Abstract

Prediction of Indian Ocean interannual variability may be limited by the systematic biases in coupled GCMs or by a lack of resolution of the processes involved. In particular, little is known about the impact of ocean resolution on simulated climate variability. The simulation of Indian Ocean climate and dipole is investigated in Hadley Centre coupled models with different horizontal and vertical ocean resolutions.

The mean state of the Indian Ocean is found to improve only slightly when horizontal resolution is increased from 1.25° to ⅓° and when vertical resolution is increased from 20 to 40 vertical levels due to a small reduction of the Maritime Continent warm bias. However, improvements in the simulation of the dipole are more substantial. All versions of the model realistically simulate dipole onset between April and June, peak in September to October, and then rapidly decay between October and January. The SST anomalies are accompanied by realistic equatorial easterly wind anomalies with thermocline shoaling in the east and deepening in the southwest.

In the model with the 1.25° ocean and 20 vertical levels, the dipoles do not terminate completely but persist through the austral summer and then frequently reinvigorate the following year. This unrealistic behavior is eliminated when the ocean vertical resolution is increased from around 20 m in the thermocline to 10 m in the whole of the top 135 m and when Java is represented (even at 1.25° resolution). It is hypothesized that the improvement is due to the resolution of the separation between the thermocline and the surface and also due to the small reduction of the Maritime Continent warm bias.

1. Introduction

The climate of the Indian Ocean region is determined by a multitude of interactions that arise both within and between different components of the earth system: the atmosphere, ocean, and land surface. Consequently, the accurate simulation in numerical models of Indian Ocean climate is a major challenge. Sea surface temperatures (SSTs) in the Indian Ocean are some of the warmest on the planet and their interannual variability is linked with the climate of the surrounding continents. Indian Ocean SST anomalies (SSTAs) have been linked with Australian summer and winter rainfall (e.g., Joseph et al. 1991; Ansell et al. 2000), with the East African short rains (e.g., Hastenrath et al. 1993; Black et al. 2003), and with southern and East African rainfall during the southern African summer (e.g., Latif et al. 1999; Goddard and Graham 1999) and perhaps farther afield (Saji and Yamagata 2003a).

A key component of Indian Ocean variability is the Indian Ocean dipole, described by Saji et al. (1999) and Webster et al. (1999). The positive phase of the dipole involves cool SSTAs in the southeast tropical Indian Ocean and warm SSTAs farther west. The dipole is initiated during boreal spring [March–May (MAM)] or summer [June–August (JJA)], grows during summer, peaks in autumn [September–November (SON)] and terminates rapidly during the austral summer [December–February (DJF)]. There have been many studies of the mechanisms responsible for dipole variability, but as yet there is no clear consensus as to which processes are most important (e.g., Saji et al. 1999; Webster et al. 1999; Murtugudde et al. 2000; Allan et al. 2001; Baquero-Bernal et al. 2002). Some of the variation in opinion is related to the different models that have been used, some of which are very idealized. This situation highlights the need for a detailed investigation of dipole variability in comprehensive GCMs that attempt to represent all the processes that may be important.

The aim of this study is to evaluate the simulation of Indian Ocean climate and dipole variability in two coupled GCMs: the Third Hadley Centre Coupled Ocean–Atmosphere General Circulation Model (HadCM3) and a related model known as Hadley Centre Coupled Eddy-Permitting Model (HadCEM), which has the same atmospheric component but includes a higher-resolution ocean component. A comparison of the results from HadCM3 and HadCEM is used to assess the impact on the simulation of Indian Ocean climate of enhanced ocean resolution. The seasonal phase locking of the dipole highlights the importance of the evaluation of the simulation of the seasonal cycle. Model errors in the Pacific are not studied in this paper but the impact of ocean resolution on the simulation of Pacific and ENSO variability has been looked at by Schneider et al. (2003) and Roberts et al. (2004).

The paper begins (section 2) with a review of previous work highlighting the different processes that may play a role in dipole variability. The models with different ocean resolutions are described in section 3, together with the datasets that are used to evaluate model performance. In section 4, the model climatologies and key aspects of the interannual variability are assessed. The simulated evolution of the Indian Ocean dipole is analyzed in section 5. Because this study focuses on the Indian Ocean basin, the analysis is restricted to those dipole events that are not associated with strong El Niños. Section 6 discusses the specific impact of vertical resolution on the termination of simulated dipole events and conclusions are presented in section 7.

2. Review of Indian Ocean dipole processes

A convenient index for dipole variability is the anomaly in the zonal SST gradient [10°S–10°N, 50°–70°E, minus 10°S–0°, 90°–110°E; Saji et al. (1999)], referred to here as the dipole index (DI). [This index has previously been called the dipole mode index (DMI) but this can be confused with the dynamic monsoon index.] The positive phase of the dipole during SON is frequently associated with El Niño (e.g., Allan et al. 2001). The correlation of the DI with the Pacific peaks in the Niño-3.4 region in October with a correlation coefficient of just over 0.5, dependent on time period and SST dataset. However, not all dipole events are associated with El Niño. In this study we focus on those dipole events that are not associated with strong El Niño events. The reason is that we wish to focus on the processes occurring within the Indian Ocean basin. Interactions with the Pacific are not studied in this paper since they are influenced by model errors in the Pacific as well as model errors in the Indian Ocean. Figure 1 shows a composite analysis, based on observations and reanalysis, of events that exhibit a large dipole index and a small or negative Niño-3.4 index during SON. The Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) dataset and the National Centers for Environmental Prediction (NCEP) 10-m wind and precipitation anomalies are shown from initiation during MAM to termination during DJF. [The events included in this composite are slightly different to those defined as non–El Niño dipoles by Yamagata et al. (2002), who included 1977 rather than 1963, since HadISST is used rather than the Global Sea Ice and Sea Surface Temperature (GISST) dataset and because the data is temporally smoothed and linearly detrended.]

