A coupled ocean–atmosphere general circulation model is used to investigate the modulation of El Niño–Southern Oscillation (ENSO) variability due to a weakened Atlantic thermohaline circulation (THC). The THC weakening is induced by freshwater perturbations in the North Atlantic, and leads to a well-known sea surface temperature dipole and a southward shift of the intertropical convergence zone (ITCZ) in the tropical Atlantic. Through atmospheric teleconnections and local coupled air–sea feedbacks, a meridionally asymmetric mean state change is generated in the eastern equatorial Pacific, corresponding to a weakened annual cycle, and westerly anomalies develop over the central Pacific. The westerly anomalies are associated with anomalous warming of SST, causing an eastward extension of the west Pacific warm pool particularly in August–February, and enhanced precipitation. These and other changes in the mean state lead in turn to an eastward shift of the zonal wind anomalies associated with El Niño events, and a significant increase in ENSO variability.
In response to a 1-Sv (1 Sv ≡ 106 m3 s−1) freshwater input in the North Atlantic, the THC slows down rapidly and it weakens by 86% over years 50–100. The Niño-3 index standard deviation increases by 36% during the first 100-yr simulation relative to the control simulation. Further analysis indicates that the weakened THC not only leads to a stronger ENSO variability, but also leads to a stronger asymmetry between El Niño and La Niña events. This study suggests a role for an atmospheric bridge that rapidly conveys the influence of the Atlantic Ocean to the tropical Pacific and indicates that fluctuations of the THC can mediate not only mean climate globally but also modulate interannual variability. The results may contribute to understanding both the multidecadal variability of ENSO activity during the twentieth century and longer time-scale variability of ENSO, as suggested by some paleoclimate records.
The thermohaline circulation (THC) is a planetary-scale pattern of ocean currents and an important component of the climate system. It plays an essential role in maintaining the mean climate by transporting a large amount of heat, O(1015 W), from low to high latitudes (e.g., Broecker 1991; Weaver et al. 1999). Substantial changes in the THC and associated ocean heat transport would induce major climate change (e.g., Manabe and Stouffer 1999). Coupled modeling studies (e.g., Vellinga and Wood 2002; Dong and Sutton 2002; Dahl et al. 2005; Zhang and Delworth 2005) have shown that freshwater forcing of the North Atlantic can lead to a weakened THC, a cooling in the North Atlantic and warming in the South Atlantic. The climatic impacts are not restricted to the Atlantic region but extend globally through atmospheric teleconnections (e.g., Dong and Sutton 2002; Zhang and Delworth 2005) and through global oceanic wave–mediated adjustment (Goodman 2001; Johnson and Marshall 2004; Timmermann et al. 2005).
The amplitude of the El Niño–Southern Oscillation (ENSO) phenomenon in the tropical Pacific varies on decadal–multidecadal time scales (e.g., Gu and Philander 1995; An and Jin 2000; Fedorov and Philander 2001), and there is evidence from paleoclimate records of variability on longer time scales (e.g., Stott et al. 2002; Cobb et al. 2003). It is generally believed that low-frequency modulation of ENSO activity is related to the changes in the background mean state of the tropical Pacific (e.g., Gu and Philander 1995; Wang and An 2001, 2002). It is known that ENSO-like oscillations in intermediate coupled models are sensitive to the specific basic states of the ocean thermal structure (e.g., Zebiak and Cane 1987; Kirtman and Schopf 1998). However, there is no clear consensus about the exact relationship between the changes in the mean state and the modulation of ENSO amplitude, nor about the chief causes of the changes in the mean state (e.g., Rodgers et al. 2004; Yeh and Kirtman 2004; Kirtman et al. 2005). A recent multimodel analysis indicates an inverse relationship between the ENSO strength and the strength of mean easterly stress in the tropical Pacific (e.g., Guilyardi 2006).
Recent interdecadal variability in the North Atlantic, as observed by Kushnir (1994), and as simulated using coupled general circulation models (CGCMs; e.g., Timmermann et al. 1998; Delworth and Mann 2000) may have had an influence on the interdecadal variability of ENSO, as conjectured by Timmermann (2003). Several recent studies (e.g., Dong and Sutton 2002; Lu and Dong 2005; Zhang and Delworth 2005; Timmermann et al. 2005; Dong et al. 2006; Sutton and Hodson 2007) have demonstrated the influence of the Atlantic Ocean on tropical Pacific climate. This influence is transmitted by large-scale atmospheric circulation changes associated with anomalous diabatic heating in the tropical Atlantic. In addition to atmospheric bridges, oceanic waves can also transmit thermocline signals from the North Atlantic to the tropical Pacific (Goodman 2001; Timmermann et al. 2005). Both of these processes could have potential impact on ENSO variability through their impact on the thermocline of the tropical Pacific. Timmermann et al. (2005) suggested that a weakened THC leads to a weakening of ENSO variability. However, they focused exclusively on the role of oceanic teleconnections between the North Atlantic and tropical Pacific. In contrast, Dong et al. (2006) showed that a warming in the North Atlantic and cooling in the South Atlantic, such as may be associated with an enhanced THC, leads to reduced ENSO variability in a CGCM, and indicated that this influence is predominantly due to atmospheric teleconnections. The time scales associated with these teleconnections are very different. An atmospheric influence can propagate from the tropical Atlantic into the tropical Pacific in a matter of days or weeks, whereas the oceanic teleconnection is associated with a time scale of a few decades (Goodman 2001; Timmermann et al. 2005).
Is there a modulation of ENSO properties by THC change in a coupled ocean–atmosphere GCM? If there is, how is the ENSO variability modulated by THC change? What are the physical mechanisms behind this modulation? What are the roles of atmospheric and oceanic teleconnections? This study aims to address these questions by analyzing two coupled GCM experiments with idealized freshwater perturbations in the North Atlantic, and by comparing ENSO characteristics in these two perturbation experiments with those in a control simulation. We focus on detailed analysis of experiments with a single model; this approach is complementary to the multimodel analysis of Timmermann et al. (2007).
