall, starts the model is up to 100 times more ef®cient (in terms of central processing time per model year for a given Ϫ hk ϭ kth layer thickness at which detrainment domain and horizontal resolution) than existing ®xed- starts level primitive equation ocean models (Wallcraft 1991).

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2540 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

An important attribute of the NLOM is the ability to was required to simulate the bifurcation of the Kuroshio allow isopycnal outcropping via ventilation of model at the Shatsky Rise by way of upper ocean±topograph- interfaces. This process, called hydromixing (Wallcraft ical coupling. They found that this coupling via baro- 1991), prevents nonpositive layer thickness, allows clinic instability and particular features of the Shatsky overturning cells in the vertical, and permits the exis- Rise topographic complex were essential to the bifur- tence of relatively thin layers. In practice, when a layer cation and the existence of the northward branch even ϩ becomes thinner than a speci®ed thicknesshk , ¯uid is though the simulated Kuroshio never directly impinged entrained into it from the layer below, although there on the topography. Of particular interest to this study is no net transfer of ¯uid between layers. Instead, if is to determine whether the upper ocean±topographical there is local entrainment or detrainment or net transport coupling described in Hurlburt et al. (1996) and Hurl- through the lateral boundaries within a layer, the net burt and Metzger (1998) can affect the separation lat- mass ¯ux from these processes is instantaneously bal- itude of the EKWC and other features of current systems anced by domainwide vertical mixing, distributed ac- in the JES, which is characterized by a signi®cantly cording to the weight function Wk(␾, ␪). The mixing different ¯ow regime with volume transports typically velocity, ␻k, in hydromixing is relatively simple and an order of magnitude less than those observed in the depends only on the net mass ¯ux just described; the Kuroshio. layer thickness; a reference vertical mixing velocity, Others have used layered models in the JES with some ϩ ␻Ä k; a layer thickness at which to start entrainment,hk ; success. Seung and Kim (1997) used a version of the Ϫ a layer thickness at which to start detrainment,hk (typ- Miami Isopycnal Coordinate Ocean Model (Bleck et al. ically deactivated by setting to a large value); and the 1992) to investigate the renewal time of the upper in- weight function, Wk(␾, ␪). In the simulations for this termediate water. Their results indicated particularly study, Wk ϭ 1, a uniform distribution, although more strong mixing and isopycnal outcropping offshore from sophisticated forms have been used in global versions the Siberian coast during winter. These results were in of the NLOM with considerable success (Shriver and agreement with those of Seung (1997), who applied the Hurlburt 1997). ventilation theory developed by Luyten et al. (1983) to Hydromixing is an important aspect of the NLOM the JES. Seung and Kim (1995) used a 2.5-layer, re- design since it allows layers to outcrop via interface duced-gravity model that included through¯ow forcing ventilation. However, this design and restricting the bot- and/or analytic wind stress curl. When buoyancy ¯ux tom topography to the lowest layer also enhances model via a mixed layer was included, the intermediate layer ef®ciency because positive layer thickness is maintained outcropped north of the polar front, causing cyclonic for all layers everywhere, a requirement for the modal circulation in that region and southward displacement decomposition used in the semi-implicitization of ex- of the separation latitude of the EKWC. Vasilev and ternal and internal gravity waves. From a dynamical Makashin (1992) developed a transport streamfunction point of view, once a layer has reached the thickness model to show the impact of deep convection on the ϩ ofhk , it is dynamically inactive over the outcropped ventilation of deep water in the northern JES. region in the sense that the resulting uniformly thick layer can no longer contribute to a pressure gradient. 3. Design of the model simulations The impact of isopycnal outcropping on the circulation in the JES is discussed in more detail in section 5. Implementing the NLOM for the JES region was rel- The NLOM has been used to investigate ocean dy- atively straightforward. The model domain extends from namics in both regional and global models. Hurlburt et 34.25Њ to 49ЊN. The model domain and bottom topog- al. (1996) performed a series of simulations of the Pa- raphy are shown in Fig. 1. Unvaried model parameters ci®c Ocean, including the JES, which demonstrated the are given in Table 1. Several simulations were performed importance of baroclinic instability and realistic bottom to determine the minimum number of layers to include topography for simulating the Kuroshio/Oyashio and and to determine the sensitivity of the model solutions Kuroshio Extension. Most of their simulations included to various combinations of layer thickness. These results six active layers with a relatively thick surface layer (not shown) demonstrated the need for at least four layers, (135 m). That layer structure was optimal for the Paci®c with interface depths of 60 m, 135 m, and 250 m, re- Ocean because it allowed the formation of a realistic spectively (Table 1). The relatively thin top layer (60 m) equatorial undercurrent, but was too thick in the JES to allows isopycnal outcropping via interface ventilation to permit the formation of a robust NB via interface ven- occur over much of the basin, which, as discussed in tilation (Hurlburt et al. 1996). In addition, the horizontal section 5, is critical for robust simulation of the NB along 1 grid resolution ( ⁄8Њ, or about 14 km) was apparently too the coast of Honshu. The depth of the second interface coarse to achieve the upper ocean±topographical cou- (135 m) is chosen to coincide with the approximate sill pling needed for realistic separation of the EKWC, al- depths of the in¯ow/out¯ow straits, and also plays a role though that resolution was suf®cient for the coupling in the dynamics associated with isopycnal outcropping. mechanism to occur in the region of the Kuroshio. Hurl- The depth of the third interface (250 m) approximates 1 burt and Metzger (1998) reported that ⁄16Њ resolution the mean depth of the thermocline which separates the

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2541

TABLE 1. Unvaried model parameters. calculation, but instead represent a compromise for the Parameter Value(s) De®nition wide range of limited observations cited in the literature. 1 A modi®ed version of the ⁄12Њ ETOP05 bottom to- C 2 ϫ 10Ϫ3 Bottom drag coef®cient b pography (National Oceanic and Atmospheric Admin- Ck 0 kth interfacial stress coef®- cient istration 1986) was used to represent the coastline ge-

Cm 1 Coef®cient of additional in- ometry and bottom topography. The maximum depth terfacial friction associated was set to 3750 m and the minimum depth at 200 m with entrainment (near the shelf break). As such, the 200-m isobath was g 9.8msϪ2 Acceleration due to gravity ϩ used as the model boundary. For the JES, this is a rea- hk 50m(kϭ 1) Thickness of layer k at 40m(kϭ 2, 3) which entrainment starts sonable ®rst approximation as much of the basin is char- Ϫ hk Deactivated Thickness of layer k at acterized by narrow continental shelf regions, and shelf which detrainment starts dynamics are quite different than those in deep water. (deactivated by making very large) The ETOP05 bottom topography was ®rst interpolated 1 Ϫ3 to Њ, then modi®ed for better agreement with detailed ␳kϩ1 Ϫ ␳k 0.8/0.32/0.42 kg m Strati®cation between layers ⁄32 1 1 k and k ϩ 1 topographic charts (Fig. 1). The ⁄8Њ and ⁄16Њ versions of k 60/135/250/bottom (m) Mean depth at the bottom of H the model geometries were formed by subsampling the ͸ i layer k 1 1 iϭ1 ⁄32Њ model grid, and the ⁄64Њ model geometry was formed ␻Ä 0.04 cm sϪ1 kth interface reference verti- 1 k by interpolation from the ⁄32Њ model grid. To reduce the cal mixing velocity generation of energy at scales that are poorly resolved by the model, the topographic depths are smoothed twice using a center-weighted nine-point real smoother. Most of the simulations used in this study are forced warm saline surface waters from the cool homogeneous with the HR monthly wind stress climatology. The in- deep water. In the context of water masses, the top three dividual wind stress components from this climatology layers approximate the location of the surface and inter- have a native grid resolution of 2.0Њ and were inter- mediate water, while the abyssal layer represents the Ja- polated to the various model grid resolutions with cubic pan Sea Proper Water. The maximum depth in the model spline interpolation. To dampen the presence of 4Њ wave- domain is 3750 m. length noise associated with the wind stress on the orig- Communication with the Paci®c Ocean occurs via inal 2Њ ®eld, the wind stress ®elds were smoothed once in¯ow through Tsushima Strait and out¯ow through the with a center-weighted nine-point real smoother. Two Tsugaru and Soya Straits at the model boundaries. A simulations, (8NEC and 8NEO; Table 2), were forced review of the literature (see Preller and Hogan 1998) with a monthly mean wind stress climatology formed indicates signi®cant variability in the distribution and from the European Centre for Medium-Range Weather amplitude of ¯ow through the straits (e.g., Miyazaki Forecasts 1979±93 10-m reanalysis winds (Gibson et 1952; Yi 1966; Toba et al. 1982). Here, in¯ow through al. 1997). Overall, the ECMWF climatology has more Tsushima Strait is divided into two separate in¯ow ports, spatial structure and stronger wind stress and wind stress one each for the western and eastern channels of Tsu- curl than the HR climatology. However, both wind sets shima Strait. The mean volume transport of the total are generally characterized by positive (negative) wind in¯ow is 2.0 Sv (Sv ϵ 106 m3 sϪ1), but a seasonal signal stress curl in the northern (southern) part of the JES, is superimposed such that the maximum in¯ow and out- and both show signi®cantly stronger wind stress during ¯ow of 2.66 Sv occurs in July and the minimum of 1.34 the winter months (associated with high atmospheric Sv occurs in January (33% peak deviation from the pressure over Siberia) relative to the summer months. mean). At any given time the in¯ow is segmented so that 75% enters through the western channel and 25% through the eastern channel. Vertically, two-thirds of 4. Linear simulations the total in¯ow (for both channels) enters the basin The framework for this study is to use an ensemble through layer 1, while the remaining one-third enters of simulations that systematically increases in dynam- through layer 2. The straits are closed in layers 3 and ical complexity. The simplest member of this ensemble 4. In¯ow through Tsushima Strait is instantaneously and is a linear 1.5-layer, reduced-gravity (RG) model, which exactly balanced by out¯ow through the Tsugaru and has one internal vertical mode and a bottom layer that Soya Straits (the Tartar Strait is neglected) using a mod- is in®nitely deep and at rest. As such, it includes only i®ed Orlanski (1976) boundary condition. The out¯ow the lowest order dynamics. Furthermore, it reveals the is distributed such that two-thirds of the ¯ow exits deterministic ocean model response to external forcing through Tsugaru Strait and the remaining one-third and provides a baseline comparison for other simula- through Soya Strait. The vertical distribution for the tions, which have added features like nonlinearity, bot- out¯ow is the same as for the in¯ow (⅔ exits through tom topography, multiple vertical modes, ¯ow instabil- layer 1, ⅓ through layer 2). These boundary conditions ities, and isopycnal outcropping. Figure 3 shows the are not based on any single observation or dynamical mean sea surface height (SSH) from three 1.5-layer,

