Simulations from a 49-year, realistically forced numerical model experiment indicate that decadal variability of temperature and salinity along the equator originates from subsurface spiciness anomalies in the South Pacific. Through western boundary and interior pathways in the thermocline, the subsurface anomalies in the South Pacific are first transferred westward and then northward, eventually appearing along the equator. The large spiciness anomalies in the South Pacific are formed in the eastern subtropics where large unstable salinity gradients are present in conjunction with weak stratification and strong mixing during winters. Our analysis shows that positive anomalies are generated in late winter by diapycnal mixing across isopycnal surfaces that are not exposed to the surface, i.e., through the injection process, in agreement with Yeager and Large (2004). In addition, we show that spiciness anomalies can also be created along isopycnals that outcrop to the surface through the subduction process, although this process alone is not enough to explain a significant part of the decadal variability along the equator based upon an active tracer experiment. Both the injection and subduction processes are responsible for forming positive subsurface anomalies in the eastern subtropical South Pacific, while negative anomalies there can be generated by subduction of negative surface anomalies and accumulation via isopycnal advection.
 There have been extensive studies on the decadal variability of the North Pacific and its relation to the El Niño–Southern Oscillation (ENSO), with a few different hypotheses proposed to explain the oceanic connections between the tropics and the extratropics. The mean advection hypothesis [Gu and Philander, 1997] assumes that extratropical sea surface temperature (SST) anomalies are first subducted into the thermocline, then advected to the equator within the subsurface branch of the subtropical cell (STC [McCreary and Lu, 1994]), and finally upwelled to the surface in the eastern equatorial Pacific to affect the Pacific cold tongue. Recent studies, however, indicate that temperature anomalies subducted into the pycnocline in the subtropical North Pacific may not reach the equator with any appreciable amplitude [Schneider et al., 1999a, 1999b; Nonaka and Xie, 2000]. These anomalies, although important for generating subsurface variability in the subtropical gyre [Deser et al., 1996; Zhang et al., 2001], can either be strongly dissipated, dispersed in the form of planetary-scale oceanic waves, or become obscured by wind-forced variations at low latitudes [Schneider et al., 1999a, 1999b; Liu and Shin, 1999; Nonaka and Xie, 2000].
 Another mechanism, the perturbation advection hypothesis, is proposed by Kleeman et al. , who argue that STC variability can affect the equatorial thermal structure. In this scenario, changes in the extratropical wind field alter the STC strength, rather than its temperature, thereby changing the amount of cool subtropical thermocline water that eventually upwells in the eastern tropical ocean. Recent evidence suggests that the perturbation advection hypothesis is quite efficient and plays a primary role in equatorial variability [Kleeman et al., 1999; McPhaden and Zhang, 2002; Nonaka et al., 2002; Solomon et al., 2003].
 In addition, Lysne et al.  suggests a wave mechanism, in which the thermal anomalies propagate from the central North Pacific to the western boundary as long Rossby waves, southward along the coast as coastal Kelvin waves, and eastward along the equator as equatorial Kelvin waves. However, as noted in their paper, this wave propagation mechanism does not exclude ventilated thermocline dynamics as an important link between decadal variability in the extratropics and the tropics. In reality, all of these, and possibly other processes are likely to function simultaneously to contribute to decadal changes in the equatorial thermocline.
 The South Pacific Ocean contributes as much as 70% to the water mass in the Equatorial Undercurrent [Lindstrom et al., 1987; Blanke and Raynaud, 1997], and much of the water from the South Pacific can reach the equator directly (i.e., without encountering western boundary processes) through interior pathways of the STC due to the absence of a potential vorticity island which is present in the North Pacific created by the Intertropical Convergence Zone. Therefore decadal variability of the temperature and salinity anomalies in the South Pacific can be more efficiently communicated to the equatorial ocean. On the basis of analyses of observational data, Luo and Yamagata  have shown pronounced subsurface temperature anomalies moving from the South Pacific to the equatorial ocean. Luo et al. [2003a] used a coupled ocean-atmosphere model to confirm their observational findings. By analyzing the output of a data-assimilating ocean general circulating model (GCM), Giese et al.  have suggested that the large-scale climate change that took place across the Pacific Ocean in 1976 was forced by subsurface changes in the tropical South Pacific. Model sensitivity experiments by Yang et al.  have shown that the Southern Hemisphere contributes more than the Northern Hemisphere to equatorial variability in both absolute value and efficiency under both the mean and perturbation advection hypotheses. By using a global ocean GCM driven with 40-year realistic surface forcing, Yeager and Large  show that spiciness anomalies on the σt = 25.5 isopycnal in the eastern subtropics of both hemispheres can be traced along mean geostrophic streamlines to the western boundary, where decadal salinity variations at 7°S are about twice as large as those at 12°N. Some fraction of the signals in the Northern and Southern hemispheres appear to continue along the boundary and converge on the equator.