In section 2, the model and experimental design are described. The changes in the THC due to anomalous freshwater perturbations are described in section 3. The mean climate response, especially over the tropical Pacific is documented in section 4. The modulation of ENSO characteristics and the changes in asymmetry between El Niño and La Niña caused by the weakening of the THC, and the physical mechanisms involved, are studied in section 5. The discussion and conclusion are given in sections 6 and 7, respectively.
2. Coupled model and experiment design
The model we use is the third version of the Hadley Centre Coupled Ocean–Atmosphere General Circulation Model (HadCM3; Gordon et al. 2000). The resolution is 2.5° × 3.75° latitude–longitude with 19 vertical levels for the atmospheric component and 1.25° × 1.25° latitude–longitude using an Arakawa B-grid with 20 levels for the oceanic component. The two components are coupled once a day. The use of a rigid lid in the ocean model means freshwater flux at the ocean surface does not lead to changes in the volume of water in the column. To close global salinity budget, the surface freshwater flux is converted to an equivalent salinity flux (Gordon et al. 2000). The model does not require flux adjustments to maintain a stable climate. The mean climate and its stability in a 1000-yr control simulation are discussed in Gordon et al. (2000).
Two “waterhosing” experiments each 200 yr long have been performed. The waterhosing experiments study the sensitivity of the THC to an external source of freshwater (Stouffer et al. 2006). An additional freshwater flux of 0.1 and 1 Sv (1 Sv ≡ 106 m3 s−1) is applied for 100 yr to the North Atlantic Ocean. The hosing flux is applied uniformly over the Atlantic between 50° and 70°N. The external freshwater flux in then switched off after model year 100 and integration continues. A control integration, with no external water flux forcing, runs in parallel to the perturbation integrations. The initial state is taken from an 1800-yr control simulation of the coupled model. The response of THC and surface air temperature to these idealized freshwater perturbations for different models has been documented in Stouffer et al. (2006). This paper focuses on one model, HadCM3, and studies the impact on the mean climate in the Pacific and modulation of ENSO characteristics due to THC changes induced by these anomalous freshwater fluxes.
3. Changes of THC due to anomalous freshwater fluxes
A THC index is defined as the averaged meridional overturning circulation (MOC) over the latitude band 30° to 60°N at a depth of 996 m. The coupled control simulation under preindustrial greenhouse gas conditions maintains a stable climate, and the THC strength is statistically steady with a mean of 17.8 Sv (Fig. 1). Because of the slow adjustment of the deep ocean, the THC does not reach equilibrium with the external freshwater forcing during the period of freshwater input. The THC decreases more rapidly within the first few decades. A freshwater flux of 0.1 Sv leads to slightly weakened THC while the 1-Sv freshwater flux leads to an 86% decrease of the THC strength over years 50–100. The 0.1-Sv experiment shows a minimum THC intensity before the 100th year of the hosing and a recovery even before freshwater forcing stops. This recovery is related to a switch of the major convection sites to a point north of the hosing area, while overshooting is due to the fact that deep convection at the original sites resumes quickly after the termination of the freshwater input and that north of the hosing region does not switch off immediately (Stouffer et al. 2006). The 1-Sv experiment shows a slow recovery starting from year 150 from a near-shutdown condition. The maximum of the meridional streamfunction shows similar behavior, indicating that the features discussed are robust and are not sensitive to specific definition of the THC index.
The time mean (first 100 yr) pattern of the MOC in the Atlantic is given in Fig. 2, where positive values stand for an anticlockwise circulation. It reveals a maximum value of about 18 Sv for the control simulation (Fig. 2a). This strength is very close to current observations of North Atlantic Deep Water (NADW) formation of 17 Sv in the North Atlantic based on chlorofluorocarbon inventories (Smethie and Fine 2001). The MOC shows an outflow of NADW of about 14 Sv at the equator and an inflow of Antarctic Bottom Water (ABW) of 6 Sv into the North Atlantic. The sinking occurs in a broad region between 50° and 70°N, down to a depth of ∼2500 m with strong sinking taking place near 65°N. In response to the 0.1-Sv freshwater perturbation, the mean MOC weakens slightly (Fig. 2b) by 1–2 Sv. In response to the 1-Sv freshwater perturbation, the entire MOC over the Atlantic weakens significantly. The maximum overturning streamfunction is about 6 Sv, showing a decrease of about 12 Sv from the control simulation.
4. Mean climate response to freshwater perturbations
a. Changes in surface climate
1) Atlantic responses
In the 1-Sv water-hosing experiment, North Atlantic SSTs decrease in a range from 1.0° to 6.0°C (Fig. 3a) with the largest cooling (∼6.0°C) occurring over the region north of 45°N and smallest cooling (∼1.0°C) in the western subtropics. Because of decreased northward ocean heat transport (Stouffer et al. 2006), the southern Atlantic warms by about 1.0°–1.5°C. This cooling of the North Atlantic and warming of the South Atlantic is consistent with previous coupled modeling studies (Vellinga and Wood 2002; Dong and Sutton 2002; Dahl et al. 2005; Zhang and Delworth 2005) and with the multimodel results of Timmermann et al. (2007). The interhemispheric asymmetry leads to a strong cross-equatorial gradient of SST anomalies and a southward shift of the intertropical convergence zone (ITCZ) over the tropical Atlantic (Fig. 3c), which is a direct response to the underlying SST anomaly dipole (e.g., Moura and Shukla 1981). Corresponding to this southward shift of the Atlantic ITCZ is a decrease of precipitation north of the equator (∼3 mm day−1), in the Sahel region of Africa (∼0.5 mm day−1) and northern South America (∼1–4 mm day−1), and an increase of precipitation south of the equator (∼3–4 mm day−1) and over southern Africa (∼1–2 mm day−1). As shown in Fig. 3b, the northeast trade winds become stronger, while the southeast trade winds become weaker with anomalous northerly winds dominating near the equatorial latitudes in the Atlantic. The southward ITCZ shift is also associated with anomalous anticyclonic circulation in the midlatitudes and cyclonic circulation in the high latitudes of the North Atlantic, corresponding to a significant enhancement of the North Atlantic Oscillation index (Hurrell 1995). The climatological responses to the 0.1-Sv freshwater perturbation are similar to those in the 1-Sv experiment, but with much smaller magnitude (not shown). North Atlantic SSTs decrease by a maximum of ∼0.5°C, while SST anomalies are generally small (∼0.1°C) over the tropical Pacific (not shown).