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2542 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

TABLE 2. List of simulations.

Simulation Resolution Model layers A (m2 sϪ1) Wind forcing Generic description 8LHO 1/8Њϫ45/256Њ 1.5 50 HR Linear RG HR winds only 8LPO 1/8Њϫ45/256Њ 1.5 50 HR Linear RG ports only 8LHP 1/8Њϫ45/256Њ 1.5 50 HR Linear RG ports and HR winds 8NHO 1/8Њϫ45/256Њ 4.0 50 HR Nonlinear HR winds only 8NEO 1/8Њϫ45/256Њ 4.0 50 ECMWF Nonlinear ECMWF winds only 8NPO 1/8Њϫ45/256Њ 4.0 50 HR Nonlinear ports only 8NHP 1/8Њϫ45/256Њ 4.0 50 HR Nonlinear ports and HR winds 8NHP3 1/8Њϫ45/256Њ 3.0 50 HR As above with three active layers 8NEP 1/8Њϫ45/256Њ 4.0 50 ECMWF Nonlinear ports and ECMWF winds 8NFB 1/8Њϫ45/256Њ 4.0 50 HR Flat bottom version of 8NHP 8NRG 1/8Њϫ45/256Њ 3.5 50 HR RG version of 8NHP 16NHP 1/16Њϫ45/512Њ 4.0 15 HR 1/16Њ analog of 8NHP 32NHP 1/32Њϫ45/1024Њ 4.0 5 HR 1/32Њ analog of 8NHP 32NHPb 1/32Њϫ45/1024Њ 4.0 5 HR 32NHP with modi®ed topography 32NFB 1/32Њϫ45/1024Њ 4.0 5 HR Flat bottom version of 32NHP 32NRG 1/32Њϫ45/1024Њ 3.5 5 HR RG version of 32NHP 64NHP 1/64Њϫ45/2048Њ 4.0 2 HR 1/64Њ analog of 8NHP reduced-gravity simulations that differ only in the type there is no ¯ow through the boundary. These are evi- of external forcing applied. For these simulations, lin- dent, but small in Fig. 3b. earity was achieved by reducing the amplitude of the The mean SSH from simulation 8LPO (Table 2), wind and/or through¯ow forcing by a factor of 103. The which was driven by seasonally varying through¯ow result is very small local time derivatives, small non- forcing (only) is examined (Fig. 3a). It is uncommon to linear advective terms, and small amplitude variations consider a ``port-driven'' solution in the context of Sver- in the interface depth. Dynamically, the linear solutions drup dynamics, but this simulation simply shows the in Fig. 3 are essentially the same as Munk (1950) with impact of ¯ow through the JES in a linear fashion, and a Sverdrup (1947) interior calculated using realistic depicts the TWC as a Munk (␤1/3) boundary layer that wind and/or through¯ow forcing and coastline geom- hugs the western boundary until it is forced to separate etry, except that horizontal friction is applied every- from the coast of Korea and ¯ow eastward to satisfy where and the SSH ®eld includes a component that bal- the constraints of the out¯ow ports. Consistent with the ances the wind stress but contributes nothing to the ve- distribution of the out¯ow ports (described in section locity ®eld. However, this component does allow de- 3), about two-thirds of the ¯ow separates at the latitude partures from SSH ϭ const along the boundary where of the Tsugaru out¯ow port, and the remaining one-third

1 FIG. 3. Mean SSH from ⁄8Њ 1.5-layer reduced-gravity linear simulations forced by (a) seasonally varying through¯ow (8LPO), (b) monthly HR winds (8LHO), and (c) both through¯ow and HR winds (8LHP). Linearity was achieved by reducing the amplitude of the external forcing by 103. For plotting, the SSH has been scaled upward by the same factor. Contour interval is 1 cm.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2543 separates at the latitude of the Soya out¯ow port. The current in the JES, is totally absent and the large over- northeastward ¯ow along the Siberian coast (Fig. 3a) is shoot of the EKWC beyond its observed separation lat- needed to satisfy the constraint of the Soya out¯ow port. itude means the NKCC is absent as well. This is contrary to the observed cyclonic circulation in In the following sections, the dynamical complexity the northern part of the JES and in particular with the of the simulations is systematically increased by in- observed southwestward ¯ow along the Siberian coast cluding some combination of nonlinearity, multiple in- associated with the LCC (Fig. 2). This also results in a ternal modes, realistic bottom topography, isopycnal severe overshoot of the EKWC, which is observed to outcropping, and ¯ow instabilities. This modular ap- separate from the coast at about 37Њ±38ЊN. Indeed, proach to investigating the dynamics provides a road ``overshooting'' of boundary currents has been a prob- map for identifying the crucial processes needed to sim- lem with previous modeling efforts in the JES, partic- ulate the JES circulation in a robust and realistic fashion ularly those with relatively coarse horizontal grid res- and the major changes needed to correct the de®ciencies olution (e.g., Seung and Kim 1993). seen in the linear solutions. The mean SSH from simulation 8LHO, which was forced with the HR monthly wind stress climatology 5. Impact of nonlinearity, external forcing, and (only), is shown in Fig. 3b (Table 2). This solution is isopycnal outcropping characterized by two large gyre systems, an anticyclonic gyre in the southern half of the basin associated with The linear simulations discussed in the previous sec- negative wind stress curl and a cyclonic gyre in the tion demonstrated that most of the large-scale circula- northern half basin associated with positive wind stress tion features in the JES can be reproduced with minimal curl. Unlike simulation 8LPO, this solution has realistic dynamical ingredients when the model includes (in terms of the ¯ow direction) southwestward ¯ow through¯ow and wind forcing. A major exception is along the Siberian coast. Similar to simulation 8LPO, unrealistic southwestward ¯ow along the coast of Hon- the 8LHO simulation has a northward Munk ␤1/3 bound- shu due to the wind stress curl in the HR climatology. ary current on the western side of the anticyclonic gyre In the following sections, the model complexity is pro- in the southern part of the basin. This Munk boundary gressively increased by adding nonlinearity, bottom to- layer also hugs the coast up to about 41.5ЊN, but here pography, multiple internal modes (i.e., more layers), the separation is controlled by the overall pattern of the interface ventilation, and diapycnal mixing. The model wind stress curl, and the boundary between the two sensitivity to external forcing is also examined. The gyres is approximately the location of the zero wind dynamical consequences of these changes, which can stress curl line. The anticyclonic gyre in the southern be examined in a systematic fashion, include the pos- part of the basin results in unrealistic southwestward sibility of barotropic and baroclinic ¯ow instabilities, ¯ow along the coast of Honshu. Hence, in the linear the formation of inertial jets and boundary layers, and sense, the HR climatology acts counter to realistic north- isopycnal outcropping. eastward ¯ow along the coast of Honshu associated with As in the previous section, we begin by investigating the NB of the TWC (Fig. 2). the impact of the external forcings, individually and When the linear solution is forced with both winds collectively, but here using a series of nonlinear sim- and ports, as in simulation 8LHP (Fig. 3c), the result is ulations. The mean SSH and surface layer currents from 1 a linear superposition of simulations 8LPO and 8LHO, a sequence of 4-layer, ⁄8Њ simulations forced with sea- and in any given region the solution is altered depending sonal through¯ow and/or HR monthly winds are ex- on which of the individually forced solutions dominates. amined (Fig. 4). In addition to nonlinearity, these sim- For instance, in the area off the Siberian coast, simu- ulations include multiple internal baroclinic modes and lation 8LHP shows southwestward ¯ow because the realistic bottom topography. The addition of these dy- southwestward ¯ow from simulation 8LHO is stronger namical ingredients results in O(1) changes from the than the northeastward ¯ow that resulted from simu- linear solutions, providing much more realistic depic- lation 8LPO. The net result is realistic ¯ow to the south- tions of the current systems. Even when driven by sea- west, which is the model's representation of the LCC. sonal through¯ow forcing only (simulation 8NPO), This result suggests that the LCC is a predominantly most of the major surface layer currents, nonlinear re- wind-driven current. Although simulation 8LHP shows circulation gyres, and realistic mesoscale circulation an unrealistic latitude for the EKWC separation and a features occur throughout the JES (Fig. 4a). In contrast southwestward boundary current along the coast of to the linear simulations, the currents are inertial in char- Honshu, it is noteworthy that the rudiments of most of acter and the fronts meander, forming rings and eddies. the basic current systems are present in a solution that For instance, in simulation 8NPO, the EKWC ¯ows (only) contains the lowest order dynamics. In particular, northward along the Korean coast until it separates from simulation 8LHP shows a cyclonic (anticyclonic) gyre the coast near 41ЊN, which is farther north than the in the northern (southern) part of the basin, a polar front observed separation latitude. However, unlike simula- separating the two gyres, the southward ¯owing LCC, tion 8LPO, after separation from the coast the EKWC and the EKWC. However, the NB of the TWC, a major forms a large meander before ¯owing southward and