 There have been different hypotheses on the generation of the temperature (or spiciness) anomalies in the South Pacific. Luo and Yamagata  propose an air-sea interaction process within the tropics: when a decadal positive temperature anomaly occurs in the eastern equatorial ocean the atmospheric response excites a negative wind stress curl in the western tropical South Pacific, which causes the thermocline to shoal and induce a negative temperature anomaly there. Whereas Capotondi et al.  suggest that the tropical center of thermocline variability is associated with first-mode baroclinic Rossby waves forced by anomalous Ekman pumping. A recent study by Yeager and Large , however, has shown a subsurface injection process at work in late winters that generates positive anomalies in the eastern subtropics. Meanwhile, Yeager and Large  suggest that significant interannual spiciness variability is expected wherever large unstable salinity gradients are present in conjunction with weak stratification and strong mixing in winters.
 For this study we use a sigma-coordinate ocean model [Gent and Cane, 1989; Chen et al., 1994a] to continue to explore the possibility that decadal variability along the equator originates from the South Pacific. The model is computationally efficient; thus experiments can be performed using relatively high horizontal and vertical resolutions. Moreover, the surface mixed layer, which is a key component of the subtropical upper ocean circulation, is accurately represented by this model. The model has already been proven to be a useful tool in analyzing the subtropical cells and subduction pathways in the Pacific, Indian and Atlantic Oceans [Chen et al., 1994b; Rothstein et al., 1998; Inui et al., 2002; Lazar et al., 2001, 2002; Kroger et al., 2005].
 Our model results indicate a strong connection between the South Pacific and the equator on decadal timescales, as shown in other modeling and observational studies. Moreover, in remarkably good agreement with Yeager and Large , we find largest spiciness anomalies in the South Pacific formed in the eastern subtropics where large unstable salinity gradients are present in conjunction with weak stratification and strong mixing in winters; that is, the subsurface injection process is at work. In addition to the injection process, we show that the subduction process also contributes to the generation of anomalies in the eastern subtropics, which is not apparent in the modeling of Yeager and Large . There are two differences between the injection and subduction processes in the context of this paper. Injection is expected where an isopycnal is not exposed to the surface but large unstable salinity gradients are present in conjunction with weak stratification, whereas subduction is expected where an isopycnal outcrops and sea surface temperature/salinity anomalies are found. In addition, the injection process may generate only positive anomalies in winters, whereas the subduction process can produce both positive and negative anomalies, depending upon the surface anomalies and not limited in winters.
 This paper is arranged as follows. A description of the model and experiment configurations is given in section 2. Verification of the model results against the observations is addressed in section 3. Section 4 presents an analysis of the propagation of decadal spiciness anomalies in the Pacific Ocean. Section 5 investigates the origin and generation of these decadal signals in the South Pacific, and a summary in section 6 concludes the paper.
2. Numerical Model
 The reduced gravity, multilayer ocean model that we employ is a version of the primitive equation, sigma-coordinate model developed by Gent and Cane , with the embedded hybrid vertical mixing scheme of Chen et al. [1994a]. This mixing model incorporates physics of the widely used the Kraus and Turner  mixed layer model and the Price et al.  dynamical instability model, as well as an instantaneous convective adjustment. By combining the advantages of these two models, the hybrid scheme simulates the three major physical processes of oceanic vertical turbulent mixing in a computationally efficient manner. The surface mixed layer entrainment and detrainment are related to atmospheric forcing using a bulk mixed layer model; shear flow instability (important for equatorial regimes) is accounted for by partial mixing controlled by the gradient Richardson number; and free convection in the thermocline is simulated by an instantaneous adjustment. One of the main features of this model is its ability to accurately reproduce the surface mixed layer, a key factor for this study.
