Using a four-box model with a throughflow system, it is shown that changes in the Tsushima Warm Current could induce thermohaline variability comparable to the ones observed in the northern deep part of the East/Japan Sea. When the current becomes stronger, salt flux to the northern surface area becomes larger. The vertical stability, then, weakens to produce warm and salty deep water through deep convection a few years after the maximum of the inflow. If the inflow salinity varies synchronously as well, deep convection could occur more easily. Interdecal variation in the atmospheric temperature also could induce thermohaline variability. When the air temperature is low, cold and fresh deep water would be formed. Therefore, neither the intensification of the current during 90s nor atmospheric cooling could explain the cold and salty deep convection observed during 2000–2001. Only when both effects are combined, the observed deep convection could be properly explained.
 The East/Japan Sea (hereinafter referred to as EJS) surrounded by Korea, Japan and Russia is a mid-latitudes marginal sea. The EJS is connected to the western North Pacific through four channels, but the channels are less than 140 m deep, and the deep cold water found below the thermocline located at about 100 m is formed within the EJS. Recently, by analyzing forty year long temperature time series obtained in the northern deep part of the EJS, Watanabe et al.  reported an interdecadal variability of about 15 years long period. Dissolved oxygen (O2) and phosphate (PO4) also show similar characteristics. They argued that the variability is synchronous to that in the North Pacific Intermediate Water, and since the intermediate water cannot flow into the EJS, the atmospheric variability such as sea level pressure that drives the North Pacific variability drives that in the deep EJS as well.
 If the EJS is completely isolated from the North Pacific, the atmosphere must be the only way of producing the variability in the EJS. The dominant flow in the EJS is the Tsushima Warm Current (hereinafter referred to as TWC), a throughflow of about 2.6 Sv of warm and salty Pacific Water entering through the Korea Strait and leaving through the Tsugaru and Soya Straits. This warm and salty TWC is responsible for the net heat loss to the air in the EJS [Hirose et al., 1996; Han and Kang, 2003], and the salt transported by the current could play an important role in deep water mass formation off Vladivostok [Gamo et al., 1986; Sudo, 1986; Senjyu and Sudo, 1994; Seung and Yoon, 1995; Yoshikawa et al., 1999]. From a time series of the TWC volume transport estimated using sea level difference across the Korea Strait, Takikawa and Yoon  found a spectral peak at 15 years long period. Therefore, it is not hard to speculate that the TWC could have strong effect on the thermohaline circulation (hereinafter referred to as THC) of the EJS, whose strength is known to be less than 1 Sv [Kang et al., 2003]. Nof  has investigated the effect of salinity change in the TWC, which could induced by reduction in Changjian river discharge, on the deep water mass formation in the EJS, but the effect of long term variability in the TWC on the THC has not been investigated previously. In this paper, using a simple four box model with a throughflow, we investigated the effects of the TWC on the interdecadal variability of the THC in the EJS.
2. A Four Box Model
 The model is consisted of four well mixed boxes as in the works of Griffies and Tziperman , and a throughflow (Figure 1). The throughflow is consisted of an inflow that represents the TWC and connected to the southern surface box (Box 1), and two outflows, one of which represents the Tsugaru Current and is connected to the southern surface box and the other of which represents the Soya Current and is connected to the northern surface box (Box 2). The total area and average depth of the EJS are 1.01 × 106 km2 and 1684 m respectively [Korea Ocean Research and Development Institute, 2002]. Therefore we set the depth to 1500 m, for simplicity, and to exclude the shallow area the total volume of the model ocean to 1.3 × 1015 m3, 75% of the total volume of the EJS. The permanent thermocline, which is found at about 100 m, is the boundary between the surface and deep boxes that are 100 m and 1400 m thick, respectively. Since deep water mass formation occurs over a small area [Winton, 1995], the northern boxes (Boxes 2 and 4), which represent the area of deep convection off Vladivostok [Seung and Yoon, 1995], occupy 10% of the total similar to earlier box models on the thermohaline circulation [Park, 1999].
