## Abstract

The North Pacific Subtropical Countercurrent (STCC) has a weak eastward velocity near the surface, but the region is populated with eddies. Studies have shown that the STCC is baroclinically unstable with a peak growth rate of 0.015 day^{−1} in March, and the ~60-day growth time has been used to explain the peak eddy kinetic energy (EKE) in May observed from satellites. It is argued here that this growth time from previously published normal-mode instability analyses is too slow. Growth rates calculated from an initial-value problem without the normal-mode assumption are found to be 1.5 to 2 times faster and at shorter wavelengths, due to the existence of (i) nonmodal solutions and (ii) sea surface temperature front in the mixed layer in winter. At interannual time scales it is shown that because of rapid surface adjustments, the STCC geostrophic shear, hence also the instability growth, is approximately in phase with surface forcing, leading to EKE modulation that peaks approximately 10 months later. However, the EKE can only be partially explained by this mechanism of modulation by baroclinic instability. It is suggested that the unexplained variance may be caused additionally by modulation of the EKE by dissipation.

## 1. Introduction

East of Taiwan and Luzon, a weak (a few centimeters per second) and shallow (~100 m) zonal current, called the North Pacific Subtropical Countercurrent (STCC), penetrates for thousands of kilometers into the Pacific Ocean. The STCC was predicted by Yoshida and Kidokoro (1967), although the wind-driven mechanism that they proposed is now believed not to be the principal one for the formation of the STCC (Kobashi and Kubokawa 2012). The first observation of the STCC was by Uda and Hasunuma (1969) and later confirmed by White et al. (1978) and others. The portion of the STCC from 130°E to 180° is believed to actually consist of two bands: the northern STCC between 21° and 25°N and the southern STCC between 19° and 21°N (Kobashi et al. 2006) (Fig. 1a). Farther east, between 175°E and 160°W, there is another zonal current along ~26°N, but this latter current will not be considered in this work. These STCCs are supported by subsurface density fronts produced by low potential vorticity (PV) mode waters formed in the Kuroshio and Kuroshio Extension. A review of the STCC is given by Kobashi and Kubokawa (2012).

The STCC region (defined here as 17°–27°N, 130°E–180°) is populated with eddies that typically have diameters of 150–300 km and sea surface height (SSH or *η*) anomalies of ±0.1 m (Chelton et al. 2011). Eddies propagate westward at speeds of approximately 0.05 m s^{−1}, and the horizontal velocities *u* within them have speeds of ~0.1–0.3 m s^{−1}. We define *x*, *y*, and *z* as the zonal, meridional, and vertical (positive upward, *z* = 0 at the mean sea surface) directions, and the corresponding velocity components are *u*, *υ*, and *w* and primes denote fluctuations. Using altimetry data (1992–97), Qiu (1999, hereafter Q99) showed that eddy kinetic energy (EKE) in the STCC peaks in May and is weakest in November–December. The EKE [(*u*′^{2} + *υ*′^{2})/2] seasonally correlates with a 2-month lag with the vertical shear of the zonal geostrophic current computed from monthly climatological temperature *T* and salinity *S* data (Levitus and Boyer 1994). To explain the lag, Qiu used a 2.5-layer quasigeostrophic (QG) model (two active layers overlying a quiescent, infinitely deep layer) with an eastward current in the upper layer of depth *H*_{1} = 150 m and velocity *U*_{1} = 0.03 m s^{−1} (0.01 m s^{−1}) representing the STCC in spring (fall). The second layer has *H*_{2} = 300 m and *U*_{2} = −0.03 m s^{−1}, the same for spring and fall, representing the westward-flowing North Equatorial Current (NEC). The system is baroclinically unstable with a maximum growth rate *σ*_{max} ~ 0.016 day^{−1} in spring at a wavelength of 2*π*/*k*_{max} ~ 300 km, while the value in fall is lower, *σ*_{max} ~ 0.005 day^{−1}, and the corresponding wavelength is longer at ~350 km. Kobashi and Kawamura (2002, hereafter KK02) extended the instability analysis using a 3-layer QG model (with a finite, lowest layer). Despite the different *H*_{1} and *H*_{2} used (125 and 575 m, respectively), a similar maximum growth rate ~0.015 day^{−1} for March was obtained.