 The model domain covers the Pacific basin from 55°S to 63°N and from 100°E to 70°W. The model land boundaries are taken roughly along the 200 m isobath. The latitudinal resolution is 0.3° within 10°S and 10°N and gradually increases poleward to 1° at and beyond 20°S and 20°N, and the longitudinal resolution is ∼0.3° near the western boundary and ∼1° in the ocean interior. The active model of the reduced gravity ocean is bounded below by the 27.4σt isopycnal, which is assumed to overlie an infinitely deep, motionless ocean. The model upper ocean thus has an average depth of about 1000 m in the subtropics and is divided into 17 vertical layers.
 All atmospheric forcing fields are from the Comprehensive Ocean-Atmosphere Data Set (COADS) [Da Silva et al., 1994], including wind speed, surface air temperature, relative humidity, precipitation rate, fractional cloudiness, and incoming short-wave radiation. Bulk formulae are used to calculate latent and sensible heat fluxes. For the spin-up experiments, a correction term to the heat flux is used; the model SST is restored to the Levitus  monthly climatology with a relaxation time of 90 days. Note that this restoration technique is removed in the main experiments. The model fresh water flux includes two terms; the difference in evaporation and precipitation, and essentially a restoring boundary condition on the sea surface salinity (SSS), by which the model top level salinity is restored to the Levitus  monthly seasonal climatology with a relaxation time of 90 days in spin-up experiments and 150 days in the main experiments.
 The annual-mean climatological temperature, salinity, and layer thicknesses taken from Levitus  data are used for the model initial conditions. The southern boundary is closed at 55°S with temperature, salinity, and layer thickness relaxed to monthly Levitus  climatology values in a buffer zone near the boundary. The western boundary is closed; there is no Indonesian Throughflow.
 The model is first spun up with the COADS climatological forcing fields for 20 years, and then integrated with the COADS monthly forcing fields from January 1945 to December 1993; we call this the baseline experiment. Recent experiments indicate that it takes approximately 50 years for the model to adjust in the Atlantic Ocean [Kroger et al., 2005]. In order to investigate the adjustment timescale of the model in the Pacific Ocean, we perform the following confirmation experiment in which the model is continued to run from the December 1993 fields of the baseline experiment with the climatological forcing for another 20 years and then with the monthly forcing from January 1945 to December 1993. Shown in Figures 1a and 1b are the domain-averaged kinetic energy and temperature, respectively, for the entire integrated period. The kinetic energy reaches a stationary value after only a few years, which is a typical timescale for baroclinic adjustment of the velocity field to the initial fields [e.g., Smith et al., 2000; Luo et al., 2003b]. Compared with the baseline experiment, the kinetic energy shows no significant difference from the confirmation experiment. The average temperature, however, shows significant difference between the baseline and confirmation experiments. In the baseline run, the average temperature increases by 0.7°C; whereas in the confirmation run, the temperature is in a very stationary annual cycle during the 20-year spin-up period and increases only 0.2°C during the 49-year realistic forcing period. This suggests that the model is still adjusting during the baseline run, but reaches equilibrium during the confirmation run. Note that there is a temperature trend of about 0.04°C per decade, but this is acceptable for our purposes.
 All variables presented below (temperature, salinity, velocity, etc.) are taken from the confirmation experiment, in which 588 monthly means of these variables from January 1945 through December 1993 are saved. Their monthly anomalies are obtained by subtracting a 49-year monthly climatology. Yearly means/anomalies of these variables are obtained by averaging their monthly mean/anomaly values in a year. In addition, an isopycnal depth is computed from a linear interpolation of density at adjacent levels, and the temperature and salinity anomalies on that depth are then calculated from the linear interpolation of their values in the adjacent levels.
 Note that the temperature and salinity variability on the time averaged isopycnal can be partitioned into two components attributable to different physical mechanisms; “spiciness” is due to changes in water mass characteristics on isopycnal surfaces, and “heave” is due to vertical and lateral displacement of isopycnal surfaces [Doney et al., 2003]. Spiciness describes the temperature and salinity of water of a given density, with hot and salty water having a high spiciness [Munk, 1981]. In the upper ocean, with an approximate constant equation of state, spiciness is a passive tracer. Upon surfacing, however, spiciness anomalies affect air-sea interaction due to their temperature signals [Schneider, 2004]. Our analysis is performed on time varying isopycnal surfaces to obtain monthly and interannual temperature and salinity anomalies that exclude changes associated with pure isopycnal heave; spiciness is therefore calculated on an isopycnal surface.