 The temperature and salinity in each box are determined by advection from neighboring boxes, and in the upper two boxes by surface fluxes and the throughflow. In addition local convective adjustment is utilized. This convective adjustment, however, does not have significant effect on the meridional overturning because the strength of the overturning Ψ depends linearly on the vertically averaged density difference between the southern and northern boxes. By incorporating convective adjustment, however, one can identify preferable condition for deep mixing in high latitude. The temperature and salinity of the surface boxes are restored to reference values T* with time scale γT = 15 days, and S* with time scale γS = 150 days, respectively. The equations governing the box model are then
 Here, subscripts in Arabic numbers represent box numbers, “s” the surface and “d” the deep boxes, and V and H are the volume and thickness of a box, respectively. The thermal expansion coefficient α = 2 × 10−4 °C−1, and haline expansion coefficient β = 8 × 10−4 psu−1, ρ0 = 1035 kg m−3, Ti* and Si* are reference temperature and salinity, respectively. C is a proportionality constant whose value was set to yield Ψ = 0.7 Sv when T1 = T1*, T2 = T3 = T4 = T2*, S1 = S1*, S2 = S3 = S4 = S2*. The volume transport Qin, temperature Tin and salinity Sin of the inflow are given externally. It is unclear how much of the inflow flows into the convection area. Since about 70% water introduced into the EJS by the TWC leaves the area through the Tsugaru Strait and the remaining 30% through the Soya Strait, the fraction of water flows into the northern surface box f should be a constant between 0 and 0.3. In this paper, by running the model with a few different values, we investigated the sensitivity of the circulation to f.
 In this paper we explore the property of the solution by changing parameters within the proximity of the observations rather than investigating the property of the solution analytically. In the EJS, the haline forcing is much weaker than that of the thermal forcing, and a haline mode circulation, in which water mass formation occurs in low latitudes, would not occur. Therefore, we only consider the variability of a thermal mode circulation, in which water mass formation occur in high latitudes. We, first, ran the model with a steady inflow (Qin, Tin, and Sin) and reference temperature (Ti*) and salinity (Si*) to obtain an equilibrium solution. The Euler forward method with a short enough time step was used. We then applied a sinusoidal perturbation to the inflow or the reference temperature. The amplitude of the perturbation is adjusted in such a way that the amplitude of the variability in the deep northern box T4′ becomes comparable to that reported by Watanabe et al. . Parameters used to obtain the equilibrium state are Qin = 2.6 Sv, Tin = 12°C, Sin = 34.3 psu, T1* = 10°C, T2* = −1.2°C, S1* = 34.2 psu, S2* = 33.7 psu, f = 0.1, and the equilibrium state, which is comparable to the observations [Chang et al., 2004; Kang et al., 2003], is T1 = 9.97°C, T2 = T3 = T4 = 0.23°C, S1 = 34.23 psu, S2 = S3 = S4 = 34.07 psu, and Ψ = 0.72 Sv. If we change the parameters we would obtain a different equilibrium state, but the behavior of the solution under the perturbation does not change significantly as along as the parameters remain within the proximity of the above values.
 In Figure 2, the anomalies of temperature and salinity from long term means at each box (Ti′ and Si′, i = 1, 4) when the inflow transport oscillates by 20% with 15 years long period are shown. Without the perturbation in the inflow Qin′, the circulation does not show any variability, and we can conclude that Qin′ could be conducive to generating the variability in the model. As the inflow becomes stronger, the temperature of both surface boxes increases due to intensified warm advection. The ratio of the advection to the volume is larger in the northern surface box than in the southern one, and the temperature of the former rises more rapidly. The deep boxes, of course, show weaker variability than the surface boxes because they are not under the direct influence of the throughflow. The southern deep box does not show any meaningful change because the ratio of the advection to the volume is too small. T1′ and T2′ follow Qin′ rather closely and all three quantities reach their respective maxima and minima almost synchronously within a window of 6 months wide. T4′ lags behind T2′ because it is at the downstream. The same applies to the salinity (Figure 2b). Therefore both meridional temperature and salinity differences become weaker (stronger) when the inflow intensifies (weakens), in general, and −αΔT and βΔS are nearly in opposite phase (Figure 2c). The former, however, is significantly larger than the latter and Ψ′ follows −αΔT mainly.
 Convective adjustment becomes active about two years after the maximum of Qin′, and becomes inactive about half year before the minimum of Qin′. At the onset of the convective adjustment (around year 11 in Figure 2b), the water at the surface is warmer but saltier enough than the water below to make the water column statically unstable. S2 drops suddenly (at around year 16 in Figure 2b) because of vertical mixing with the fresher deep water. The deep box is fourteen times thicker than the surface box and the vertical mixing cannot cause noticeable change in S4. As Qin′ weakens the surface water becomes cooler to make the water column statically unstable. Except for a few months during the early stage the vertical thermal gradient is the main cause of the convective adjustment. The weakening of the inflow at the same time reduces salt transport to high latitudes and at around year 16 in Figure 2c, the surface water eventually becomes fresh enough to overcome the destabilizing thermal effect and the period of convective adjustment ends about one year before the inflow starts to intensify. Salt input from the deep water due to the convective mixing stops and S2 drops sharply again while stabilizing the water column further. After the inflow starts to intensify, the northern surface water becomes saltier but warmer enough to maintain the static stability of the water column.
