The Japan (East) Sea (JES) is a region of intense air-sea interactions. Recent observations show deep open ocean convection occurring south of Vladivostok near the subpolar front to a depth of roughly 1000 m during the winter of 2000, with even more intense convection during the extremely cold winter of 2001. In this region, wintertime cold continental air outbreaks, combined with strong very cold mountain-gap winds give rise to deep convection and Intermediate Water Formation. We have reproduced the 2000 event in a three dimensional numerical model and examine its onset and evolution, thus shedding some more light on this important process. The study demonstrates that the confluence of warm and cold water masses at the subtropical front south of Vladivostok is the preferred location of deep convection events in JES, as the confluence-induced downwelling assists in the deeper penetration of the convective turbulence generated by strong wintertime cooling.
 Deep Convection (DC) occurs at a few select locations in the global ocean leading to intermediate (IW) and deep water (DW) formation. IW/DW formation in the primary ocean basins is of great importance to the meridional circulation in the global ocean and plays a crucial role in the impact of oceans on climate. Labrador Sea, Greenland Sea and Weddell Sea are known to be regions of such DC and DW formation [e.g., Marshall and Schott, 1999]. However, DC occurs in semi-enclosed seas such as the Mediterranean Sea, the Sea of Okhotsk and the Sea of Japan as well, and the resulting IW formation and deep ventilation are also of importance albeit primarily to the physics and dynamics of these local seas.
 Observations of DC are very few, since it is hard to be at the right place and at the right time when it occurs. Consequently, observational data cannot be relied upon entirely to understand the initiation, evolution, termination and the impact of DC events, and appeal must be made to numerical models of DC [e.g., Marshall and Schott, 1999]. Some preconditioning of the underlying ocean is essential for deep convection to occur. Consequently, DC is observed to be associated with cyclonic circulation, which is conducive to upward doming of isopycnals, so that the pycnocline is readily eroded by strong air-sea fluxes.
 The subpolar front that separates the East Korea Warm Current water masses from the relatively cold ambient water masses of the Japan Basin is home to some of the most intense mesoscale activity observed in the global oceans; the region abounds in eddies and meanders that span a wide spectrum of scales. While Russian historical hydrographic data suggest that DC and IW formation occur during winter south of Vladivostok [Seung and Yoon, 1995], they do not occur every winter despite intense surface cooling. The primary reason appears to be the presence of cold but fresh water adjacent to the coast. It has long been thought that the dense water formation in JES occurs south of Vladivostok [e.g., Senjyu and Sudo, 1993] but definitive evidence could not be found until the surveys undertaken during the Office of Naval Research – sponsored observational program. Extensive ship measurements were made during the summer of 1999 and the winter of 2000, in addition to other observations such as deep PALACE drifting buoys [Riser and Danchenkov, 2004]. The winter 2000 and 2001 surveys concentrating on the region north of the subpolar front but south of Vladivostok captured DC events (defined as those extending to a depth of 1000 m or more) during both winters [Talley et al., 2003]. The winter of 1999–2000 was not anomalously cool [Kim et al., 2002], but profiles of oxygen from late February through March 2000 showed a double mixed layer in this region, with depths of 500 m and 1100 m (near 41.0°N, 132.3°E [Talley et al., 2003]). As can be seen in Figure 1, which shows both the AVHRR data and model-simulated SST, this location is very near a warm-core feature associated with the subpolar front and the location where the cold Primorye (or Liman Cold) Current separates from the coast. Further observations during the much colder winter of 2001 determined three stations with deep mixed layers, located between 40 and 41°N and 131°30′ and 132°E [Talley et al., 2003]. These mixed layers extended to between 800 and 1000 m. Surface and subsurface float tracks indicated that water in this region came from the Primorye Current.
 It is well known from studies in regions such as the Labrador Sea that DC in the global ocean tends to occur in the overlap region between regions of strong surface cooling and regions of weak ambient stratification [LabSea Group, 1998]. However, in JES, DC appears to be confined to only a few isolated spots south of Vladivostok. The precise mechanism for the deep penetration of convection south of Vladivostok is largely unknown. Therefore, to shed more light on DC processes in JES, we undertook a numerical simulation of the deep convection during the winter of 2000. Our results indicate that while strong convection is quite widespread off the Siberian coast in JES, open ocean DC occurs preferentially away from the coast at a few spots along the subtropical front between the cold fresh waters advected from the north and the salty warm waters to the south. The primary reason for this localization is the confluence between the warm eddy and the colder Liman Current water that gives rise to strong downwelling velocities along the front, which facilitate the deep penetration of convection into the water column. This is consistent with the suggestion of Ryabov  that dense water in JES is formed in the frontal zone where cold fresh and warm salty waters meet south of Peter the Great Bay although his conjecture was that it occurred in waters on the warmer side of the font. Danchenkov and Aubrey  recognized the need for the salty waters of the Tsushima current and proposed the location of dense water formation to be in the region found by Talley et al.  and the model simulations, but again emphasized the mechanism of surface cooling of the warm waters of the meandering Tsushima Current.
