1. Introduction
The study of potential changes that occur in El Niño–Southern Oscillation (ENSO) when the mean climate conditions are changed can be of interest both for reconstructing paleoclimate conditions (Rosenthal and Broccoli 2004) and for impact-oriented projections of future climate under continuing anthropogenic greenhouse warming (Houghton et al. 2001).
Although the great majority of numerical GCMs produce an ENSO-like dominant mode of tropical Pacific interannual variability (Latif et al. 2001), the specific properties (pattern, amplitude, and frequency) of ENSO anomalies are model dependent. Furthermore, the models’ sensitivity to climate change differs widely, making projections of ENSO change quite uncertain (Collins and CMIP Modelling Group 2005). Differences in resolution, formulation, and parameterization, of both the atmosphere and the ocean components, have been shown to be important factors (e.g., Guilyardi et al. 2004).
Here we address the question of which climate mean-state properties are most likely to significantly affect the ENSO, and in which way. Theoretical studies of the ENSO based on models of intermediate complexity (henceforth ICMs; Neelin et al. 1998) suggest a basic set of mean-state parameters (mean zonal wind stress, depth, and strength of the equatorial thermocline) to which ENSO is expected to be sensitive. However, such predictions need to be compared with results from GCMs, which have a more complex (and hopefully realistic) behavior.
We follow the approach of previous modeling studies (Timmermann et al. 1999; Otto-Bliesner et al. 2003; Peltier and Solheim 2004; An et al. 2004) that attempt to relate modeled changes in the ENSO to changes in the simulated mean climate, with the help of the insight gained from previous analyses of the model (cf. references in section 2), ENSO theory (Philander 1990), and ICM-based studies of ENSO (Fedorov and Philander 2001; Wang and An 2002; An et al. 2004). We describe the changes in the model climatologies to which the ENSO is likely to be sensitive and attempt to relate them with the changes in the simulated ENSO.
We have analyzed model data derived from monthly average climate diagnostics as simulated in three integrations of the Third Hadley Centre Coupled Ocean–Atmosphere GCM (HadCM3) model in its standard configuration. All three setups are for steady-state conditions, and the integrations have been carried forward for many centuries to ensure statistical sampling of their internal variability and residual drifts. The first integration, dubbed “CTL” hereafter, constitutes the baseline “control” integration (Gordon et al. 2000) that simulates an “unperturbed” nineteenth-century climate. The second integration is set up to simulate Last Glacial Maximum (LGM) conditions at the peak of the glaciation 21 000 yr ago (Hewitt et al. 2003). Finally, the third integration is greenhouse forced (GHS) with a fourfold CO2 atmospheric concentration relative to CTL, but otherwise identical to it (Thorpe 2004). ENSO in an early part of this integration is discussed in Collins (2000), but the analysis is limited to the Niño-3 index and does not assess the actual SST variability in the model integration. Immediately from an initial EOF analysis (Toniazzo 2002b) significant differences with CTL are seen, and they require explanation.
In the present paper, after a description of the HadCM3 integrations and their main characteristics (section 2), we show the principal differences in tropical Pacific climatologies (section 3). In section 4, we discuss the differences in the simulated ENSO. Section 5 focuses on ENSO-related oceanic and atmospheric anomalies in the equatorial Pacific, which contribute to anomalous SST tendencies in ways that depend on the mean climatological conditions and are at least partly responsible for the changes in the ENSO. In section 6, we summarize our results, compare them with results from other models, and draw conclusions.
2. Model integrations
HadCM3 is a first-order finite-difference numerical model comprising an atmosphere component on a 3.75° × 2.5° latitude–longitude grid with 19 hybrid (p − z* or p − σ) levels, and an ocean component on a 1.25° × 1.25° regular latitude–longitude grid with 20 vertical (z−) levels. As usual, surface, boundary layer, cloud, and convection processes, among others, are parameterized using local or single-column bulk formulations. The reader is referred to Pope et al. (2000) for details.
We have analyzed a 100-yr section from each of the three HadCM3 integrations mentioned in the introduction. The CTL integration has a nineteenth-century climate setup (atmospheric CO2 concentration at 280 ppm) that serves as a baseline for most anomaly or perturbation experiments with HadCM3. The integration is more than 3000 model years long and shows no significant drifts. This numerical model has been shown to be adequate in representing many properties of the observed mean climate. Having been used for over 5 yr in many experiments, its virtues and shortcomings are by now well known (Slingo et al. 2003, and references therein). Some of the biases that affect the ENSO include poor resolution of the complex island topography and local circulation patterns in the Maritime Continent (MC; Neale and Slingo 2003), an easterly low-level wind bias in the equatorial Pacific (especially in the west) associated to a western extension of the cold equatorial SST “tongue,” a split tropical convergence zone that tends to straddle the equator, and a rather zonal structure of the South Pacific convergence zone (SPCZ; Davey et al. 2002). In the ocean model, the equatorial current structure is represented rather well considering the limited resolution used, but the undercurrent is rather weak and thermocline wave activity is damped (Gordon et al. 1995). Nevertheless, the model produces an ENSO with many realistic properties (Latif et al. 2001; Collins et al. 2001; Toniazzo 2002a), characterized by a Niño-3 index with a slightly short main periodicity of broadly 3 yr, a rather narrow, but not extreme, phase locking on the winter season, and a realistic rms amplitude of 0.93°C (0.96°C for positive anomalies) as compared to the Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST; Rayner et al. 2003) value of 0.80°C (0.91°C for positive anomalies). The model El Niño SST anomaly (SSTA) patterns are similar to observed ones, albeit too narrow in the meridional direction; anomalous precipitation over the equator is produced, but overall convective activity is less responsive than observed and tends to stay locked over the eastern side of the SPCZ and near the Maritime Continent (Spencer and Slingo 2003).
The setup of the LGM simulation is characterized by representations for surface albedo, paleotopography, and ice cover corresponding to the ICE-4G(VM2) model of Peltier (1994), and a reduced concentration of atmospheric CO2 at 200 ppm. Ocean topography is the same as CTL, except for changes in the land–sea mask, which allow for extended coastlines in Indonesia and the North Atlantic. Its spinup has been realized with the help of Haney forcing of SSTs (Haney 1971), but subsequently the integration has been continued without flux correction for over 1000 model years in close-to-equilibrium top-of-the-atmosphere (TOA) balance (−0.3 W m−2) and without important drifts. Properties of the mean modeled LGM climate, as well as details of the model setup, are documented in Hewitt et al. (2001, 2003).
The GHS simulation setup is identical to CTL, except for a fourfold atmospheric concentration of CO2 (i.e., 1120 ppm). Although the spinup was spread over 70-yr model time with a CO2 increase of 2% per annum, and the simulation at fixed CO2 concentration has been integrated for over 1000-yr model time, TOA balance is not achieved and remains at around +1 W m−2, with a persistent warming of the deep ocean and reduction of the polar ice caps (which eventually disappear completely). Still, the climatological drift in the tropical atmosphere and in the upper (ice free) oceans is negligibly small, over 100 yr, compared to the amplitudes of annual and interannual variability. For our purposes, this solution can also be considered to represent an equilibrium climate.
One-hundred-year sections from the HadCM3 integrations were used for the analysis. When results were thought to be potentially dependent on sampling (e.g., the Fourier spectra or EOFs), their robustness was controlled with the help of an additional 100 yr of data from each run. Unfortunately, failures in the data storage systems of the Hadley Centre have caused “gaps” of a few months in the available datasets. For time series analysis, values for the anomalies of up to three consecutive missing months were obtained by spline interpolation. For LGM, due to missing years, the time series from the ocean submodel had to be truncated, leaving a maximum length of only 74 yr.