2) Remote responses
In addition to the above-mentioned local responses, the weakened Atlantic THC also induces significant remote responses. These remote responses show considerable similarity to the results of Zhang and Delworth (2005). On the largest scales, the surface temperature anomalies (Fig. 3a) show a global dipolar pattern with cooling over the Northern Hemisphere and warming over the Southern Hemisphere. Cold anomalies occur over Africa, Eurasia, the North Pacific, North America, and the Maritime Continent, and warm anomalies (0.5°–1.0°C) occur over the eastern tropical Pacific cold tongue south of the equator, the south Indian Ocean, and in the central tropical Pacific (Fig. 3a). In the eastern tropical Pacific there is an anomalous meridional SST dipole, similar to that seen in the tropical Atlantic and also found by Zhang and Delworth (2005) and Timmermann et al. (2007) and, in the context of a different experiment, Wu et al. (2005). The pattern of wind stress anomalies in this region is also similar to that seen in the tropical Atlantic, with anomalous southward cross-equatorial flow, northeasterly wind stress anomalies north of and on the equator, and northwesterly anomalies south of the equator. As in the tropical Atlantic, these wind anomalies are likely to be, at least in part, a response to the underlying SST dipole. Near the equator both the wind and SST anomalies could be enhanced through the coupled wind–evaporation–SST (WES) feedback (Wu et al. 2005). According to the WES mechanism, changes in absolute wind speed play an important role in modifying the rate of evaporation, which in turn forces changes in SST. However, especially near the equator, there are other important influences on SST, such as anomalous upwelling and Ekman advection. A full heat budget for the ocean mixed layer is beyond the scope of this study, but inspection of the pattern of anomalous evaporation (not shown) suggests that near the equator in the east Pacific WES feedback is not a dominant factor. However, anomalous evaporation does appear to be an important (possibly dominant) factor farther west. A reduction in evaporation is located in the central tropical Pacific around 10°–20°N, where there is a positive SST anomaly. In this region the mean northeasterly trade winds are weakened by anomalous southwesterlies, suggesting that reduced absolute wind speed is the cause of the reduced evaporation. In addition, the cooling (negative SST anomalies) in the region of the Maritime Continent and the western tropical Pacific is associated with enhanced evaporation, but in this case the controlling factor is changes in specific humidity (dry atmosphere) rather than changes in wind speed.
In the northern extratropics the weakened THC excites a large-scale stationary wave pattern. Upper-air fields show that the associated anomalies are equivalent barotropic (not shown). A notable feature is an anomalous cyclonic circulation in the lower troposphere over the North Pacific (Fig. 3b), corresponding to a strengthened Aleutian low. This feature is associated with a local cooling of ∼1°–2°C, probably due to enhanced turbulent heat fluxes and/or anomalous Ekman upwelling. The weakened THC also leads to large-scale circulation changes in the western tropical Pacific and Indian Ocean. These changes are associated with a large-scale zonal divergence over the Maritime Continent, and weakened easterlies in the western and central tropical Pacific.
There are also significant changes in precipitation over the Indian and Pacific Oceans (Fig. 3c). Significantly enhanced precipitation is found over the western Indian Ocean, and large parts of the tropical, subtropical, and midlatitude Pacific, while reduced precipitation is found over the eastern Indian Ocean, the Maritime Continent, south and southeast Asia, and Australia. The reduced precipitation in the latter regions is associated with weakened Asian and Australian summer monsoons (not shown). The changes of precipitation over the tropical Indo-Pacific are closely related to the pattern of SST anomalies with suppressed (enhanced) precipitation over negative (positive) SST anomalies. The anomalous latent heating associated with the precipitation anomalies is also likely to play an important role (along with anomalous SST gradients; e.g., Lindzen and Nigam 1987) in driving the wind anomalies, which (as discussed above) influence SST. Hence it appears that the mean response of the tropical Indo-Pacific to weakening of the THC is substantially shaped by coupled ocean–atmosphere feedbacks. Dong et al. (2006) reached a similar conclusion, and argued that coupled feedbacks were particularly important in shaping the response in the west Pacific and Indian Ocean regions. Following Dong and Sutton (2002), and further supported by Zhang and Delworth (2005) and Wu et al. (2005), Dong et al. (2006) argued that the initial trigger for the development of anomalies over the tropical Pacific is atmospheric Rossby waves propagating westward from the tropical North Atlantic. The Rossby waves are excited by anomalous latent heating associated with anomalous precipitation. For this study, Fig. 3c shows negative precipitation anomalies in the tropical North Atlantic, the Caribbean Sea, and over northern South America, which are likely to play the primary role in forcing of the relevant Rossby waves. Sutton and Hodson (2007) show that the propagation of atmospheric Rossby waves into the east Pacific is particularly favored in boreal summer and autumn.
b. Changes in the tropical Pacific Ocean
The changes in surface climate shown in Fig. 3 are related to changes in the subsurface ocean. Here we briefly discuss the mean changes in the tropical Pacific Ocean, as a prelude to our discussion of the changes in ENSO characteristics. Figure 4 shows the changes in annual mean ocean temperature as a function of depth in the tropical Pacific. The section along the equator (Fig. 4a) shows a cooling of ∼0.8°C in the western tropical Pacific around the depth of the mean thermocline, associated with a slight shallowing of the thermocline. This feature is likely to be a dynamical response to the weakened trade winds over the western tropical Pacific as shown in Fig. 3b. Figure 4a also shows changes in near-surface temperature on the equator that appear decoupled from the changes in the thermocline. The shallow extent and spatial pattern of the surface anomalies, with positive anomalies in the west (where the trades weaken) and negative anomalies in the east (where the trades strengthen), suggest that these temperature changes are likely to be a response to anomalous evaporation controlled by the anomalous wind speed.