 We calculate the spiciness anomalies as follows. For year y (1945–1993) and tracer X (where X can be either temperature or salinity), the difference between the yearly averaged tracer value Xyσ and the 49-year mean 〈X〉yσ on an isopycnal surface σ is defined as the spiciness anomaly X′yσ, i.e., X′yσ = Xyσ − 〈X〉yσ. Because the depth of an isopycnal surface changes with time, both Xyσ and 〈X〉yσ vary from year to year. A similar procedure is used for calculating monthly anomalies on an isopycnal surface in which Xyσ now represents the monthly value and 〈X〉yσ the 49-year climatology of that month. In addition, all monthly and interannual temperature and salinity anomalies, as obtained above, are then detrended.
where ρ, u, and v are the density and horizontal components of the velocity, σ is an isopycnal, and η is the sea surface level. The Bernoulli function represents geostrophic streamlines that measure geostrophic flow away from the equator.
3. Model Validation
 Since the σt = 25.5 kg/m3 surface lies in the core of the main thermocline, and temperature and salinity anomalies as well as their propagation on this surface are particularly robust in the model, we will focus on this surface and start by comparing the model results with the observations.
 The spatial patterns of the mean depth of the 25.5σt surface are shown in Figure 2a for the model and Figure 2b for the observations, based on the World Ocean Atlas 2001 archive produced at the National Ocean Data Center (hereinafter referred to as the WOA01 data set). In both model and observations, in the Northern Hemisphere this isopycnal outcrops at midlatitudes around 35°N and reaches a maximum depth in the recirculation area in the western North Pacific between 20° and 30°N. In the Southern Hemisphere this isopycnal outcrops in the far eastern South Pacific between approximately 10° to 35°S and everywhere else along 30°S, and reaches a maximum depth in the central and western South Pacific around 20°S. However, the modeled depth of the surface in the eastern equatorial ocean is about a few tens of meters deeper than the observations, which is due to a too diffuse thermocline in the model. Even with this, the simulated thermocline here is sharper than that in most previous studies that use reduced gravity models [e.g., Ginis et al., 1998; Large and Gent, 1999].
 The standard deviation of the depth of the 25.5σt surface (Figure 2c) can be used as a proxy for thermocline variability. The interannual variability of this surface is a maximum in the eastern equatorial ocean as well as in the areas in which this isopycnal surface outcrops. The maximum in the eastern equatorial ocean is associated with the ENSO, whereas the large values of variability in the outcropping areas are likely associated with temperature/salinity anomalies subducting on this isopycnal surface and propagating within the main thermocline. In addition, there are twin zones of high variability along the northern and southern tropics, associated with the large variability of local forcing. These features are basically consistent with the observations [Lysne and Deser, 2002] and other modeling studies [Capotondi and Alexander, 2001; Capotondi et al., 2003].
 The distribution of the mean temperature and salinity on the 25.5σt surface is shown in Figures 3 and 4, respectively. Comparison of the model results with WOA01 shows that the model captures the main features of temperature and salinity on this surface, that their maxima appear to be in the western and central subtropical gyres in both Hemispheres, and their largest gradients are in three areas: the northeastern North Pacific, the southeastern South Pacific and near the equator.
 The interannual variability of temperature (not shown) and salinity (Figure 4c) on this isopycnal surface are dominated by twin lobes of high variability on both sides of the equator that extend from local maxima in the subtropics, where the root mean square (RMS) values of temperature and salinity exceed 0.25°C and 0.08 practical salinity unit (psu), respectively. Since the lobes follow isolines of isopycnal depth toward the equator (Figure 2b), the pattern of variability on the surface is related to the advection of anomalies along mean geostropic streamlines. The westward and equatorward weakening of the RMS temperature and salinity implies an eastward and poleward source of variability with subsequent propagation and dispersion. A similar pattern of salinity variability is described by Yeager and Large , but their model shows larger variability (maximum salinity over 0.14 psu) than does our model.