3. Climatological differences
In this section, we show the differences in the mean and seasonal-mean climatology of the tropical Pacific, with a focus on changes in the SSTs, in the strength, seasonality, and patterns of tropical Pacific atmospheric convection and atmospheric circulation, and in the thermal and dynamical structure of the upper equatorial Pacific.
a. Mean climatology
The net global-average TOA radiation budget balances a total net incoming shortwave forcing of 235, 240, and 245 W m−2 in LGM, CTL, and GHS, respectively. The warmer GHS climate is due to increased atmospheric opacity alone, while cooler LGM conditions are also affected by surface-albedo changes in the Northern Hemisphere. The surface perturbation corresponding to the GHS climate relative to the CTL climate is larger than that between the CTL and LGM climates, with a tropical-mean surface temperature difference of around 4.2°C in the former case and of only 1.5°C in the latter. Midtropospheric temperatures are also affected differently (6.9° and 3.2°C at 500 mbar for GHS–CTL and CTL–LGM, respectively). Consistently with a “cold finger” effect (Broecker 1997), LGM has generally drier conditions (3.37 mm day−1 30°S–30°N average compared to 3.51 mm day−1 in CTL), while GHS is slightly wetter (3.55 mm day−1), presumably due to increased convective heating offsetting larger midtropospheric radiative cooling.
Figure 1 summarizes the changes in climatology, with progressively warmer and wetter conditions in CTL and GHS. Temperature differences are characterized by a more pronounced warming in the eastern parts of the equatorial Pacific, with a reduction in the zonal SST contrast. Also, in both the CTL–LGM and GHS–CTL differences an area of reduced warming is found in the southeastern subtropical Pacific. The western parts of the basin also tend to warm less. Precipitation tends to increase mainly over or near the equator.
Comparing CTL with LGM, the temperature difference pattern shows warm SSTs in the eastern equatorial Pacific and in the tropical Atlantic. They are associated with stronger easterlies in LGM, linked to a southward extension of the cold anticyclone over the Laurentide ice sheet (Hewitt et al. 2003) and probably with anomalous orographic forcing from the ice sheet over the eastern North Pacific via a mechanism described in Timmermann et al. (2004b). The increased differential SST between the tropical Pacific and tropical Atlantic also tends to encourage a strengthened interbasin atmospheric flow.
A reduced-average interbasin atmospheric flow implies a westerly low-level wind difference in the eastern equatorial Pacific (Fig. 1, arrows in the upper-left panel). Increased east equatorial SSTs in CTL relative to LGM are associated with reduced equatorial upwelling and reduced evaporation. The cyclonic low-level circulation difference encourages convective activity in the Gulf of Panama (Fig. 1, bottom left), which is associated with a Gill-type (Gill 1980) warming response in the midupper troposphere around 90°W, 15°N, and 15°S (black contours). The corresponding upper-level warm anticyclones (arrows in the lower-left panel) reinforce the subtropical westerlies, especially in the Northern Hemisphere. Correspondingly, vertical wind shear is reduced in the east and west Pacific convection areas and is stronger in the central Pacific (contours in the upper-left panel), inhibiting deep convection.
The surface warming appears to be well correlated with the change in cloud radiative forcing (not shown). The latter quantity is mainly associated with a shortwave (SW) component. It is negatively correlated with the changes in 500-hPa vertical pressure velocity, ω, and positively with relative humidity, and reflects an overall reduction in low-cloud cover in CTL compared to LGM. In the southeastern Pacific, however, negative cloud forcing associated with the stratocumulus region is stronger and more extended in CTL than in LGM. This is associated with relatively cool SSTs and the westward contraction of the SPCZ.
Turning to GHS, we also see strong negative cloud forcing associated with a relatively cool Southern Ocean and a contraction of the SPCZ relative to CTL (Fig. 1, right). However, the warming pattern for GHS–CTL in the tropical Pacific appears generally to be associated with changes in midtropospheric moisture (not shown), with no clear relation with cloud forcing except in the southeastern Pacific. The increase in humidity, and its radiative effect, is larger in normally drier regions than in humid regions such as the main convective areas, where instead it tends to induce increased precipitation. The warming in the northeastern Pacific is associated with a weakened anticyclone, and weaker northeasterly trades. Along the equator, the reduction in zonal wind stress is accompanied by reduced upwelling and latent heat loss. This induces an eastward displacement of Maritime Continent convection to the western equatorial Pacific (colors, lower-right panel). The upper-tropospheric response resembles a double-anticyclone Gill-type pattern over the western Pacific, associated now with reduced wind shear over the whole area of the equatorial Pacific, which thus becomes more prone to deep convection.
In GHS, the SPCZ is contracted west- and equatorward, while convective activity in the intertropical convergence zone (ITCZ) north of the equator is reinforced. In LGM, convection is more concentrated north and south of the equator, split between the ITCZ and the SPCZ, and reduced precipitation occurs over the Maritime Continent. Both the CTL–LGM and GHS–CTL differences show enhanced convection over the equator, although with different zonal distributions. Compared to the other integrations, CTL has a larger meridional SST gradient north of the equator in the central-western Pacific, and its ITCZ is narrower and is associated with stronger trade winds in the western Pacific. Convection over the Maritime Continent is also greatest in CTL, in spite of the generally larger precipitation rates of GHS. This is mainly due to the eastward displacement, in GHS, of the convective activity usually associated with this region. GHS also shows increased local wind stress and wind shear in the southeastern Pacific, associated with convergence over the central equatorial Pacific. A similar change is seen in the Gulf of Panama, coincident with reduced convection despite the warmer SSTs.
The changes in low-level circulation are reflected in the annual-mean wind stress forcing of the ocean (Fig. 2). On the equator, the maximum wind stress is progressively farther west, with reduced strength in GHS, as expected. The meridional distribution shows the reinforcement of the southern anticyclone and the weakening of the northern anticyclone, which appear systematic across the three integrations, although much more pronounced between GHS and the other two. The zonally integrated wind stress is seen to be comparable in CTL and LGM.
Consistent with the changes in wind stress, equatorial SSTs and depth-average ocean temperatures show reduced zonal SST contrast in GHS. The vertical thermal structure of the equatorial Pacific (Fig. 3) indicates a progressive shoaling of the thermocline and of the zonal current system. In GHS, the weakened zonal wind stress allows the equatorial thermocline to rise, in qualitative accord with theoretical scaling laws (Pedlosky 1996, p. 341). The subtler changes between LGM and CTL cannot be explained in this way. Probably, the different zonal distribution of the wind stress plays a role. Where the thermocline is very shallow, its position may also be affected by stronger surface shortwave flux, half of which is penetrative. The warmer climates have a somewhat weaker vertical stratification (Fig. 3, right) in the thermocline. Near the surface, GHS has increased stratification, in part related to the reduced depth of the thermocline. The weaker trades also affect the South Equatorial Current, which is shallower and slower (by about 25%) in GHS.
b. Seasonal cycle
Figure 4 shows a time–longitude plot of the seasonal cycle along the equatorial Pacific in the monthly average SSTs, zonal wind stress, and 0–360-m depth-average ocean temperatures for the three climatologies.
Over the equator, the wind stress becomes progressively weaker from LGM to GHS, and its seasonality changes, with a single broad maximum in late summer and autumn taking the place of two distinct peaks around August and December. In terms of wind stress work (i.e., the integral of the stress over the equatorial area), the winter maximum found in LGM and CTL disappears in GHS, and the corresponding spring relaxation of the trades, which tends to mark the termination of El Niño events in CTL, is replaced by a gentler transition in late January–February.
The changes in seasonal climatology of equatorial SST trace the changes in strength, location, and seasonality of the wind stress, with cold eastern Pacific SSTs forming in August and persisting into spring in LGM, and a single, westward-shifted and relatively short-lived minimum in autumn for GHS. The depth-average ocean temperature shows a greater persistence of warm-pool water over the equator in GHS (cf. the 17.5°, 19.5°, and 21.5°C isotherms for LGM, CTL, and GHS, respectively, in the rightmost panel in Fig. 4).
The change in the wind stress seasonality over the equator is related to the different patterns of convective activity in the three climates (Fig. 5). Equatorial wind stress is associated with the trade winds emanating from the subtropical highs in the eastern parts of the ocean basin, and converging on three main areas of convection, the ITCZ in a zonal band north of the equator, the MC between New Guinea and Indonesia, and the SPCZ across Melanesia and Polynesia. Southeasterlies are seasonally prevalent when the ITCZ is active, and northeasterlies are prevalent when MC and the SPCZ convection are active. In the central and eastern Pacific, there is an additional easterly contribution to the total wind stress in winter from the low-level flow associated with the southern anticyclone, which retains a cross-equatorial southeasterly component, and a westerly contribution in autumn associated with convergence over the Gulf of Panama, which is responsible for the wind stress “dip” in October. Figure 5 shows that the changed seasonal cycle in GHS mainly depends on the weakening MC and SPCZ convection, along with the weakening of the northern anticyclone, and persistent convergence over the ITCZ in northern winter.