Away from the equator, larger anomalies in subsurface temperature are found (Figs. 4b–e). In particular, there are negative temperature anomalies, associated with a shallower thermocline, to the south (center around 10°S) and positive temperature anomalies, associated with a deeper thermocline, to the north (center around 10°N). Figure 4f shows that the anomalies in thermocline depth are up to 40 m. This asymmetric change in subsurface temperature across the equator is similar to that found by Zhang and Delworth (2005). Interestingly, Fig. 4e shows that in the eastern tropical Pacific the subsurface temperature anomalies have opposite sign to the surface temperature anomalies. As on the equator, this suggests that different processes are responsible.1 It was already argued in section 4a that the changes in SST in this region (the meridional dipole pattern) are likely to be substantially controlled by anomalous evaporation. By contrast, we might expect the changes in thermocline depth to be controlled by anomalous Ekman pumping, associated with the anomalous wind stress. We therefore calculated the anomalous Ekman pumping (not shown) and compared the result to the anomalous thermocline depth shown in Fig. 4f. We found a broad correspondence in some major features; for example, in the eastern tropical Pacific there is anomalous Ekman downwelling around 10°N, and anomalous Ekman upwelling around 10°S, corresponding to the asymmetric changes in the thermocline depth shown in Fig. 4e. However, there is no detailed correspondence between features in the anomalous Ekman pumping and features in the anomalous thermocline depth (Fig. 4f). This finding is not a surprise since the ocean’s response to anomalous Ekman pumping is not purely local, being significantly shaped by adjustment processes involving Rossby and Kelvin waves (e.g., Anderson and Gill 1975; Gill 1982) as well as advection. Anomalous ventilation may play an important role in explaining the subsurface temperature changes off and on the equator (e.g., Gu and Philander 1997). However, the meridional asymmetry of the thermocline changes suggests that the purely oceanic mechanism of adjustment to Atlantic THC change that was discussed, for example, by Timmermann et al. (2005) and Goodman (2001) is unlikely to play a key role in our simulations. Rather, as we have already argued, it is most likely that atmospheric teleconnections are the primary means via which information about the change in the Atlantic THC reaches the Pacific. This conclusion is confirmed by the fact that anomalies for the first decade computed between the 1-Sv experiment and the control (not shown) show similar characteristics of the atmospheric circulation and thermocline anomalies to that seen in the 100-yr mean anomalies.
c. Change of annual cycle in the tropical Pacific
There is evidence from both observations (Gu and Philander 1995) and model simulations (e.g., Guilyardi 2006) that the amplitude of ENSO is inversely related to the strength of the annual cycle in the tropical Pacific. It is therefore of interest to examine the changes in the annual cycle that are induced by the weakening of the THC (Fig. 5). We find that the changes in the annual cycle have many similarities to the changes found by Dong et al. (2006, their Fig. 4), but—as expected—with opposite sign and larger amplitude. The response of zonal wind stress and precipitation is strongest in the boreal summer and autumn seasons (Figs. 5a,d). These are the precisely the seasons in which Sutton and Hodson (2007) found the direct impact of Atlantic SST anomalies on the Pacific basin to be greatest, the reason being that in these seasons the latent heating anomalies are largest and farthest north, so they are more effective at exciting atmospheric Rossby waves that can propagate into the Pacific. As discussed in connection with Fig. 3, the direct impact appears to be amplified by a coupled feedback; such a feedback is suggested by Figs. 5a,c,d, which show an association between westerly (easterly) wind stress anomalies, positive (negative) SST anomalies, and positive (negative) precipitation anomalies over the central (far western) tropical Pacific in boreal summer and autumn.
Associated with the weaker easterly stress in the central tropical Pacific, the 20°C isotherm is 8–12 m shallower in the western tropical Pacific (∼135°E–160°W) in the perturbation experiment (Fig. 5b). The seasonal variation of the thermocline depth is consistent with it responding to the seasonal variation of zonal winds stress. Thus, following the onset of westerly stress anomalies in summer, the thermocline shallows in the west Pacific and deepens in the east Pacific, until midwinter when the anomalous wind stress weakens. In spring, a negative thermocline depth anomaly appears to propagate eastward into the central Pacific, while a negative SST anomaly develops in the eastern Pacific. The SST anomaly is a response to wind anomalies north of the equator that are forced directly from the Atlantic (Wu et al. 2005; Dong et al. 2006; Sutton and Hodson 2007). Subsequently an easterly wind stress anomaly develops on the equator at around 135°W, which is likely to be a response to the SST anomaly, and appears to force shallowing of the thermocline and further cooling of SST in the far east Pacific in boreal summer. Note finally that the annual cycle of anomalous SST shown in Fig. 5c corresponds to a weakening of the mean annual cycle in the tropical Pacific and is consistent with the findings of Wu et al. (2005) and Xie et al. (2007) with regards to the impact of SST anomalies in the tropical North Atlantic on the tropical Pacific. In fact, in parallel experiments to those analyzed here, four out of five coupled GCMs show a weakened annual cycle in the eastern tropical Pacific in response to a weakening of the THC (Timmermann et al. 2007).