 As the cross frontal exchange coefficient f becomes larger (smaller), both the mean and variation of warm advection from the southern surface box to the downstream intensify. Subsequently the mean temperature and salinity in Boxes 2, 3, and 4 becomes higher (lower) as well as the variability. The relation between the temperature and salinity in each box, however, do not change significantly.
 Since a part of the TWC originates from the warm and salty Kuroshio Current [Nitani, 1972; Lie et al., 1998], as the TWC intensifies (weakens), the temperature and salinity might rise (fall) as well. In Figure 3, results from a case in which Sin varies by 0.025 psu along with Qin are presented. The overall behavior of the solution does not change, but the variation in salt transport becomes larger, and so do the amplitudes of the salinity anomalies. (The contribution of the meridional haline gradient, which is opposite to the thermal gradient, to the overturning then becomes larger and variation in the overturning becomes smaller.) In addition, convective adjustment driven by the destabilizing vertical haline gradient becomes more active (between years 7 and 10 in Figure 3b). These tendencies, of course, enhance along with the amplitude of Sin′.
 When Tin varies along with Qin, the amplitude of the variability becomes larger in all boxes. The effect of the perturbation manifests most strongly in Box 1, and the amplitude of T1′ becomes larger than that of T2′, if Tin′ becomes large enough. Otherwise a sinusoidal perturbation of a few degrees Celsius in Tin would not cause notable change in the overall property of the solution.
4. Discussion and Conclusion
 A four box model with a throughflow system shows that interdecadal variability in the Tsushima Current transport such as that reported by Takikawa and Yoon  could induce variability in deep northern temperature similar to the one reported by Watanabe et al. . Since the atmospheric variability may be able to induce the oceanic variability [Watanabe et al., 2003], we conducted an experiment in which a sinusoidal perturbation of 15 years long cycle and 0.2°C amplitude is applied to the restoring temperature (Figure 4). Since the size the EJS is comparable to the atmospheric Rossby deformation radius, the same perturbation is applied to T1* and T2*. A variation in the atmospheric thermal condition induces variability of the same frequency in all boxes as in the cases explained earlier. The amplitude of the variability, of course, is larger at the surface boxes that are in direct contact with the air.
 If we just focus on the deep northern temperature, results from the case with the inflow variability (Figure 2a) are similar to those from the case with the air temperature variability (Figure 4a). The difference between the two cases can be found in high latitude salinity and its change during the convective adjustment. When the inflow varies, T4 and S4 oscillate almost in phase, and the onset of the convective adjustment accompanies a sharp decrease of the surface salinity. Meanwhile, when the atmosphere varies T4 lags S4 behind by 6 to 7 years, and the onset, which occurs a few years after the air temperature starts to decline (year 11 in Figure 4), accompanies a sharp increase of salinity. Unfortunately, however, a salinity time series of high accuracy is not available yet, and we cannot verify the above differences using observational data directly.
 During the winter 2000–2001, cold and salty bottom water formation was observed in the northern part of the EJS [Kim et al., 2002; Senjyu et al., 2002]. As Kim et al.  argued if the deep convection was initiated only by cold winter conditions, the newly formed water would be fresher than the surrounding water because the climatological mean surface salinity is lower than that of the deep water. Therefore, the salinity of the surface water must be higher during the period of the water mass formation for the newly formed water to be salty. Takikawa and Yoon  showed the TWC transport gradually increased from 1990s to 2000s. The model shows that the surface salinity of the northern area, then, would increase while lowering the vertical stability. Around the maximum of the inflow (Only if the variation in the inflow is considered about two years after as shown in Figure 2, and if that in the inflow salinity is considered as well a few years before as shown in Figure 3.), deep convection would occur due to the destabilizing effect of salinity. The deep box is much thicker than the surface box and the effect of the convective adjustment cannot be detected in the lower box, but if the newly formed water does not mix with the ambient water instantaneously, salty but warmer water would be observed at the bottom. The bottom water formation observed during 2000–2001 cannot be explained by the inflow variability alone, either. If a cold winter occurs when the inflow is strong so that the northern surface water would become saltier than the water below, deep convection would occur easily to produce cold and salty deep water. Only when both the effects of the variability in the Tsushima Warm Current and that of the atmospheric conditions are combined, the deep convection during the winter 2000–2001 could be explained. The relative importance of the inflow to the atmospheric condition is an important issue that requires further studies.
 This research has been funded by KORDI projects PE93500, PE97604, and PP07401.