Observational, Theoretical and Computational Research on the Climate System
|© Springer-Verlag 2007|
Source of low frequency modulation of ENSO amplitude in a CGCM
|(1)||Division of Science Education/Institute of Science Education, Chonbuk National University, Jeonju, Korea|
|(2)||Korea Ocean Research and Development Institute, Ansan, Korea|
|(3)||Laboratoire d’Etude en Géophysique et Océanographie Spatiale, 14 av. Edouard Belin, 31400 Toulouse, France|
|(4)||School of Earth and Environmental Sciences, Seoul National University, Seoul, Korea|
|(5)||Climate Environment System Research Center (CES), Seoul National University, Seoul, Korea|
Received: 18 January 2006 Accepted: 15 December 2006 Published online: 26 January 2007
We study the relationship between changes in equatorial stratification and low frequency El Niño/Southern Oscillation (ENSO) amplitude modulation in a coupled general circulation model (CGCM) that uses an anomaly coupling strategy to prevent climate drifts in the mean state. The stratification is intensified at upper levels in the western and central equatorial Pacific during periods of high ENSO amplitude. Furthermore, changes in equatorial stratification are connected with subsurface temperature anomalies originating from the central south tropical Pacific. The correlation analysis of ocean temperature anomalies against an index for the ENSO modulation supports the hypothesis of the existence of an oceanic “tunnel” that connects the south tropical Pacific to the equatorial wave guide. Further analysis of the wind stress projection coefficient onto the oceanic baroclinic modes suggests that the low frequency modulation of ENSO amplitude is associated with a significant contribution of higher-order modes in the western and central equatorial Pacific. In the light of these results, we suggest that, in the CGCM, change in the baroclinic mode energy distribution associated with low frequency ENSO amplitude modulation have its source in the central south tropical Pacific.
Decadal climate variability in the atmosphere and ocean over the tropical Pacific Ocean has attracted considerable scientific interest and debate in the past decade. It is not yet clear whether the midlatitudes independently undergo decadal changes which may affect tropical Pacific decadal variability (hereafter, TPDV) through atmospheric teleconnections (Pierce et al. 2000) or oceanic teleconnections (Gu and Philander 1997; Zhang et al. 1998; Luo and Yamagata 2001; McPhaden and Zhang 2002) or whether the TPDV is intrinsic to the tropics (Jin 2001). Regarding the tropical–extratropical oceanic interactions there are number of competing hypotheses. It is still open question whether propagation of sub-surface thermal anomalies from the extratropics are effective (Hasegawa and Hanawa 2003) or not (Schneider et al. 1999) in modifying the tropical Pacific mean state. The issue of whether the origin of decadal signals comes from the North Pacific (Zhang et al. 1998) or the South Pacific (Luo and Yamagata 2001) has also been explored. From a purely oceanic perspective, there is now an overall consensus that the South Pacific may be a major contributor to the TPDV (Chang et al. 2001; Luo and Yamagata 2001; Giese et al. 2002) due in part of the absence of an Intertropical Convergence Zone (ITCZ) and to the relatively large temperature and salinity variability in the South Eastern Pacific subduction region (Yeager and Large 2004) which is favorable to the entrance of spiciness anomalies in the tropics.
Despite the wealth of potential sources of TPDV and El Niño/Southern Oscillation (ENSO) decadal variability, it is unclear how TPDV is associated with the low frequency modulation of ENSO amplitude. Flügel and Chang (1999), Yeh et al. (2004), Flügel et al. (2004), Kleeman et al. (2003) all argue that the low frequency modulation of ENSO amplitude is entirely driven by atmospheric noise and unrelated to TPDV based on the analysis of coupled models. On the other hand, Fedorov and Philander (2000) suggested that TPDV forces changes in the ENSO statistics (see also Kirtman and Schopf 1998). In addition, there are some evidences that TPDV is inherently related to ENSO through non-linearities (Jin et al. 2003; An 2004; Rodgers et al. 2004; Cibot et al. 2005).