Figure 5 shows mean temperature, salinity and density along 10°S for the model and the WOA01. Consistent with the observations, modeled SST along 10°S decreases to the east (from 29°C in the west to 25°C in the east), the thermocline shoals from west to east, and the 20°C isotherm depth decreases from about 220 m at the dateline to about 120 m at 110°W. For salinity in both model and observations, the fresh water pool in the western tropics (as bounded by the 35.5 psu isohaline) extends down to about 100 m. The core of the high-salinity water (maximum of about 36.1 psu) is located just on the upper thermocline, and below the thermocline the salinity decreases to less than 35 psu. Another fresh water pool is located in the eastern tropics, and is separated from the western fresh pool by saltier water in the central tropics.
 A direct comparison of spatially averaged SST anomalies in the eastern and central equatorial Pacific (9°N–9°S, 90°W–180°W), where equatorial upwelling is most intense, is shown in Figure 6. As also evident in the observations, the evolution of the model temperature shows a clear shift in the mid-1970s; that is, the temperature increases about 0.6°C from the early to the late 1970s. This increase of SST cannot be explained by an increase in surface heating [McPhaden and Zhang, 2002], and is likely related to variability in the eastern subtropical South Pacific. This will be discussed in sections 4 and 5. The reasonable agreement between the model and observations suggest that the model is reliable for examining subsurface water pathways from the subtropics to the equator.
4. Propagation of Spiciness Anomalies
 From the variability of temperature and salinity on the 25.5σt surface (Figure 4c), as well as this isopycnal's depth contours (Figure 2b), it is suggested that equatorward and westward advection of temperature and salinity variability originates from the eastern subtropics of both hemispheres. In addition, this variability is transmitted to the western boundaries near their bifurcation points (about 12°N and 10°S, respectively) and then to the equator. However, the influence on the equator from the South Pacific appears to be stronger than that from the North Pacific. The intensification of the variability in the central tropics of the South Pacific (around 145°W) could be due to air-sea interaction and/or convergence of anomalies along which the spiciness anomalies are advected. The local maximum of variability on the western equatorial ocean (around 140°E) could result from a local generation process and/or from a convergence of anomalies from both hemispheres, as suggested by Yeager and Large .
 The spatial evolution of spiciness anomalies on the 25.5σt surface during a 9-year period is shown in Figure 7. The period from 1967 through 1975 captures large salinity anomalies on this surface in both hemispheres. In the North Pacific, significant positive anomalies can be observed around (135°W, 28°N) in 1967 propagating southwestward toward the western subtropics and tropics with some weakening. About 6 years later in 1973, the anomalies arrive at the western boundary at ∼12°N. Via western boundary pathways, the anomalies can reach the western equatorial ocean with further weakening but there is no evidence that this signal continues along the equator. The subduction of these spiciness anomalies from the northeastern subtropics to the western boundary is basically in agreement with Zhang et al. , who demonstrate that there is a major decadal (warm and salty) subduction event in the early 1970s in the North Pacific, and the time needed to propagate from the subtropical outcrop sites into the western boundary is about 5 years.
 In the South Pacific, preexisting positive anomalies are intensified in the eastern tropics/subtropics during 1968 and 1969. Then the anomalies move northwestward and are intensified in the tropics around 1971. After reaching the equator through the western boundary and interior pathways in 1974, the signal propagates eastward along the equator in the Equatorial Undercurrent. This decadal propagation event in the South Pacific is also shown by Yeager and Large . Since the model reveals a strong connection between the South Pacific and the equator, we will focus on the South Pacific.
 The anomaly structure in a depth-longitude plane further demonstrates the decadal evolution. Zonal sections of salinity anomalies along 10°S are shown in Figure 8 for the same period. In 1967–1968, positive anomalies appear at 120 m around 100°W, and then move deeper and westward with some strengthening. By 1971–1972 its center is located at 200 m around 160°W, finally reaching the western boundary in 1974–1975.