The 60-day *e*-folding growth in late winter has been used by Q99 and others to explain the large EKE observed in spring (through summer). The rationale is that “the 2-month lag is the length of time the unstable waves need to fully grow” (Q99, p. 2479). On the other hand, it has long been known to forecasters (Petterssen 1955; Buzzi and Tibaldi 1978; more recently, for the ocean, see Yin and Oey 2007) that several (2–3) *e*-folding times are often required for perturbations to grow to matured weather events. In the case of the STCC, a simple estimate suffices to show that the 60-day growth is too slow. Instability analysis calculates growths of infinitesimal initial perturbations (Pedlosky 1979). Suppose the SSH perturbations are |*η*′| ≈ 10^{−2} m, which is approximately 10% of the observed values (see, e.g., Fig. 1 of Q99, showing *η*′ ~ 0.07–0.2 m in the STCC), then it would require 2–3 *e*-folding times [i.e., *e*^{n} ~ (0.07–0.2)/0.01; *n* = 2–3], or 120–180 days, for the perturbations to grow to the observed eddy amplitude, and the months of peak EKE would be in July or September, contrary to the observed peak month in May. A main goal of this manuscript is to resolve this discrepancy between theory and observation. Pedlosky (1964) noted that discrete normal modes [obtained by assuming a separable solution of the form *ψ*(*z*)*F*(*t*)*e*^{i(kx+ly)}, as in the above cited instability analyses] are, in general, incomplete and need to be supplemented by a (generally) continuous spectrum of nonmodal modes. An extreme example is the Couette problem, for which the normal-mode solution is a trivial one (Case 1960; Farrell 1982). For the Eady (1949) problem, Farrell (1982) showed that the nonmodal solution can produce growths exceeding the exponential growth in the early stages and also for wavenumbers near the neutral or decaying Eady modes (Farrell 1982, 1984, 1985). Descamps et al. (2007) demonstrated strong initial growth with a realistic case study and attributed the subsequent decreased growth to nonlinearity (cf. Orlanski and Cox 1973). In this work, instead of a normal-mode approach, we compute the unstable evolution of perturbations by direct numerical integration of an initial-value problem. The resulting growth rates are compared with those calculated using the same background state but based on normal modes. We then examine the relative importance of the sea surface temperature (SST) front on instability growths, as suggested by KK02 and Kobashi and Xie (2012), who pointed out its potential relevance to STCC variability, as well as interannual variability.

## 2. Numerical model, instability experiments, and observational data

We use the parallel version of the Princeton Ocean Model (POM; Blumberg and Mellor 1987) implemented by Jordi and Wang (2012) and developed by Oey et al. (2013) for the North Pacific Ocean: 16°S–70°N and 98°E–73°W. The horizontal resolution is 0.1° × 0.1° and there are 41 terrain-following (sigma) levels. The model is extensively described in Oey et al. (2013). Smagorinsky (1963) horizontal viscosity is used with the (nondimensional) coefficient = 0.1, which yields values of 10–100 m^{2} s^{−1} in the STCC; the horizontal diffusivity is made 10 times smaller. All surface fluxes are zero. Monthly *World Ocean Atlas* (*WOA*) data from the National Oceanographic Data Center (NODC; http://www.nodc.noaa.gov/OC5/WOA05/pr_woa05.html) for *T* and *S* are first used to diagnostically (i.e., *T* and *S* are kept fixed) integrate the model to steady state, which yields 12 monthly current fields in balance with the corresponding densities *ρ*. Starting from these balanced states, the calculation of instability is posed as an initial-value problem for each of the 12 months from January to December. Spatially random (i.e., grid scale) temperature perturbations with an amplitude |*T*′| of 0.01°C that decay exponentially from the surface with an *e*-folding depth of 500 m are seeded into each of the 12 balanced fields, and each evolution is tracked prognostically (i.e., with *T* and *S* freely evolving) for 360 days. The exponential decay follows the decay of the amplitude function of the fastest growing normal mode (see the appendix), which was also found by Tulloch et al. (2011, their Fig. 4p). However, the growth rates are found to be insensitive to other *z*-dependent forms (different *e-*folding depths, constant and sinusoidal) of the initial perturbations. For the purpose of showing faster nonmodal growths, the demonstration of any of these solutions suffices. To separate the effects of the SST front from the total solution, another set of experiments is carried out, as will be discussed in the appropriate subsection below. The above experiments track the unstable (if one exists) evolution of perturbations from a basic state that has energy contained in long but finite wavelengths (approximately >1000 km). Pierrehumbert and Swanson (1995) pointed out that eddies in nature rarely proceed from small perturbations of a nearly undisturbed jet. For a comparison with the more classical approach, another set of experiments, with zonally invariant basic flow, is also conducted in a zonal periodic channel 3°–33°N and 130°E–180° with flow-relaxation sponge zones (*T* and *S* fields are relaxed to climatology; Oey and Chen 1992) of 3° latitude width along the northern and southern boundaries, where free-slip conditions are also imposed. Zonally averaged, monthly *T* and *S* from *WOA* are specified, and the model is again integrated until the velocity is in steady balance with the density field. Perturbations are introduced and tracked prognostically, as before, for each of the 12 months. This latter set of experiments closely mimics the conventional instability analysis in which zonally parallel basic flow is perturbed, with the important difference that normal modes are not assumed. The experiment is called “Model//.”

Observational data used for analysis are weekly SSH anomalies (SSHAs) at ⅓° resolution from 1993 to 2010, from the Archiving, Validation, and Interpretation of Satellite Oceanographic data (AVISO; http://www.aviso.oceanobs.com/) project.

## 3. Validation of the modeled basic flow

The < shaped SSH contour (Fig. 1a) between 19° and 27°N is characteristic of the existence of the STCC; the southern and northern branches are as described above. The simulation also shows other familiar features besides the STCC: poleward shelf flows in the Taiwan Strait and the East China Sea (Isobe 2008); the Kuroshio along the East China Sea and Japan’s southern continental slopes (Taft 1972; Wijffels et al. 1998); recirculation gyres south of Japan and a southwestward countercurrent east of the Ryukyu Islands (Hasunuma and Yoshida 1978); the Kuroshio Extension (Niiler et al. 2003); and the westward-flowing NEC that bifurcates near 13°–14°N off the Philippines (Nitani 1972; Toole et al. 1990).