2. Model and Data Description
 The model simulations described here come from a model derived from the University of Colorado version of the Princeton sigma-coordinate model [Kantha and Clayson, 2000; see also Blumberg and Mellor, 1987]. Full details of the model can be found in Luneva and Clayson . The mixed layer model uses second moment turbulence closure [Kantha and Clayson, 1994], and includes recent developments in turbulence [Kantha, 2003]. The model makes use of the POM structure with the following modifications: sigma-coordinate levels for the upper 100 m (11 levels) and 27 z-levels at the shelf break and deep sea. The Adams-Bashworth scheme is used for the momentum equations and a predictor-corrector scheme for the transport equations. The horizontal resolution of the model is 6 km, the time step is 15 min. The model topography is derived from the 1/12° ETOPO5 data base and the model is initialized with hydrography from National Oceanography Data Center (NODC) archives. Currently sea ice is not explicitly simulated within the model, and no attempt was made to adjust the salinity of the model fields to account for ice formation and melting. The inclusion of these processes that affect the freshwater flux in the northern portion of the JES is a subject of future research. Inflow data for the Tsushima Strait inflow were provided by D. Nechaev et al. (A semi-implicit baroclinic model designed for variational data assimilation, submitted to Ocean Modelling, 2004), based on their mass inflow, temperature, salinity, and sea surface height outputs from a variational model.
 The model was initialized with data from the Khromov cruises during the summer of 1999. Hydrographic data (from Steve Riser of the University of Washington) from the PALACE floats deployed during the ONR-sponsored observational program were also assimilated into the model until 25 September 1999. The model is forced by ECMWF model data with a temporal resolution of 6 hours but a coarse spatial resolution of 1.25° × 1.25°.
3. Deep Convection During 2000
 The modeled sea surface temperature (SST) during the first week of March 2000 (Figure 1) agrees reasonably well with AVHRR-derived composite sea surface temperatures for that week. Figure 1 shows the strong winds and the cold air temperatures prevalent during this week off the coast of Siberia that are responsible for vigorous convective cooling in the water column. The modeled currents at 800 m depth are also shown with the PALACE buoy tracks overlaid. The general pattern of the modeled deep currents is consistent with data.
 The model grid point (132.3°E, 41.3°N) at which DC occurs during the first week of March is shown by a large red dot in Figure 1. This location is very close to that where Talley et al.  observed DC during the winter of 2000. It should be noted that DC depths in the model are determined from TKE; when oxygen is used as a tracer (as in Talley et al. ) oxygen penetration to 1000 m is seen at precisely where Talley et al.  observed it in their profiles.
 The temporal evolution of the temperature and salinity profiles at the DC point and an adjacent neighbor located immediately to the west, just inside the warm eddy, are shown in Figure 2. It is noteworthy that DC does not occur at this adjacent point (although lateral mixing of the newly convected water does). The reason for this can be seen in Figure 3, which shows the TKE as well as the vertical velocities in the water column along a longitudinal section at 41.26°N (the temperature and salinity distributions are also shown). The strong subduction east of the temperature front is responsible for the deep penetration of convection, while along the rest of the transect east of the front, where the vertical velocities are weak, the convection is strong but not deep. The location of DC as observed by Talley et al.  and shown in the model is therefore strongly tied to the confluence of the waters of the eddy and the colder waters from the Liman Cold Current and the consequent subduction velocities along the front. When inflow conditions through Korea Strait were changed in the model, the position of the warm anticyclonic eddy feature and the frontal location shifted, and so did the location at which DC occurs (see Luneva and Clayson  for details). Figure 2 also shows the restratification process after DC. Note that this process is very similar at both points, with lateral advection playing an important role.
 The extent to which isopycnal shallowing plays a role in the DC during this time period can be seen in Figure 4. Although strong convection occurs over much of the shallowest depth of the σ = 27.34 isopycnal, the deepest convection is generally not within this area, even after deepening caused by previous convection. Instead, it is confined to isolated spots to the east of the front.
 While water column preconditioning by cyclonic flow and intense air-sea cooling due to winter-time cold air outbreaks are the ingredients for deep convection events in many regions in the global ocean, the presence of cold fresh water at the surface along the Siberian coast suppresses deep convection, even though strong cooling occurs during cold air outbreaks. This study demonstrates that the confluence of warm and cold water masses at the subtropical front is the preferred location of DC events in JES. This is primarily due to the fact that downwelling induced by the confluence assists in the deeper penetration of the convective turbulence. The model study shows that when prevailing circulation changes, the location of the deep convection also changes, always staying at the confluence of the warm and salty and cold and fresh water masses south of Vladivostok. DC occurs preferentially on the colder side of the front and not on the warmer side as conjectured by some.
 We acknowledge with pleasure the support by the Office of Naval Research under Grants N00014-00-1-0611 and N00014-03-1-0989. Thanks to Dr. S. Riser for the PALACE buoy data, Dr. L. Talley for the ship data, and Dr. D. Nachaev for the inflow model data. This is contribution 441 of the Geophysical Fluid Dynamics Institute.