5. Modulation of ENSO variability by freshwater perturbations in the North Atlantic
The changes in the mean state discussed in the previous section have the potential to influence ENSO variability (Gu and Philander 1995; Guilyardi 2006). In this section we present analysis of the changes in ENSO simulated in our model experiments, and in the next section we attempt to explain these changes in terms of the changes in the mean state.
a. Enhanced ENSO variability
What then is the effect on ENSO variability of the weakened Atlantic THC? Shown in Fig. 6 is the standard deviation of monthly mean SST anomalies for the control experiment, and the changes in standard deviation, relative to the control, for the two water-hosing experiments. For the control simulation, the maximum variability occurs in the eastern and central tropical Pacific with a magnitude of 1.0°–1.2°C, similar to that found in observations (not shown). However, this variability extends too far westward into the west Pacific warm pool region. This westward extension of ENSO variability is a likely consequence of errors in the mean climate of the model in this region, which has a cold tongue extending too far westward (Collins et al. 2001).
Relative to the control experiment, the amplitude of SST interannual variability over the tropical Pacific is enhanced in the two waterhosing experiments (Figs. 6b,c). The largest increase occurs over the tropical central and eastern Pacific, while in the warm pool region the change is small. This increase is stronger for the 1-Sv experiment with ∼30%–50% increase relative to the control while it shows about 10% increase for the 0.1-Sv experiment. The standard deviation of monthly Niño-3 SST index anomalies for the first 100-yr control simulation is 0.81°C while it is 0.94°C for the 0.1-Sv experiment and 1.1°C for the 1-Sv experiment, showing increases of 16% and 36% (Fig. 7). The error in the estimate of Niño-3 index standard deviation based on 100-yr sections drawn from an 1800-yr control simulation is 0.072°C with a mean standard deviation of 0.87°C. The increase of Niño-3 SST index standard deviation for the 1-Sv experiment relative to the control exceeds 4 times the error bar, and this increase passes an F-test at the 95% confidence level. These results indicate that stronger ENSO variability is associated with a weakened THC due to the 1-Sv anomalous freshwater perturbation.
In view of the temporal evolution of the THC that is seen in Fig. 1 for the 1-Sv experiment (rapid weakening in the first 50 yr and gradual recovery after year 100), it is interesting to note the associated evolution of ENSO variance. The standard deviation of monthly Niño-3 SST index anomalies is 1.05°C for the first 50 yr and 1.12°C for the second 50 yr, corresponding to increases of 25% and 45% relative to the control, and indicating an increase in ENSO variance as the THC weakens. The standard deviation for the second 100 yr is 0.96°C (a 19% increase), compared to 1.1°C for the first 100 yr (a 36% increase), suggesting that ENSO variance declines as the THC recovers. Note that the North Atlantic area mean (0°–70°N) cooling is ∼0.6°C weaker in the second 100 yr by comparison with the first 100 yr.
Figure 7 also indicates that El Niño and La Niña events in the control simulation have a similar magnitude and this is reflected by a skewness of 0.0 for the Niño-3 SST index. This is in contrast with a skewness of 0.74 based on observations from 1903 to 2002 (Rayner et al. 2003) and indicates that the skewness of the Niño-3 SST index is not correctly simulated in the control simulation. However, with freshwater perturbations, the skewness of the Niño-3 SST index becomes positive and is 0.43 for the 1-Sv experiment, implying that El Niño events become relatively stronger. The changes in skewness indicate that the weakened THC not only leads to stronger ENSO variability, but also leads to asymmetry between El Niño and La Niña events with warm events becoming stronger. This relationship is in agreement with observational studies, which show that the asymmetry between El Niño and La Niña was stronger in the last few decades when ENSO amplitude was stronger (e.g., Jin et al. 2003; An and Jin 2004).
One of the most important features of observed ENSO events is the phase locking to the annual cycle. Most observed ENSO events begin in northern spring and peak from November to January. The annual cycle of standard deviations of interannual variability of Niño-3 SST is shown in Fig. 8a. The three model simulations indicate the strongest variability occurring in the northern winter, in agreement with observations (not shown). However, the weakest interannual variability occurs in summer, in contrast to the observed minimum in northern spring. The standard deviation in all months is higher in the two perturbation experiments than in the control, indicating the enhanced SST variability due to the weakened THC. The similar seasonal cycle suggests that the weakened THC hardly affects the phase locking of ENSO. Figure 8b shows the Niño-3 SST anomaly power spectra. The control simulation shows a dominant peak at 3.5 yr and a broad peak at 5–6 yr. This behavior is very similar to the power spectrum of observations (Guilyardi 2006). For the two perturbation experiments, the power spectra indicate enhanced power at 3–3.5 yr with reduced power at longer time scales. This result hints that in HadCM3 ENSO becomes more periodic due to the weakened THC; however the changes are not statistically significant.
b. Enhanced ENSO asymmetry
In this section, we investigate the impact of the weakened THC on the asymmetry between El Niño and La Niña. Because the ENSO variance change in the 1-Sv experiment is larger and this change is statistically significant relative to the control simulation, we will focus on the 1-Sv experiment. Shown in Fig. 9 are the composite SST anomalies for El Niño and La Niña events based on the control simulation, and their asymmetry, defined as the sum of El Niño and La Niña composites. Events are selected for the El Niño (La Niña) composites when the December Niño-3 SST anomaly is greater (less) than 1.5 (−1.5) times the Niño-3 SST standard deviation. As Fig. 9 indicates, both positive SST anomalies associated with El Niño and negative SST anomalies associated with La Niña extend too far westward by comparison with observations (not shown). Both the El Niño and La Niña composites exhibit maximum values on the equator in the eastern and central equatorial tropical Pacific from 100° to 170°W with similar magnitudes. As a result, the asymmetry between El Niño and La Niña is weak, as shown in Fig. 9c.