In the *y–z* sectional contours of *u* and *T* (Fig. 1b), the southern and northern branches of the STCC and their vertical shears are seen. The subtropical density front is primarily controlled by *T* (e.g., White et al. 1978), although salinity gradients tend to weaken the front (Kobashi and Kubokawa 2012). The eastward current exists in the upper 100 m where isotherms (depth *Z*_{T} measured from the free surface) slope upward from south to north, ∂*Z*_{T}/∂*y* > 0. This supports a positive shear by the thermal wind: ∂*u*/∂*z* ≈ (*N*^{2}/*f*)∂*Z*_{T}/∂*y*, in such a way that *u* ≈ 0.03 m s^{−1} near the surface (Aoki et al. 2002). Beneath the eastward current is the broad, westward return flow of the subtropical gyre that merges with the NEC in the south. The main thermocline deepens northward, that is, ∂*Z*_{T}/∂*y* and ∂*u*/∂*z* < 0. The core of this subsurface flow is 400–500 m thick, and *u* ≈ −0.03 m s^{−1}. These features are in good agreement with previous observations (e.g., Kobashi et al. 2006). The NEC is quite strong near the surface, *u* ≈ −0.16 m s^{−1}, and below it, for *z* < −300 m, a weak eastward current of a few centimeters per second is simulated; these features are also consistent with observations (e.g., Qiu and Joyce 1992).

The seasonal variation of the STCC has been examined (plots not shown). The surface current is stronger in late winter–early spring because the temperature is vertically more mixed near the surface (50 m) and ∂*Z*_{T}/∂*y* > 0 as mentioned above, creating strong meridional gradients and zonal current in the thermal wind balance. In summer–fall, the near-surface layer is stratified, isotherms flatten, and the surface STCC weakens. The PV (and ∂PV/∂*y*) distributions are consistent with the previous analyses of hydrographic observations and numerical models (Aoki et al. 2002; KK02; Kobashi et al. 2006; Yamanaka et al. 2008) with characteristic low PV mode waters north of the STCC in the potential density *σ*_{θ} ranges of 25.3–25.6 (subtropical mode water) for the northern front and additionally 25.7–26.2 (central model water) for the southern front. The resulting ∂PV/∂*y* distribution is such that its *y–z* integral from 19° to 25°N and the 0–400-m layer is ≈0 from February to April, which is close to satisfying the necessary condition for linear instability (Pedlosky 1979), consistent with the peak growth rates *σ* found during those months (Q99; KK02).

## 4. Growth rates

These are shown in Fig. 2 for different cases. In order to estimate growth rate from satellite data (*σ*_{AVISO}), eddies are first identified based on the Okubo–Weiss parameter *W* (Okubo 1970; Weiss 1991) using the velocity (*u, υ*) calculated from weekly SSHA from 1993 to 2010 assuming geostrophy:

Here, *s*_{n} = *∂u*/*∂x − ∂υ*/*∂y* and *s*_{s} = *∂υ*/*∂x* + *∂u*/*∂y* are the normal and shear components of strain, respectively, and *ω* = *∂υ*/*∂x − ∂u*/*∂y* is the relative vorticity. Eddies are connected regions within which the vorticity dominates the two strain terms, that is, *W* < 0. The identification method follows Isern-Fontanet et al. (2006), who define a vortex as a region where *W* is smaller than −0.2 times the standard deviation of *W*. Various eddy properties are then calculated:

Using *E* is similar to computing EKE by first high-passing SSHA to filter out long wavelengths and then using the resulting geostrophic velocities; it yields slightly higher EKE than the Fig. 5 of Q99 but the same seasonal variation with a maximum in May–June and a minimum in December–January. The growth rate is then computed from (e.g. Farrell 1982)

and composited from 1993 to 2010 to obtain the solid curve in Fig. 2. Since at any time the observed eddies are of finite amplitude, *σ*_{AVISO} does not correspond to the linear growth rate in the mathematically defined sense (Pedlosky 1979). One expects a weaker growth rate (Orlanski and Cox 1973; Descamps et al. 2007) than the linear growth rate *σ*_{KK02} from KK02 (shown as dotted line connected by triangles in Fig. 2); nonetheless *σ*_{AVISO} shows a seasonal peak in March and minimum and negative in fall, in remarkable consistency with *σ*_{KK02}.

The Eady (1949) (see also Gill 1982) maximum growth rate

where Ri = *N*^{2}/[(∂*u*/∂*z*)^{2} + (∂*υ*/∂*z*)^{2}] is the Richardson number and *N*^{2} = −*g*(∂*ρ*/∂*z*)/*ρ*_{o} is the buoyancy frequency, is computed from the monthly basic flow fields and averaged in the STCC region from the surface to *z* = −400 m. The resulting *σ*_{Eady} shows a statistically barely significant seasonal variation with a maximum in February and a minimum in August; its 12-month mean (0.0152 ± 10^{−3} day^{−1}) is shown in Fig. 2 as the thin horizontal line, which is close to the peak growth rate of 0.015–0.016 day^{−1} computed by Q99 and KK02. The Eady model assumes constant shear and *N* so the good agreement is fortuitous; nonetheless the classic model anticipates many of the gross instability properties detailed in Q99, for example, dependence of the growth rate on stratification and shear. The Eady model is included here for comparison only as in Smith (2007) and Tulloch et al. (2011).

