a. Observational background

The Pacific North Equatorial Countercurrent (NECC) is a major surface current in the tropical ocean, transporting more than 20 Sv (Sv ≡ 10⁶ m³ s⁻¹) of water eastward out of the warm pool region. It is centered near 5°N in the western Pacific and shifts poleward as it flows eastward to about 7°N in the central Pacific (Donguy and Meyers 1996, their Fig. 2). Its northern boundary is well defined by the adjacent westward flowing North Equatorial Current, but the location of its southern boundary is not always obvious. In the central Pacific, although it is clearly bounded near the surface by the westward South Equatorial Current (SEC), its deeper portions merge with the North Subsurface Countercurrent (NSCC; Wyrtki and Kilonsky 1984, their Fig. 2). In the western ocean, its southern edge is located as far south as 2.5°N, and may merge with the Equatorial Undercurrent (EUC) at depth. Likewise, its bottom boundary is not clear everywhere because of the NSCC. Generally, the Pacific NECC extends only slightly below the depth of the 20°C isotherm (Z₂₀), that is, not much deeper than 200 m (Wyrtki and Kilonsky 1984, their Fig. 2; G. Johnson 1999, personal communication).

Estimates of the mean NECC transport based on expendable bathythermograph (XBT) data tend to vary considerably, partly due to uncertainties in determining the locations of its northern, southern, and bottom boundaries. Using XBT data from five frequently repeated tracks, Donguy and Meyers (1996) reported an increase of mean NECC transport toward the west, with values of 12 Sv, 14.2 Sv, and 21 Sv in the eastern, central, and western Pacific, respectively. These estimates, however, are somewhat lower than those from earlier studies (e.g., Wyrtki and Kilonsky 1984; Taft and Kessler 1991), and the authors attributed this difference to selection of NECC boundaries and other factors.

b. Theoretical background

Since Sverdrup’s (1947) seminal paper, the NECC is believed to be a direct response to atmospheric forcing by wind stress curl associated with the intertropical convergence zone (ITCZ). According to Sverdrup theory, the depth-integrated, zonal geostrophic current (i.e., the transport per unit latitude) is given by the full Sverdrup transport minus the Ekman transport

View Expanded

where x and y are eastward and northward coordinates, respectively; subscripts x and y indicate partial derivatives; x_e is the location of the eastern boundary; f is the Coriolis parameter, β = f_y; and τ = (τ^x, τ^y) is the annual-mean wind stress. An estimate for the NECC geostrophic transport is then given by

View Expanded

where M⁺_G = max(M_G, 0) is the positive part of M_G, and latitude limits are chosen as y₁ = 2.5°N and y₂ = 10°N. As mentioned earlier, the choice of y₁ = 2.5°N is reasonable for the southernmost edge of the NECC; but note that the transport value is sensitive to this choice. Since M_G includes contributions from all eastward flows in this latitude band (the NSCC and part of the EUC, as well as the NECC), U_G is not a precise measurement of the total NECC transport. Indeed, since the transport of the NSCC is of the order of 10 Sv (Johnson and McPhaden 1999), U_G could be a considerable overestimate of the actual NECC transport.

The zonal Ekman transport in Eq. (1) is negligible (less than 2 m² s⁻¹, except for a small region east of 150°W and south of 5°N), therefore we will focus on the contribution of (curlτ)_y to M_G. Figure 1 plots the (curlτ)_y and M_G fields for the Florida State University (FSU; top) and European Centre for Medium-Range Weather Forecasts (ECMWF; middle) climatologies described in section 2c. Because of the tilt in the (curlτ)_y fields, M_G tends slightly poleward as it flows eastward, consistent with the observed NECC structure. Note that the corresponding plots of FSU and ECMWF fields in Fig. 1 are very similar. Landsteiner et al. (1990), however, showed that M_G is sensitive to different wind stress products when shorter record lengths are used (their Fig. 4), and concluded that more accurate winds are needed to improve ocean modeling. The total NECC transports (U_G) based on these climatologies are plotted in the bottom panel of Fig. 1. They are also similar to each other, both increasing monotonically toward the west and attaining maximum values that approach 25 Sv.

c. Present research

Given the success of Sverdrup theory, one might expect that it would be easy to simulate the NECC in numerical models. In fact, simulated NECCs tend to be weak (e.g., Philander et al. 1987; Grima et al. 1999) and/or poorly formed (see Fig. 9). Understanding why numerical models have this difficulty is the motivation for this research. We seek to answer the following questions: What processes determine the spatial structure and total transport of the Pacific NECC in ocean models? How sensitive are NECC simulations to forcing by different wind products? Do the deficiencies in simulated NECCs result from wind or model error? If it is wind error, can the nature of the error be characterized?