4. Interannual Pacific SST anomalies and ENSO indices
Figure 6 shows the leading EOF of Pacific monthly mean SST anomalies and its amplitude time series for the three simulations LGM, CTL, and GHS. The leading EOF is in all cases ENSO-like in pattern and periodicity, with a center of action in the equatorial Pacific, accounting for more than 20% of the total variance not related to the seasonal cycle, and having an amplitude of above 1°C and a main periodicity of 2–4 yr. The center of action is progressively displaced toward the west from LGM to CTL and from CTL to GHS, moving the anomaly away from the Niño-3 region. The maximum amplitude increases, but the area affected by the largest warm anomaly is reduced. The dominant period in the Fourier decomposition of the EOF1 time series appears to decrease from around 4 yr to around 2 yr.
The Niño-3 and Niño-4 index time series (Fig. 7) support the results from EOF decomposition. The Niño indices are in fact strongly correlated with the EOF1 time series. The standard deviation of the Niño-3 index steadily decreases and is equal to 1.1°, 0.93°, and 0.83°C, for LGM, CTL, and GHS, respectively. The changes in the standard deviation of the Niño-4 index, which suggest weaker variability in CTL, are not as significant. The Niño-3.4 index (5°S–5°N, 170°–110°W), on the other hand, has a very similar standard deviation in all three integrations (1.01°, 0.98°, and 1.02°C for LGM, CTL, and GHS, respectively).
The changes in the standard deviation of ENSO indices between LGM and CTL are not statistically robust, as CTL has a significant interdecadal variability whereby its ENSO activity increases, and the 1 − σ intervals of the running 30-yr standard deviation of the respective indices overlap. The changes between LGM and GHS, by contrast, are significant but are related to the pattern shift, and not to the overall strength of the ENSO.
The visual impression of an increasing frequency of the ENSO oscillation that can be obtained from the Niño-3 indices in Fig. 7 is confirmed by the power spectra in Fig. 8, which indicate a statistically significant shift toward shorter periods. A similar change is also present in the Niño-4 index power spectrum going from LGM to CTL and to GHS. In the warmest climate, there is no power in periods at or longer than 4 yr. The power in the CTL and GHS indices peaks between 2 and 3 yr, while in LGM ENSO variability peaks between 3 and 4 yr. This shift appears to be robust and is also observable if different sections of the runs are used, or if different windows are applied in the Fourier analysis. Moreover, and perhaps more importantly, El Niño events, which are defined by a peak in the Niño-3 or in the Niño-3.4 index exceeding 1.5 standard deviations, tend to be more numerous over any 30-yr period (or longer) in the warmer integrations.
These results show a continuous change in the pattern of the SST variability associated with the ENSO, with El Niño events becoming more frequent in the warmer climates. The changes in standard deviation suggest weaker variability in CTL. This conclusion however does not reflect the changes in the frequency distribution of Niño-3.4 amplitude, and especially its tails (Fig. 9). In GHS, there is a lack of very large warm anomalies, while in LGM they are more frequent. This result does not depend on the pattern change and is also found for the Niño-3 or Niño-4 SST indices. In GHS, the skewness and the kurtosis of all indices are negative (except for the skewness of the Niño-4 index), while they are positive for both indices in the other two climates, and largest in LGM. As a result, changes in the average strength of El Niño events depend on the chosen threshold used to define them. With the 1.5 standard deviation criterion (which is commonly used), they are not significant between our three integrations.
Analysis of the seasonal distribution of Niño-3/-4 variability (Fig. 10) shows that Niño-3 variability tends to lock onto the seasonal cycle in early winter, November to January, when in fact 9 out of 13 events (defined as anomalies greater than 1.5 standard deviations of the index) reach their peak. While there is some variability in spring and early summer (Fig. 10), “events” between June and October are rare. In CTL, Niño-3 variability follows a similar pattern, although winter events, which are still the great majority, are more evenly spread between October and March. The termination of El Niño events takes place around June. In contrast, in GHS Niño-3 activity is at a maximum in August when most of the events (8 out of 18) reach their peak. The Niño-3 variance, however, is fairly constant from August to January, with January a second local maximum, and seven events, like in CTL, peak between November and February. Moreover, the August events do not terminate until March and stay above the Niño-3 standard deviation, mostly with a secondary peak, during winter. In summary, Niño-3 events appear to tend to develop and terminate earlier in the warmer climate conditions, and with a sharper onset locked on the seasonal cycle.
Niño-4 variability tends to follow a similar pattern of change. The seasonal locking of this index is stronger than for Niño-3 in all runs, and Niño-4 events tend to occur later in the year than Niño-3 events in all cases. In GHS, however, Niño-3 variability is secondary and does not reflect the actual amount of variability in the equatorial Pacific. All Niño-3 events in GHS are related to a Niño-4 event that takes place in the same year.
The temporal relationship between Niño-4 and Niño-3 anomalies is illustrated in Fig. 11 by the lagged correlations of the increments of the two indices [i.e., the time series of the quantities Niño-4(t + 1) − Niño-4(t) and Niño-3(t + 1) − Niño-3(t)]. Lagged correlations above the autocorrelation function of Niño-3 increments (dotted lines) can be deemed significant. In all cases, a significant correlation at positive lags between 1 and 5 months for the Niño-4 increments is seen, suggesting either direct westward signal propagation or a Niño-3 trigger for the Niño-4 anomaly. Significant differences however are seen between the three climates at negative lags (i.e., Niño-4 leading Niño-3), especially for positive Niño-3 anomalies. Positive Niño-4 increments tend to lead Niño-3 progressively more in the warmer climates, with a clean double-peak structure for the GHS case. This suggests that equatorial SST anomalies tend to be mainly westward propagating in LGM and CTL, while an increasing eastward-propagating component appears in the warmer climates. Such an impression is also obtained from inspection of Fig. 12, showing the zero-level contour in a Hovmoeller plot of 5°S–5°N equatorial SST anomalies for a 20-yr section from each model simulation. Another confirmation is found by tracking the maximum SST anomalies in time for the composite El Niño event (Fig. 13), defined as the anomaly of the average over all years when the Niño-3 index exceeds 1.5 standard deviations in extended winter—November to March. (We have assured ourselves that other choices yield results consistent with those discussed throughout this work.) In the months leading to the peak, LGM shows a weak westward propagation of the maximum SST anomaly. CTL anomalies appear broadly stationary between August and November, followed by clear westward propagation well into the Niño-4 region. In GHS, the largest SST anomaly is attained within the Niño-4 region, as already seen. In the months around the peak, it tends to move westward, as in the other climates. Eastward propagation however appears before the peak of the event (from May to August), and in the decay stage (from February to May). Note also from Fig. 13 that the peak in maximum amplitude is reached latest in LGM (February) and earliest in GHS (November), as was suggested also by the canonical Niño indices.
5. Generation of ENSO-related SST tendencies
The amplitude, pattern, propagation properties, and frequency of ENSO SST anomalies are related to the processes providing amplification or damping of such anomalies. Here we discuss some of the processes envisaged by ENSO theory (Neelin et al. 1998) to produce such mechanisms: the alteration of subsurface ocean temperatures induced by anomalies in the depth of the equatorial thermocline; changes in upwelling and surface advection; and anomalies in surface fluxes determined by evaporation and cloud response.
SST growth generated by surface processes (upwelling and advection) is usually associated with westward propagation in what is called the “surface” or “SST” mode, while SST tendencies generated by thermocline depth anomalies generally cause eastward propagation of the SST anomaly (“thermocline mode”; Jin and Neelin 1993).