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2544 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

1 FIG. 4. Mean SSH from 4-layer ⁄8Њ nonlinear simulations forced by (a) seasonally varying through¯ow (8NPO), (b) HR monthly climato- logical winds (8NHO), and (c) both through¯ow and HR winds (8NHP). Contour interval is 1 cm. bifurcating into two distinct branches (Fig. 4a). The and the ¯ow along the polar front and it affects the northern branch joins the eastward ¯ow along 40ЊN con- shape of the eddy that forms where the EKWC separates tributing to the polar front, while the other branch joins from the Korean coast. Except for the separation latitude the NB, which ¯ows along the coast of Honshu. The of the EKWC, which is farther to the north than ob- latter current is totally absent in linear simulation 8LPO. served, simulation 8NHP reasonably replicates all of the In the northern part of the JES the currents are char- major current systems depicted in Fig. 2. Perhaps the acterized by overall cyclonic circulation, even in the single largest improvement in simulations 8NPO and absence of wind forcing. Some of the ¯ow from this 8NHP is the formation of the NB of the TWC along gyre exits to the Paci®c Ocean via Soya Strait, while the coast of Honshu, which was either missing (simu- the remainder forms a nonlinear recirculation gyre, in- lation 8LPO) or ¯owing in the wrong direction (simu- cluding the LCC along the Siberian coast even though lations 8LHO and 8LHP) in the linear solutions. Using wind forcing is absent. a simpli®ed model with port and wind forcing, Yoon Simulation 8NHO is identical to simulation 8NPO (1982) attributed the formation and persistence of the except that it is forced by the HR monthly wind stress NB to topographical control. This hypothesis was sup- climatology (only). Again, the impact of nonlinearity is ported in numerical studies by Kawabe (1982) and Ichi- evident. The mean SSH and surface layer currents (Fig. ye (1984). In the simulations described here, a different 4b) are characterized by cyclonic circulation in the mechanism is offered as an alternative explanation for northern part of the basin and anticyclonic circulation the formation of this boundary current. in the southern part, although ¯ow along the coast of In these simulations isopycnal outcropping is the key Honshu is quite weak and unrealistically southwestward dynamical mechanism responsible for the formation of west of the Noto Peninsula. The polar front, which is the NB of the TWC. In regions of outcropping, the also weak, marks the boundary between the two gyres. model interfaces ventilate via strong diapycnal mixing, The cyclonic gyre in the northern part of the basin is a process described in section 2. Isopycnal outcropping associated with positive wind stress curl, which pro- and ventilation can occur in response to mechanisms duces isopycnal outcropping and locally intense open such as (i) Ekman suction, (ii) coastal upwelling in re- ocean upwelling due to Ekman suction offshore from sponse to longshore winds or Kelvin wave propagation, the Siberian coast. (iii) geostrophic response of an interface to a current When both seasonal through¯ow and wind forcing (the mechanism for isopycnal outcropping north of the are included, as in simulation 8NHP (Fig. 4c), the im- polar front in simulation 8NPO, a simulation with no pact of each forcing mechanism is visible in Fig. 4. As wind forcing), or (iv) formation of a cyclonic eddy. In evidenced by the similarity of Figs. 4a and 4c, the sea- general, interface ventilation in the NLOM is a conse- sonal through¯ow forcing clearly dominates the mod- quence of having O(1) variations in layer thickness and eled circulation, particularly the meandering and bifur- is enhanced by including a relatively thin top layer in cation associated with the EKWC. However, the wind- the model (see Table 1). These conditions permit the driven component clearly strengthens the northern gyre thin top layer (60-m mean depth in these simulations)

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2545

1 FIG. 5. Mean SSH from ⁄8Њ nonlinear simulations forced by HR monthly climatological winds and seasonally varying through¯ow for (a)

4-layer case (H1 ϭ 60 m) and (b) 3-layer case where the top two layers from (a) have been combined (H1 ϭ 135 m). The removal of the thin top layer in (b) inhibits isopycnal outcropping and hence the formation of the Nearshore Branch along the coast of Honshu. Contour interval is 1 cm. to outcrop over much of the basin for most of the time. Consistent with this explanation and the wind forcing, In simulations without a thin top layer, such as simu- the ¯ow immediately below in layer 2 (which does not lation 8NHP3 (Table 2), the NB is nonexistent, and ¯ow outcrop in this region) is in the opposite direction. In south of the polar front is weakly anticyclonic (Fig. 5). contrast, Kim and Yoon (1994) used a multilevel model 1 This anticyclonic ¯ow is similar to the ¯ow depicted in with ⁄6Њ resolution to describe the seasonal variability the linear solutions that included wind forcing (e.g., of the NB, and reported that it drifted offshore due to Figs. 3b,c). In simulation 8NHP (Figs. 4c and 5a), how- the ␤ effect and developed into a meandering phase from ever, the NB is a well-de®ned boundary current along summer to autumn. the entire coast of Honshu. The meridional segments of From section 2, the outcropped layers do not attain this current remain as eastern boundary currents in the zero-layer thickness, but rather outcrop to a speci®ed mean (rather than propagating westward as nondisper- layer thickness that controls activation of vertical mix- sive Rossby waves) because the vertical mixing term is ing. In the JES, this depth (50 m) is chosen to be a larger in magnitude than the mass divergence term in nominal mixed layer depth for layer 1. The result, how- the continuity equation [Eq. (3)]. Restated, this means ever, is that layers with an outcropped bottom interface that the rate of upward vertical mixing in the region of are baroclinically inactive (i.e., do not contribute to the outcropping is faster than the rate of interface deepening pressure gradient) because they are isopycnal and have produced by westward Rossby wave propagation. Fun- minimal variation in layer thickness in outcropped re- damentally, this is an extreme example of the geo- gions. The interface ventilation for simulations 8NPO, strophic current mechanism for outcropping (``iii'' 8NHO, and 8NHP is examined using maps of the frac- above), which occurs because there is a surface layer tion of time that each interface ventilates (Fig. 6). From that has insuf®cient volume to carry the portion of the Fig. 6a, the top interface from simulation 8NPO ven- TWC transport within that layer, unless it is con®ned tilates over 90% of the time north of the polar front. near the southeastern boundary between the Tsushima Simulation 8NHO (Fig. 6b) on the other hand, produces and Tsugaru Straits. This is discussed further shortly. much less interface ventilation except in a small region

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2546 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