Our approach is to force an ocean model with different wind products, and to compare solutions with new estimates of observed NECC structures and transports determined from XBT and Pacific Ocean Analysis (POA) data, the latter being a data-assimilation product of model and observed thermal fields from the NOAA/National Centers for Environmental Prediction. The ocean model used here is a 4½-layer system with active thermodynamics and mixed layer physics, essentially a general circulation model of intermediate complexity. Solutions are forced by climatological and interannual versions of FSU and ECMWF winds.

Our results support the idea that wind inaccuracies cause poor NECC simulation, and identify the problematic aspects of the wind fields and the ocean response. One of our main results is that when solutions fail to develop a realistic NECC structure, they do so in a distinct manner by developing a discontinuity in the central ocean. This failure is traceable to wind inaccuracies in two specific regions: (curlτ)_y in the ITCZ region and τ^x in the near-equatorial region. Only when the forcing in these two regions is properly prescribed do solutions develop a NECC with both realistic spatial structure and transport.

a. Ocean observations

Climatological, zonal geostrophic currents were determined from the tropical Pacific XBT profiles (Kessler 1990) as follows. First, the average annual cycle of temperature for the period 1970–1987 was constructed from the approximately 175 000 irregularly spaced profiles, by interpolating and averaging them onto a regular grid with a resolution of Δx = 5°, Δy = 2°, Δz = 10 m, and Δt = 1 month. The depth of Z₂₀ was then determined from the gridded temperatures, and the zonal geostrophic current u_ob was estimated from Z₂₀ by

View Expanded

that is, by assuming that Z₂₀ defines the interface at the bottom of a reduced-gravity system. The depth-integrated current above Z₂₀ is

M_obu_obZ₂₀(4)

and the total NECC transport is

View Expanded

where M⁺_ob = max(M_ob, 0), and y₁ and y₂ are defined as in Eq. (2). Equation (3) requires choosing a value for the parameter Δρ/ρ. By matching the transports derived from (5) with geostrophic transports found from the full vertical resolution above Z₂₀, Kessler and Taft (1987) found that a value of Δρ/ρ = 4 × 10⁻³ best fit the data, and we therefore use that value here.

Note that these definitions for M_ob and U_ob only include transports above Z₂₀, and so differ from the corresponding Sverdrup fields M_G and U_G, which include deeper eastward flows (e.g., the NSCC). The value of U_ob should also be smaller than some of the earlier XBT estimates that included part of the NSCC.

The POA dataset consists of monthly mean estimates of the state of the tropical Pacific starting in 1980, and therefore provides information on interannual variability, as well as climatological mean. This dataset is not based entirely on observations, but rather is model output in which surface and subsurface temperature observations, including much of the XBT data, have been assimilated (Behringer et al. 1998). The assimilation improves the model’s subsurface thermal structure considerably (e.g., Ji and Smith 1995; also compare Figs. 3 and 9 below), and its climatological mean is very similar to the one obtained from the XBT data (section 3a).

To be consistent with estimates using the XBT data, NECC transports are also determined from POA’s Z₂₀ field, using eqs.(3)–(5) and the same choice of the value for Δρ/ρ. We chose the period 1980–1996 to obtain a POA climatology because of the record length of the XBT data, but the climatology is not much changed if the POA data from 1997 (a strong El Niño year) is included. Finally, the POA data were also smoothed in x with a 5-point boxcar filter since the POA data have a finer zonal resolution (Δx = 1.5°) than the XBT data.

b. Ocean model

The ocean model consists of four active layers with variables of thicknesses h_i, velocities v_i, salinities S_i, and temperatures T_i (layer index i = 1, 2, 3, or 4), overlying a deep, inert ocean where the pressure gradient vanishes (a 4½-layer system). Each of the layers represents water generated primarily by a specific process, and hence corresponds mostly to a single water-mass type: Layer 1 is the surface mixed layer, determined by Kraus–Turner (1967) physics; layer 2 is the seasonal thermocline; layers 3 and 4 represent thermocline and upper-intermediate waters, respectively. To simulate the processes of upwelling, subduction, and diapycnal mixing, fluid is allowed to transfer across the interfaces between adjacent layers. When this occurs, mass, heat and salt remain conserved. It can be regarded as an extension of the Lu et al. (1998) 3½-layer model, modified to include a seasonal-thermocline layer and the active thermodynamics described in Han et al. (1999).