The composite El Niño and La Niña SST anomalies and their asymmetry for the 1-Sv experiment are illustrated in Fig. 10. This shows positive SST anomalies of above 3.0°C over the eastern tropical Pacific for the El Niño composite and of about −2.0°C for the La Niña composite, indicating that the magnitude of SST anomalies associated with El Niño is stronger, as illustrated by Fig. 10c. A possible mechanism responsible for this asymmetry is changes of the anomalous wind stress pattern associated with El Niño in the perturbation experiment relative to that in the control simulation (e.g., Kang and Kug 2002; Wu and Hsieh 2003), while nonlinear dynamical heating may also play a role (e.g., Jin et al. 2003; Dong 2005). This issue will be discussed further in the next section.
c. Changes to the evolution of ENSO events
Insight into the processes that carry the memory of the ENSO evolution can be gained by examining the temporal evolution of upper layer (top 200 m) ocean heat content (OHC) variation and its relationship with SST and wind stress anomalies. Here, year 0 refers to the year when the events mature (in December). Year −1 is the previous year and year +1 is the decay year. Composite time–longitude sections of these variables for El Niño events are shown in Fig. 11 for the control and 1-Sv experiments.
For the control experiment, positive SST anomalies (Fig. 11a) begin to develop around March (year 0) simultaneously from 90°W to 180°. The OHC evolution (Fig. 11b) indicates that positive OHC anomalies (about 0.25°C) occur in the western Pacific 10–12 months before the peak in Niño-3 SST anomalies. These OHC anomalies amplify and migrate to the central Pacific and finally into the eastern central Pacific, marking the onset of El Niño. Concurrent with the positive SST anomalies in the eastern tropical Pacific are positive zonal wind stress anomalies in the central and western tropical Pacific (Fig. 11c). Because of the well-known coupled instability (Bjerknes 1969) the warming is further amplified in the central and eastern tropical Pacific, reaching a peak anomaly of about 2.0°C in December (year 0) or January (year +1). Concurrent with the development of positive SST anomalies in the central and eastern tropical Pacific, negative OHC anomalies develop in the western tropical Pacific (Fig. 11b). These negative OHC anomalies are seen to migrate eastward and play a role in the eventual termination of the warm event, leading to negative SST anomalies in the central and eastern tropical Pacific (Guilyardi et al. 2003).
The composites for the perturbation experiment exhibit many features that are similar to the control experiment but also show some notable differences. First, OHC anomalies (0.5°C) in the western tropical Pacific 10–12 months before the peak of El Niño are stronger (Fig. 11e). Second, the peak amplitude of the anomalies in all three fields (i.e., not just SST) is larger. Third, the zonal wind stress anomalies associated with El Niño in the perturbation experiment are 20°–30° longitude eastward of those in the control simulation.
In contrast to the composite El Niño events, Fig. 12 shows that composite La Niña events are very similar (both in terms of magnitude and spatial pattern) in the control and perturbation experiments. This similarity can also be seen by comparing Figs. 9 and 10. The contrast with El Niño events suggests that the impact of a weakening of the THC on ENSO is nonlinear, being dominated by the increased amplitude of, and other changes in, El Niño events.
In the following section we attempt to explain the changes in ENSO properties found in our experiments, with reference to the previously discussed changes in the mean state. In particular we argue that our results may be substantially explained by a mechanism similar to that which Wang and An (2002) used to explain the observed decadal shift in ENSO properties that occurred in the mid-1970s.
Several authors have shown that ENSO properties changed in a coherent manner related to the mid-1970s “climate shift” (e.g., Wang and An 2002; Wu and Hsieh 2003; Jin et al. 2003; Zhang and Busalacchi 2005). These studies reveal greater amplitude and nonlinearity found during 1981–99 than during 1961–75. Spatial asymmetry (for both SST and wind stress anomalies) between warm El Niño and cold La Niña events was significantly enhanced in the later period when the “thermocline feedback” (vertical advection of temperature anomalies by mean upwelling) played an important role in ENSO growth, in contrast with the earlier period when ENSO growth was dominated by local feedback (advection of mean temperature by anomalous zonal and vertical currents; Jin et al. 2003; An and Jin 2004). Wu and Hsieh (2003) showed that there was an eastward shift in the westerly anomalies during warm ENSO episodes in the later period, but no shift in the easterly anomalies during the cold ENSO episodes, thereby enhancing the asymmetry between El Niño and La Niña. Many of these observed changes are similar to the changes in ENSO properties found in our simulations of the response to a weakened THC.
In seeking to explain the observed changes in ENSO properties, Wang and An (2002) argued that the decadal changes of the background equatorial wind, and associated upwelling, played a central role. In particular, they showed that the decadal change in mean winds was associated with anomalous zonal convergence in the central Pacific, leading to an eastward displacement of the mean convergence and mean latent heating of the atmosphere. These changes then modified the structure of the ENSO coupled mode in such a way as to displace the equatorial westerly wind anomalies eastward, and to increase the amplitude of ENSO variability.
The change in mean equatorial winds that we find in response to a weakened THC (Fig. 3b) is similar to that which Wang and An (2002) found to be characteristic of the mid-1970s climate shift. In particular, Fig. 3b shows anomalous zonal convergence in the central Pacific. This convergence is associated in August–February with enhanced SST, and precipitation (Fig. 5). Thus, during the part of the year when ENSO events grow to largest amplitude, the change in the mean state in our simulations is characterized by an eastward extension of the west Pacific warm pool. Following Codron et al. (2001) and Wang and An (2002), this extension is likely to be the cause of the eastward displacement of the westerly wind anomalies associated with El Niño events that we find (Fig. 11). It also favors larger-amplitude El Niño events. The first reason is that, in the perturbation experiment with a weaker THC, SST anomalies in the central Pacific grow on a warmer mean state. For a given magnitude of SST anomaly, the higher absolute SST will lead (as a consequence of the Clausius–Clapeyron relation) to higher rates of evaporation, precipitation, and latent heating. The higher rate of latent heating will then drive stronger zonal wind anomalies, boosting the amplitude of El Niño.