Fig. 1.

Anomalous SST (HadISST), 10-m wind (NCEP), and precipitation (NCEP) for a composite of Indian Ocean dipole years not accompanied by strong El Niños: Years 1961, 1963, 1967, and 1994. Boxes for dipole index shown for SON SST.

Fig. 1.

Anomalous SST (HadISST), 10-m wind (NCEP), and precipitation (NCEP) for a composite of Indian Ocean dipole years not accompanied by strong El Niños: Years 1961, 1963, 1967, and 1994. Boxes for dipole index shown for SON SST.

The composite shows that prominent cool SSTAs are found in the southeast Indian Ocean, in the region that is used to define the eastern lobe of the dipole index. By contrast, the main warm SSTAs in the west are located southeast of the region that is generally used to define the western lobe of the dipole index. This is because we have excluded strong El Niño years. The basinwide warming associated with El Niño strengthens the warm lobe and shifts it to the west and north onto the equator (not shown). [Saji and Yamagata (2003b) shifted the western box 10° to the east to better capture this nonENSO variability.]

The SSTA dipole is accompanied by collocated precipitation anomalies (south of the equator in this composite) and low-level anticyclones in each hemisphere that can also be seen in the figure. This circulation is similar to the Gill atmospheric response to anomalous tropical heating and, since the strongest SST and precipitation anomalies are south of the equator, the surface wind response is also strongest south of the equator. There are a number of possible processes by which the atmospheric response to these SSTAs could feed back positively, reenforcing the SSTAs, leading to a coupled ocean–atmospheric process. Which of these processes dominates is uncertain; hence we now review the various ideas that have been proposed.

There has been much discussion as to whether the Indian Ocean has internal modes of coupled variability, independent of ENSO (e.g., Saji et al. 1999; Allan et al. 2001), or whether the Indian Ocean is a largely passive integrator of heat, forced by the atmosphere with modes of variability controlled by the Pacific (e.g., Baquero-Bernal et al. 2002). Much of the interannual variability of the Indian Ocean has been reproduced by simplified models that do not account for the full dynamics of the ocean (e.g., Behera et al. 2000; Baquero-Bernal et al. 2002). There is also evidence that Rossby and Kelvin waves and salinity anomalies play an important role in influencing the interannual surface variability (e.g., Webster et al. 1999; Murtugudde et al. 2000; Feng et al. 2001; Rao et al. 2002; Xie et al. 2002; Gualdi et al. 2003; Annamalai et al. 2003). In this case, routine subsurface measurements may be required in order to initialize fully coupled, ocean–atmosphere general circulation models for seasonal predictions—an expensive task. It is not straightforward to identify the relative importance of the physical processes involved from observations alone due to the short observational record, uncertainties in measuring surface heat fluxes, and since subsurface ocean properties are only sparsely measured. Therefore assessing the accuracy of GCMs is crucial and, since results of published work come to different conclusions, some models may not have been tested sufficiently rigorously.

The dominant mechanisms by which the atmospheric anomalies associated with the Indian Ocean dipole positively feed back onto the ocean, maintaining the dipole, are uncertain. A number of possibilities have been proposed and these are outlined.

  1. The wind–evaporation–SST feedback occurs when strengthened southeasterly winds off Sumatra lead to enhanced evaporation and latent cooling of the sea surface. This process was found to be important in studies of observations and models by Behera et al. (1999), Yu and Rienecker (1999), Baquero-Bernal et al. (2002), and Li et al. (2003). The wind–evaporation–SST feedback is only a positive feedback when the climatological winds are southeasterly and the feedback reverses sign when the climatological winds off Sumatra are northwesterly, during the austral summer. This could explain why the dipole terminates rapidly in DJF. However, Fischer et al. (2005) found the wind–evaporation–SST feedback to be negative in their fully coupled ocean–atmosphere GCM since cool SSTs can also lead to less evaporation.

  2. The coastal upwelling (or entrainment) feedback occurs when southeasterly wind anomalies along the coast of Sumatra force upwelling of cooler, deeper water. This process was thought to be important in the observational study by Saji et al. (1999) and its importance was confirmed in a modeling studies by Behera et al. (1999) and Murtugudde et al. (2000). However, it is not so obvious to see how the seasonality of this feedback could determine the seasonality of the dipole.