1 FIG. 6. Maps of the percent of time that an interface ventilates from 4-layer ⁄8Њ nonlinear simulations. (a±c) show interfaces (1±3) for through¯ow forced simulation (8NPO); (d±f) show interfaces (1±3) for HR wind forced simulation (8NHO), and (g±i) show interfaces (1±3) for simulation forced by both through¯ow and winds (8NHP). Units are percent of time; contour interval is 10%. Note 0% is a separate color.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2547

Finally, where all three interfaces ventilate, between about 42.5Њ±45ЊN along 138ЊE (Figs. 6i and 7), there is no contribution to the pressure gradient except from the barotropic mode. Hence, the mean SSH shows little variation there (Fig. 4c), and the current velocities are weak (although current transports can be large because the ¯ow is barotropic). Analysis of the simulations in Fig. 4 revealed the impact of the wind-driven circulation in the JES when HR monthly winds are used. However, different wind forcing can produce greatly different results, particularly with respect to separation of the EKWC. For example, Kim and Yoon (1996) achieved realistic separation of the EKWC in a 1.5-layer, reduced-gravity model. In that FIG. 7. Meridional cross section of mean interface depth along study the overshoot problem was diminished by using 1 138ЊE from 4-layer, ⁄8Њ simulation forced by both winds and ports the Na et al. (1992) wind stress, which is characterized (8NHP). The horizontal lines at depths 50 m, 90 m, and 130 m indicate ϩ by strong positive wind stress curl over the subpolar the cumulative depths of thehi (Table 1). For the thickness of a ϩ gyre. The strong positive wind stress curl resulted in given layer, vertical mixing is turned on when hi Յ hi . strengthening of the subpolar gyre, thereby moving the separation latitude of the EKWC to the south. To assess in the central subpolar gyre. For a given simulation, the the impact that a different wind climatology has on the effect of outcropping due to the external forcing is cu- NLOM, two simulations (8NEO and 8NEP) were forced mulative over all layers, and that is re¯ected in the in- by a monthly wind stress climatology formed from the terface ventilation maps for simulation 8NHP (Figs. 1979±1993 ECMWF 10-m reanalysis winds (see section 6g±i). 3). Except for the wind forcing, these simulations are By measuring the net mass ¯ux across each interface, identical to simulations 8NHO and 8NHP (Table 2). the strength of the mean meridional overturning cell in The mean SSH from simulation 8NEO (Fig. 8b) is simulation 8NHP is determined to be about 1.7 Sv over signi®cantly different than the mean SSH from simu- the entire basin. Of that, 0.36 Sv of exchange occurs lation 8NHO (Fig. 4b). Most striking is the large anti- with the abyssal layer. This exchange rate gives a res- cyclonic gyre in the center of the basin. Another sig- idence time for renewal of the deep water of about 130 ni®cant difference is the continuation of the southward years, close to that estimated by Kim and Kim (1997) ¯ow associated with the subpolar gyre south of 42ЊN. and Seung and Kim (1997). As Fig. 6i indicates, nearly In essence, this simulation contains a well-de®ned all of the abyssal layer outcropping occurs in the north- NKCC as well as the LCC. The NKCC ¯ows southward western part of the basin, where the third interface is to about 37.5Њ, where it separates from the coast, and ventilated over 90% of the time. Previous studies (Se- ¯ows eastward forming part of a large-scale cyclonic njyu and Sudo 1994; Seung and Yoon 1995a) indicate circulation that encompasses most of the basin, includ- that this is a likely area for deep convection, and that ing northeastward ¯ow along much of the coast of Hon- this convection produces cyclonic circulation of the in- shu. That the separation of the NKCC occurs near the termediate and deep water north of the polar front, an same latitude as the observed separation latitude of the indication which is also supported by numerical studies EKWC is signi®cant because it demonstrates that re- 1 (Seung and Kim 1993; Seung and Yoon 1995b). alistic separation of the EKWC at ⁄8Њ resolution may be To rephrase an earlier statement, the uniformly thin substantially in¯uenced by the wind-driven circulation. outcropped layers contribute very little to the pressure When the model is forced with seasonal through¯ow gradient and are dynamically inactive in that sense. As and the ECMWF monthly wind stress climatology (sim- such, a robust current (the NB) persists in the region ulation 8NEP), the separation latitude of the EKWC is where the thin top layer is not outcropping, and although much more realistic (Fig. 8c) than its HR counterpart the pressure gradient in layer 1 is due to the gradients (simulation 8NHP). In simulation 8NEP,the EKWC sep- of layer thickness anomaly in all four layers, the gra- arates from the coast near 37ЊN, but the core of the dients in layer 1 are essential for the formation of the current ¯ows northward to about 39ЊN due to the in¯u- NB, and the ¯ow is reversed in layers below the surface ence of the through¯ow forcing. After separation from layer as in the wind-forced contribution. The mean layer the coast, the EKWC ¯ows eastward along the polar structure can be depicted in meridional cross sections front until it bifurcates into two branches in response of interface depth. One cross section (Fig. 7), for sim- to the large anticyclonic gyre that was observed in sim- ulation 8NHP along 138ЊE, demonstrates that the top ulation 8NEO. The northern branch of this bifurcation interface in simulation 8NHP does not ventilate south ¯ows eastward forming part of the southern boundary of 39ЊN, the location of the NB. Likewise, the second of the subpolar gyre. Some snapshots of the SSH (not interface ventilates along the polar front at about 40.5ЊN. shown) show the southern branch of the bifurcation far-

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2548 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

1 FIG. 8. Mean SSH from 4-layer ⁄8Њ nonlinear simulations forced by (a) seasonally varying ports only (8NPO), (b) ECMWF climatological winds only (see text) (8NEO), and (c) both ports and ECMWF winds (8NEP). Contour interval is 1 cm. ther offshore than the NB, suggesting that this branch and Metzger 1998) show the importance of numerical might represent the highly variable Offshore Branch of model resolution on the mesoscale eddy ®eld by dem- the TWC. In these cases, the NB is fed more directly onstrating the effect of eddy variability in the Kuroshio by in¯ow through the eastern channel of Tsushima Extension on the Kuroshio bifurcation at the Shatsky Strait, although it tends to merge with the Offshore Rise. The mechanism through which the eddy ®eld af- Branch east of 136ЊE where the continental shelf be- fects the mean circulation is complex, but shows that comes very narrow and deepens considerably. the mean ¯ow is related to the bottom topography. Hol- The signi®cance of simulation 8NEP is that it clearly loway (1992) introduced the concept of topostress as a illustrates the model sensitivity to the atmospheric forc- means of parameterizing the topographic in¯uence on ing, particularly with respect to separation and branch- circulation. Holloway et al. (1995) included this topos- ing of the EKWC. It is reasonable, and indeed likely, tress parameterization in a relatively coarse resolution that other wind stress climatologies give signi®cantly model of the JES, and showed that it had the effect of different solutions than the two discussed here. How- increasing the strength of the NKCC and of separating ever, the sensitivity of the circulation to atmospheric the EKWC close to the observed latitude. Because of forcing is not the focus of this paper. Hence, in the NLOM ef®ciency, it was possible to progressively in- following section, we limit our discussion to simulations crease the horizontal grid resolution, thereby enhancing forced with the HR monthly wind stress climatology, the mesoscale eddy ®eld, to determine to what extent and investigate the role of horizontal grid resolution and the mean ¯ow, mesoscale eddy ®eld, and bottom to- the impact of bottom topography. Given the comparison pography are related. 1 of simulations 8NHP and 8NEP, this choice may seem Simulations 8NHP and 64NHP form the ⁄8Њ (14 km) 1 counterintuitive. However, as will be shown, by pur- and ⁄64Њ (1.7 km) end members of a sequence of sim- posefully choosing a wind stress climatology that is ulations that are identical except for the horizontal grid known a priori to give unrealistic results (e.g., overshoot resolution and the corresponding decrease in eddy vis- of the EKWC), the impact of certain dynamical pro- cosity afforded by the increase in grid resolution (Table cesses can be evaluated independent of any contribution 2). Snapshots of SSH for each grid resolution (Fig. 9) from the atmospheric forcing, and hence, in a more clear and maps of mean SSH (Fig. 10) are examined. As and concise fashion. described in section 5, the separation latitude of the EKWC in simulation 8NHP occurs farther to the north than observed. When the resolution is increased to 6. Impact of horizontal grid resolution and bottom 1 Њ (simulation 16NHP, Figs. 9b and 10b), there is mod- topography ⁄16 est improvement in the separation latitude of the EKWC Studies of the JES eddy ®eld from observations have in that the current no longer hugs the Korean coastline, suggested a relationship between the upper ocean cur- but there is still overshoot of the EKWC beyond 40ЊN. rents and the bottom topography (Toba et al., 1982; An Also, there is a marginal increase in the number of ed- 1 et al. 1994; Lie et al. 1995). Numerical studies (Hurlburt dies when the resolution is increased to ⁄16Њ, particularly