Because the model is thermodynamically active, surface boundary conditions include forcing by heat and freshwater fluxes, as well as wind stress (section 2c). For climatological solutions, monthly mean shortwave radiation, longwave radiation, precipitation, air temperature, and specific humidity from the da Silva et al. (1994) dataset are used to calculate the heat and buoyancy fluxes along with model T₁. For interannual solutions, the Seager and Blumenthal (1994) approach is used to determine air temperature and humidity.

The model basin is an idealized representation of the tropical Pacific Ocean from 30°S to 30°N. The northern and southern boundaries are closed (to ensure that the system conserves mass), and S₁ and T₁ are relaxed back to the Levitus (1982) monthly mean climatology from 30° to 20° (tapering off linearly from 25° to 20°) with a timescale of one month. The grid resolution is Δx = 1° and Δy = 0.5°, and the integration time step is usually 1 hour.

When forced by climatological winds, the model is spun up from a state of rest for 4 years and year 5 is used for analysis. For interannual winds, the fourth year of the FSU climatology run is used as the initial state, and interannual winds are introduced gradually over a period of 15 days.

For comparison with the XBT and PAO results based on Z₂₀, we use the depth-integrated, zonal geostrophic current over layers 1 and 2,

M_moh₁u₁h₂u₂τ^yf,(6)

and the associated NECC geostrophic transport,

View Expanded

where M⁺_mo = max(M_mo, 0), and y₁ and y₂ are defined as in Eq. (2). These definitions essentially assume that the model’s Z₂₀ field is Z_mo = h₁ + h₂. (A more precise definition might be Z_mo = h₁ + h₂ + αh₃, with α = 0.5 say, which fits better with layer 3 being the thermocline layer and Z₂₀ being roughly the midthermocline depth. This choice, however, does not increase M_mo much because almost all of the model NECC is confined within layers 1 and 2.) Finally, the model output is smoothed in x with a 7-point boxcar filter for comparison with observations.

c. Winds

Two gridded wind products are used to force the model: FSU monthly mean pseudostresses for the period of 1961–1996, and ECMWF monthly mean wind stresses for 1985–96. Unless specified otherwise, the FSU wind stress is estimated from its pseudostress assuming an air density ρ_a = 1.2 kg m⁻³ and a drag coefficient C_D = 1.3 × 10⁻³. The ECMWF monthly mean wind stresses are based on its 12-hourly operational wind product at 10-m height, with C_D calculated according to the Large and Pond (1981) algorithm. Climatological monthly mean wind stresses are formed from both datasets using the entire records mentioned above.

During this study, we also considered other wind products. For example, the solution forced by Hellerman and Rosenstein (1983) winds was similar to that forced by FSU climatology. For interannual winds, we experimented with the European Radar Sensor (ERS) winds, which gives similar results as ECMWF winds for the period during which ERS data was available.

a. Observed NECC

The structure of the mean NECC determined from the XBT data is shown in the top panels of Fig. 2. The Z₂₀ field shoals from west to east and has a well-defined ridge (7°–10°N) and trough (2°–5°N) structure. The M_ob field shows a NECC between the ridge and trough that shifts poleward from about 4°N in the far west to 7°N in the east. There is a strong westward flow south of the NECC, the northern branch of the SEC. The total NECC transport U_ob (thick curve in the bottom panel) shows a gradual increase toward the west, with a maximum of 23 Sv near 140°E.

The Z₂₀ and M_ob fields determined from the POA data for the 1980–96 period (middle panels of Fig. 2) are very similar to the XBT fields. The most obvious difference is that the ridge in the Z₂₀ field is somewhat broader in the east in the POA data. The SEC also reaches farther west along 2°N and the northern boundary of the NECC seems to tilt more in the POA fields. The total NECC transport U_ob determined from the POA data (thin curve in the bottom panel) also agrees with the XBT estimate (thick curve).