Such processes are linked in a feedback loop responsible for the growth and decay of El Niño events that can be divided into an atmospheric component, which creates anomalous wind stress and surface radiation fluxes in response to SST anomalies, and an oceanic component, which responds to the wind anomalies, in particular the zonal component over the equator.
a. Atmospheric sensitivity and surface fluxes
Figure 14 shows diagnostics originally introduced by Timmermann et al. (1999) as a proxy for the strength of the atmospheric feedback responsible for growth of the ENSO mode. The “atmospheric sensitivity” index is defined as the covariance of equatorial wind stress and Niño-3.4 SST anomalies, divided by the variance of Niño-3.4 SSTA. Between CTL and GHS, there is an increase in atmospheric sensitivity. A steadily increasing atmospheric sensitivity index is also found in the transient greenhouse-forced integration (orange curve in Fig. 4) leading from CTL to GHS, which also displays an increase in frequency and a westward displacement of the ENSO SST pattern (Toniazzo 2002b). The difference between CTL and LGM is less clear cut, because of the large interdecadal variations of the wind stress sensitivity in CTL.
Figure 14 (right) shows that the differences in atmospheric sensitivity index correspond to different zonal distribution of wind stress anomalies over the equator, with a greater response in the east for LGM and in the west for GHS. Because the low-level wind is affected both by direct thermal forcing of the boundary layer and by convective activity, the pattern differences depend on the SST anomaly pattern associated with Niño-3.4 anomalies as well as on the different climatologies. To investigate the importance of the latter, we have conducted a simple atmosphere-only experiment (using the atmospheric component of HadCM3) forced with SSTs obtained from the climatology of GHS, and with the same CO2 concentration, plus El Niño anomalies obtained from the composite El Niño of CTL (applied in the Pacific only). The result, for a single El Niño event, is shown as the thin solid line in Fig. 14 (right). Despite the obvious differences, the wind stress response appears also in this case locally larger, more concentrated, and to some extent shifted westward compared to CTL.
Figure 15, for the peak of the composite El Niño event, shows that the spatial relation between SST anomalies and precipitation anomalies differs in the three climate integrations. Almost irrespective of the shift in SST anomaly pattern, the largest convective anomalies occur in all cases at the edges of the climatological convective regions, where convection is more easily triggered by SST anomalies. As the main area of tropical Pacific convection changes from a zonally extended SPCZ in LGM to a more zonally confined region in the western Pacific in GHS, anomalous convective activity moves from a zonal band south of the equator to a narrower, near-equatorial region around the date line in GHS. As a result, the maximum SST anomalies are widely separated from the maximum precipitation anomalies in LGM, while they are nearly collocated in GHS. This contributes to increased precipitation anomalies. The wind anomalies become more confined, more coherently zonal in direction, and more symmetric about the equator. The greater sensitivity of equatorial Pacific convection to SST anomalies in GHS is probably related to weaker climatological wind shear and greater moisture availability. Precipitation is not as well correlated with low-level convergence in LGM and CTL, and anomalous moisture sources are more remote. To some extent, similar differences hold also between CTL and LGM.
The changes in strength and location of climatological convective activity, and the associated cloud cover, also affect the response of radiative and evaporative surface fluxes to SST anomalies over the equatorial Pacific (Fig. 16). The SW radiation flux at the surface depends on the prevailing climatological cloud cover conditions. In LGM and CTL, the boundary layer in the eastern Pacific is capped by thin, warm stratocumulus. Warm SST anomalies imply reduced boundary layer stability and destruction of some stratocumulus clouds, resulting in a positive SW feedback. By contrast, in the trade cumulus areas over the equatorial Pacific “cold tongue” farther west the SW feedback is negative. In GHS, with increased convective cloud over the equator, the SW feedback to warm SST is generally smaller in magnitude, and negative everywhere. This behavior is again qualitatively confirmed in the atmosphere-only experiment mentioned previously.
The surface longwave feedback (not shown), which depends on the changes in cloud-top temperature, generally has the opposite sign than the SW feedback, and is smaller in magnitude. In GHS, it is quite negligible, probably due to both the generally higher climatological cloud tops and the greater clear-sky downward longwave. Evaporation depends nonlinearly both on the wind speed and on the SSTs. With increasing SSTs and weakening winds the differences in anomalous latent heat loss per unit Niño-3.4 SST anomaly (also not shown) are small between the three climates. Sensible heat fluxes are never important.
The zonal distribution of the total surface flux response to ENSO anomalies (Fig. 16, right) mainly reflects the behavior of the SW component. The implied damping is of the order of a few tenths of degree C per month per degree C Niño-3.4 anomaly, comparable to the positive feedbacks dependent on upwelling and advection (see below), and therefore its timing and its zonal distribution affect the magnitude, pattern, and evolution of the SST anomalies.
During the growth phase of an El Niño, surface flux damping is negligible everywhere over the equatorial Pacific in LGM and in CTL. In the eastern side of the basin, surface fluxes sometimes have a reinforcing effect, especially in LGM. Around the peak of the composite event, damping is still confined to the west of the equatorial Pacific, with negligible effect elsewhere. Damping becomes large and effective only in winter, when SPCZ convection moves out into the central Pacific, particularly east of the date line, where SST anomalies are largest. By contrast, with increased climatological CAPE over the equator, in GHS anomalous convection and surface flux damping are significant and approximately stationary, located around the date line, for the whole duration of the event, competing with the reinforcing wind feedback locally. We have already noted that the equatorial wind stress associated with northeasterlies emanating from the northern anticyclone is weak in GHS compared with LGM and CTL. As the seasonal cycle moves convective activity away from the ITCZ and toward the SPCZ (Fig. 5), climatological wind stress weakens, while surface flux damping increases, thus causing the termination of the event. This can help explaining the earlier and sharper seasonal locking of ENSO in GHS (Fig. 10).
b. Thermocline anomalies
Climatological upwelling produces a positive anomaly in SST tendency if the temperature contrast between the upwelled water and the surface water is smaller (i.e., less negative) than average, which in turn can be associated with a deepening of the thermocline. If the latter is due to positive wind stress anomalies over the equator, this mechanism gives rise to an eastward-propagating “mode” of SST anomaly growth (Jin and Neelin 1993). To assess whether it is effective in the HadCM3 integrations we compare anomalies in subsurface-to-surface ocean temperature contrast with SST anomalies and with anomalies in the depth of the thermocline. The latter is diagnosed for each month and location as the depth at which the vertical potential temperature gradient is largest (excluding the first two model layers). Note that a bulk warming of the ocean does not imply subsurface influence on the evolution of surface anomalies, hence using the depth of a fixed isotherm, or the temperature along a particular depth, as proxies for the thermocline depth can be misleading.
Figure 17 shows such relations. Results from composites (graphs) and lag correlations with Niño-3 (small Hovmoeller plots) are shown, along with representative El Niño events plotted against the subsurface temperature contrast anomalies and thermocline depth anomalies (large Hovmoeller plots). Positive anomalies in subsurface vertical temperature contrast are consistently associated with thermocline depth anomalies and with positive trends in SST anomalies only in GHS.
The lack of correlation between large El Niño surface anomalies and the thermocline suggests that the canonical thermocline mode is generally ineffective in LGM and CTL, while more active in GHS, which is consistent with the evidence from propagation characteristics of anomalies discussed above. This does not mean that subsurface temperature anomalies are absent or unimportant in LGM and CTL. Because of phase mixing, noise, and the effect of strong surface feedbacks that generate a largely independent mode of SST growth, composites and correlations tend to lead to an underestimation of the importance of subsurface anomalies for the initiation of El Niño events. Single events also differ widely from one another, especially in LGM where, for example, the strongest event in the 100-yr section is thermocline led and eastward propagating. The examples shown in Fig. 17 are representative of the “typical,” or most common, El Niño events.
In all integrations, warm anomalies at depth are seen to accompany El Niño events, often propagating eastward with typical speed exceeding the advective velocity of the undercurrent. In a large fraction of all events in LGM and CTL, these anomalies surface in the eastern equatorial Pacific and trigger or reinforce El Niño events. In about a third of all cases, however they fail to directly produce significant surface anomalies. The majority of El Niño events in LGM and in CTL are then either generated at the surface, with a neutral or even cool thermocline (when they occur the year after the “thermocline event”), or appear to be of a mixed nature. Although there are episodes of clear eastward propagation connected to the thermocline, SST anomalies appear to be reinforced in a stationary or westward-propagating mode that is not coupled with the thermocline.
In GHS, by contrast, thermocline anomalies surface much farther west and affect SSTs across the whole basin. All warm and cold events are led by the thermocline, and they all show eastward propagation, especially for the small anomalies. Near the peak of the event, SST growth appears to occur in a stationary or westward-propagating mode, decoupled from the thermocline anomaly and qualitatively similar to the other two integrations.