  3. The Bjerknes, equatorial wind–thermocline–SST feedback is the dominant feedback in the eastern Pacific. This feedback was thought to be important during the dipole of 1997 by Yu and Rienecker (1999) and was found to be the only positive feedback by Fischer et al. (2005) in SON in their fully coupled ocean–atmosphere GCM. However, this model suffers from a shallow bias of the equatorial thermocline in the east due to equatorial wind biases, making thermocline–surface coupling more dominant. Here, the term “Bjerknes feedback” is used to describe an equatorial process, as distinct from the role of Kelvin waves and upwelling along coasts.

  4. Meridional advection of cold, higher-latitude air will lead to surface heat flux anomalies. This feedback will be sensitive to the climatological meridional SST gradient and is therefore also dependent on the season. This feedback will be weak during austral summer since the meridional SST gradient is weak in the summer hemisphere. This positive feedback process has received little attention by those studying the Indian Ocean dipole.

  5. The easterly equatorial wind anomalies will lead to poleward Ekman currents, advecting warm water into the southwest, warming the SST. This feedback is also sensitive to the meridional SST gradient and hence the season.

The influence of oceanic Rossby and Kelvin waves on surface variability in the Indian Ocean is another controversial aspect of the dipole. Webster et al. (1999) proposed that the propagation and reflection of these waves influenced the thermocline depth in the tropical Indian Ocean and was linked with the surface variability during 1997. They proposed that westward propagating, downwelling, off-equatorial Rossby waves contributed to the downwelling of the thermocline in the west, linked to the warm SSTA in the west, while thermocline upwelling was linked to the cool SSTA in the east. The following year, an equatorial, downwelling, Kelvin wave reversed the thermocline gradient following a positive dipole, leading to quasi-biennial behavior, which was reproduced in an ocean modeling studies by Rao et al. (2002) and Feng and Meyers (2003).

In their highly simplified model, Li et al. (2003) found that reflected, downwelling, Kelvin waves lead to the warming in the southeast after the termination of the dipole but that the timing of the termination was due to the reversal of the wind–evaporation–SST feedback. However, both Li et al. (2003) and Loschnigg et al. (2003) found the quasi-biennial behavior of the dipole to be due to interactions with the Asian summer monsoon, rather than with long ocean waves directly.

The interactions between the dipole and the Indonesian Throughflow are currently unknown. Models and observations suggest that the throughflow warms the Indian Ocean and that the throughflow is sometimes weak during El Niño due to reduced sea surface height gradients (e.g., Meyers 1996; Murtugudde et al. 1998; Wajsowicz and Schneider 2001). However the throughflow is also influenced by Indian Ocean winds and was strong during the weak El Niño of 1994 [Meyers (1996): 1994 was a strong Indian Ocean dipole year]. Annamalai et al. (2003) proposed that weak throughflow, leading to southeast Indian Ocean cooling, can trigger a dipole in spring or summer, which in turn strengthens the throughflow. However the throughflow is far from uniform, and Song and Gordon (2004) found that the response of the Indian Ocean is highly sensitive to the depth of the throughflow anomalies, with throughflow intensification at the thermocline level actually leading to cool Indian Ocean SSTs in an ocean-only model. A clear relationship between the variability of the throughflow and the Indian Ocean has not been found in the Hadley Centre coupled models analyzed in this study and is not covered further in this paper. However, this might be a very fruitful area for future research.

3. Models and evaluation data

a. Hadley Centre coupled GCMs

This study initially uses two coupled ocean–atmosphere GCMs: HadCM3 (Gordon et al. 2000) and HadCEM (Roberts et al. 2004). Both models have the same atmospheric component, HadAM3 (Pope et al. 2000), which has preindustrial greenhouse gases, a horizontal resolution of 2.5° latitude by 3.75° longitude, and 19 hybrid vertical levels. The HadCM3 ocean component has a horizontal resolution of 1.25° and 20 vertical levels. The vertical resolution is 10 m in the top 35 m, approximately 20 m at 50-m depth and approximately 50 m at 100-m depth. The coastlines have the resolution of the atmospheric component of the model, 3.75° × 2.5° so that the boundary conditions between the atmosphere and the ocean and land are straightforward. HadCEM has a horizontal resolution of ⅓° and 40 vertical levels. The vertical resolution is 10 m in the top 135 m and hence has better resolution of the thermocline. The coastlines have the ocean resolution of ⅓° and a tiling procedure is employed so that a single atmospheric grid box can be attached to both ocean and land. Bernie et al. (2005) have shown that higher vertical resolution is needed near the surface of the ocean to capture diurnal and intraseasonal variability more accurately, but this is not studied here. The resolutions of both models are summarized in Table 1, which also includes the low-resolution version of HadCEM (HadCEML). This will be described further in section 6.

Table 1.

Resolutions of the coupled models; N is number of levels.

Resolutions of the coupled models; N is number of levels.
Resolutions of the coupled models; N is number of levels.