There are intriguing similarities and differences between the M_ob and M_G fields. Sufficiently far from the equator (y ≥ 4°N) and away from the western boundary, their structures are much the same. Near the equator, however, M_ob and M_G differ markedly in that the SEC establishes a well-defined southern boundary in M_ob (Fig. 2) but not in M_G (Fig. 1). This difference is understandable since the SEC is forced by near-equatorial τ^x and is significantly influenced by momentum advection (Philander and Pacanowski 1980; McCreary 1985), and hence cannot be fully described by Sverdrup theory. This is one indication of the influence of equatorial dynamics on the NECC structure.

Note that both U_ob curves are similar to the U_G curves (bottom panel of Fig. 2). This agreement would seem to indicate a consistency between observation and theory, but, as noted in section 1b, U_G includes a contribution from the NSCC that is not included in the definition of U_ob, so it should be larger than U_ob. The fact that it is not larger is an indication that the wind products used to obtain the U_G curves are in error, specifically that their regions of positive (curlτ)_y are too weak.

The good comparison between mean XBT and POA results gives us some confidence that the POA data may be used to show observed interannual variability of the NECC. Figure 3 shows M_ob fields determined from annual-mean POA data each year from 1988 to 1991. Although the annual-mean NECCs from different years have similar spatial patterns, they differ in magnitude and also near the NECC southern boundary, where equatorial dynamics are expected to become important.

b. Climatological NECC in the model

Figure 4 shows the spatial structures of the model NECCs forced by FSU (top panels) and ECMWF (middle panels) climatological winds. The overall structures of the Z_mo fields are similar to those for Z₂₀ in Fig. 2, except that the ridges forming the northern boundary of the NECC at 7°–10°N are not as well defined in Z_mo. In contrast, M_mo fields differ strikingly from M_ob, with the model NECC (nearly) developing a discontinuity near the date line. The discontinuity seems to be related to the westward flowing SEC farther to the east, and is slightly worse in solutions forced by FSU winds than by ECMWF winds. The bottom panel of Fig. 4 shows the corresponding U_mo curves from the climatological experiments; they are quite similar east of 150°E. In general, however, U_mo is about 5 Sv smaller than U_ob in the central Pacific.

The poorly formed ridges in Z_mo and weak transports of U_mo indicates that the wind stress curl is weak in both climatologies, whereas the discontinuity, apparently linked to the SEC to its east, suggests that near-equatorial τ^x is too strong. To investigate the latter possibility, we first compared the near-equatorial τ^x fields from the two climatologies, defining an equatorial-band average τ^x

View Expanded

where L_y = 10°. The top-left panel of Fig. 5 shows that τ^x_eq(x) from the FSU climatology (heavy-dotted curve) is about 20% larger than that for ECMWF (dashed curve), consistent with the fact that the discontinuity is somewhat worse in the FSU forced solution. Then, we removed a fraction of the near-equatorial winds from the FSU climatology by subtracting γτ^x_eq(x) from τ^x at each latitude, where 0 < γ < 1. Since τ^x_eq(x) is independent of y, this modification of the forcing does not affect the (curlτ)_y field that is the Sverdrup part of the flow. In a test run using the FSU climatological winds with γ = 0.2, the discontinuity became less as the SEC weakens; the solution is much improved when γ is increased to 0.4 (top-right panel of Fig. 5). This improvement suggests that τ^x_eq(x) is too strong in the climatologies, but does not rule out the possibility that the (curlτ)_y field is too weak.

It is worth noting that the magnitude of the total NECC transport U_mo did not seem to be affected by τ^x_eq(x): As a fraction of τ^x_eq(x) was removed from the forcing field, U_mo remained much the same (bottom panel of Fig. 5). This implies that the model NECC transport mainly depends on the strength of the (curlτ)_y field, consistent with Sverdrup theory.

c. Interannual NECC in the model

Two experiments were carried out, in which the model was forced by ECMWF and FSU monthly mean winds for the period of 1985–96. These solutions provide evidence that wind error, rather than model error, is the main cause of the discontinuity in the spatial structure of the model NECC.

Figure 6 shows the annual-mean transport per unit latitude fields of M_mo from both solutions for the years of 1988–91, which is to be compared to the observed estimates in Fig. 3. The model NECC forced by ECMWF winds is poorly simulated before 1988 (not shown) because (curlτ)_y was very weak; it still shows a trace of discontinuity near the date line in 1988 (top-left panel). The solution compares well with observations for the remaining years, except that the magnitudes were somewhat weaker. In contrast, the discontinuity remains in the solution forced by FSU winds (right panels). The responses of both solutions for the subsequent 1992–96 period are similar in pattern to those of 1991 (bottom panels).