The modeled growth rate is calculated using the Okubo–Weiss vortex energy method described above for AVISO but applied to the prognostic simulations for each of the 12 calendar months, then averaged over the STCC region and from *z* = 0 to −400 m, as well as for the first 60 days. The resulting *σ*_{Model} (red solid curve in Fig. 2) is insensitive to averaging periods less than ~60 days before finite-amplitude disturbances begin to appear. Q99 (his Fig. 10) estimated shears from observations such as Δ*U* = *U* at the surface minus *U* at *z* = −400 m for instability analysis, while KK02 (their Fig. 6) used the surface minus “the central depth of the westward flow” that lies slightly shallower than *z* = −400 m (their Fig. 7a) (see appendix; Fig. A2). The 400-m averaging depth is also based on Roemmich and Gilson (2001), who, with regard to observed temperature anomalies *T*′, concluded that “...the features slant westward with decreasing depth…on average from 400 m up to the sea surface.” It also agrees with the analysis of Argo data by Qiu and Chen (2010), which shows significant coherence of *T*′ confined in the top 400 m (see their Fig. 7), consistent with Roemmich and Gilson (2001). The *σ*_{Model} is further checked by computing the eddy kinetic energy plus eddy potential energy (EPE), as well as the barotropic (BT) and baroclinic (BC) conversion terms, as in Oey (2008):

Then,

The two methods yield growth rates that are within ±5% of each other. In the latitudinal range from 17° to 27°N, we find that BC ≫ BT, which supports the hypothesis of Q99 and KK02 that baroclinic instability (BCI) dominates in the STCC.

The same methods of computing growth rates are applied to the instability simulations of the zonally parallel basic flow. The corresponding *σ*_{Model//} is shown as the dashed–dotted line connected by triangles in Fig. 2. Also, the red dashed line with open circles, *σ*_{Model0–100m}, is simply *σ*_{Model}, but computed for the upper 100 m. Finally, *σ*_{NM} is the growth rate computed based on normal modes using the same monthly *WOA* data used in Model// (see the appendix).

It is apparent that during the period of large growth (0–400 m) from January through April *σ*_{Model//} > *σ*_{Model} > *σ*_{Eady} > *σ*_{KK02} ≈ *σ*_{NM}. We interpret the stronger growths (*σ*_{Model} or *σ*_{Model//}) as resulting from the existence of nonmodal, rapidly growing solutions that are absent from the normal-mode formulation of Q99 and KK02, but which are consistent with the results of Farrell (1982, 1984, 1985) and Descamps et al. (2007). This interpretation for *σ*_{Model//} > *σ*_{NM} is clear, since the difference between them is that *σ*_{NM} is from the normal mode, while *σ*_{Model//} is from the initial-value integration. On the other hand, the interpretation of *σ*_{Model//} > *σ*_{KK02} (≈*σ*_{Q99}) is less clear, because the larger *σ*_{Model//} could arise from differences in the climatologies being used. In the appendix, we show that the background states used in Q99 and KK02 are comparable to that used to compute *σ*_{Model//}. Therefore, differences in the background states cannot account for the fact that *σ*_{Model//} > *σ*_{KK02} (≈*σ*_{Q99}).

It is interesting that during strong growth months in winter, *σ*_{Model} is bounded by *σ*_{Model//}, though the difference is insignificant (it barely exceeds the ±5% uncertainty between the different methods of calculating *σ*’s). From April to August, the *σ*_{Model//} drops more rapidly than the *σ*_{Model}; the reason is unclear.

In March, the *σ*_{Model} (=0.023 day^{−1}) exceeds *σ*_{KK02} and *σ*_{NM} by about 50%, which may be compared with the 38%–45% increase in the initial growths calculated by Farrell (1982) for the Eady problem and 50% increase (also compared to *σ*_{Eady}) calculated by Descamps et al. (2007) for the initial growth of a realistic winter cyclone across the North Atlantic Ocean. It is clear that *σ*_{Model} is a more realistic estimate (than *σ*_{NM}); however, the resulting *e-*folding time of approximately 43 days is still slow. It is clear from Fig. 2 that the most rapid growth is near the surface where (*σ*_{Model0–100m})^{−1} ≈ 28 days in March. The 2–3 *e*-folding times because of this rapid surface growth are consistent with the 2–3-month period that is required for perturbations in late winter to grow into the observed EKE in late spring and summer.