The model initialization procedures are described in Gordon et al. (2000) and Roberts et al. (2004). All three models are coupled at the beginning of the runs with no prior separate spinup. The oceans are initially at rest with temperature and salinity specified from the Levitus and Boyer (1994) and Levitus et al. (1994) climatologies and realistic atmospheric conditions are imposed. For each model, 100 years of integration is analyzed, beginning from times when the models are no longer drifting.

b. Reference data

The observational, reanalysis, and data assimilations are collectively described here as reference data. The results from the coupled models are compared with the HadISST SST and sea ice dataset from 1948 to 2002 (Rayner et al. 2003), with ocean temperature from the simple ocean data assimilation (SODA) from 1950 to 2000 (Carton et al. 2000), with the Climate Prediction Center’s Merged Analysis of Precipitation (CMAP) blended precipitation estimates from 1979 to 2000 (Xie and Arkin 1996), and with NCEP reanalysis winds, wind stresses, and precipitation anomalies from 1948 to 2002 (Kalnay et al. 1996). Xie et al. (2002) showed that the SODA interannual thermocline variability in the Indian Ocean is realistic by comparing it with expendable bathythermograph (XBT) measurements and they also showed that there are smooth transitions across the XBT lines indicating that the assimilation is not overfitted to observations. SODA actually has lower horizontal and vertical resolution than HadCEM but is unlikely to suffer from the same model errors due to the assimilation of observations and the realistic surface forcing. The errors of HadCM3 and HadCEM have large components driven by the errors of the atmospheric component. The climatologies from SODA shown in this paper have also been compared with Levitus and Boyer (1994) observations (not shown). The accuracy of the NCEP precipitation anomalies for some of the years studied is demonstrated by comparison with CMAP (not shown).

All of the reference datasets are linearly detrended at every grid point with the trend taken from 1950 to 1998.

4. Coupled model climates and variabilities

An accurate simulation of the mean state is important since regional climate anomalies can depend nonlinearly on the climate mean state. Some sources of nonlinearities of climate anomalies are briefly described.

  1. Advection is of course a nonlinear process and wind or current anomalies in the presence of erroneous mean temperature gradients will lead to erroneous temperature anomalies.

  2. The mean depth of the oceanic thermocline is important since, if the thermocline is shallow, the surface temperature will vary substantially as the thermocline depth varies due to upwelling and entrainment. This, then, increases interaction between the atmosphere and the ocean.

  3. The evaporation from the ocean depends on the absolute wind speed (which is nonlinearly related to the vector wind) and nonlinearly on the sea surface temperature via the Clausius–Clapeyron relation. The evaporation influences the surface temperature through latent heat fluxes and influences atmospheric humidity and, hence, precipitation.

a. Annual mean climates

The annual mean climatologies of the SST, precipitation, and the curl of the wind stress, together with the standard deviation of the SST, are shown in Fig. 2 for two reference datasets: HadCM3 and HadCEM. The standard deviation is calculated from seasonal mean anomalies with the seasonal cycle removed. The curl of the wind stress is shown since, off the equator and away from coastal boundaries, it controls the thermocline depth. Both models have a warm SST bias over the Maritime Continent warm pool and both models have equatorial Pacific cold tongues that extend too far west toward this warm pool. These errors are improved slightly with ocean resolution but not completely, suggesting that a large component of these errors is due to errors in the atmospheric component of the model (which is the same for both models). The precipitation of the coupled models, shown in Fig. 2, indicates a wet bias over the ocean of the Maritime Continent with unrealistically sharp land–sea contrasts. The dry islands are consistent with the dry bias of the atmospheric model (Neale and Slingo 2003) and the wet bias over the ocean is consistent with the warm SST bias. The coupled model climate in the Maritime Continent is relevant since the conditions over the Maritime Continent will directly influence the temperature gradient and winds over the Indian Ocean.

Fig. 2.

Observed (HadISST, NCEP, and CMAP) and modeled climatological, annual mean SST, precipitation, and curl of surfacewind stress (τ) and standard deviation (std dev) of SST. The std dev is calculated from the seasonal anomalies from the seasonal cycle.

Fig. 2.

Observed (HadISST, NCEP, and CMAP) and modeled climatological, annual mean SST, precipitation, and curl of surfacewind stress (τ) and standard deviation (std dev) of SST. The std dev is calculated from the seasonal anomalies from the seasonal cycle.

The easterly bias in the tropical Pacific, described by Neale and Slingo (2003), contributes to overactive upwelling of cold water and the cold SST bias in the equatorial Pacific and a positive feedback that amplifies the error. However this cold bias is not solely attributable to the atmospheric errors. It is reduced with the increasing ocean resolution between HadCM3 and HadCEM since HadCEM begins to capture tropical instability waves in the equatorial Pacific that enhance the mixing between the cold equatorial and warmer off equatorial waters (Roberts et al. 2004).

Figure 2 shows that the model Maritime Continent warm bias extends into the east of the Indian Ocean, leading to erroneous equatorial and meridional SST gradients in the east, which will influence (and be influenced by) the climatological winds. These strong climatological SST gradients may prevent reversal of these gradients during Indian Ocean dipole events (as is observed) and may lead to unrealistically strong anomalous temperature advection in the ocean and atmosphere and erroneous surface fluxes.