In contrast to the large spatial differences that exist between the two M_mo fields, the total NECC transport curves U_mo forced by these two wind products can sometimes be very similar because U_mo is a meridional integration of M_mo. One such example is shown in the top panels of Fig. 7: The U_mo curves of 1989 have similar magnitudes for ECMWF and FSU forced solutions (dashed and dotted curves in top-right panel, respectively), as do their U_G curves (top-left panel). Both U_mo curves of 1989, however, are smaller than U_ob estimated from the POA data for the same period (solid curve in the top-right panel). The annual-mean U_mo becomes much closer to U_ob during 1991 in the FSU solution (dotted and solid curves in the middle-right panel), and by 1993 the annual-mean U_mo in the ECMWF solution is also close to U_ob (dashed and solid curves in the bottom-right panel). Note that the improvement in U_mo fields from 1989–93 corresponds to increases in U_G fields shown in the left panels, suggesting that it resulted from improved (curlτ)_y fields due to either better wind observations and/or better assimilation techniques.

The improvement of U_mo towards U_ob during early 1990s is encouraging. However, a good comparison of U_mo and U_ob does not necessary guarantee a realistic spatial structure for the model NECC. For example, the U_mo curve of 1991 from the FSU solution compares well with U_ob (middle-right panel of Fig. 7), but its NECC spatial structure remains poor (bottom-right panel of Fig. 6). This is an indication that the simulation of NECC structure involves more than Sverdrup dynamics.

The fact that our model is able to produce a realistic NECC structure after 1988 when forced by ECMWF monthly mean winds suggests that some error in the FSU winds is likely responsible for its poor simulation. To illustrate why the model NECCs have such different spatial structures when forced by the two wind products, we consider the year 1989: The U_G curves are very similar for the two wind products during this year (top-left panel of Fig. 7), but their M_mo fields are strikingly different (upper-middle panels of Fig. 6). We then compare these wind products with in situ data from the Tropical Atmosphere Ocean (TAO) Array (McPhaden 1993) for the same period. (The TAO wind stresses were determined from buoy winds adjusted to 10-m height and the same ρ_a and C_D as for the FSU data.)

Figure 8 shows that τ^x from the FSU data is similar to that from TAO near the equator (bottom panel), but is much larger than TAO at 5°N (bottom panel). The significant reduction of τ^x from the equator to 5°N in the TAO data indicates the location and strength of the ITCZ, but this feature is missing in the FSU winds. This difference suggests that the failure of the FSU solution was due to a weak ITCZ. The ECMWF winds also show a weak ITCZ, as indicated by the small change of τ^x from the equator (bottom panel) to 5°N (top panel). However, the ECMWF winds are much weaker than the TAO winds near the equator. The fact that the ECMWF solution has a realistic spatial structure must then result from its near-equatorial τ^x being much weaker than that of the TAO and FSU winds. In other words, for the ECMWF winds weak (curlτ)_y in the region of the ITCZ is matched by weak τ^x in the near-equatorial region. Consistent with the relatively weak ITCZ indicated by Fig. 8, the total NECC transport U_mo of 1989 is rather small for both ECMWF and FSU solutions (top-right panel of Fig. 7).

It is worth noting that the TAO buoy winds go into both FSU and ECMWF analyses; particularly beginning about 1991, they become a significant constraint on those analyses. Hence, it is curious that the analyses disagree considerably even in the late 1990s, which leads to questions about how in situ winds are weighted in the analyses. It is also possible that analysis procedures involved in producing wind products are very different; therefore the quality of the fields also varies.

d. Solutions in other numerical models

It should be mentioned that poor NECC simulations also occur in other numerical models. Figure 9 shows the vertically integrated, zonal geostrophic flows (calculated the same way as M_ob for the POA data) from a solution to GFDL’s Modular Ocean Model (version 1; MOM1) forced by mean Hellerman and Rosenstein (1983) winds combined with anomalies from the FSU interannual winds using the same C_D as here. This solution is, in fact, a twin integration to the one shown in Fig. 3, except without any data assimilation. Similar to our FSU solution, the NECC in MOM1 has distinct relative minima in the central ocean during 1988 and 1990, although it is never totally interrupted by the westward flow. The MOM1 solution also has much weaker NECC transports (dash–dotted curves in the right panels of Fig. 7) than ours. The reason for these differences is probably due to model differences, but the precise processes causing them are not clear.