It is known that concurrent changes in the amplitude and frequency of ENSO from before the mid-1970s to after the late-1970s were linked to TPDV (Trenberth and Hurrell 1994; Wang 1995). Recently, Moon et al. (2004, hereafter, M04) suggested that the climate shift of the 1970s is associated with an increased contribution of the high-order baroclinic modes in the tropical Pacific, which has the potential to impact the equatorial wave dynamics towards larger amplitude and longer period ENSO events as observed during the post-1976 period. The underlying idea behind this hypothesis is that changes in equatorial stratification as measured by changes in the baroclinic mode energy distribution could explain some aspect of the ENSO amplitude modulation. Recent results (i.e., Dewitte et al. 2006) based on the analysis of a coupled general circulation model (CGCM) simulation support this hypothesis. The source of the change in baroclinic energy distribution remains, however, unclear due in part to the biases in simulated mean state as observed in many CGCMs. Because of these biases the simulated ENSO mode in CGCMs usually tends to resemble the theoretical ‘ocean basin mode’ (Neelin and Jin 1993; An and Jin 2000) due to weak upwelling feedback loop in the eastern tropical Pacific (van Oldenborg et al. 2005). This in particular leads to over energetic ENSO variability in the quasi-biennial band (AchutaRao et al. 2004; Guilyardi et al. 2004; Rodgers et al. 2004; Dewitte et al. 2006). Because the processes leading to ENSO modulation may involve the interaction with the mean state, it is important to rely on CGCMs that simulate it with some degree of realism. While efforts are undertaken to improve the simulated mean state in CGCMs, another strategy consists in designing anomaly coupling strategy that avoids such difficulty. Although limiting the interaction between tropical Pacific mean state and ENSO, such coupling strategy applied to CGCM may be useful to investigate the processes associated with the ENSO modulation, and those involving equatorial wave dynamics in particular. The intent of this paper is therefore to reexamine connections between TPDV and the low frequency modulation of ENSO amplitude with a focus on changes in the equatorial stratification using a CGCM with an anomaly coupling strategy to prevent or reduce climate drift.
Recent model studies have suggested that the ENSO modulation is simultaneous with TPDV, meaning that TPDV forces the ENSO modulation and vice versa (Rodgers et al. 2004; Cibot et al. 2005). For instance in the CGCM of Cibot et al. (2005), the thermocline decadal mode associated with the ENSO modulation has a center of action both in the South Central Pacific and along the equator, and is forced through Ekman pumping associated with the ENSO modulation. The equatorial component of this mode is able to produce changes in ENSO on low frequencies through the non-linear equatorial dynamics (Dewitte et al. 2006). No equatorward propagation of subsurface temperature anomalies from the South-Western Pacific could be observed. In the model of Rodgers et al. (2004), the ENSO modulation appears also to be related to the nonlinearity/asymmetry of the equatorial system, with a thermocline decadal mode comparable to the one of Cibot et al. (2005). Note that both models have a tendency simulate a biennial ENSO and have comparable biases in the mean states, which includes too diffuse thermocline. Taking advantage of the anomaly coupling procedure we revisit the issue of the source of the ENSO modulation in the tropics. Our assumption is that the changes in mean equatorial stratification can produce ENSO modulation through equatorial nonlinear dynamics (Dewitte et al. 2006). Therefore we aim here at identifying the source of the low frequency changes in equatorial stratification and how they are connected to the tropical-extratropical oceanic teleconnections.
As a synthetic measure of the low frequency change of the equatorial stratification, we compute the wind projection coefficients from the vertical mode structures as derived from the model density structure. This assumes that most changes of the ENSO characteristics are due to changes in the equatorial wave dynamics and associated changes of the energy distribution of the baroclinic mode. Dewitte (2000) and Yeh et al. (2001) have shown that the energy distribution on the baroclinic mode controls to a large extend the characteristic of ENSO variability in intermediate coupled models of the tropical Pacific. In the light of their theoretical result and following M04 hypothesis, we investigate whether the change in the baroclinic mode energy distribution along the equator is related to the ENSO modulation and, if it is, the source of such change from a full-physics coupled model.
The paper is organized as follows. Section 2 is devoted to the coupled model description. Section 3 describes the variability of the first three baroclinic modes associated with the low frequency modulation of ENSO amplitude. Section 4 seeks for the origin of the changes in baroclinic mode energy distribution. Section 5 provides concluding remarks.