KK02 (paragraphs 26–29 and their Fig. 6; see also Kobashi and Xie 2012) pointed out that in addition to being maintained by the subsurface front, the STCC also depends on the strength of the SST front in the mixed layer (Fig. 3a, black contours). The SST front is strong from December to April and weak from June to October, as seen in the contours of ∂SST/∂*y* in Fig. 3b. In particular, in February and March, the region of large |∂SST/∂*y*| extends southward into the main portion of the STCC. As a result, KK02 shows that the contribution of the SST front to the vertical shear of the STCC is large in late winter to spring and is weak in other months, especially from summer to fall (Fig. 3c, black). To study how the presence of the SST front may change the growth rate, we design a set of experiments in which the SST front is removed from the monthly climatology and the above perturbed simulations and growth rate calculations are repeated. For a given month, the frontal-removal procedure “flattens” the isotherms by merging them with the September values in the region 19°–30°N from the surface to *z* ≈ −100 m, but leaving temperatures unchanged outside this region; an example for March is shown in Fig. 3a (gray contours). September is chosen as the reference month because it is when the surface front is weakest (Fig. 3b). In the absence of the SST front, the vertical shear weakens, especially in late winter and spring (Fig. 3c, gray). The growth rate for the “no SST front case” *σ*_{ModelNF} is computed for the surface 100 m (shown in Fig. 2 as the green dashed line with open circles) to compare with *σ*_{Model0–100m}. There is about a 30% drop in the growth rate when the SST front is removed. The drop is anomalously large for March. An examination of the eddy energetics maps [using (4.6) and (4.7)] reveals that this is caused by the anomalously stronger SST front (Fig. 3b) and BC (not shown) in the region 19°–25°N, 135°–150°E.

## 5. Spatial scales

Q99 deduced a zonal wavelength *λ*_{x}|_{Q99} ≈ 300 km for the most unstable wave assuming an isotropic eddy field. KK02 found using observed EKE spectra from altimetry that the meridional scales are generally smaller than their zonal scales. Monthly zonal wavelengths of the most unstable waves computed from the normal-mode analysis of KK02 are reproduced in Fig. 4c (dotted line), showing *λ*_{x}|_{KK02} ≈ 330 km in March, increasing to a maximum ≈370 km in June. In the present case, unstable perturbations grow into eddies during each of the model 360-day run for each calendar month. Figure 4a gives an example for the March case using *W* and currents at *z* = −25 m, showing eddies with 100–300-km diameters and current speeds of 0.1–0.2 m s^{−1} in agreement with observations (Chelton et al. 2011). To determine the dominant wavelengths (*λ*_{x}, *λ*_{y}), spectral energies *E*_{W} of 10-day-averaged maps of *W*_{ω} (=*W* with signed vorticity within each eddy) in zonal and meridional directions are calculated. Then (following KK02),

The 360-day evolution of *λ*_{x} for each of the 12 calendar months is shown in Fig. 4b, and 0–60- and 60–120-day averages of *λ*_{x} are compared with *λ*_{x}|_{KK02} in Fig. 4c [The term *λ*_{y} is similar (not shown), and as in KK02 its values are generally smaller. Also, the same calculation is repeated on the *W* using the AVISO data, and good agreements with Chelton et al. (2011, section 4.2) are found.]. During the initial growing stage (*t* < 60 days), *λ*_{x} ≈ 225–270 km irrespective of the months, though waves with the fastest growth rates in February–March tend to be shorter (Fig. 4b). These wavelengths are notably shorter than those of the most unstable waves of Q99 and KK02 using normal modes and are in fact close to their short-wave cutoff wavelengths: *λ*_{min}|_{Q99} ≈ *λ*_{min}|_{KK02} ≈ 224–273 km (spring–fall; see Fig. 11b of Q99 and Fig. 11a of KK02). For comparison, for the Eady problem (Pedlosky 1979), *λ*_{min}|_{Eady} = *2π* (*NH*/*f*)/2.4 ≈ 200 km using *N* ≈ 10^{−2} s^{−1} (averaged in the STCC from 0 to 400 m using *WOA*), *H* ≈ 400 m, and *f* ≈ 5 × 10^{−5} s^{−1}. In the Eady problem, Farrell (1982, 1984) found growing nonmodal solutions near the neutral normal modes. In addition to faster growths, the shorter wavelength *λ*_{x} near the neutral normal modes of Q99 and KK02 found here during the initial growth stage is more evidence of the existence of the nonmodal solution.

Because of the idealized nature of the instability experiments, the time-asymptotic eddy state for each of the 12 independent calendar months does not have an observed correspondence. For example, because of the faster growth, March eddies tend to be the most energetic and numerous, but in reality they evolve into May as observed. Nonetheless, Fig. 4b shows that with time, eddies with larger scales emerge, as can be anticipated from the theory of geostrophic turbulence (Rhines 1979). The asymptotic scales of around 300–315 km are comparable to the longest of the eddy scales estimated by Chelton et al. (2011) from altimetry data.