A second factor that may help to explain the increased amplitude of El Niño in response to a weakened THC is the sensitivity of SST to zonal wind stress anomalies. Because the mean thermocline slopes upward from west to east, the eastward displacement of the zonal wind stress anomalies associated with El Niño will tend to enhance the effect of “upwelling feedback” (vertical advection of mean temperature by anomalous upwelling), thereby favoring the growth of larger SST anomalies. In addition, anomalies in the mean thermocline depth (Fig. 5b) could also play a role. In particular, the shallower mean thermocline found in the central and eastern Pacific in March–June could make SST in the growth season for ENSO events more sensitive to perturbations in the thermocline depth (vertical advection of temperature anomalies by mean upwelling or thermocline feedback); such perturbations may be forced remotely by wind anomalies over the west Pacific. An enhanced role for thermocline feedback could also provide some of the explanation for the enhancement of El Niño relative to La Niña. The reason is that enhanced thermocline feedback is associated with stronger nonlinear dynamical heating (advection of the anomalous temperature gradient by an anomalous current), which tends to amplify El Niño and to damp La Niña (Jin et al. 2003; Dong 2005). The fact that the eastward shift of zonal wind stress anomalies is seen for El Niño events and not for La Niña events is another factor that explains the asymmetric response (Kang and Kug 2002; Wu and Hsieh 2003).
We have already noted a number of similarities between the changes in the tropical Pacific that we have found in response to a weakened THC and the changes that were observed associated with the mid-1970s climate shift. The comparison is summarized in Table 1. The degree of agreement is striking, but of course there are some differences. For example, some analyses suggest that an increase in the vertical stratification (and hence stronger thermocline feedback) played a larger role in the real world, with a similar structure to that found in our simulations (Fig. 5a) but with anomalies about 3 times larger (Moon et al. 2004; Zhang and Busalacchi 2005). However, this comparison is based on annual mean anomalies whereas we have shown there is a large seasonal cycle in thermocline depth (Fig. 5b).
Wang and An (2002) suggested that the decadal changes in equatorial Pacific zonal winds, which triggered the changes in ENSO, were a remote response to changes in the North Pacific. Our results suggest an alternative possibility: that the changes were remotely forced from the Atlantic. Clearly there was no shutdown of the THC in the twentieth century. However, there was prominent decadal variability in Atlantic SST, some of which is likely to have been related to variability in the THC (e.g., Delworth and Knutson 2000; Delworth and Mann 2000; Knight et al. 2005). In our previous study (Dong et al. 2006) we simulated the effects on the Pacific of decadal variability in Atlantic SST and concluded that the Atlantic Ocean could have been a significant factor in driving recent decadal variability of ENSO amplitude. The results of the present study support this hypothesis.
Spencer et al. (2007) recently presented additional evidence that the amplitude of ENSO in HadCM3 is sensitive to equatorial wind perturbations. In this case the wind perturbations were induced by a flux correction procedure. The pattern of changes in the equatorial zonal winds was somewhat different to that found in the present study (in particular it was shifted eastward), but a key common feature is anomalous westerlies in the central Pacific in boreal summer, autumn, and early winter. Furthermore these anomalous westerlies are associated with increased eastward extension of the west Pacific warm pool and give rise to an increase in the standard deviation of the Nino-3 SST index, in this case by ∼60%. Together with the results of this study, the results of Spencer et al. (2007) suggest that ENSO variance in HadCM3 is particularly sensitive to equatorial wind perturbations, especially perturbations that can change the eastward extent of the west Pacific warm pool in the seasons when ENSO events grow to their largest amplitude. This behavior is consistent with the hypothesis of Wang and An (2002) concerning the cause of the changes in ENSO properties in the mid-1970s. It will clearly be of interest to examine whether other models show similar sensitivity.
Returning to the influence of the Atlantic THC on the tropical Pacific, there is some support for our findings from paleoclimate data (Stott et al. 2002; Cobb et al. 2003). In a study of Dansgaard–Oeschger (D–O) cycles Stott et al. (2002) argued that mean El Niño–like conditions are associated with cold periods in the North Atlantic. The cold North Atlantic periods are commonly believed to be a consequence of a weakened THC, and the mean El Niño–like conditions could be a consequence of the increased amplitude of El Niño events, relative to La Niña events, that we find in our simulations. Also, coral records from the tropical Pacific (Cobb et al. 2003) suggest that there was stronger ENSO variability at the height of the seventeenth-century “Little Ice Age,” a cold period over the North Atlantic. Our results are also in agreement with paleoclimate modeling studies (e.g., An et al. 2004; Peltier and Solheim 2004), which indicate that glacial boundary conditions lead to a weakened THC and enhanced ENSO variability.
Most climate models predict a weakening of the Atlantic THC for the twenty-first century when forced by increasing levels of greenhouse gas concentrations (e.g., Schmittner et al. 2005). According to our results, and also the multimodel results of Timmermann et al. (2007), this weakening in THC might lead to enhanced ENSO variability. However, there is an obvious need to be cautious in making predictions. There are many others processes internal to the Pacific that might also play a role in modulating the amplitude of ENSO under anthropogenic greenhouse forcing (Guilyardi 2006).