The atmospheric component of coupled model is the Center for Ocean-Land-Atmosphere studies (COLA) atmosphere GCM with T42 and 18 vertical levels. The ocean model is adapted from the Geophysical Fluid Dynamic Laboratory (GFDL) modular ocean model version 3 (MOM3). The ocean model’s spatial domain is global in longitude and extends from 75°S to 65°N in latitude. The grid has 240 uniformly spaced points in longitude. There are 134 points in latitude with higher resolution near the equator expanding to a maximum of 0.5 resolution at 11°N and 11°S.
The component models are anomaly coupled ones in terms of heat, momentum and fresh water (Kirtman et al. 2002). With this coupling strategy, the component model monthly climatologies must be specified at the sea surface. In the case of the ocean, the model climatology is determined from an uncoupled extended integration forced by observed momentum flux, with surface relaxation of temperature and salinity to their observed values. In the case of atmosphere, the model climatology is determined from a multi-decadal simulation with specified observed SST. The coupling frequency of the model is once a day with daily mean values being exchanged between the ocean and the atmosphere. The purpose of the anomaly coupling strategy is to prevent or significantly reduce climate drift. In this sense, it can be interpreted as being analogous to flux corrections. The procedure, however, does not prevent the model from developing anomalies on decadal or longer time scales.
We then integrate the coupled model for 330 years for the subsequent analyses. The variability and climatology simulated by the model have been described in Kirtman et al. (2002). Overall, they are rather realistic although the mean equatorial Pacific SST is lower than observations by 0.5–1.0°C and the cold tongue is too strong and extends too far to the west. Like many other CGCMs (AchutaRao et al. 2004), the model ENSO is biased to the biennial time scale with SSTAs somewhat weaker and extending too far to the west compared to observations. However, the ENSO modulation is realistic (see below), which also motivates this study. The simulated equatorial stratification is stronger in comparison with observations. The mean thermocline simulated by the model is deeper (shallow) in the west (east) (Kirtman et al. 2002). The reader is invited to refer to Kirtman et al. (2002) for more details about the model characteristics and skill in simulating the tropical Pacific mean state and variability.
3 ENSO modulation and equatorial vertical stratification changes
We first show the low frequency modulation of ENSO amplitude and frequency simulated in a CGCM. Figure 1a shows the time series of the NINO3.4 SST index for the entire simulation period with its local (Fig. 1b) and corresponding global (Fig. 1c) wavelet spectrum. The Nino3.4 SST index is defined as the time series of sea surface temperature anomalies (SSTAs) averaged over (170°E–240°E; 5°N–5°S). The ENSO events in the model simulation are irregular and the standard deviation of the NINO3.4 SST index is 0.6°C. As mentioned above, the ENSO variability simulated in the CGCM is dominated by timescales between 2 and 4 years (Fig. 1b, c), which is shorter than in observations (2–7 years). Despite this over energetic peak in the quasi-biennial timescale, the wavelet analysis of the NINO3.4 SST index exhibits low frequency modulation on decadal timescales (Fig. 1b).
Fig. 1 a The Nino3.4 SST index for the entire simulation period. b The local wavelet power spectrum of (a) using the Morlet wavelet, normalized by variance (0.60°C2). The thick contour line delineates the regions of greater than 95% confidence from a red-noise process. The dashed line regions indicate the boundary of the cone of influence where edge effects become important. c The global wavelet power spectrum. The dashed line is the 95% confidence spectrum. d Scale-averaged wavelet power over the 2–7 year band
As a measure of the low frequency modulation of the ENSO amplitude, we use an index computed from the interannual (2–7 years) wavelet variance of the NINO3.4 SST index (hereafter, N34Var index), following the method described in Torrence and Webster (1999). This method provides a robust representation of the ENSO variance by selecting a preferred timescale (i.e., 2–7 years for ENSO; Cibot et al. 2005). The temporal evolution of the N34var index is characterized by decadal to multi-decadal fluctuations with periods of strong and weak ENSO amplitude (Fig. 1d). In addition, when the ENSO amplitude is strong, the dominant period of ENSO tends to increase as well. For instance, this can be seen during model year 2550–2580, 2620–2640 and 2720–2740. The ENSO modulation simulated in the CGCM has comparable amplitude with the observations. We computed the N34Var index based on the monthly mean SST from the National Climatic Data Center, the so-called ‘Extended Reconstructed SST version 2’ (ERSST.v2, Smith and Reynolds 2004) for the period of 1880–2004. The root mean square of N34Var is 0.16 and 0.24°C2 in the CGCM and the observation, respectively.