For LGM and CTL, more robust eastward propagation is seen in the development of cold anomalies, when the thermocline is raised with respect to its mean position, and has a smaller zonal slope. A higher sensitivity of the thermocline to wind anomalies may arise from reduced stability (Fedorov and Philander 2001). However, Fourier analysis of the Hovmoeller fields of 2°S–2°N depth anomalies (not shown) does not show an increase of thermocline activity, in terms of total power in nonadvective, eastward-propagating components, between LGM and GHS. This suggests that the main reason for the stronger effect of thermocline anomalies on near-surface waters in GHS is its smaller climatological depth (see Fig. 3; Fedorov and Philander 2001), especially in the central Pacific. Increased numerical dissipation of equatorial Kelvin waves (Gordon et al. 1995) with increasing depth and zonal slope could also be a factor adversely affecting the thermocline feedback in LGM and CTL.
c. Upwelling and zonal advection
Given the dominance of the surface westward mode of SST growth, the response of the surface ocean to wind stress anomalies clearly must be very important for the ENSO in HadCM3. Unfortunately, due to high-frequency variability and nonlinearity, monthly mean data are not adequate to diagnose them properly. From budget calculations for the upper-ocean layer, we find residuals that are particularly significant, comparable in magnitude to the total, around and after the peak of El Niño events. Here we provide estimates for upwelling and zonal-current feedbacks based on correlations with monthly mean wind stress, which are not useful for budget calculations but are better behaved, and show systematic and interpretable differences between the three integrations.
Figure 18 shows temperature advection anomalies associated with equatorial wind stress anomalies. In the Niño-4 region, increased subsurface stratification is accompanied by reduced anomalies in upwelling and increased anomalies in surface current (Fig. 18, left), consistently with reduced vertical mixing. Such relation still holds in the Niño-3 region when comparing CTL and LGM. In the case of GHS, there is a significant thermocline feedback that implies a significant negative correlation between anomalies in subsurface stratification and in wind stress, and the reduced subsurface ocean stability with weakened easterlies encourages upwelling and vertical mixing of momentum (not shown).
In terms of implied advective SST tendencies, the upper panels in Fig. 18 confirm a predominant thermocline effect (advection of temperature anomaly field, WΔT ′, where W is the climatological upwelling) in GHS across most of the basin, by virtue of its shallower thermocline. In CTL and LGM, instead, the increase in surface stability associated with warm surface anomalies in the central Pacific may be responsible for both a greater sensitivity of the upwelling anomaly W ′ to wind anomalies in the east (not shown) and the confinement of the net positive SST tendency to the Niño-3 region.
Zonal advective anomalies (Fig. 18, bottom) are always largest in the west, as may be expected by the upwelling-dependent downstream intensification of the South Equatorial Current and by the localization of the largest negative zonal SST gradient in the central Pacific. While advective SST tendencies associated with positive wind stress anomalies are positive over most of the basin for LGM and CTL, in GHS both the reduced climatological zonal SST gradient and the localization of the largest SST anomalies in the Niño-4 region implies positive anomalous SST tendency contributions at and west of the date line, and negative elsewhere. LGM, by contrast, shows the largest, positive advective (vertical and zonal) SST tendencies associated with wind stress anomalies.
Positive advective and upwelling feedbacks are consistent with westward propagation of the surface anomalies (Philander 1990, their section 6.2; Jin and Neelin 1993). They appear to be dominant in LGM and in CTL. In GHS, the thermocline feedback is dominant in the eastern Pacific, while advective tendencies are dominant in the central and western Pacific. Also in this case, when comparing with the right-most panel in Fig. 17, consistency with theoretical expectations is found.
6. Summary and discussion
a. Summary
The ENSO that is found in the GHS integration of the HadCM3 model is more rapid, more regular, and more localized over the central equatorial Pacific, with a maximum SST anomaly that is shifted westward, compared to the ENSOs in the LGM and CTL integrations (section 4, Figs. 6 –9). The phase locking of the oscillation is earlier in the seasonal cycle (section 4, Fig. 10). Differences of the same sign, although smaller, are observed between the CTL and the LGM integrations. Additionally, however, the ENSO in GHS differs from the other cases in that it shows clear evidence of eastward propagation of the anomalies (section 4, Fig. 12; section 5, Fig. 17), although the maximum SST anomalies still tend to propagate westward like in the other cases (section 4, Fig. 13; section 5, Fig. 17) The amplitude of the ENSO appears to be slightly larger in LGM and in GHS than in CTL.
Anomalies in the depth of the thermocline of the equatorial Pacific are always accompanying ENSO events, and usually initiating them, but over the development of an El Niño event they appear to feed back on SSTs, via climatological upwelling, only in the case of GHS (section 5, Fig. 17). This is in theoretical agreement (e.g., Philander 1990, their section 6.2) with the observed eastward propagation of SST anomalies in GHS. Such difference with the colder climate integrations is likely to depend on the smaller climatological depth and zonal slope of the equatorial Pacific thermocline (section 2, Fig. 3), in accord with results from models of intermediate complexity (ICMs; Fedorov and Philander 2001). Moreover, the reduced depth of the thermocline in the central Pacific appears to allow subsurface anomalies to produce SST anomalies farther west than in the LGM and CTL integrations (section 5, Fig. 17 and Fig. 18, top).
The sensitivity of surface winds to anomalous SST forcing (as measured by the “atmospheric sensitivity” index) appears to increase (section 5, Fig. 14), and is particularly large in GHS. The wind response appears to be increasingly stronger in the west and weaker in the east (Fig. 14), tracing qualitatively the zonal distribution of climatological wind stress. In GHS, it is also narrower and more symmetric about the equator, and more confined zonally (Fig. 15). We have verified by means of an atmosphere-only experiment that the increased atmospheric sensitivity and the changed pattern partly depend on the mean climatology with greater propensity to convection in the central equatorial Pacific.
We have attempted an estimate of the surface ocean response to changed wind stress (Fig. 18). There appears to be a shift toward smaller upwelling and somewhat larger zonal current anomalies, which combine with climatological SST gradients and with SST anomalies to produce a weaker reinforcement of SST anomalies in the east in CTL compared to LGM (Fig. 18). In GHS, they produce a strengthened reinforcement in the west, while in the east the thermocline feedback is counteracted by a negative advection feedback. The different magnitude and pattern of advective SST tendencies are likely to depend on the increasing surface stability from LGM to GHS. The upper panels in Fig. 18 highlight the negligible contribution, on average, of thermocline anomalies compared to upwelling anomalies in LGM and CTL. Positive advective and upwelling feedbacks are associated with westward propagation as found in LGM and CTL (Philander 1990; Jin and Neelin 1993). The prevalence of westward-propagating SST anomalies in LGM and CTL, and in the central and western Pacific for GHS, is qualitatively consistent with our estimate of the feedbacks.
Surface flux anomalies are also seen to change in pattern and strength across the equatorial Pacific (Fig. 16). The main differences are in the SW component, the response of which depends on climatological cloud cover. In LGM and CTL, SW anomalies damp SST anomalies in the western and central Pacific but reinforce them in the eastern Pacific during the growth and peak phases of ENSO events. This appears to be related to loss/gain of stratocumulus cover for warm/cold SST anomalies. Such effect is greater in LGM, but absent in GHS, where the equatorial Pacific is relatively warm. Total surface flux damping of SST anomalies is significant only where convection takes place. This results in larger and earlier damping in GHS, with a pattern that is stationary and nearly collocated with the reinforcing wind anomalies. In CTL and LGM, reinforcing feedbacks and damping take place in different areas until late in the development of the SST anomaly. LGM in particular has very small surface flux damping rates.
In summary, El Niño events in HadCM3 are generally initiated by an oscillation of the thermocline, which appears to be an important component of the ENSO. Eastward-propagating SST anomalies associated with the thermocline occur when the thermocline is raised sufficiently close to the surface to affect the subsurface vertical temperature gradient, but generally its coupling with the atmosphere appears to be weak, and the amplification of surface anomalies is mainly associated with surface processes (upwelling and advection). Strong surface flux damping occurs in areas where there is convective activity.