7. Concluding remarks
Our idealized experiments show that a substantially weakened THC leads to significant and persistent remote responses outside the Atlantic. The mean responses include an asymmetric change of SST across the equator in the eastern tropical Pacific (associated with a weakened annual cycle), cooling and suppressed precipitation over the Maritime Continent, and substantial changes in surface wind stress over the Indo-Pacific basin. Coupled ocean–atmosphere feedbacks play an important role in the development of these remote responses, and many of the features are similar to those found by Zhang and Delworth (2005). In addition to the mean changes in the tropical Pacific, we find a substantial impact on the mean annual cycle and—the focus of this study—on the characteristics of ENSO. The weakened Atlantic THC causes an enhancement (>30%) in ENSO variability, and stronger asymmetry between El Niño and La Niña. A schematic diagram of the major elements of the mechanism via which Atlantic THC change influences ENSO variance in HadCM3 is shown in Fig. 13. These elements are as follows:
The weakened THC in the Atlantic generates an anomalous cross-equatorial SST gradient with negative SST anomalies over the tropical North Atlantic and positive SST anomalies in the tropical South Atlantic (Fig. 3a). This is associated with a southward shift of Atlantic ITCZ (Fig. 3c), and negative diabatic heating anomalies in the tropical North Atlantic. The heating anomalies excite atmospheric Rossby waves, which propagate into the tropical northeast Pacific.
The eastern tropical Pacific response to 1 above is an asymmetric change of SST with a warming to the south and cooling to the north (Fig. 3a). Over the western Pacific and Maritime Continent SST cools, while over the central equatorial Pacific it warms, particularly in August–February (Figs. 3a and 5c). The warming is associated with westerly wind stress anomalies and enhanced precipitation (Figs. 3b,c and 5a).
The mean changes are associated with an eastward extension of the Pacific warm pool, which causes an eastward shift of the zonal wind stress anomalies associated with El Niño events (Fig. 11). In addition, the westerly wind anomalies cause a slight shoaling of the thermocline in the west, reducing the mean zonal tilt of thermocline. When the wind anomalies weaken in boreal spring, this shoaling signal propagates eastward into the central and eastern equatorial Pacific (Fig. 5b).
The eastward shift of the zonal wind stress anomalies associated with El Niño events leads to enhanced ENSO variability (Fig. 6c), and a stronger asymmetry between El Niño and La Niña events (Figs. 10 and 11). The enhanced ENSO variance is attributable to a stronger Bjerknes feedback: first, the response of zonal wind stress to SST anomalies is increased as a consequence of the warmer mean SSTs in the central tropical Pacific; and second, the SST response to zonal wind stress anomalies is increased by enhanced upwelling and thermocline feedbacks. The stronger asymmetry between El Niño and La Niña is a consequence of the eastward shift of wind stress anomalies associated with El Niño (but not La Niña), and an increased role for nonlinear dynamical heating.
Our finding that a weakening of the THC leads to an enhancement of ENSO variance in the HadCM3 model is supported by the multimodel study of Timmermann et al. (2007). They show that in five coupled GCMs (of which HadCM3 is one), freshwater hosing of the North Atlantic leads to enhanced ENSO variability. It will be a task of future work to establish to what extent the mechanisms we have found in the HadCM3 model are also responsible for the ENSO changes simulated in the other models. However, it appears that in all models the influence of the Atlantic THC on the tropical Pacific is primarily mediated by atmospheric teleconnections, as hypothesized by Dong and Sutton (2002), rather than by oceanic teleconnections. Timmermann et al. (2005) showed that a dominance of oceanic teleconnections would lead to a decrease in ENSO variability in response to THC weakening, rather than an increase.
One aspect of the changes in ENSO due to a weakened THC highlighted by Timmermann et al. (2007) is that, in four of the five models analyzed, increases in ENSO variability are correlated with a weakened annual cycle in the eastern tropical Pacific. A similar inverse relationship has been noted in observations (Gu and Philander 1995; Fedorov and Philander 2001; Timmermann et al. 2007) and in other model experiments (e.g., Guilyardi 2006). Our results may provide some insight into this correlation. In particular, we have argued that westerly wind anomalies over the central and western Pacific in boreal summer and autumn (Fig. 5a) cause eastward extension of the warm pool in the period of August–February (Fig. 5c) and thereby cause both the increase in ENSO variability and a weakening of the annual cycle in SST. The weakening of the annual cycle is a simple consequence of the fact that the August–February period includes the part of the year (September–October–November) when mean SSTs are coldest in the central/eastern equatorial Pacific. In fact, the weaker annual cycle in HadCM3 is also associated with negative SST anomalies in the March–April–May period (Fig. 5c), when the mean SSTs are warmest. These boreal spring SST anomalies appear to be controlled by off-equatorial processes not directly related to the equatorial zonal wind anomalies (Xie et al. 2007; Wu et al. 2005). Developing a full understanding of the weaker annual cycle is therefore an important subject for further work.
This work was supported by the EU DYNAMITE (003903-GOCE) and ENSEMBLES (GOCE-CT-2003-505539) projects at the National Centre for Atmospheric Science. R. Sutton is supported by a Royal Society University Research Fellowship. We thank Jonathan Gregory for performing the coupled model simulations at the Hadley Centre for Climate Prediction and Research and for making them available to us. We would also like to thank Axel Timmermann for insightful discussion. We thank Drs. L.-X. Wu and R. Zhang, and an anonymous reviewer, for their constructive comments and suggestions. We also thank Drs. E. Guilyardi, J. Gregory, and M. Latif for their comments and suggestions on the early version of the paper.
Corresponding author address: Buwen Dong, Walker Institute for Climate System Research, University of Reading, and National Centre for Atmospheric Science—Climate, Reading, RG6 6BB, United Kingdom. Email: firstname.lastname@example.org
Note that there are regions where the SST changes appear to be consistent with, and are therefore probably controlled by, the changes in the thermocline depth: for example, around 10°S in the western tropical Pacific.