As mentioned in the introduction, M04 argued that an increase in ENSO amplitude and period after the late 1970s was associated with an increase of the vertical stratification at upper levels, which is associated with an increase in the contribution of higher-order baroclinic modes. Motivated by this result, we investigate the relationship between oceanic vertical stratification and changes in ENSO amplitude in the CGCM simulation. Using the ocean temperature and salinity, we calculated the vertical profile of buoyancy frequency N 2(z) along the equator from surface to 350 m for the entire simulation period. Here, is the Brunt-Väisälä frequency.
To isolate the structure of the oceanic stratification associated with ENSO amplitude modulation, we apply a linear correlation analysis to the vertical profile of N 2(z) along the equator against the N34Var with a significance test based on “random-phase” test by Ebisuzaki (1997) (Fig. 2). Note that this non-parametric test has been used in this study since our data might be highly serially correlated. Significant changes in vertical stratification, which are associated with changes in ENSO amplitude, occur in the western and central equatorial Pacific. The correlated buoyancy frequency N 2(z) against the N34Var corresponds to the structure of vertical stratification during periods of high ENSO amplitude. The stratification is intensified in the western and central equatorial ocean from surface to 100 m and is weakened from 100 to 200 m during periods of high ENSO amplitude. Although not significant, this tendency of oceanic stratification is reversed in the eastern equatorial Pacific, i.e., the vertical stratification is weakened within shallow levels (~60 m). Note that these changes in stratification in the western and central equatorial Pacific correspond to temperature changes around 0.5°C on average (not shown). In spite of significant correlation coefficients as in Fig. 2, however, we do not exclude that the vertical stratification modulation could be one of the mechanism contributing to the decadal variability of ENSO amplitude.
In order to estimate the contribution of vertical structural changes in the temperature and salinity responsible for the vertical stratification shown in Fig. 2, we calculate N 2(z) using a climatological temperature and salinity for the entire simulation period, and then apply the same correlation analysis with the N34Var (Fig. 3). Note that the mean salinity simulated in the model exhibits some biases which consist of a saltier warm pool associated with the reduced precipitation due to cold SST biases in the western tropical Pacific (not shown).
Figure 3a, b is the same as Fig. 2 except for the climatology salinity (temperature). The change of vertical salinity structure does not make a difference in the profiles of N 2(z) associated with the N34Var (Fig. 3b). In spite of the discrepancy of simulated vertical salinity structure in the model the calculation indicates that changes in the vertical stratification associated with ENSO amplitude modulation are preferentially due to the change of vertical temperature structure, which is consistent with the results using ocean assimilation data (Fig. 1 of M04).
Here, we show that, in the CGCM, an intensification of the stratification is associated with an increase in the contribution of the higher baroclinic modes in a straightforward manner. We analyze the projection coefficient (P n ) of the wind stress onto the baroclinic modes. The projection coefficient,
as defined as in Lighthill (1969), is indicative of surface momentum flux projecting onto the baroclinic modes. Here, A n (z) is the vertical structure function which is derived from the vertical mode decomposition of temperature and salinity profiles from the CGCM. The vertical mode decomposition was calculated similar to Dewitte et al. (1999).
Figure 4a shows the amplitude (root mean square) of the low-pass (>7 years) filtered projection coefficients corresponding to the first three baroclinic modes along the equator. Remarkably, Fig. 4a indicates that on decadal time scales the atmospheric forcing preferentially projects onto the higher baroclinic modes compared to the gravest baroclinic mode. Simply put, this result suggests that the decadal variability in the tropical Pacific may be associated with higher modes of baroclinic waves, which are associated with the long ‘memory’ of the equatorial Pacific system. The projection coefficients onto the higher baroclinic modes, i.e., the second and third mode, have large amplitude from the western equatorial Pacific to the date line. In the far eastern tropical Pacific, the amplitude of the third baroclinic mode is strong, which, in part, reflects modal dispersion processes due to the rising thermocline (cf. Dewitte et al. 1999). It is noteworthy that similar characteristics of baroclinic mode on decadal timescales are found in another CGCM simulations (Dewitte et al. 2006), indicating that the above results are not limited in the CGCM used in this study. In particular, Dewitte et al. (2006) showed that the contribution of the second and third modes on the decadal variability of surface currents peaks up in the western equatorial Pacific.