The standard deviation of the SST, shown in Fig. 2, shows that the modeled equatorial Pacific variability extends too far west. This is because the area of the cold tongue and ocean upwelling extends too far west. This is likely to influence Indian Ocean variability because the atmospheric response to large equatorial Pacific SSTAs is forced from too far west, much closer to the Indian Ocean. In the Pacific and southern Indian Ocean, the high extratropical SST variability is too large and too close to the equator in both models. In the tropical Indian Ocean the observed SST variability is low, but, in the models, the high Southern Hemisphere variability encroaches on the Tropics. The observed variability associated with the Indian Ocean dipole is in the southeast Indian Ocean to the south of Java. There is also a local variability maximum here in the models but it is too intense. The large off-equatorial SST variability of both models could be related to the large horizontal temperature gradients, as described in section 4a.

The climatological wind stress curl, given in Fig. 2, shows more biases that will be relevant to interannual variability. The reanalysis curl has two minima (maxima of negative curl) in the southern tropical Indian Ocean, a broad minimum in the southwest as the southeasterly trade winds turn westward, and a sharper minimum in the southeast associated with the wind curving around the islands of Java and Sumatra. Both models have these two minima combined into one, located in a broad region in the southeast. This cyclonic low-level circulation is consistent with the forcing of the atmosphere from the SST warm bias. The location of the minima of the wind stress curl will lead to errors in the location of upwelling in the Indian Ocean, which influences SST variability.

The annual mean climatologies of the depth of the 24°C isotherm in the ocean (z24) and the ocean temperature at a cross section of the Indian Ocean at 10°S are shown in Fig. 3 for SODA, HadCM3, and HadCEM. The depth of the 24°C isotherm is used as a proxy for the thermocline depth in the Indian Ocean rather than the 20°C isotherm, commonly used in the Pacific, since the Indian Ocean is warmer than other ocean basins. Also, the warm bias in both of the coupled models means that the 24°C isotherm falls within the thermocline in all datasets in the Indian Ocean whereas the 20°C isotherm is below the thermocline in some regions in the models. Despite the basinwide depth bias, it is still useful to compare the spatial patterns of the thermocline topography.

Fig. 3.

SODA and modeled climatological, annual mean depth of 24°C isotherm (z24) and ocean temperature sections at 10°S in theIndian Ocean. Depth contour interval: 10 m; shading on deeper side. Temperature contour interval: 1°C; (≤23°C: dashed, ≥24°C: solid).

Fig. 3.

SODA and modeled climatological, annual mean depth of 24°C isotherm (z24) and ocean temperature sections at 10°S in theIndian Ocean. Depth contour interval: 10 m; shading on deeper side. Temperature contour interval: 1°C; (≤23°C: dashed, ≥24°C: solid).

There are two regions of upwelling in the Indian Ocean that have been identified as being relevant to the interannual variability (e.g., Murtugudde and Busalacchi 1999; Xie et al. 2002). First, there is the region of open ocean upwelling in the southwest, at around 10°S, 60°E and, second, there is the region of occasional upwelling south of Java. The upwelling south of Java is absent in HadCM3 since Java is not represented in this coarse-resolution model.

The SODA thermocline dome in the southwest is collocated with the minima in the wind stress curl seen in Fig. 2. The modeled thermocline domes south of the equator are too far east, coinciding with the incorrect locations of the modeled wind stress curl minima. The influence of the thermocline dome errors on the surface variability errors is not obvious in the standard deviation of SST in Fig. 2 but will be discussed further in sections 5 and 6.

As an example of the structure of the thermocline, the ocean temperature annual climatology through a section of the Indian Ocean at 10°S is shown in the second row of Fig. 3. This latitude is chosen since it is where Rossby waves propagate along the thermocline, which may be involved in the dipole variability (Masumoto and Meyers 1998; Rao et al. 2002; Xie et al. 2002). The thermocline dome in the southwest is shifted eastward in the models. These sections also show that the HadCM3 thermocline is too deep and too diffuse whereas in HadCEM, with enhanced vertical resolution, the thermocline has more realistic gradients and is shallower. The depth and stratification of the thermocline affect the speed and amplitude of the waves that propagate along it and can influence the interactions with the surface.

b. Climatological seasonal cycle

The Indian Ocean dipole is strongly phase locked with the seasonal cycle, so to simulate the dipole accurately the GCM must simulate certain aspects of the seasonal cycle accurately. The Asian and austral monsoons and the associated reversal of the winds in much of the Indian Ocean provide dramatic changes throughout the year in the Indian Ocean. Off the equator the extremes in the seasonal cycle of low-level winds occur during the monsoon seasons of DJF and JJA. The reanalysis and modeled climatological winds for these seasons are shown in Fig. 4. These show that the model errors are small in comparison to the strength of the seasonal cycle and the seasonal reversals. In particular, both models accurately simulate the reversal of the winds along the shores of Java and Sumatra.

Fig. 4.

Observed (NCEP) and modeled climatological wind in the Indian Ocean (nonlinear vector length scale).

Fig. 4.

Observed (NCEP) and modeled climatological wind in the Indian Ocean (nonlinear vector length scale).

Further details of the seasonal cycle are displayed by plotting monthly means and standard deviations of area averages of the zonal SST gradient (the total value of the dipole index), wind stress, and 24°C depth in the locations shown in Fig. 5. The means and standard deviations are shown in Fig. 6. These locations have been chosen because the equatorial wind stress in the central Indian Ocean and the wind stress along Sumatra are related to dipole variability, as has the equatorial thermocline depth in the east (e.g., Murtugudde and Busalacchi 1999; Yu and Rienecker 1999; Webster et al. 1999).