## 6. Interannual variation

Qiu and Chen (2010) suggested that in the STCC, EKE is modulated by vertical geostrophic shear due to interannually varying Ekman convergence that acts on the meridional surface density gradients. The corresponding expression can be derived by integrating the heat equation across the surface layer from *z* = −*δ*_{E} to 0 (Chang et al. 2010):

where **u**_{3} = (*u*, *υ*, *w*) is velocity and *Q* is the vertical diffusive heat flux. We assume that *δ*_{E} ≈ constant (or varies slowly)^{1} and that in the thin surface layer **u**_{3} can be approximated using the wind-driven Ekman theory, then (6.1) becomes

where **k** is the *z-*unit vector, **τ**_{o} is the kinematic wind stress, **u**_{E} = −**k** × **τ**_{o}/(*fδ*_{E}) is the Ekman depth-averaged velocity, *Q*_{s}*δ*_{E} is the surface heat flux (positive upward, i.e., heat loss from the ocean), and *α*_{N}*Tδ*_{E} parameterizes the corresponding heat flux at *z* = −*δ*_{E}, that is, across the base of the layer, as a Newtonian cooling term with coefficient *α*_{N}. The second term on the RHS is due to Ekman pumping and will be neglected here, assuming that the surface layer is well mixed, ∂*T*/∂*z* ≈ 0, and/or the scale of the STCC front is much smaller than the scale of the wind, so that the second term is smaller in magnitude than the first term. Take the “grad” of (6.2):

We call **E**_{kcon} the Ekman forcing term and **G**_{s} the geostrophic shear vector through the thermal wind equation −**∇***T* = *f*/(*gα*)**k** × ∂**U**_{o}/∂*z*, where **U**_{o} is the geostrophic velocity and *α* is the thermal expansion coefficient. In the STCC, |∂/∂*x*| ≪ |∂/∂*y*|, and setting *Q*_{s} = 0 and *α*_{N} = 0 leads to

which is (3) of Qiu and Chen (2010), where for brevity we use the same symbols for scalar quantities: *G*_{s} = −∂*T*/∂*y* and *E*_{kcon} = −∂(*υ*_{E}*G*_{s})/∂*y* the Ekman convergence due to the dominance of ∂*υ*_{E}/∂*y* at interannual time scales (Qiu and Chen 2010). As seen from (6.4), (6.5) remains valid if *Q*_{s} is also included (cf. Qiu and Chen 2013).

Using reanalysis wind and SST data, Qiu and Chen (2010) calculated *E*_{kcon} and showed that it leads EKE (from AVISO) by approximately 9 months:

where the notation →_{9 mo} means “leads by 9 months.” Qiu and Chen (2010, p. 221) concluded that the 9-month lead “represents the time required for the STCC-NEC shear to adjust to the time-varying Ekman forcing plus the time for instability…to fully grow.” On the other hand, we note that (6.5) implies that *E*_{kcon} leads *G*_{s} by *π*/*2* or 10 ± 3 months, since the period of *E*_{kcon} is 3.3 ± 1 yr (see Fig. 10a of Qiu and Chen 2010)^{2}. Thus,

where means “in phase with.”

This implies a very rapid growth that, while it may be consistent with the existence of nonmodal solutions and SST front, evidence below argues for a different balance than (6.5).

It was shown above that *σ*_{AVISO} estimated from (4.4) yields the correct seasonal variation, even though the value is underestimated (Fig. 2). Interannual variations of the growth rate arise because in some years the seasonal conditions are more favorable for instability than other years (Qiu and Chen 2010). Figure 5a shows *σ*_{AVISO} and *E* [EKE computed using (4.3)] averaged in 19°*–*25°N and 130°E–180°. Years of large (weak) *E* from 1995 to 1998 and 2003 to 2007 (1993–95, 1999–2002, and after 2008) agree with the EKE in Fig. 3b of Qiu and Chen (2010; see also Fig. 15a of Qiu and Chen 2013). For convenience, we temporarily set *E* = EKE of Qiu and Chen (2010), as this does not change the logic. The lagged correlation^{3} Corr(*σ*_{AVISO}, EKE, 10) = 0.67 is maximum when *σ*_{AVISO} leads EKE by 10 months (Fig. 5a):

Since STCC eddies are produced by baroclinic instability, for which growth rate and shear are approximately in phase (e.g., Qiu and Chen 2010), we obtain

The contradiction arises because (6.5) (which leads to 6.7) is incomplete and should be replaced by (6.3):

The Newtonian cooling term with *α*_{N} in per second parameterizes (crudely) the effects of unresolved eddies and turbulence arising from, for example, mixing due to surface waves (e.g., Polton and Belcher 2007), enhanced turbulence in fronts (D’Asaro et al. 2011), or from the effects of rapid submesoscale dynamics (Boccaletti et al. 2007). At interannual time scales |∂/∂*t*| ≪ *α*_{N}, then instead of (6.7), we have

Physically, small-scale turbulent eddies are effective and geostrophic shear responds instantaneously (compared to interannual time scales) to surface forcing by Ekman convergence and/or heat flux. Using (6.12), we obtain using (6.9)

To verify relation (6.12), we compare *E* against *σ*_{AVISO} and *E*_{kcon} that is independently estimated by Qiu and Chen (2010; their Fig. 10a) using National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) monthly reanalysis data. They are plotted in Fig. 5a and their maximum lagged correlations are shown; they indicate that

Thus, *E*_{kcon} leads *σ*_{AVISO} (↔*G*_{s}) by a short but nonzero time of about 1 month, and the latter in turn leads *E* by another 10 months. Theory (6.3) and the data are now self-consistent.