Fig. 4 a The amplitude for decadal (>7-year) P n (n = 1,2,3) along the equator of the first mode (solid), the second mode (dashed), and the third mode (dot dot dashed) for the entire simulation period. Units are nondimensional. b Simultaneous linear correlation coefficients of P n (n = 1,2,3) against the N34Var along the equator. Thin lines, styles are the same as in (a), indicate the p-values at the 0.05% significant level based on random-phase test
In order to show the contribution of the baroclinic modes during periods of high and low ENSO amplitude, the simultaneous correlation coefficients between the N34Var and P n (n = 1,2,3) along the equator are shown in Fig. 4b. Consistent with results of M04, the projection coefficients for the second and third baroclinic mode significantly increases in the western and central equatorial Pacific during periods of high ENSO amplitude, however, that of the first baroclinic mode decreases. This suggests that the atmospheric forcing projects onto the higher baroclinic modes in the western and central equatorial Pacific during periods of high ENSO amplitude. In other words, the higher-order baroclinic modes are more favored during periods of high ENSO amplitude than during periods of low ENSO amplitude. On the other hand, the gravest baroclinic mode is more favored during periods of low ENSO amplitude or stable ENSO activity than during periods of high ENSO amplitude or unstable ENSO activity.
Because the vertical stratification is intensified in the western and central equatorial ocean during periods of high ENSO amplitude (Fig. 2), an intensification of the stratification is also associated with an increase in the contribution of the higher baroclinic modes. We are arguing that the intensification of the stratification in the western and central equatorial Pacific is associated with a redistribution of surface energy with an increase in the contribution from the higher baroclinic modes (Fig. 4b), resulting in an increase in ENSO amplitude. M04 discussed the impact of changes in the baroclinic mode energy distribution on the ENSO amplitude and showed that the relative contribution of the baroclinic modes is associated with an increase of ENSO amplitude and period using simple coupled model experiments. However, it remains an unsettled question whether changes in equatorial stratification which leads to an increase of contribution of the higher order baroclinic modes and ultimately ENSO amplitude modulation have their source in the off equatorial regions. In the next section we examine this issue by diagnosing the source of changes in equatorial stratification.
4 Oceanic teleconnections
In this section, we relate the changes in equatorial stratification diagnosed above with some aspects of the large scale tropical oceanic variability in order to identify the sources of the ENSO modulation in the CGCM. We first analyze the simulated thermocline variability associated with the low frequency of ENSO modulation. Figure 5 shows the standard deviation of the 20°C isotherm depth anomalies (hereafter, z20 anomalies) on decadal timescale (>7 years), which represent the decadal variability in thermocline depth simulated in the CGCM. We used a 7-year running mean to obtain z20 anomalies on decadal timescale. Note that modifying the averaging period has little qualitative impact on the results. There are two centers of large thermocline variability on decadal timescales both in the western North and South off equatorial Pacific. Decadal thermocline variability is also trapped in the eastern equatorial Pacific, however, its magnitude is relatively weaker than that centered off equatorial in both Hemispheres. The thermocline variability located in the South western Pacific shows more confined spatial structure from the west to the date line, on the other hand, that of the western North Pacific are more elongated from the eastern to the western off equatorial Pacific. This spatial structure of decadal thermocline variability at both the western North and South Pacific is consistent with that of the observed Pacific decadal-interdecadal variability (Zhang et al. a class="anchor-link" href=#CR43">1997; Tourre et al. 2001) and, as so, may be part of the mechanism that produces TPDV as suggested by Yu and Boer (2004). In the following, rather than analyzing the TPDV of the model and the mechanism that produces it, we concentrate on the decadal thermocline variability associated with low frequency ENSO amplitude modulation, which is considered here as a component of the whole TPDV. Our analysis indicates that in the CGCM decadal ENSO modulation has its source in the South Pacific, as shown below.