Fig. 5.

Locations for the area averages of SST, 24°C depth, and wind stress.

Fig. 5.

Locations for the area averages of SST, 24°C depth, and wind stress.

Fig. 6.

Reference (NCEP, HadISST, and SODA) and modeled climatological mean and standard deviation for each month of the year of the dipole index, the southeasterly wind stress along the coast of Sumatra (τSE), the zonal equatorial wind stress in the center of the ocean (τeq), and the depth of the equatorial 24°C isotherm in the east Indian Ocean (z24). The locations of the area averages are shown in Fig. 5.

Fig. 6.

Reference (NCEP, HadISST, and SODA) and modeled climatological mean and standard deviation for each month of the year of the dipole index, the southeasterly wind stress along the coast of Sumatra (τSE), the zonal equatorial wind stress in the center of the ocean (τeq), and the depth of the equatorial 24°C isotherm in the east Indian Ocean (z24). The locations of the area averages are shown in Fig. 5.

The climatological mean zonal SST gradient (or dipole index, first row in Fig. 6) is negative throughout most of the year due to the warm pool of the Maritime Continent. The gradient weakens during the transitional seasons of MAM and SON and is even observed to reverse in April. However the reversal is not reproduced by either of the models. This is due to the persistence of the warm bias in the east throughout the seasonal cycle. The observed warmest waters migrate to north of the equator in the east in April, but the modeled warmest waters remain on and south of the equator (not shown). The standard deviation of the dipole index rises between May and October when dipoles occur. Therefore the climatological gradient is prone to reversal in October when the climatological gradient is weak and the variability is high. The reversal of this zonal SST gradient is an important feature of the dipole, for example, for the interaction with East African rainfall (Black et al. 2003). Although the dipole variability is strong in October in HadCM3, the mean dipole index is still strongly negative, so reversal is less likely to occur. The standard deviation of the dipole index falls between October and December, related to the termination of the dipole, discussed further in sections 5 and 6.

The mean and standard deviation of the southeasterly component of the wind along Sumatra are also shown in Fig. 6. The period of increase of dipole variability coincides with the period when the climatological winds along Sumatra are southeasterly and the period of decrease with reversal to northwesterly winds along Sumatra. The models reproduce the reversal related to the austral summer reasonably well although the onset of northwesterlies is slightly early; they are too strong and last for too long, into April.

The mean and standard deviation of the equatorial zonal wind stress between 70° and 90°E are also shown in Fig. 6 (following Saji et al. 1999). Climatological westerlies are present in the reanalysis on the equator during the transitional seasons of MAM and SON, which lead to the Yoshida–Wyrtki jets (e.g., Siedler et al. 2001), and easterlies during the monsoon seasons of JJA and DJF, similar to the seasonal cycle of the dipole index. In SON the modeled westerlies are delayed, especially in HadCEM. This leads to errors in the thermocline depth in the east since these equatorial winds force Kelvin waves in the thermocline that propagate to the east.

The behavior of the thermocline in the east (between 90° and 100°E) in all datasets is similar at 10°S and at the equator. The equatorial thermocline is studied here but the same comments hold for the thermocline at 10°S in the east. The thermocline shoals in the east during March in SODA and in the models just before the equatorial westerlies start. Then the thermocline deepens in May when the climatological zonal winds turn westerly. Between July and November the HadCEM thermocline shoals dramatically, unlike SODA, which is consistent with the persistence of equatorial easterly winds in HadCEM. Despite the large errors in the climatological thermocline depth, the seasonal cycle of the standard deviation is modeled remarkably accurately, peaking between October and December, following the variability at the surface by one month.

To relate the interannual variabilities of these fields, correlations between anomalies for each month of the year are plotted in Fig. 7. The correlation between anomalies of all the variables are high during the active dipole seasons when the mean winds are southeasterly and low (or negative) when they are northwesterly. However, the HadCM3 correlations remain fairly strong and positive throughout the seasonal cycle. In particular, the high correlation between the DI and the 24°C depth throughout the year for HadCM3 implies that the surface does not become decoupled from the thermocline during the austral summer. The causes and consequences of this model error will be discussed further in sections 5 and 6.

Fig. 7.

Observed (NCEP, HadISST, and SODA) and modeled correlations for each month of the year between the southeasterly wind stress along the coast of Sumatra (τSE), the zonal equatorial wind stress in the center of the ocean (τeq), and the depth of the equatorial 24°C isotherm in the east Indian Ocean (z24). The locations of the area averages are shown in Fig. 5.

Fig. 7.

Observed (NCEP, HadISST, and SODA) and modeled correlations for each month of the year between the southeasterly wind stress along the coast of Sumatra (τSE), the zonal equatorial wind stress in the center of the ocean (τeq), and the depth of the equatorial 24°C isotherm in the east Indian Ocean (z24). The locations of the area averages are shown in Fig. 5.

The correlations between the dipole index and the equatorial zonal wind stress anomalies and between these wind stresses and the 24°C depth in the east are measures of the strength of the Bjerknes equatorial feedback. This appears to be slightly stronger in the models than the reference data.