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2549

FIG. 9. Snapshots of SSH (color) and currents (vectors) on 15 Jan for 4-layer nonlinear simulations with realistic bottom 1 1 1 topography forced by through¯ow and HR winds. Resolution is (a) ⁄8Њ (8NHP), (b) ⁄16Њ (16NHP), (c) ⁄32Њ (32NHP), and (d) 1 1 ⁄64Њ (64NHP). Realistic separation of the EKWC is not achieved until ⁄32Њ resolution. Contour interval of SSH is 1 cm.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2550 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

1 1 1 FIG. 10. Mean SSH from the same suite of simulations shown in Fig. 9. Resolution is (a) ⁄8Њ (8NHP), (b) ⁄16Њ (16NHP), (c) ⁄32Њ 1 1 (32NHP), and (d) ⁄64Њ (64NHP). At least ⁄32Њ resolution is required for realistic separation of the EKWC.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2551 south of the polar front. It is not until the resolution is those from simulation 32NHP, which is another indi- 1 increased to ⁄32Њ (simulation 32NHP, Figs. 9c and 10c) cation that the mesoscale variability in simulation that the EKWC realistically separates at about 38ЊN. 64NHP has nearly converged with respect to that in Also, the eddy ®eld at this resolution is ubiquitous, simulation 32NHP. forming predominantly warm (cold) rings on the north In Fig. 11 we have seen the increase in the kinetic (south) side of the Polar Front (Fig. 9c), although the energy of the mean ¯ow and the accompanying reversal simulations at all four resolutions are eddy-resolving in the direction of the abyssal circulation as the model 1 1 and demonstrate eddies in Fig. 9. resolution is increased from ⁄8Њ to ⁄32Њ. This is the result Separation of the EKWC also occurs near 37Њ±38ЊN of strengthened baroclinic instability (actually a mixed in simulation 64NHP (Figs. 9d and 10d). Indeed, the baroclinic±barotropic instability) in the higher resolu- change in the solution when the resolution is increased tion simulation. Baroclinic instability is very ef®cient 1 1 from ⁄32Њ to ⁄64Њ is qualitatively much smaller than the at transferring energy from the upper layers into the change in solution that accompanies any of the other abyssal layer. Maps of surface and abyssal eddy kinetic changes in grid resolution (and eddy viscosity). The energy (EKE) (Figs. 12 and 13, respectively) demon- largest differences appear in the region of the Yamoto strate how dramatically the surface and abyssal eddy Basin. In that region, simulation 32NHP is characterized energy levels increase with increasing grid resolution 1 by a strong cyclonic gyre but simulation 64NHP is char- up to ⁄32Њ. Those maps also show that the EKE maxima acterized by a strong cyclonic/anticyclonic gyre pair. in the surface and abyssal layers are geographically col- However, comparison of several 10-yr means (not located, a classic signature of baroclinic instability (Hol- shown) from either simulation 32NHP or 64NHP (i.e., land and Lin 1975). from a single simulation) indicate that this is an area of There is a clear correlation between the unrealistic substantial interdecadal variability, even though these separation of the EKWC (which is accompanied by 1 simulations used climatological monthly mean forcing. weak anticyclonic deep circulation) at ⁄8Њ and realistic This is one possible reason for the differences in this separation of the EKWC (and the reversal in the deep 1 region. However, in all other regions of the JES, the circulation) at resolution of ⁄32Њ and higher. This sug- fact that the mean SSH in simulations 32NHP and gests that baroclinic instability and the inclusion of re- 64NHP is quite similar suggests that the simulation is alistic bottom topography are the critical factors in¯u- 1 nearly converged for mesoscale variability at ⁄32Њ res- encing the surface circulation in the JES. Hence, to in- olution. vestigate the role of baroclinic instability involving the Clearly, a profound change in the circulation dynam- barotropic mode, it is useful to compare simulations 1 1 ics has occurred between ⁄16Њ and ⁄32Њ resolution, par- with ¯at or realistic bottom topography, and reduced ticularly with respect to the separation latitude of the gravity simulations that exclude baroclinic instability EKWC. What are the dynamics responsible for this fun- involving the barotropic mode. To examine the impact damental change? Examination of the abyssal circula- of the bottom topography (or lack thereof) on the sur- tion (Fig. 11) suggests that the strength and direction face layer circulation, a sequence of simulations that of the abyssal circulation can strongly in¯uence the sur- include realistic bottom topography, a ¯at bottom, and face circulation. In simulation 8NHP (Fig. 11a), the deep a lowest layer that is in®nitely deep and at rest (e.g., currents are relatively weak and are largely anticyclonic. reduced-gravity simulations) were performed at both 1 1 In simulation 16NHP (Fig. 11b), localized areas of ⁄8Њ and ⁄32Њ resolution (Fig. 14; Table 2). eddy-driven deep mean ¯ows develop and are mainly The circulation features and dynamics of simulation associated with the topographic highs and lows. How- 8NHP (which included realistic bottom topography) ever, in simulation 32NHP (Fig. 11c), there is a profound were discussed in sections 5 and 6, but the mean SSH change in the abyssal circulation compared to the coars- from that simulation is repeated (Fig. 14a) for ease of er resolution simulations. In particular, the eddy-driven comparison. The mean SSH from reduced gravity sim- deep mean ¯ows are much stronger and occur over most ulation 8NRG (Fig. 14c) is strikingly similar to that from of the basin. Furthermore, there is a reversal in the cur- simulation 8NHP. Both show overshoot of the EKWC, 1 rent directions from anticyclonic at ⁄8Њ resolution to cy- a robust NB of the TWC and realistic meandering of 1 1 clonic at ⁄32Њ and ⁄64Њ resolution. Indeed, in simulation the polar front. The presence of a robust NB in simu- 32NHP (Fig. 11c) there is southward abyssal ¯ow along lation 8NRG, which does not include bottom topogra- the coast of Korea to about 38ЊN, the observed sepa- phy, is a further indication that bottom topography is ration latitude of the EKWC. Near 129ЊE, 37.5ЊN the not required for the formation of this current. Both also strength of the southward ¯owing current is 2.4 cm sϪ1, show a robust LCC off the Siberian coast. The similarity which is in close agreement with current meter mea- of the mean SSH from simulations 8NHP and 8NRG surements (Lie et al. 1989) that indicate a mean current indicates that the bottom topography and baroclinic in- speed of about 3 cm sϪ1 in nearly the same location. stability (involving the baroclinic and barotropic With the exception of the gyres in the Yamato Basin, modes), in general, have little if any in¯uence on the the strength and pathways of the mean abyssal layer surface circulation since simulation 8NRG has a lowest currents in simulation 64NHP (Fig. 11d) are close to layer that is in®nitely deep and at rest. However, sim-

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2552 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

FIG. 11. Kinetic energy per unit mass of the mean abyssal ¯ow (color) and mean abyssal currents (vectors) for the 1 1 1 same simulations shown in Fig. 9: (a) ⁄8Њ resolution (8NHP), (b) ⁄16Њ resolution (16HP), (c) ⁄32Њ resolution (32HP), and 1 2 Ϫ2 Ϫ1 (d) ⁄64Њ resolution (64NHP). Contour interval of KEM is 0.25 log10 m s , and is also expressed in cm s (mean speed) and cm2 sϪ2 (KEM per unit mass).

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2553

FIG. 12. Surface layer EKE per unit mass (color) and bottom topography (black contours) for the same simulations shown in Fig. 1 1 1 1 9: (a) ⁄8Њ resolution (8NHP), (b) ⁄16Њ resolution (16NHP), (c) ⁄32Њ resolution (32NHP), and (d) ⁄64Њ resolution (64NHP). Contour interval 2 Ϫ2 Ϫ1 2 Ϫ2 of EKE is 0.125 log10 m s , and is also expressed in cm s (mean speed) and cm s (EKE per unit mass). Contour interval of bottom topography is 400 m.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2554 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

FIG. 13. Abyssal layer EKE per unit mass (color) and bottom topography (black contours) for the same simulations shown in Fig. 1 1 1 1 9: (a) ⁄8Њ resolution (8NHP), (b) ⁄16Њ resolution (16NHP), (c) ⁄32Њ resolution (32NHP), and (d) ⁄64Њ resolution (64NHP). Contour interval 2 Ϫ2 Ϫ1 2 Ϫ2 of EKE is 0.125 log10 m s , and is also expressed in s (mean speed) and cm s (EKE per unit mass). Contour interval of bottom topography is 400 m.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2555