In LGM, thermocline anomalies usually only surface far in the east, where they cannot propagate much farther eastward. Occasionally, when the thermocline is raised compared to its climatological depth farther west, it concurs to generate larger El Niño events, but its large climatological depth in the Niño-4 region prevents significant surface effects there. Climatological wind stress and upwelling are large in the east, implying stronger positive surface feedback. The general ineffectiveness of surface flux damping on the equator, related to very cold eastern-equatorial SSTs that do not easily allow convection to move away from the climatological convergence zones, permits large SST anomalies, in excess of 5°C, to be sustained.
In CTL, the situation is similar, but the sensitivity to thermocline-depth anomalies is weaker, probably because of the reduced climatological wind stress and upwelling in the eastern Pacific. In the central and west Pacific, however the wind stress is larger, allowing for a stronger surface feedback. Surface anomalies are never eastward propagating. With the increased proximity of climatological convection to the equator, surface flux damping is also larger. Strong events are not as large as those in LGM.
In GHS, the thermocline is sufficiently close to the surface to allow thermocline modes throughout the equatorial Pacific. Subsurface anomalies generate significant SST anomalies in the central Pacific, which then propagate eastward. At the same time, however, surface amplification takes place, generating a larger, independent, stationary or westward-propagating SST anomaly. The surface feedbacks are stronger than in the other integrations but also more localized in the central/western Pacific. The same is true for surface flux damping. Such strong, localized atmospheric response is associated with the greater susceptibility to convective activity over the equator, which in turn is related with smaller climatological wind shear and reduced meridional SST contrast. Increased stability of the oceanic surface layers may also be a contributing factor for localized SST growth. Larger growth rates would favor El Niño events of large amplitude, but surface flux damping associated with convection limits the value of equatorial SSTs (at about 35°C).
The earlier and sharper phase locking of the ENSO in GHS relative to the LGM and CTL (Fig. 10) is related to the relative weakening of the northeasterly trades emanating from the northern anticyclone in the tropical east Pacific and associated with winter convection over the Maritime Continent and in the SPCZ (Fig. 5) that imply a single, summer maximum of equatorial wind stress and hence a progressive weakening of the positive surface SST feedbacks in autumn and winter.
Our conclusion from the present analysis is that in HadCM3 the main climatological factors affecting the ENSO, in strength and pattern, are the depth of the equatorial thermocline, the distribution and the seasonal cycle of climatological convection in the tropical Pacific, and the sensitivity of atmospheric convection to equatorial SST anomalies.
b. Discussion and conclusions
We have presented an in-depth analysis of the ENSO in three steady-state integrations of HadCM3, showing significant shifts in ENSO-related variability when the mean climatology is changed. This work also updates the analysis of an early part of the GHS integration by Collins (2000), which was based on the Niño-3 index alone and hence failed to detect the changes in ENSO properties with respect to CTL.
The changes in the mean climatology and in the ENSO that have been presented here are clearly model dependent. For example, the relatively cold tropical Atlantic SST in LGM may be related to the stronger Atlantic overturning circulation (Hewitt et al. 2003), which is not seen in other models (Peltier and Solheim 2004; Timmermann et al. 2004b). Paleoclimatological evidence points toward a slightly reduced ENSO variability in the LGM (Tudhope et al. 2001) but is not compelling, and consensus has still to be reached within the paleoclimatological community (e.g., Rosenthal and Broccoli 2004). Significant uncertainties also affect the cloud response to warming (e.g., Williams et al. 2003). Discussing such uncertainties would exceed the scope of the present work, however, which is concerned primarily with the relations between changes in the mean climatology and changes in ENSO, as represented in the HadCM3 model.
The differences we have found in some climatological aspects of the three integrations help demonstrate the westward pattern shift seen between LGM, CTL, and GHS as related with a predominance of surface feedbacks reinforcing SST anomalies, and with their shift toward the western Pacific related to the mean zonal wind stress and to the atmospheric sensitivity to local SST changes. In a qualitative sense, we also can argue that the greater regularity of the ENSO in GHS is related with the smaller climatological depth of the thermocline, which is always active, as opposed to LGM or CTL where it reinforces surface anomalies occasionally, with the rapid growth of the anomaly, and with the reduced temperature contrast between the equatorial “cold tongue” and the surrounding convective areas, which implies increased surface flux damping that does not allow SSTs to grow much beyond a certain threshold. We also can see a relation between the seasonal cycle and the preferred period for El Niño growth in the three integrations.
Our analysis does not explain, however, the small changes in the overall amplitude of the ENSO, and the increase in its frequency. Drawing direct inferences from simple ENSO paradigms, like the “recharge oscillator” of Jin (1997), might help our understanding but may equally be misleading, unless evidence can be shown that the implied mechanisms are indeed dominant for the ENSO in the GCM (see discussion in Neelin et al. 1998). For example, we do not find support for a simple recharge oscillator–type ENSO in HadCM3. Moreover, the changes between integrations are many and are complex and require careful consideration. Clearly, more work in complementary modeling and in sensitivity studies is necessary to assess the causes of the changes in the model ENSO. While this exceeds the scope of the present paper, here we briefly compare our results with previous studies using other GCMs and ICMs.
Timmermann et al. (1999) observe an increase in ENSO amplitude and frequency under greenhouse forcing in the ECHAM4/OPYC3 GCM. This GCM predicts changes in atmospheric mean state and seasonal cycle that are consistent with ours (Timmermann et al. 2004a). However, they do not find increased atmospheric sensitivity and attribute the changes in the ENSO to an increased vertical temperature contrast across the equatorial thermocline. Otto-Bliesner et al. (2003) and Peltier and Solheim (2004) find increased LGM ENSO variability, with no significant change in frequency, from a simulation with the National Center for Atmospheric Research (NCAR) Climate System Model version 1.4 (CSM1.4) GCM. Various aspects of the mean-state changes between the control (with trace gas concentrations of 1990) and the LGM integration of this model are the opposite of those seen in HadCM3 between LGM and CTL (and similar to those between GHS and CTL): weakened trade winds, smaller zonal and meridional SST gradients, and a raised thermocline. An et al. (2004) use each of these mean-state changes in turn to perturb the background climatology of an ICM of the Cane–Zebiak (CZ) type (Cane and Zebiak 1985). They find that the raised thermocline and the reduced meridional SST contrast both promote amplification of the ENSO. The effect of the latter change is attributed to reduced damping by meridional advection.
In HadCM3, we also find a slightly reduced meridional SST gradient in LGM compared to CTL, which may contribute to its slightly larger ENSO variability. Moreover, between CTL and GHS, with a larger reduction in meridional SST gradient, we can see a significant decrease in heat loss from the equatorial Pacific due to oceanic meridional advection during El Niño events. Here we also argue however that a reduced meridional and zonal SST gradient makes the equator more prone to anomalous convective activity, potentially increasing the model’s atmospheric sensitivity, a quantity related to the crucial (and poorly constrained) “coupling strength” parameter μ of the Cane–Zebiak model. Greater values of μ promote growth of the leading ENSO mode (Jin and Neelin 1993). At the same time, convection is also associated with greater surface flux damping. In general, by reducing the effective Bjerknes feedback [parameter R in Eq. (13a) of Neelin et al. (1998)], this would contribute to reduced amplitude and increased frequency, opposing the effect of meridional SST gradients.
The effects of the raising thermocline and of weakening zonal winds are more complex. The presence of a thermocline oscillation with a long period and of strong surface feedbacks with dominant westward propagation of SST anomalies suggest that the ENSO in CTL corresponds to a mixed state involving both local and remote modes as described by Fedorov and Philander (2001), probably near the middle of the diagrams on the right-hand side of Fig. 11 in their paper. This would be consistent with the slightly strong winds and weak thermocline of CTL. A reduction in zonal wind stress by 25% and in mean thermocline depth by 34%, as in GHS compared to CTL, would then imply according to Fig. 11 of Fedorov and Philander (2001) a net transition toward a local (“SST”) mode in GHS, with larger amplitude and higher frequency. Although we do not see such an extreme behavior, at a qualitative level our results are not necessarily inconsistent with this and are in line with those of An et al. (2004). Certainly, the raised thermocline in GHS favors fast and local feedbacks over remote and delayed feedbacks.