### a. How can σ_{AVISO} lead *E* by so long (~10 months)?

The explanation for seasonal instability growth and EKE is relatively straightforward: March growth leads to high EKE in May and summer after several *e-*folding times. The interpretation at interannual periods is not as simple, and (6.14) does not imply a growth time of ~10 months! A plausible scenario is, in years of favorable *E*_{kcon} that leads to stronger *G*_{s}, more eddies are produced and survived. Following the first March of such strong growth, the EKE peaks in summer, and eddies survive some 10 months later through the following March when again strong growth begins, boosting the EKE, so on and so forth. This idea is consistent with observations in Chelton et al. (2011, their Figs. 11 and 13) that “…eddies with large amplitude or horizontal scale generally have longer lifetimes.” To support the idea, we calculate the AVISO (*λ*_{x}, *λ*_{y}) the same way as for the model (e.g., Fig. 4). The AVISO *λ*_{x} is found to correlate well with EKE: Corr(AVISO *λ*_{x}, *E*, −2) = 0.67, and lags it by 2 months.

### b. A different mechanism?

While the above scenario seems reasonable and the theory culminating in (6.14) self-consistent, a deeper thought on the data indicates some unsatisfactory aspects. To see, we plot the original (i.e., not low passed) time series of *E* and *σ*_{AVISO} as two-dimensional (year, calendar month) contours in Fig. 5b. One expects strong late-winter growth to lead to large summer *E* of the same year, but Fig. 5b shows that there is not always a clear connection, for example, large winter *σ*_{AVISO} in 1993 and 1994 but weak summer *E* (also in 2000 and 2001). Moreover, comparing years of large *E*, for example, 1996–97 and 2003–04, the apparently stronger *σ*_{AVISO} of the latter actually leads to a weaker *E*. The correlation between *σ*_{AVISO} averaged from February to April and *E* averaged from May to July is Corr(*σ*_{AVISO_Feb–Apr}, *E*_{May–Jul}, 0) ≈ 0.56; if the low-passed time series of Fig. 5a is used, an even lower Corr(*σ*_{AVISO}, EKE, 3) = 0.41 is obtained. These moderate correlations suggest that a substantial portion of the variance cannot be explained by the instability process alone. At long time scales, modulation of the EKE by dissipation is likely to be important (Zhai et al. 2008), such that years of high (low) EKE may be in part contributed by weaker (stronger) summertime dissipation in those years. The idea is also applicable at the seasonal time scale. As demonstrated by Zhai et al. (2008), dissipation reduces the time lag between eddy production (e.g., BCI) and EKE^{4}. Therefore, in addition to the stronger growth by nonmodal solutions and near-surface BCI, dissipation may also contribute to the relatively short lag between the peaks of March BCI and May EKE. The details depend on the relative contribution of BCI and dissipation of EKE. Much is yet to be learned, which is left for a future study using observations (e.g., Argo and satellite data) and models.

## 7. Conclusions

This manuscript revisits the problem of the instability of the STCC. We argue that the previously estimated peak growth rates of 60 day^{−1} in March are too slow in explaining the large EKE in May. Instead of the normal-mode approach, we obtain faster growths by posing an initial-value instability problem and numerically track the evolution of unstable perturbations. The resulting stronger growths near the short-wave cutoff wavelengths of neutral normal modes agree well with nonmodal growths previously found in the literature. Additionally, we show that the existence of the SST front in late winter also leads to strong growth rates. At interannual time scales, we formulate a self-consistent theory of modulation of EKE by BCI based on the idea that the STCC geostrophic shear rapidly adjusts to surface forcing, leading to growths of longer-lasting, larger, and more energetic eddies. Finally, we note that years of strong BCI and EKE do not necessarily coincide and suggest that near-surface dissipation may play a role.

## Acknowledgments

We thank two anonymous reviewers and Editor Mike Spall, whose comments improved the manuscript. YLC is supported by National Science Council under NSC 101-2119-M-003-005. Startup funding from National Taiwan Normal University is acknowledged. Support for LYO from Taiwan’s Foundation for the Advancement of Outstanding Scholarship and the National Science Council under fund NSC 100-2119-M-008-036-MY3 is acknowledged.

### APPENDIX

#### Normal-Mode Linear Instability Analysis

Linear baroclinic instability calculations based on the QG equations and with background shear and stratification from various climatological datasets have been conducted by Smith (2007) and Tulloch et al. (2011) for the global ocean. In the STCC, rough values of *σ* ≈ 0.01–0.02 day^{−1} may be inferred from their plots, which are in general agreement with the values of Q99 and KK02. Here, we also conduct normal-mode calculations employing the same monthly *WOA* datasets used in the instability integrations posed as initial-value problems; their growth rates can then be directly compared to provide further evidence for the existence of nonmodal modes. To that end, we use the simplest case of the basic flow that is zonally parallel (i.e., Model//).