1 FIG. 14. Mean SSH from ⁄8Њ nonlinear simulations with (a) realistic bottom topography (8NHP), (b) a ¯at bottom (8NFB), and (c) a 1 reduced gravity version (8NRG). Panels (d)±(f) are the same except for ⁄32Њ resolution. Realistic separation of the EKWC is only achieved in simulation 32NHP. Contour interval is 1 cm. ulation 8NFB (Fig. 14b) shows notable deviations from nondispersive Rossby wave propagation) ef®ciently simulations 8NRG and 8NHP and therefore a substantial transfers energy to the abyssal layer, generating closed impact of the barotropic mode. The relatively high abys- abyssal circulation eddies with a quarter wavelength off- sal EKE in the southern half of the basin in simulation set from the surface layer. When strongly sloping bottom 8NFB (not shown) indicates that baroclinic instability topography is present, the f/h contours of the topog- is mainly responsible. Baroclinic instability is an ef®- raphy tend to inhibit baroclinic instability by inhibiting cient process for surface to abyssal KE transfer and it the formation of the associated abyssal circulation ed- is the only mechanism that can explain the larger EKE dies [because the conservation of potential vorticity lim- and stronger abyssal currents in 8NFB, because it is the its the crossing of f/h contours depending on the relative only possibility that was either absent or inhibited in amplitudes of the relative and planetary vorticity; e.g., the other two simulations. see section 5.5 of Hurlburt (1986)]. In addition, Hurlburt In baroclinic instability involving the barotropic et al. (1990) contains a clear illustration of a baroclin- mode, the abyssal circulation plays an essential role. ically unstable event and a comparison to one that is Rapid vortex stretching and compression associated barotropically unstable (see their section 3 and Figs. 4 with rapidly moving upper ocean eddies (compared to and 5). Thus, the upper ocean circulation of simulation

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2556 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

8NHP with realistic bottom topography is more like that burt and Metzger 1998), a situation which exists in the of reduced gravity simulation 8NRG than ¯at bottom JES. simulation 8NFB. The cycle of upper ocean±topographical coupling is The mean SSH ®elds for simulations 32NHP,32NFB, summarized in Fig. 15. Baroclinically unstable surface and 32NRG are examined (Figs. 14d±f, respectively). currents ef®ciently transfer energy to the abyssal layer. 1 At ⁄32Њ resolution all three simulations show striking The resulting abyssal ¯ow follows the f/h contours of differences from each other as well as changes from the bottom topography to conserve potential vorticity 1 their ⁄8Њ counterparts. For example, simulation 32NHP (Fig. 15b). The mean surface layer currents are strongly shows separation of the EKWC from the Korean coast in¯uenced by the bottom topography even though they 1 near 37ЊN, unlike its ⁄8Њ counterpart (8NHP) which do not impinge on the bottom topography (Fig. 15a). shows EKWC separation at 40Њ±41ЊN, as do both of the The EKWC separates from the coast of Korea where reduced-gravity simulations (8NRG and 32NRG). Sim- the southward abyssal ¯ow near the coast is interrupted ulation 32NFB also shows EKWC overshoot of the ob- by anticyclonic ¯ow around the ridge near 129.5ЊE, served separation latitude and at least weak northward 38.4ЊN in the bottom topography. Offshore, near 130ЊE western boundary current ¯ow as far north as 42ЊN, the latitude of the eastward ¯ow associated with this while 32NHP shows southward ¯ow along this coast as deep anticyclonic eddy coincides with the southern far as 37ЊN. A similar situation is found in the abyssal boundary of another strong southward abyssal current. layer (not shown). The upper ocean circulation of sim- While the topographic steering is strongest where upper ulation 32NHP exhibits many other departures from ocean and abyssal currents intersect at large angles, the simulations 32NFB and 32NRG, which are clearly re- end result is upper ocean and abyssal currents that are lated to the bottom topography as discussed next. nearly parallel in many locations as seen in Fig. 15. The upper ocean±topographical coupling that allows Figure 16 demonstrates unequivocally that the bottom the EKWC to separate near the observed latitude re- topography in this region is critical for separation of the quires stronger baroclinic instability and larger ampli- EKWC to occur near the observed latitude of 37Њ±38ЊN. tude abyssal layer relative vorticity (generated by vortex Figures 16a,b are simply enlargements of Figs. 15a,b 1 compression and stretching) than occurred at ⁄8Њ or in the region of the EKWC. Figures 16c,d are the same 1 ⁄16Њ resolution. This is possible only when mesoscale except from simulation 32NHPb, which is identical to 1 ¯ow instabilities are very well resolved. At least ⁄32Њ simulation 32NHP except that all of the bottom topog- resolution was needed in the JES simulations. This res- raphy shallower than 2000 m (excluding the continental olution is substantially ®ner than the ®rst internal Ross- slope) in the region where the EKWC separates from by radius of deformation, which is resolved even in the the coast, particularly the north±south oriented ridge, 1 ⁄8Њ simulations. The coupling mechanism relies on the has been set to a constant depth of 2000 m. This mod- fact that baroclinic instability is very ef®cient at trans- i®cation removes the topographic features required to ferring energy down to the abyssal layer and that en- constrain the deep mean ¯ows in this region (Fig. 16d), ergized deep ¯ows tend to follow the f/h contours in thereby removing their in¯uence on the surface circu- the abyssal layer as required by the conservation of lation. As a result, in simulation 32NHPb the EKWC potential vorticity. Using the continuity equation, Hurl- overshoots too far to the north, similar to simulation burt and Thompson (1980) show how abyssal currents 8NHP (although simulation 32NHPb is clearly more can steer upper ocean currents in a two-layer model realistic than 8NHP where the bottom topography was since not modi®ed). The similarity of 1 Њ simulations 8NHP (realistic to- ١h , (4) ⁄8 ´ ١h ϭ v ´ v 1g 1 2g 1 pography) and 8NRG (reduced gravity) suggests that where vkg is the geostrophic velocity in layer k and the realistic bottom topography plays no role in realistic left-hand side of (4) is the geostrophic advective con- separation of the EKWC nor in the surface layer JES tribution to the mass divergence term in the continuity circulation in general. However, that is an incorrect con- 1 equation (3) for layer 1. The geostrophic balance for clusion. Only when the results from the ⁄32Њ simulations the internal mode is given by are examined does the importance of the bottom to- pography become clear since only the 1 Њ simulation k ϫ f(v Ϫ v ) ϭ gЈ١h , (5) ⁄32 1g 2g 1 (32NHP) with realistic bottom topography shows an where gЈϭ(␳ 2 Ϫ ␳1)/␳o. In essence, if v1g k v 2g, then EKWC that separates at a realistic latitude (when HR -١h1 is a good measure of ␷ 1g, wind plus straits forcing is used). This result is signif and (5) state that if (4) 1 then in a two-layer model the currents in the lower layer icant because, if the ⁄32Њ simulations had not been per- can advect gradients of upper-layer thickness and thus formed, an incorrect conclusion concerning the role of steer upper-layer currents, especially where the currents the bottom topography would have been reached. Un- intersect at large angles. This argument formally breaks fortunately, this scenario may be more common than down in multilayer models, but the steering effect re- not, as few ocean models are economical enough at mains in situations where the barotropic and ®rst bar- present to run with such high resolution on a basin-scale oclinic modes are dominant (Hurlburt et al. 1996; Hurl- domain.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2557 graphical coupling via mesoscale ¯ow instabilities (mixed baroclinic±barotropic). . 15. Bottom topography and mean currents for (a) the surface layer, and (b) the abyssal layer. This ®gure illustrates the dynamics of upper ocean±topo IG F

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2558 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30

FIG. 16. Bottom topography (color) and mean currents from simulation 32NHP in the region where the EKWC separates from the coast of Korea for (a) the surface layer, and (b) the abyssal layer. The same is shown in (c) and (d) for simulation 32NHPb, which has the topographic high located near 39ЊN, 130ЊE removed.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2559