The temporal-spatial evolution of z20 anomalies is obtained by time-lagged correlation with the N34Var. Based on this, the evolution of the z20 anomalies is displayed at 1.5 year intervals in Fig. 6a–h. At a lag of 6 years (Fig. 6a), there are weak negative z20 anomalies in the central-south tropical Pacific (180°E–150°W). As time progresses, the negative z20 anomalies significantly amplify and develop from the central-south tropical Pacific to the western tropical Pacific (140°E–170°E) between a lag of 6 years and a lag-zero (Fig. 6a–e). After lag-zero, negative z20 anomalies seem to progress eastward to the central and eastern equatorial Pacific, ultimately, changing sign (Fig. 6e–h). In the Northern Hemisphere the positive z20 anomaly appears in the central-north Pacific at a lag of 4.5 years, however, these z20 anomalies are relatively stationary and also do not spread to the equatorial western Pacific. After lag-zero, the amplitude of positive z20 anomalies in the central-north tropical Pacific gradually decay without any particular indication of propagation or migration. It is apparent that the z20 anomalies from the North Pacific do not play a dominant role in the phase reversal of the equatorial thermocline anomalies. This might be due to the potential vorticity barrier related to the ITCZ acting to prevent z20 anomalies from reaching the equatorial region on decadal timescale (McCreary and Lu 1994; Luo and Yamagata 2001). This result suggests that subsurface temperature anomalies originating from the South Pacific reach the equatorial wave guide on decadal timescales, consecutively impacting the low frequency modulation of ENSO amplitude. To further investigate the ‘propagation’ of temperature anomalies associated to the ENSO modulation, the cross sections of the upper ocean temperature anomalies associated with ENSO amplitude modulation along the line (180°, 20°S–140°E, 0°) and the equator are shown in Fig. 7. Figure 7 is the same as in Fig. 6 except for vertical sections of upper ocean temperature anomalies.
Fig. 6 Lead-lag correlation maps between the 20°C isotherm depth anomalies over the tropical Pacific and N34Var from the -6.0 year lead to +4.5 year lag at 1.5 year intervals. Contour interval is 0.05 and shading indicates 95% confidence level by random-phase test
Fig. 7 The same as in Fig. a class="anchor-link" href=#Fig6">6 except for the upper ocean temperature anomalies both along the transect (180°, 20°S–140°E, 0°) and the equator. Contour interval is 0.05 and shading indicates 95% confidence level based on random-phase test. Positive lag means that N34Var leads upper ocean temperature
Figure 7 shows evidence that changes in vertical temperature structure of the equatorial Pacific are associated with oceanic teleconnections from the South Pacific. In other words, changes in the stratification of the equatorial western Pacific are more likely due to subsurface thermal anomalies originated from the South Pacific. Prior to the strongest ENSO amplitude (from lag -4.5 year to lag-zero), cold temperature anomalies amplify at upper levels in the western equatorial Pacific which is associated with the spreading of temperature anomalies from the South Pacific equatorward and westward (Fig. 7a–d). The transactional analysis shows that cold temperature anomalies amplify at upper levels around 10°S–15°S from lag -4.5 year to -3 year between 100 and 200 m depth, and then these anomalies become stronger, spreading toward the equator from lag -1.5 year to lag-zero. In the meantime, the negative z20 anomalies amplify and develop from the central-south tropical Pacific to the western equatorial Pacific with an east/west contrast in the equatorial Pacific as shown in Fig. 6b–e. Here we are arguing that changes in the vertical temperature structure of the western and central equatorial Pacific are due to the spreading of subsurface temperature anomalies from the South Pacific. For the same period from lag -4.5 year to lag-zero (Fig. 7a–e), the vertical stratification represented by dT/dz intensifies in the western and central equatorial ocean from the surface to 100 m and weakens from 100 to 200 m (not shown). On the other hand, the vertical stratification is weakened within very shallow levels and is intensified from 50 to 200 m in the eastern equatorial Pacific.