 The model captures the decadal spiciness signal from the eastern tropical/subtropical South Pacific to the equator not only for the period from 1967 through 1975 but also for the entire integration. Combining Figures 7 and 8 shows that a negative anomaly existed since the early 1970s. A time-longitude map of temperature anomaly is shown in Figure 9 to illustrate the propagation of the anomalies. (In the eastern equatorial ocean, the spiciness becomes obscured due partly to high-frequency motions and strong mixing processes there; the salinity signal appears to be weak while the temperature signal remains strong.) Figure 9 (left) shows temperature anomalies on the 25.5σt surface from 120°W to 155°E along 10°S. (Note that the x axis in Figure 9 (left) is reversed so that east is on the left-hand side.) Figure 9 (middle) shows temperature anomalies on the 25.5σt surface along the equator from 140°E to 120°W. Figure 9 (right) shows SST along the equator from 130°W to 80°W. The evolution of SST (in Figure 9 (right)) shows decadal variations: a warm period from the late 1970s to the middle 1980s, a cold period until the early 1990s, and then another warm period thereafter. These surface temperature changes are preceded by temperature anomalies in the subsurface South Pacific. The temperature anomalies on the 25.5σt surface first appear in the tropical eastern and central South Pacific and propagate to the west, and the anomalies then reach the equator through both western boundary and interior pathways. The anomalies then propagate relatively quickly across the ocean with the Equatorial Undercurrent, reaching the surface near 130°W to affect the SST in the eastern equatorial ocean.
 We therefore suggest that, through western and interior pathways in the thermocline, the subsurface spiciness anomalies in the eastern subtropical South Pacific are first transferred westward and then northward, eventually appearing along the equator. The anomalies then propagate eastward along the equator in the Equatorial Undercurrent, and eventually upwell to the surface in the eastern equatorial ocean.
 To further demonstrate the mean advection pathways connecting the South Pacific with the equator, we below present the particle trajectories and the Bernoulli function. Figure 10a shows a kinematic picture of water particle behavior on the 25.5σt surface by calculating flow trajectories from the simulated mean velocity field. After 2 years, water particles have made long journeys that allow us to trace their pathways. The arrows in Figure 10a indicate the terminus of the trajectories. The trajectories are dominated by geostrophic flow and the horizontal structure of the STC is clearly depicted, as are the tropical and subtropical recirculation gyres. In particular, eastern subtropical water moves northward and westward to feed the Equatorial Undercurrent through the western boundary and interior pathways, and much of the water reaches the equator through interior pathways, helped by the absence of a potential vorticity island (i.e., the absence of an Intertropical Convergence Zone) in the South Pacific. On the other hand, flow patterns in Figure 10b can be separated into regions with blocked (outcropping or land boundaries) and closed Bernoulli function contours (recirculation). Recirculating flows occur in the southwestern part of the subtropical gyre and eastern tropical Pacific centered at 5°S. In contrast, subtropical/tropical exchange flows originate from the eastern ocean around 15°–30°S feeding the Equatorial Undercurrent.
5. Origins and Generation of Spiciness Anomalies
 The existence of a linkage between the tropical/subtropical South Pacific and the equator on decadal timescales has been shown here as well as in other observational and modeling studies. Some suggest that the temperature anomalies in the tropical ocean are due to displacements of the thermocline as a whole, associated with purely mechanical forcing by winds [Chang et al., 2001; Luo and Yamagata, 2001; Luo et al., 2003a; Capotondi et al., 2003]. In the scenario proposed by Luo and Yamagata , when a decadal positive temperature anomaly occurs in the eastern equatorial ocean, the atmospheric response excites a negative wind stress curl in the western tropical South Pacific, which causes the thermocline to shoal and induce a negative temperature anomaly there. Capotondi et al.  suggest that the tropical centers of thermocline variability in both hemispheres are associated with first-mode baroclinic Rossby waves forced by anomalous Ekman pumping. However, our analysis performed on time varying isopycnals excludes the possibility that the temperature and salinity anomalies considered here are primarily due to thermocline displacements. In fact, our study indicates that there is a mismatch between changes in isopycnal temperature/salinity and local wind stress curl/Ekman pumping in both time and space (not shown).
 Through a global ocean GCM driven with 40-year realistic surface forcing, Yeager and Large  find largest spiciness anomalies in the South Pacific near the 25.5σt surface originating from the eastern subtropics. They propose a subsurface injection process through which positive anomalies are generated in late winters by diapycnal mixing across subducted isopycnal surfaces. In addition, it is suggested that significant interannual spiciness variability is expected wherever large unstable salinity gradients are present in conjunction with weak stratification and strong mixing in winters.