7. Summary and conclusions have suggested topographic control as a mechanism for the NB, but the nonlinear simulations performed here An ensemble of layered ocean simulations with in- suggest isopycnal outcropping as an alternate expla- creasing dynamical complexity has been used to inves- nation. tigate the circulation dynamics in the JES. The simu- All major current systems in the JES are realistically 1 lations used horizontal resolution that ranged from ⁄8Њ 1 simulated at ⁄8Њ except that the EKWC separates from 1 to ⁄64Њ for each variable and vertical resolution that the Korean coast farther to the north than the observed ranged from 1.5-layer reduced-gravity to four-layer ®- latitude when the model is forced with the HR monthly nite depth with realistic bottom topography. All include wind stress climatology. More realistic separation of the a free surface and realistic coastline geometry. Most EKWC can be achieved when the model is forced with were forced by prescribed seasonally varying transports a different wind climatology, such as the climatology through the Tsushima, Tsugaru, and Soya Straits and by formed from the 1979±1993 ECMWF 10-m reanalysis the Hellerman and Rosenstein (1983; HR) monthly wind winds used in this study. The more realistic separation stress climatology, but two were forced with a monthly latitude of the EKWC is due to the stronger positive climatology formed from the 1979±1993 ECMWF 10-m wind stress curl north of the separation latitude in this reanalysis winds. The modularity of the NLOM allowed wind product. This result suggests that great care must certain dynamical processes to be included, excluded, be exercised in the interpretation of model results, since or varied, revealing the impact of speci®c dynamical the external forcing might provide a realistic depiction processes on circulation features in the JES. of the current systems without including all the relevant Linear versions of the model revealed the impact of dynamics. Since the HR climatology resulted in over- the wind and through¯ow forcing, separately and col- 1 shoot of the EKWC at ⁄8Њ resolution, increased realism lectively. The linear versions of the model depicted the at higher resolution is due to changes in the dynamics, TWC as a Munk (1950) western boundary layer with a not the external forcing. Sverdrup (1947) interior. In these simulations, the large- When forced with HR winds, overshoot of the EKWC scale circulation was dominated by the through¯ow 1 1 persists at ⁄16Њ resolution. However, when ⁄32Њ resolution forcing, but the LCC was only reproduced when wind is used, the EKWC separates from the coast near the forcing was included. Most of the basic current systems observed latitude between 37Њ and 38ЊN. This change can be reproduced with a linear version of the model in the behavior of the EKWC as a function of grid forced with (both) through¯ow forcing and the HR wind resolution is accompanied by much higher levels of up- stress climatology. A major exception is the absence of per ocean and abyssal EKE at higher resolution and the NB, which is observed to ¯ow northeastward along distinct changes in the abyssal circulation. Speci®cally, the Honshu coast. Instead, the linear simulations show 1 the ⁄8Њ simulation shows a weak anticyclonic abyssal unrealistic southwestward ¯ow along the Honshu coast 1 circulation, whereas the ⁄32Њ simulation shows stronger, with HR wind stress forcing present and no ¯ow along cyclonic deep ¯ows. Maps of EKE in the surface and this coast with the through¯ow forcing (only). The abyssal layers show geographical colocation of high NKCC is also missing because the separation latitude EKE in these layers, indicating the presence of baro- of the EKWC is too far north. clinic instability. This instability (actually mixed bar- Nonlinear versions of the model include higher order oclinic±barotropic) is very ef®cient at transferring en- dynamical processes, such as ¯ow instabilities, inertial ergy downward. Once the energy is transferred into the jets and boundary layers, and isopycnal outcropping via lowest layer, it energizes the abyssal ¯ow, which is con- interface ventilation when some combination of realistic strained to follow the f/h contours of the bottom to- bottom topography, the barotropic mode, and multiple pography. The abyssal circulation can then in¯uence the internal modes are included. Nonlinear simulations were surface circulation via a process described in Hurlburt much more realistic than their linear counterparts and and Thompson (1980) and Hurlburt et al. (1996), even have inertial currents and fronts that form meanders and when the upper-ocean ¯ow does not impinge on the rings at all resolutions used in this study. The O(1) bottom topography. This cycle of upper ocean±topo- variations in layer thickness allow for the possibility of graphical coupling requires suf®cient levels of baro- 1 isopycnal outcropping via ventilation of model inter- clinic instability, which were only attained at ⁄32Њ res- faces. For the NLOM, isopycnal outcropping is crucial olution and higher. Hence, this effect is missed at coarser for the formation of the NB and has substantial impact resolution. on other aspects of the JES circulation, including the The roles of high horizontal grid resolution and re- 1 LCC and water mass formation. In a ⁄8Њ simulation alistic bottom topography are strongly interrelated. The 1 forced with seasonal through¯ow and the HR monthly ⁄8Њ simulations that have a bottom layer that is in®nitely wind stress climatology, the residence time for Japan/ deep and at rest (i.e., reduced-gravity) and include HR East Sea Proper Water, the abyssal layer in the model, and through¯ow forcing are very similar to those with is about 130 years. This is consistent with observation- realistic bottom topography, including almost identical 1 ally based estimates of about 80 years by Kim and Kim overshoot of the EKWC. Corresponding ⁄8Њ ¯at-bottom (1997). Some authors (e.g., Yoon 1982; Kawabe, 1982) simulations show signi®cant differences from these, but

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC 2560 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 30 again show the EKWC overshoot. This result suggests Center. The nonworkstation computations were per- that realistic bottom topography is unimportant for re- formed under grants of computer time from the Defense alistic separation of the EKWC. However, when a cor- Department High Performance Computing Initiative. responding trio of three model simulations is performed Dr. Alan Wallcraft is recognized for substantial contri- 1 at ⁄32Њ resolution, only the case with realistic bottom bution to this effort through his work on model devel- topography depicts realistic separation of the EKWC. opment and his computer expertise. Moreover, realistic separation is dependent on a speci®c, relatively shallow topographic feature near the Korean coast. When this topographic feature is removed, the REFERENCES EKWC separates too far to the north. Hence, simply An, H., K. Kim, and H. R. Shin, 1994: On the warm eddies in the using high horizontal resolution to generate mesoscale southwestern part of the East Sea (the Japan Sea). J. Korean ¯ow instabilities is not suf®cient, in and of itself, for Soc. Oceanogr., 29, 152±163. attaining upper ocean±topographical coupling; realistic Beardsley, R. C., R. Limeburner, K. Kim, and J. Candela, 1992: bottom topography is also required, particularly to attain Lagrangian ¯ow observations in the East China, Yellow, and separation of the EKWC at the observed latitude. Japan Seas. La Mer, 30, 297±314. Bleck, R., C. Rooth, D. Hu, and L. Smith, 1992: Salinity-driven The mean pathways of the current systems from the thermocline transients in a wind- and thermohaline-forced iso- 1 1 ⁄32Њ and ⁄64Њ simulations are quite similar, as are the pycnic coordinate model of the North Atlantic. J. Phys. Ocean- surface and abyssal EKE from these simulations, with ogr., 22, 1486±1505. some notable exceptions in the southern JES. In con- Cho, Y. K., and K. Kim, 1996: Seasonal variation of the East Korea 1 Warm Current and its relation with the cold water. La Mer, 34, trast, when the resolution was increased from ⁄8Њ to 172±182. 1 1 1 ⁄16Њ or from ⁄16Њ to ⁄32Њ, the changes in the solutions were Gibson, J. K., P. Kallberg, S. Uppala, A. Hernandez, A. Nomura, and quite dramatic. This comparison suggests that statistical E. Serrano, 1997: ERA description. ECMWF Re-Analysis Pro- equilibrium for mesoscale variability has been nearly ject Report Series 1, ECMWF, Reading, Berkshire, United King- achieved at 1 Њ resolution. This convergence is signif- dom, 71 pp. ⁄32 Hellerman, S., and M. Rosenstein, 1983: Normal monthly wind stress icant since a doubling of the horizontal grid resolution over the world ocean with error estimates. J. Phys. Oceanogr., usually results in signi®cant changes in the model so- 13, 1093±1104. lution. In short, the results presented in this paper de- Holland, W. R., and L. B. Lin, 1975: On the generation of mesoscale scribe the process by which the surface circulation can eddies and their contribution to the oceanic general circulation. I. A preliminary numerical experiment. J. Phys. Oceanogr., 5, be in¯uenced by the abyssal circulation through me- 642±657. soscale ¯ow instabilities. The mechanism by which this Holloway, G., 1992: Representing topographic stress for large-scale occurs, known as upper ocean±topographical coupling, ocean models. J. Phys. Oceanogr., 22, 1033±1046. requires that mesoscale eddies be very well resolved as , T. Sou, and M. Eby, 1995: Dynamics of circulation of the Japan well as the presence of realistic bottom topography. Sea. J. Mar. Res., 53, 539±569. Hurlburt, H. E., 1986: Dynamic transfer of simulated altimeter data Well-resolved eddies, certainly those smaller than the into subsurface information by a numerical ocean model. J. Geo- internal Rossby deformation radius, are needed to gen- phys. Res., 91, 2372±2400. erate suf®cient ¯ow instabilities and the presence of the , and J. D. Thompson, 1980: A numerical study of Loop Current bottom topography serves to steer upper ocean currents intrusions and eddy shedding. J. Phys. Oceanogr., 10, 1611± 1651. and to regulate the strength and location of the ¯ow , and E. J. Metzger, 1998: Bifurcation of the Kuroshio Extension instabilities. In numerical ocean models, the coupling at the Shatsky Rise. J. Geophys. Res., 103, 7549±7566. 1 requires high horizontal grid resolution, at least ⁄32Њ in , D. N. Fox, and E. J. Metzger, 1990: Statistical inference of the JES. Thus, the role of upper ocean±topographical weakly correlated subthermocline ®elds from satellite altimeter coupling is missed in coarser resolution models, which data. J. Geophys. Res., 95, 11 375±11 409. , A. J. Wallcraft, W. J. Schmitz Jr., P. J. Hogan, and E. J. Metzger, can lead to unexplained errors in the mean pathways of 1996: Dynamics of the Kuroshio/Oyashio current system using the current systems. eddy-resolving models of the North Paci®c Ocean. J. Geophys. Res., 101, 941±976. Acknowledgments. This is a contribution to the Japan/ Ichiye, T., 1984: Some problems of circulation and hydrography of the Japan Sea and the Tsushima Current. Ocean Hydrodynamics East Sea Departmental Research Initiative (DRI) spon- of the Japan and East China Seas, T. Ichiye, Ed., Elsevier, 15± sored by the Of®ce of Naval Research (ONR) under 54. Program Element 601153N. The 6.1 Low Latitude West- Kawabe, M., 1982: Branching of the Tsushima Current in the Japan ern Boundary Currents (LLWBC) project under the Sea. II. Numerical experiment. J. Oceanogr. Soc. Japan, 38, same program element provided matching funds in 183±192. Kim, C. H., and J. Y. Yoon, 1994: A numerical study on the seasonal FY98. The 6.2 Global Ocean Prediction System project variation of the Tsushima Warm Current along the coast of Japan. contributed in earlier years under Program Element Proc. Third CREAMS Workshop, Seoul, Korea, Research Insti- 602435N. The simulations were performed on Sun tute of Oceanography, Seoul National University, 73±79. workstations, the Cray T3E and YMP at the Naval Kim, K., and J. Y. Chung, 1984: On the salinity minimum and dis- solved oxygen maximum layer in the East Sea (Sea of Japan). Oceanographic Of®ce, the Cray T3E at the Army High Ocean Hydrodynamics of the Japan and East China Seas, T. Performance Computer Resource Center, and the Cray Ichiye, Ed., Elsevier, 55±65. T3D and T3E at the Arctic Region Supercomputing , and R. Legeckis, 1986: Branching of the Tsushima Current in