The changes in the mean climatology between LGM and CTL are not consistent with results from the CSM1.4 simulations, so the results for the ENSO are difficult to compare. There is a better agreement with the results from the ECBILT/CLIO model (Timmermann et al. 2004b), which however does not resolve equatorial Pacific variability. An et al. (2004) estimate the implied changes in ENSO according to their linearized CZ model and find an increase in amplitude and frequency (Fig. 7 in their paper). (Unfortunately, they do not discuss the separate effect of single climatological changes, as for the CSM1.4 simulations.) With HadCM3, we find changes in the amplitude distribution, with a possible small increase in variability. The dominant period of the oscillation however tends to increase. The changes in atmospheric sensitivity and surface flux damping between LGM and CTL are similar, with opposite sign, to those between GHS and CTL, even if smaller (and not as statistically robust, due to the significant interdecadal variability of CTL). Moreover, the patterns of wind and SST changes are qualitatively similar to those discussed by Wang and An (2002) in relation to the 1976 “climate shift” that appears to be associated with less frequent El Niño events that have a stronger signature in the eastern Pacific. Wang and An (2002) find that the driving change is in the zonal wind stress distribution. A wind response to SST changes shifted more to the east, as in LGM compared to CTL (cf. Fig. 14), leads to a greater delay in the thermocline response to wind stress anomalies and to lower frequencies.
A limitation arising in the use of ICMs for the interpretation of results from full GCMs lies in their crude parameterization of atmospheric dynamics, particularly convection. We have shown that the changed atmospheric response plays an important role in the ENSO changes between CTL and GHS, in particular the westward shift in the pattern. Guilyardi et al. (2004) have shown that the model’s atmospheric component has a large impact on the simulated ENSO. To bridge the gap between GCMs and ENSO theory, along with insightful modeling like that of An et al. (2004), the inclusion of atmospheric processes beyond the Gill linear dynamical response in CZ models might be helpful.
Acknowledgments
I am grateful to Adam Scaife for the comments and advice in the course of this work, and to Matthew Collins for comments and suggestions. Useful comments and thorough correction of the manuscript by one of the anonymous referees were also much appreciated. I would like to thank Chris Hewitt for making the data from his Last Glacial Maximum integration available to me. My thanks go to the tropical CGAM group at Reading University for much interesting and stimulating discussion. This work was supported by the U.K. Department of the Environment, Food and Rural Affairs under Contract PECD 7/12/37 and by the Government Meteorological Research Contract.
REFERENCES
An, S-I., A. Timmermann, L. Bejarano, F-F. Jin, F. Justino, Z. Liu, and A. W. Tudhope, 2004: Modeling evidence for enhanced El Niño–Southern Oscillation amplitude during the last glacial maximum. Palaeoceanography, 19 .PA4009, doi:10.1029/2004PA001020.
Broecker, W. S., 1997: Mountain glaciers: Recorders of atmospheric water vapor content. Global Biogeochem. Cycles, 11 , 589–597.
Cane, M. A., and S. E. Zebiak, 1985: A theory for El Niño and the Southern Oscillation. Science, 228 , 1085–1087.
Collins, M., 2000: Understanding uncertainties in the response of ENSO to greenhouse warming. Geophys. Res. Lett., 27 , 3509–3512.
Collins, M., and CMIP Modelling Groups, 2005: El Niño- or La Niña-like climate change? Climate Dyn., 24 , 89–104.
Collins, M., S. F. B. Tett, and C. Cooper, 2001: The internal climate variability of HadCM3, a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn., 17 , 61–81.
Davey, M. K., and Coauthors, 2002: STOIC: A study of coupled model climatology and variability in tropical ocean regions. Climate Dyn., 18 , 403–420.
Fedorov, A. V., and S. G. Philander, 2001: A stability analysis of tropical ocean–atmosphere interactions: Bridging measurements and theory for El Niño. J. Climate, 14 , 3086–3101.
Gill, A. E., 1980: Some simple solutions for heat-induced tropical circulation. Quart. J. Roy. Meteor. Soc., 106 , 447–462.
Gordon, C., D. K. Wright, and C. M. Roberts, 1995: The simulation of the tropical oceans in models of different horizontal resolution. Hadley Centre for Climate Research Tech. Note CRTN 58, 18 pp.
Gordon, C., C. Cooper, C. A. Senior, H. Banks, J. M. Gregory, T. C. Johns, J. F. B. Mitchell, and R. A. Wood, 2000: The simulation of SST, sea-ice extent and ocean heat transport in a version of the Hadley Centre coupled model without flux adjustment. Climate Dyn., 6 , 147–168.
Guilyardi, E., and Coauthors, 2004: Representing El Niño in coupled ocean–atmosphere GCMs: The dominant role of the atmosphere model. J. Climate, 17 , 4623–4629.
Haney, R. L., 1971: Surface thermal boundary conditions for ocean circulation models. J. Phys. Oceanogr., 1 , 241–248.
Hewitt, C. D., A. J. Broccoli, J. F. B. Mitchell, and R. J. Stouffer, 2001: A coupled model study of the last glacial maximum: Was part of the North Atlantic relatively warm? Geophys. Res. Lett., 28 , 1571–1574.
Hewitt, C. D., R. J. Stouffer, A. J. Broccoli, J. F. B. Mitchell, and P. J. Valdes, 2003: The effect of ocean dynamics in a coupled GCM simulation of the Last Glacial Maximum. Climate Dyn., 20 , 203–218.
Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, 2003: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.
Jin, F-F., 1997: An equatorial ocean recharge paradigm for ENSO. Part I: Conceptual model. J. Atmos. Sci., 54 , 811–829.
Jin, F-F., and J. D. Neelin, 1993: Modes of interannual tropical ocean–atmosphere interaction—A unified view. Part I: Numerical results. J. Atmos. Sci., 50 , 3477–3503.
Latif, M., and Coauthors, 2001: ENSIP—The El Niño simulation intercomparison project. Climate Dyn., 18 , 255–276.
Neale, N. B., and J. M. Slingo, 2003: The Maritime Continent and its role in the global climate system: A GCM study. J. Climate, 16 , 834–848.
Neelin, J. D., D. S. Battisti, A. C. Hirst, F-F. Jin, Y. Wakata, T. Yamagata, and S. Zebiak, 1998: ENSO theory. J. Geophys. Res., 103 , 14261–14290.
Otto-Bliesner, B. L., E. C. Brady, S. Shin, Z. Liu, and C. Shields, 2003: Modeling El Niño and its tropical teleconnections during the last glacial-interglacial cycle. Geophys. Res. Lett., 30 .2198, doi:10.1029/2003GL018553.
Pedlosky, J., 1996: Ocean Circulation Theory. Springer Verlag, 453 pp.
Peltier, W. R., 1994: Ice-age paleotopography. Science, 265 , 195–201.
Peltier, W. R., and L. P. Solheim, 2004: The climate of the Earth at Last Glacial Maximum: Statistical equilibrium state and a mode of internal variability. Quat. Sci. Rev., 23 , 335–357.
Philander, G., 1990: El Niño, La Niña, and the Southern Oscillation. International Geophysics Series, Vol. 46, Academic Press, 293 pp.
Pope, V. D., M. L. Gallani, P. R. Rowntree, and R. A. Stratton, 2000: The impact of new physical parameterizations in the Hadley Centre climate model—HadAM3. Climate Dyn., 16 , 123–146.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of sea surface temperature, sea ice and night marine air temperature since the late nineteenth century. J. Geophys. Res., 108 .4407, doi:10.1029/2002JD002670.
Rosenthal, Y., and A. J. Broccoli, 2004: In search of Palaeo-ENSO. Science, 304 , 219–221.
Slingo, J. M., and Coauthors, 2003: How good is the Hadley Centre climate model? Research at CGAM on identifying and understanding model systematic errors, 1999–2002. University of Reading, CGAM/NCAS Rep., 60 pp.
Spencer, H., and J. M. Slingo, 2003: The simulation of peak and delayed ENSO teleconnections. J. Climate, 16 , 1757–1774.
Thorpe, T., 2004: Report on the possible long-term global climate response to a stabilisation of CO2 at 4x pre-industrial levels. Hadley Centre Rep. to the U.K. Department of Environment, Fisheries and Rural Affairs, 16 pp.
Timmermann, A., J. Oberhuber, A. Bacher, M. Esch, M. Latif, and E. Roekner, 1999: Increased El Niño frequency in a climate model forced by future greenhouse warming. Nature, 398 , 694–697.