 Similar to Yeager and Large , temperature and salinity anomalies in the South Pacific on the 25.5σt surface in our model are particularly robust, and these anomalies appear to originate from the eastern subtropical South Pacific (Figure 4c), suggesting that the subsurface injection process might be at work over the region in our model. We will examine this below.
 In the eastern South Pacific Ocean, there exists a salinity minimum at a depth of about 200 m due to air-sea exchange and advection of water from south of the subtropical gyre [Karstensen, 2004]. Figure 11 shows the salinity difference between the sea surface and 200 m (S0m − S200m) from the model and observations, respectively. The model reproduces a destabilizing mean upper ocean salinity gradient between 10° and 30° latitudes in both hemispheres with a larger vertical salinity gradient in the eastern tropics/subtropics of the South Pacific, where low mean upper ocean density stratification is found in both model and observations (Figure 12); that is, the model shows that the eastern subtropical South Pacific is characterized by large vertical salinity gradients with weak stratification in winter. Therefore, as suggested by Yeager and Large , the vertical temperature gradient there can strongly stabilize the water column only in summer. Winter cooling erodes the thermal stratification, allowing the salinity gradient to enhance the wind mixing and to produce a deep layer of near uniform density with a wide range of temperature and salinity combinations. We will give an example of how the subsurface injection process works for the region.
 The large spiciness anomalies on the 25.5σt surface originate from the eastern subtropics of the South Pacific in three regions (Figure 4c): in the west centered at (105°W, 23°S), in the east centered at (88°W, 18°S), and in the south centered at (82°W, 36°S). In the following discussions, the three central points with largest spiciness anomalies will represent the three regions for convenience. The large salinity anomalies on this surface in the western and eastern regions are also shown by Yeager and Large  but their model does not show a southern region with high variability as does our model.
 The evolution of upper ocean temperature, salinity and density during the large positive anomaly event in 1967 is shown in Figure 13 for the western point. From March, the gradually deepening mixed layer increases the temperature/salinity at depth while concentrating the upper ocean temperature/salinity gradients into a sharp, unstable thermocline/halocline at the base of the mixed layer. By August, the stable pycnocline erodes and the surface density is very close to 25.5σt, and the temperature/salinity anomaly crosses the 25.5σt surface to generate significant property anomalies on this isopycnal by October. The anomalies reach about 0.6°C for temperature and 0.2 psu for the salinity, respectively, in this injection event. Therefore, for the western region, our model results support the study by Yeager and Large , in which the subsurface injection process is at work in late winter to generate positive temperature and salinity anomalies.
 However, unlike the western point where the 25.5σt surface never outcrops, the eastern and southern points are both located in winter outcrop sites. An examination from the model results reveals that in 18 of 49 years at the eastern point, and in all 49 years at the southern point, the 25.5σt surface outcrops during austral winters. Hence the subduction process may play an important role in forming the temperature and salinity anomalies for the eastern point, and the dominant role in forming anomalies in the southern point.
 For example, during the winter of 1967 for the eastern point, our analysis shows that the 25.5σt surface outcrops and the SST anomalies reaches up to 0.8°C, suggesting that the positive anomalies on this isopycnal can be produced by the subduction of the surface anomalies. For the event during the late 1960s shown in Figures 7 and 8, therefore we suggest that both subsurface injection and subduction processes play important roles in the generation of spiciness on the 25.5σt surface in the eastern subtropical South Pacific. Note that the subsurface injection process may be still at work in years when the 25.5σt surface does not outcrop at the eastern point. Since the 25.5σt surface outcrops every single winter at the southern point, the large variability of temperature and salinity anomalies there can be merely due to the subduction process.
 Since the subsurface injection process is responsible for positive anomalies only, negative anomalies, according to Yeager and Large , are attributed to anomaly accumulation via isopycnal advection when stratification between an isopycnal and the sea surface remains relatively large during austral winters and subsurface injection events are absent. However, our results suggest the subduction of negative SST/SSS anomalies can be another process for the negative anomalies on an isopycnal over the region. Therefore we conclude that the large spiciness anomalies on the 25.5σt surface originate from the eastern subtropical South Pacific in three regions where both the injection and subduction processes are at work to generate the subsurface spiciness. The positive anomalies can be attributed to both processes, while negative anomalies can be generated by the subduction of surface negative anomalies and accumulation via isopycnal advection.