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC OCTOBER 2000 HOGAN AND HURLBURT 2561

1981±83. Progress in Oceanography, Vol. 17, Pergamon, 265± Senjyu, T., and H. Sudo, 1994: The upper portion of the Japan Sea 276. Proper Water: Its source and circulation as deduced from iso- , and J. Y. Yoon, 1996: Modeling of the wind-driven circulation pycnal analysis. J. Oceanogr., 50, 663±690. in the Japan Sea using a reduced gravity model. J. Oceanogr., Seung, Y.-H., 1997: Application of ventilation theory to the East Sea. 52, 359±373. J. Korean Soc. Oceanogr., 32, 8±16. , and K. Kim, 1997: What is happening in the East Sea (Japan , and K.-J. Kim, 1993: A numerical modeling of the East Sea Sea)? Recent chemical observations during CREAMS 93±96. J. circulation. J. Oceanol. Soc. Korea, 28, 292±304. Korean Soc. Oceanogr., 31, 164±172. , and , 1995: A multilayer model for dynamics of upper Kwizak, M., and A. J. Robert, 1971: A semi-implicit scheme for grid and intermediate layer circulation of the East Sea. J. Korean point atmospheric models of the primitive equations. Mon. Wea. Soc. Oceanogr., 30, 227±236. Rev., 99, 32±36. , and J.-H. Yoon, 1995a: Some features of winter convection in Levitus, S., 1982: Climatological Atlas of the World Ocean. NOAA the Japan Sea. J. Oceanogr., 51, 61±73. Prof. Paper No. 13, U.S. Govt. Printing Of®ce, 173 pp. , and , 1995b: Robust diagnostic modeling of the Japan Sea Lie, H.-J., M.-S. Suk, and C. Kim, 1989: Observations of south- circulation. J. Oceanogr., 51, 421±440. eastward deep currents off the west coast of Korea. J. Oceanol. , and K.-J. Kim, 1997: Estimation of the residence time for Soc. Korea, 24, 63±68. renewal of the East Sea Intermediate Water using MICOM. J. , S. K. Byun, I. Band, and C. H. Cho, 1995: Physical structure Korean Soc. Oceanogr., 32, 17±27. of eddies in the southwestern East Sea. J. Korean Soc. Shriver, J. F., and H. E. Hurlburt, 1997: The contribution of the global Oceanogr., 30, 170±183. thermohaline circulation to the Paci®c to Indian Ocean through- Luyten, J. R., J. Pedlosky, and H. Stommel, 1983: The ventilated ¯ow via Indonesia. J. Geophys. Res., 102, 5491±5511. thermocline. J. Phys. Oceanogr., 13, 292±309. Suda, K., and K. Hidaka, 1932: The results of the oceanographical Mesinger, F., and A. Arakawa, 1976: Numerical methods used in observations on board R.M.S. Syunpu Maru in the southern part atmospheric models. GARP Publ. Ser., No. 17, WMO, 64 pp. of the Sea of Japan in the summer of 1929, part 1 (in Japanese). Miyazaki, M., 1952: The heat budget in the Japan Sea (in Japanese). J. Oceanogr. Imp. Mar. Obs., 3, 291±375. Sverdrup, H. U., 1947: Wind-driven currents in a baroclinic ocean: Bull. Hokkaido Reg. Fish. Res. Lab., 4, 1±54. With application to the equatorial currents of the eastern Paci®c. Moore, D. R., and A. J. Wallcraft, 1998: Formulation of the NRL Proc. Natl. Acad. Sci. U.S.A., 33, 318±326. Layered Ocean Model in spherical coordinates. Naval Research Toba, Y., K. Tomizawa, Y. Kurasawa, and K. Hanawa, 1982: Seasonal Laboratory Rep. NRL CR 7323-96-0005, Naval Research Lab- and year to year variability of the Tsushima±Tsugaru Warm Cur- oratory, Stennis Space Center, 24 pp. [Available from Naval rent system with its possible cause. La Mer, 20, 41±51. Research Laboratory, Stennis Space Center, MS 39529.] Uda, M., 1934: The results of simultaneous oceanographical inves- Moriyasu, S., 1972: The Tsushima Current. Kuroshio: Its Physical tigations in the Japan Sea and its adjacent waters in May and Aspects, H. Stommel and K. Yoshida, Eds., University of Tokyo June, 1932 (in Japanese) J. Imp. Fish. Exp. Stn., 5, 57±190. Press, 353±369. Vasilev, A. S., and V. P. Makashin, 1992: Ventilation of the Japan Munk, W. H., 1950: On the wind-driven ocean circulation. J. Meteor., Sea waters in winter. La Mer, 30, 169±177. 7, 79±93. Wallcraft, A. J., 1991: The Navy Layered Ocean Model users guide. Na, J.-Y., J. W. Seo, and S. K. Han, 1992: Monthly mean sea surface NOARL Rep. 35, Naval Research Laboratory, Stennis Space winds over the adjacent seas of the Korean Peninsula (in Ko- Center, MS, 21 pp. [Available from Naval Research Laboratory, rean). J. Oceanol. Soc. Korea, 27, 1±10. Stennis Space Center, MS 39529.] Naganuma, K., 1977: The oceanographic ¯uctuations in the Japan , and D. R. Moore, 1997: A scalable implementation of the NRL Sea. Kaiyo Kayaku, 9, 137±141. Layered Ocean Model. Parallel Comput., 23, 2227±2242. National Oceanic and Atmospheric Administration, 1988: Digital re- Yarichin, V. G., 1980: Steady state of the Japan Sea circulation (in lief of the surface of the earth. Data Ann. 88-MGG-02. [Available Russian). Problems of Oceanography, V. V. Pokudov, Ed., Hy- online at http://www.ngdc.noaa.gov:80/mgg/global/seltopo.html.] drometeoizdat, Leningrad, 46±61. Orlanski, I., 1976: A simple boundary condition for unbounded hy- Yi, S. U., 1966: Seasonal and secular variations of the water volume perbolic ¯ows. J. Comput. Phys., 21, 251±269. transport across the Korea Strait. J. Oceanol. Soc. Korea, 1, Preller, R. H., and P. J. Hogan, 1998: Oceanography of the Sea of 7±13. Okhotsk and the Japan/East Sea. The Sea, Vol. 11: The Global Yoon, J. H., 1982: Numerical experiment on the circulation in the Coastal Ocean, Regional Studies and Synthesis, A. R. Robinson Japan Sea. Part III, Mechanism of the Nearshore Branch of the and K. H. Brink, Eds., John Wiley and Sons, 429±481. Tsushima Current. J. Oceanogr. Soc. Japan., 38, 125±130.

Unauthenticated | Downloaded 10/05/21 01:48 AM UTC