In this section, aspects of the models’ mean climate and interannual variabilities and relationships between variables have been assessed. The couplings between the wind, thermocline, and SST in selected locations have been investigated using correlations between variables and these have been related to the climatological mean and annual cycle. Next, a particular aspect of the interannual variability, the Indian Ocean dipole, is investigated and the processes involved will be related back to the climatologies and variabilities presented in this section.

5. Evolution of the Indian Ocean dipole

a. Dipole composites

Strong El Niño events are excluded from this Indian Ocean dipole composite, as we are primarily interested in Indian Ocean coupled interactions. Interactions with the Pacific are not studied in this paper since they are influenced by model errors in the Pacific as well as model errors in the Indian Ocean. This composite is created based on the dipole index (DI) and the Niño-3.4 index in October. Both indices are smoothed with a 1–2–1 binomial filter by month as the dipole index is contaminated by intraseasonal variability. The reference datasets are linearly detrended at every point, but the model datasets do not have a significant trend over 100 years and so are not detrended. Positive Indian Ocean dipoles are chosen to be in the composite if the 1–2–1 smoothed October dipole index is greater than one standard deviation and if the 1–2–1 smoothed Niño-3.4 index is less than one standard deviation in October.

The 1–2–1 smoothed dipole index defined from HadISST is greater than one standard deviation during seven Octobers since 1948. Three of these were strong El Niños (1972, 1982, and 1997) leaving a composite of four events with weak or neutral Niño-3.4 indices (1961, 1963, 1967, and 1994). These are slightly different to those defined by Yamagata et al. (2002) due to the different SST datasets, smoothing, and detrending. These events and the events included in the model composites are shown in Table 2. This shows that the HadCM3 dipole composites, excluding strong El Niños, consist of nine events during 100 years and the HadCEM composites consist of six events during 100 years. It is also noteworthy from Table 2 that many of the HadCM3 dipole years are consecutive, a feature absent from the strong observed and HadCEM dipoles. This feature will be investigated further. The HadCEML dipoles will be studied in section 6.

Table 2.

Dataset lengths and Indian Ocean dipole years with and without strong El Niños in Oct. (The dipole and El Niño are diagnosed based on the 1–2–1 smoothed dipole index and the Niño-3.4 index and their Oct standard deviations, σ)

Dataset lengths and Indian Ocean dipole years with and without strong El Niños in Oct. (The dipole and El Niño are diagnosed based on the 1–2–1 smoothed dipole index and the Niño-3.4 index and their Oct standard deviations, σ)
Dataset lengths and Indian Ocean dipole years with and without strong El Niños in Oct. (The dipole and El Niño are diagnosed based on the 1–2–1 smoothed dipole index and the Niño-3.4 index and their Oct standard deviations, σ)

b. Surface temperature and wind composites

The SST and 10-m wind composite anomalies for the seasons from MAM before the peak of the composite dipole to MAM after the peak are shown in Fig. 8 for the reference data and both models. Only anomalies significant at the 95% significance level based on a t test are plotted. For the reference data, these are the same plots as those in Fig. 1. The composites suggest that dipole events are triggered by southeasterly wind anomalies south of the equator in MAM in the reference data and HadCEM as described by Annamalai et al. (2003) and Fischer et al. (2005).

Fig. 8.

SST and 10-m wind Indian Ocean dipole composites without strong El Niños. Anomalies significant at 95% are plotted.

Fig. 8.

SST and 10-m wind Indian Ocean dipole composites without strong El Niños. Anomalies significant at 95% are plotted.

In JJA, the dipole is growing fast. All datasets show cool SSTAs in the southeast and warm SSTAs farther west, with a strengthening of the southeasterly winds off Sumatra and an anticyclonic wind anomaly farther west. This signifies the beginning of the positive feedback between the SST and the wind anomalies.

The SST and wind anomalies reach a peak in SON in all datasets. The stronger SSTA signal is in the southeast, locked to the coast of Java and Sumatra. The warm SSTA in the west is strongest south of the equator in the center of the basin, outside the box defining the western node of the dipole as shown in Fig. 6. This is typical of the dipoles that are not accompanied by strong El Niño events. El Niño is associated with warming in the west of the Indian Ocean centered on the equator, so this appears to be part of the Indian Ocean dipole if it is contemporaneous with El Niño. HadCM3 has unrealistic double cold peaks in the southeast and HadCEM is too cold on the equator in JJA and SON.

In both models, especially HadCEM in JJA, there is too much warming in the southwest. This warm anomaly bias is inconsistent with the thermocline depth errors described in section 5c. It could therefore be due to either surface flux errors or advection of erroneously large meridional SST gradients. This is explored further in section 5c.

In DJF, after the peak of the dipole, the observed and HadCEM SSTAs decay to insignificant values and most of the wind anomalies decay also. However, significant anomalies of SST and wind persist through DJF in HadCM3. These SSTAs could force the HadCM3 wind anomalies regardless of the wind direction. However, assuming that the positive feedback from the winds onto the SST requires a southeasterly basic state, these wind anomalies will not force SSTAs in DJF, that is, the positive feedback onto the SST should terminate. Therefore, either the feedback processes are quite different in HadCM3 or the HadCM3 ocean has too much memory. The causes and consequences of this persistence will be discussed further in sections 5c and 6.