The decadal variability due to spiciness is weak in the western and central South Pacific except in the limited far eastern South Pacific. Since the spiciness anomalies mean the temperature advection on isopycnals, this result indicates that the decadal variability in temperature due to thermal advection is negligible in the South Pacific. Furthermore, there is no indication of propagation from there to the equatorial region. In contrast, the spatial pattern on the time-mean s ? = 25°C isopycnal surface (Fig. 8b) shows the large variance on the western and central South Pacific. This indicates that the surface temperature anomalies in the ocean are associated with vertical thermocline displacements. Therefore, by comparing Fig. 8a and b, the subsurface temperature variability in the western and central South Pacific, which is associated with the low frequency modulation of ENSO amplitude, is likely due to vertical thermocline displacements rather than anomalous temperature advection. Further analysis on wind stress curl indicates that the vertical thermocline displacement in the western and central South Pacific is primarily due to Ekman forcing due to wind stress curl. We found that the maximum zone of s ? = 25°C isopycnal depth in the western and central South Pacific (Fig. 8b) corresponds to the regions where the variability of wind stress curl is large (not shown).
5 Discussion and concluding remarks
We investigated the connection between TPDV and the low frequency modulation of ENSO amplitude in terms of the equatorial stratification using a long-term CGCM simulation. Our analysis shows that in the model the vertical stratification is intensified at upper levels in the western and central equatorial ocean during periods of high ENSO amplitude and/or changes in ENSO activity. The change of vertical stratification during periods of high ENSO amplitude is associated with an increase in the contribution of the higher baroclinic modes, as shown by the analysis of the projection coefficients of the momentum flux. This is consistent with the results in M04 based on SODA. Dewitte et al. (2006) also found in another CGCM simulation a positive correlation between the N3VAR index and the decadal variability of the high-order modes. Whereas the changes in equatorial stratification are apparently associated with a basin-wide thermocline decadal mode forced by Ekman pumping in Dewitte et al. (2006), we find here evidence of equatorward propagation of subsurface temperature anomalies from the South Pacific directly related to the ENSO modulation. This difference between the two models may be explained to difference in mean state in the western Pacific. In particular, in Dewitte et al. (2006) the overenergetic westward climatological equatorial currents are associated to divergence that prevents the entrance of temperature anomalies from the off equatorial regions.
We now discuss the potential mechanisms associated with the propagation of these temperature anomalies. It is difficult to elucidate these mechanisms because they are associated with the ENSO modulation which involves the nonlinearities of the tropical coupled system. It is likely that the propagation of these anomalies is the result of combined processes that act constructively to make them reach the equatorial region. First, the South Western Pacific variability is forced by Ekman pumping, however, its intensity and center of action may change with changes in ENSO characteristics. In particular, as the high-order modes contribution increases, a finer meridional scale is expected for the ENSO variability, thereby, a more equatorially trapped atmospheric response. This may contribute to the equatorward propagation of the thermocline anomalies and the associated temperature anomalies. Along with this propagation, dispersion of the extra-tropical Rossby waves may also take place. On low-frequency timescales, the crossing of Rossby waves modifies the local mean stratification, which may result in a leaking energy over higher-order baroclinic modes. Note that the recent study by Kessler and Gourdeau (2006) indicates that, at annual frequency, the variability of the South Pacific subtropical gyre has simple structure that is largely accounted for by a linear response to the wind stress curl. Approximating the western boundary current anomalies to be proportional to the Rossby solution at its western edge this may provide an explanation for the equatorward propagating subsurface temperature anomalies observed in the CGCM. At last geostrophic balance on decadal timescales could explain the propagation that accumulation or depletion of heat content in the South Western Pacific results in changes of zonal temperature gradient that may induce equatorward advection of temperature anomalies. The existence and relative role of mechanisms suggested here need to be elucidated from observations and higher resolution model simulations which correctly resolves the western boundary current system of the South Pacific.
Our study illustrates the complexity of the tropical Pacific coupled system with respect to the ENSO modulation and its source. Our results suggest that the link between the changes in equatorial mean state and off-equatorial subsurface variability may be better identified if changes in baroclinic mode energy distribution are considered rather than just changes in mean thermocline depth. Because high-order baroclinic modes are sensitive to high frequency forcing (Giese and Harrison 1990), this study also raises the issue of timescale interactions for explaining the ENSO modulation. Again, progress in simulating more realistic mean state and seasonal cycle in CGCMs is needed. The large variety of behavior among the available CGCMs simulation (AchutaRao et al. 2004) should however allow testing the hypothesis of timescale interactions associated to change in baroclinic mode energy distribution. This is our current plan.