We assume that the perturbed motion is quasigeostrophic and hydrostatic and consider small disturbances to a mean state with density *ρ*(*y*, *z*), so that locally the perturbation is considered to be a superposition of separable normal modes with streamfunction *ψ*(*x*, *z*, *t*):

Here, the wavelength-scale *k*^{−1} is assumed to be much smaller than the horizontal scale of the density variations and *c* = *c*_{r} + *ic*_{i} is in general complex, so that *σ*_{NM} = *kc*_{i} is the growth rate and the system is unstable for positive *c*_{i} [normal mode (NM)]. The following eigenvalue problem for the complex amplitude *ϕ* and eigenvalue *c* is then obtained (Gill et al. 1974; Pedlosky 1979):

and the boundary conditions are

Here, subscripts *z* and *y* denote differentiations, *H* is the constant (=4500 m) water depth, *N*^{2} is the buoyancy frequency, and *U*_{o} is the basic zonal velocity in thermal wind balance with *ρ*. Both *N*^{2} and *U*_{o} are the same as the monthly background states used in the initial-value integration of instability of the zonally parallel basic flow (Model//). The *N*^{2} and *U*_{o} are *y* averaged over the STCC from 19° to 25°N, so that they vary with *z* only, and barotropic instability is not considered since in the STCC it is much weaker than baroclinic instability. It is of some interest to compare these background states with those of Q99 and KK02 to estimate if they are sufficient to explain the larger growth rate *σ*_{Model} > *σ*_{KK02} found in Fig. 2. For March, when the growth rate is at maximum (see below), *N*^{2} ≈ 10^{−4} s^{−2} near the surface averaged from 0 to 200 m (or 500 m, not sensitive; Fig. A1), which yields an equivalent, reduced gravity *g*′ ≈ 0.02 m s^{−2} or greater, depending on the depth scale being used. This value is comparable to the values used by Q99 (≈0.02 m s^{−2}; his Table 2, depth scale ~300 m) and KK02 (≈0.018 m s^{−2}; their Fig. 6a, depth scale ~350 m). Figure A2 compares monthly Δ*U*_{o} = *U*_{o}(0) − *U*_{o}(−400 m) with those of Q99 and KK02 and also show comparable vertical shears in the three cases; in particular, the stratification is weaker and shear is stronger in KK02. Therefore, differences in climatologies are unlikely to cause a stronger *σ*_{Model}.

For each monthly *U*_{o} and *N*^{2}, (A.2) and (A.3) are solved for the growth rate *σ*_{NM} and phase velocity *c*_{r}, which are shown in Figs. A3a,b for March and September. The corresponding amplitudes and phases are shown in Figs. A3c,d. The monthly maximum *σ*_{NM} is plotted in Fig. 2, which shows that in March (September) *σ*_{NM} ≈ 0.014 (0.004) day^{−1}, a little smaller than *σ*_{Q99} for spring (fall) and it occurs at a wavenumber *k* ≈ 2.3 × 10^{−5} (2 × 10^{−5}) m^{−1} that is similar to Q99. The phase velocity in September, almost equal to −0.032 m s^{−1}, is stronger (more negative) than in March, almost equal to −0.026 m s^{−1}, which is also the same behavior as the solution of Q99. The steering level (where *U*_{o} = *c*_{r}) is near *z* ≈ −200 m in the westward-flowing NEC (Figs. 1b and A1) that is stronger in fall (Chang and Oey 2012). The amplitude decreases rapidly with depth, more so in September than March (Fig. A3c), with an *e*-folding depth of about 500 m, and the phase at greater depths lead the surface (Fig. A3d).

Figure 2 shows that *σ*_{NM} < *σ*_{Model//}, which suggests the existence of the nonmodal solution in the initial-value integration with a stronger growth rate *σ*_{Model//}, since the difference is *σ*_{NM} obtained from a normal-mode calculation.

## REFERENCES

*Three-Dimensional Coastal Ocean Models, Coastal Estuarine Sci.,*Vol. 4, Amer. Geophys. Union, 1–16.

*Atmosphere–Ocean Dynamics.*Academic, 662 pp.

*Temperature.*Vol. 4,

*World Ocean Atlas 1994,*NOAA Atlas NESDIS 4, 117 pp.

*Kuroshio: Its Physical Aspects,*H. Stommel and K. Yoshida, Eds., University of Tokyo Press, 129–163.

*J. Phys. Oceanogr.,*

**38,**1426–1449.

*Geophysical Fluid Dynamics.*Springer-Verlag, 624 pp.

**112,**C09020, doi:10.1029/2007JC004205.

*Kuroshio: Its Physical Aspects,*H. Stommel and K. Yoshida, Eds., University of Tokyo Press, 165–216.

**35,**L24609, doi:10.1029/2008GL036412.

## Footnotes

^{1}

Equations for a spatially and temporally variable *δ _{E}* can be derived (Chang and Oey 2011), but the added complications are not warranted in the present case.

^{2}

Qiu and Chen (2010) apparently did not notice the *π*/2-phase shift implied by their Eq. (3). The phase-shift can be seen by comparing Fig.9 of Qiu and Chen (2013) with Fig.10 of Qiu and Chen (2010).

^{3}

Here, the notation Corr(A, B, lags) is the lagged correlation coefficient between A and B with lags in months, positive (negative) if A leads (lags) B. Unless otherwise stated, all quoted correlations are above the 95% significance level.