Timmermann, A., F-F. Jin, and M. Collins, 2004a: Intensification of the annual cycle in the tropical Pacific due to greenhouse warming. Geophys. Res. Lett., 31 .L12208, doi:10.1029/2004GL019442.
Timmermann, A., F. Justino, F-F. Jin, U. Krebs, and H. Goose, 2004b: Surface temperature control in the North and tropical Pacific during the last glacial maximum. Climate Dyn., 23 , 353–370.
Toniazzo, T., 2002a: ENSO in HadCM3. Hadley Centre Rep. to DEFRA, 29 pp.
Toniazzo, T., 2002b: Report on ENSO in a HadCM3 warming scenario. Hadley Centre Rep. to DEFRA, 10 pp.
Tudhope, A. W., and Coauthors, 2001: Variability in the El Nino-Southern oscillation through a glacial-interglacial cycle. Science, 291 , 1511–1517.
Wang, B., and S. I. An, 2002: A mechanism for decadal changes of ENSO behaviour: Roles of background wind changes. Climate Dyn., 18 , 475–486.
Williams, K. D., M. A. Ringer, and C. A. Senior, 2003: Evaluating the cloud response to climate change and current climate variability. Climate Dyn., 20 , 705–721.
Climatology differences (left) between CTL and LGM and (right) between GHS and CTL. Color coding refers to (top) SST differences (°C) and (bottom) precipitation differences (mm day−1); black contours are for differences in (top) 200–850-hPa wind shear (intervals of 2.6 m s−1, dashed lines for negative values) and (bottom) 200–500-hPa average temperature [intervals of (left) 0.15° and (right) 0.4°C]. Finally, the black arrows refer to wind velocity (m s−1) at (top) 850 and (bottom) 200 hPa.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
(left) Zonal and (right) meridional distribution of the mean zonal wind stress for the LGM (dashed line), CTL (solid line), and GHS (dashed–dotted lines) integrations.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Mean thermal structure of the equatorial Pacific for the LGM (dashed line), CTL (solid line), and GHS (dashed–dotted lines). (left) The mean position of the thermocline (defined as the depth of the maximum 4°S–4°N meridional-average vertical potential temperature gradient). (right) The 4°S–4°N, 170°–110°W area-average profiles of vertical temperature gradient.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
The oceanic seasonal cycle of the equatorial Pacific. Hovmoeller diagrams of the 4°S–4°N meridional averages of (left) zonal wind stress, (center) SST, and (right) 0–360-m depth-average ocean temperature are shown. Time increases upward; abscissa values are in degrees east. The lower parts of the contour plots refer to the LGM run, the middle parts to the CTL run, and the upper parts to the GHS run.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
The atmospheric seasonal cycle over the tropical Pacific for (left) LGM, (middle) CTL, and (right) GHS. Monthly mean pressure velocity, ω, at hybrid level 8 (about 500 hPa over sea), averaged over the areas of the MC (thin long-dashed lines), over the ITCZ (dashed–dotted lines) and over the SPCZ (dashed–triple-dotted lines) are shown together with 850-hPa zonal wind velocity (thick dashed line) and with zonal surface wind stress (thick solid line). Units are 0.01 Pa for wind stress τx, m s−1 for wind speed U850, and 0.01 Pa s−1 for pressure velocity ω.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Leading EOF of Pacific SST monthly average departures from the mean seasonal cycle for the three HadCM3 simulation runs: (top) LGM, (middle) CTL, and (bottom) GHS. (left) The EOF fields (°C), (middle) the amplitude time series (nondimensional), and (right) the power density of the Fourier transforms of the amplitude time series. The normalization is such that the amplitude time series have an rms of 1. Seventy-year sections from each run are used.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Niño-3 indices for 100-yr sections of the (top) LGM, (middle) CTL, and (bottom) GHS simulations.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Power spectra of the (top) Niño-3 and (bottom) Niño-4 indices of the LGM, CTL, and GHS simulations. The symbols represent the estimates for the normalized (nondimensional) power in each bin of 1-yr width. Vertical lines indicate statistical uncertainties. LGM (dashed line), CTL (solid line), and GHS (dash-dotted line).
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Cumulative distribution of positive Niño-3.4 anomalies.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Standard deviation of (top) Niño-3 and (bottom) Niño-4 indices.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Lag correlations of monthly Niño-4 anomaly increments vs Niño-3 anomaly increments, divided for (left) positive Niño-3 anomalies and (right) negative Niño-3 anomalies. Dotted lines are the autocorrelation functions of Niño-3 increments: (top) LGM, (middle) CTL, and (bottom) GHS.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Hovmoeller plot of 5°–5°S SST anomalies along the equator for 20-yr sections of each model simulations (left) LGM, (middle) CTL, and (right) GHS. Only the zero-anomaly contour is shown. Time increases upward.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Amplitude and position of maximum and near-maximum SST anomalies for the composite El Niño event in the three climates. All model grid points at which SST anomalies within 0.4°C of the maximum anomaly are shown; darker shades of gray are used to indicate larger anomalies, and the lines connect the maxima. (top) Amplitude, (middle) longitude, and (bottom) latitude of the anomalies as a function of time; (left) LGM, (center) GHS, and (right) GHS.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
(left) Atmospheric sensitivity indices (after Timmermann et al. 1999) for the three HadCM3 integrations (LGM: blue; CTL: black; GHS: red). The orange curve represents the transient HadCM3 integration that joins CTL and GHS with a yearly 2% increase in CO2. A sliding 20-yr window is used to calculate the correlation coefficients. (right) Regression of the 5°S–5°N wind stress response onto the 5°S–5°N, 170°–110°W average SST anomaly in the three steady-state integrations. The thinner line represents the result from a 4xCO2 atmosphere-only integration forced with the GHS SST climatology and with El Niño anomalies from the CTL composite.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
December–February (DJF) (top) SST and (bottom) precipitation fields for the three HadCM3 integrations; (left) LGM, (middle) CTL, and (right) GHS. Climatology is shown as solid contours (drawn at 1°C intervals for SSTs and at 2 mm month−1 intervals for precipitation), and El Niño anomalies are shown as colors (scales on the bottom of each panel). (top) The arrows represent 850-hPa DJF wind anomalies for each case (arrows of magnitude smaller than 1 m s−1 are not plotted), and (bottom) the broken contours represent DJF 200–850-hPa thickness anomalies (drawn at 25-m intervals).
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Sensitivity of (left) SW and (right) total surface fluxes to SST anomalies in the equatorial Pacific.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Depth of the thermocline in the equatorial Pacific and anomalies in subsurface temperature difference in the development of an El Niño event. In the graphs on the left, the thick solid and dashed lines represent the thermocline depth in the composite event and in the climatology, respectively, for April–June (scale on the left in each panel, m). The thin solid lines indicate the anomalies in vertical temperature contrast between 25 and 5 m of depth (scale on the right, °C); positive values indicate elevated 25-m temperature with respect to 5-m temperature. The small Hovmoeller contour plots to the right of the graphs show the lag correlations with the Niño-3 index of the: (left) thermocline depth (contours) and anomaly in temperature difference between the depth of the thermocline and the surface (colors); (right) SST anomaly (contours) and of the subsurface vertical temperature contrast (colors). Contour interval is 0.2. The three large Hovmoeller plots on the right-hand side show a representative El Niño event for each climate; SST anomalies are shown as black contour lines (thin solid for positive, thick solid for zero, and thin dot-dashed for negative; units are °C); colors indicate thermocline depth anomalies (interval: 10 m; yellow/red denotes positive values).
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
Sensitivity of ocean surface advective heating processes to wind stress anomalies in the equatorial Pacific. (top) The graphs show regressions onto mean zonal wind stress anomalies of SST tendencies implied by linear monthly mean anomalous upwelling terms. (bottom) Similarly, the graphs show the regressions onto zonal-mean wind stress anomalies of the linear zonal temperature advection anomalies. The regressions for the total anomalous vertical and zonal advective heating are very close to the sum of the linear terms shown. Terms associated with the advection anomaly field are on the right-hand side and those associated with the temperature anomaly field are on the left-hand side.
Citation: Journal of Climate 19, 19; 10.1175/JCLI3853.1