 Finally, we design an active tracer experiment to examine whether a synthetic SST anomaly in the eastern subtropical South Pacific can subduct and then reach the western equatorial ocean through mean advection pathways. With the climatological wind stress but SST and SSS restored to the climatological values with a relaxation time of 70 days, a control run is started from the confirming experiment and run forward 20 years to reach a new steady state. An SST anomaly run is constructed; a synthetic temperature perturbation ΔT (1.2°C maximum anomaly) is applied with a restoring SST anomaly over a circular area of 20° diameter with the form
where x represents longitudes from 70° to 90°W and y latitudes from 15° to 35°S. The SST anomaly run is further integrated for 10 years starting from the year 20 fields of the control run.
 Results indicate temperature and salinity compensating anomalies are transferred from the subtropics toward the equator in the thermocline; temperature anomalies on the 25.5σt surface for year 10 are shown in Figure 14. In agreement with the mean circulation (Figure 10), the anomalies subducted in the eastern subtropical ocean propagates northwestward as high vertical mode Rossby waves [Nonaka and Xie, 2000], however, the anomalies traced to the edge of the Equatorial Undercurrent are only ∼0.1°C, much less than the large anomalies (1.2°C) placed in the eastern subtropics. This propagation feature is basically consistent with the findings of Stephens et al. , who show that the amplitude of subducted temperature anomalies in the subtropical North Pacific decays faster than a passive tracer and the temperature anomalies appear to propagate slower than a passer tracer. Our active tracer experiment suggests that the subduction process alone seems not enough to explain a significant part of the decadal variations at the equator.
 Few studies have been directed toward understanding decadal variability of the South Pacific Ocean, partly because of the sparse observations there. In this paper we have focused on the South Pacific and explored the potential dynamical linkages between the subtropics and the equator on decadal timescales using a well-tested numerical model.
 Simulations from a 49-year, realistically forced experiment show a direct linkage between the equator and the South Pacific. Through western boundary and interior pathways in the thermocline, subsurface spiciness anomalies in the eastern subtropical South Pacific are first transferred westward and then northward, eventually appearing along the equator. The anomalies then propagate eastward along the equator in the Equatorial Undercurrent, and eventually upwell to the surface in the eastern equatorial ocean. The time for the anomalies to propagate from the eastern subtropical South Pacific into the western equatorial region is about 6–12 years, from which it is estimated that the anomalies propagate at a speed of approximately 4–8 cm/s, slower than the phase speed of the first mode baroclinic Rossby waves at this latitude. This estimation is consistent with Giese et al. .
 The large spiciness anomalies in the eastern subtropical South Pacific are formed where large unstable salinity gradients are present in conjunction with weak stratification and strong mixing in winters, and the subsurface injection process is at work in late winter to generate positive temperature and salinity anomalies, in agreement with Yeager and Large . In addition, the subduction process contributes to the generation of the anomalies because part of the large anomaly area is in winter outcrop sites of the model, although an active tracer experiment reveals that subduction of an SST anomaly in the eastern subtropics cannot work alone to explain the decadal signal observed from the 49-year realistic simulations between the eastern subtropics and the equator. Furthermore, different from the subsurface injection process which is responsible for the positive anomalies only, the subduction process can bring both positive and negative anomalies from the surface into the subsurface. Therefore we suggest that both the injection and subduction processes contribute to form the subsurface positive anomalies in the eastern subtropical South Pacific, while negative anomalies can be produced by subduction of negative SST/SSS anomalies and accumulation via isopycnal advection.
 Although the evolution of decadal spiciness anomalies from the South Pacific to the equator is quite similar to that of Yeager and Large , the anomalies appear to be weak partially due to the restoring boundary condition for SSS in our simulations. In addition, no Indonesian Throughflow in the model is allowed and this might underestimate the impact of water from the South Pacific on the equatorial ocean. The Indonesian Throughflow allows an outflow of upper ocean water estimated to be 5–15 Sv [Fine, 1985; Godfrey, 1989; Hirst and Godfrey, 1994], which primarily originates from the North Pacific [Gordon, 1986]. Instead of exiting through the Indonesian Throughflow, in the model these waters could possibly appear in the equatorial Pacific and thus impact the circulations connecting the South Pacific and the equator.
 We are grateful for the constructive criticism and comments from the two anonymous reviewers, which greatly improved the paper. This research is supported by the NASA grant NAG5-12283.