In MAM after the dipole, no significant anomalies remain in the reference data or HadCEM. However, the dipole anomalies in HadCM3 still persist and subsequently reinvigorate, as shown by the dipole index time series. The dipole index time series (as defined in Fig. 6) for each event in the dipole composite for each dataset is shown in Fig. 9 for a whole year before and a whole year after the peak of the dipole. The termination in HadCM3 is incomplete, as already shown, and some of the dipoles reinvigorate the following year. The timing of this reinvigoration coincides with the timing of the start of the climatological southeasterly winds in the southeast, in Fig. 6, and the positive feedback from the wind onto the SST under these conditions.

Fig. 9.

Dipole index time series for years with Oct DI greater than one standard deviation and Oct Niño-3.4 index less than one standard deviation; all indices 1–2–1 smoothed. Observed dipoles: 1961, 1963, 1967, and 1994.

Fig. 9.

Dipole index time series for years with Oct DI greater than one standard deviation and Oct Niño-3.4 index less than one standard deviation; all indices 1–2–1 smoothed. Observed dipoles: 1961, 1963, 1967, and 1994.

The incomplete termination and reinvigoration of the dipole in HadCM3 might be related to the ENSO variability in HadCM3, but this appears not to be the case. Examination of the years of the HadCM3 dipoles that reoccur in Table 2 and their co-occurrence with El Niño does not reveal a relationship with El Niño. Other analysis and inspection of the HadCM3 dipole index and indices of El Niño reveal a realistic correlation between the dipole and El Niño (not shown) but does not shed light on the reason for the reoccurrence, which we therefore need to explore further.

First, the reference and modeled processes at the peak of the dipole will be examined in more detail. Then, the evolution of the ocean over longer time scales will be examined and the incomplete termination in HadCM3 will be investigated further.

c. Feedbacks during the dipole peak

During the dipole, the SSTAs are influenced by (and influence) the surface flux anomalies and the ocean dynamics. To study the different processes influencing the dipole in the different datasets, 95% significant anomalies of precipitation, 24°C ocean depth, wind stress, and wind stress curl are shown in Fig. 10 and the evolution of the net downward surface heat fluxes are shown in Fig. 11.

Fig. 10.

Precipitation, 24°C depth in the Indian Ocean (z24), surface wind stress (τ), and wind stress curl (curl τ) at the peak of the Indian Ocean dipole composites without strong El Niños in SON. Anomalies significant at 95% are plotted.

Fig. 10.

Precipitation, 24°C depth in the Indian Ocean (z24), surface wind stress (τ), and wind stress curl (curl τ) at the peak of the Indian Ocean dipole composites without strong El Niños in SON. Anomalies significant at 95% are plotted.

Fig. 11.

Net downward surface heat fluxes for Indian Ocean dipole composites without strong El Niños. Anomalies significant at 95% are plotted.

Fig. 11.

Net downward surface heat fluxes for Indian Ocean dipole composites without strong El Niños. Anomalies significant at 95% are plotted.

The models do not have enhanced precipitation over warm SSTAs in the southwest, likely due to the nonlinear relationship between SST and precipitation and the strong warm bias over the Maritime Continent and, similarly, there are stronger negative precipitation anomalies in the southeast in both models than in NCEP. However, comparisons with CMAP precipitation (Xie and Arkin 1996) during the dipole of 1994 show that, although the locations of the NCEP precipitation anomalies are accurate, the modeled strong intensities in the southeast are more accurate than those of NCEP.

The 24°C depth, wind stress, and wind stress curl composite anomalies for SON at the peak of the dipole are also shown in Fig. 10. Positive depth anomalies indicate downwelling of the thermocline and negative anomalies, upwelling. Equatorial easterly wind anomalies and southeasterly anomalies along the coast of Sumatra are expected to force the upwelling in the east. This upwelling is reasonably well captured by both models and reenforces the cool SSTAs in the east. The upwelling extends southeastward along the coasts of Sumatra and Java in SODA and HadCEM, but Java is not represented in HadCM3 so the strong upwelling does not extend in a narrow band along this coast.

Across the basin, off-equatorial depth anomalies are stronger than equatorial anomalies since Kelvin waves may be acting to remove equatorial thermocline gradients. The downwelling in the center of the basin south of the equator is likely to be forced by the positive wind stress curl anomalies and acts to reenforce the warm SSTAs. The downwelling in SODA is stronger than in the models since the reanalysis wind stress curl anomaly in the southwest is stronger.

In the models, the positive wind stress curl anomalies are confined to the southeast, as are the strong SSTAs. The climatological gradients of SST are also very strong in the east around the Maritime Continent in the models, as shown in Fig. 2. Due to the nonlinear advection of temperature anomalies, the overly strong climatological SST gradients modeled will lead to overactive wind and SSTAs.

The evolution of the net downward surface heat flux anomalies for NCEP and both models are shown in Fig. 11 (e.g., longwave, shortwave, sensible, and latent heat fluxes). Positive fluxes heat the ocean and negative fluxes cool the ocean. The surface fluxes provide a negative feedback on SST according to the NCEP reanalysis, unlike the basinwide Indian Ocean warming during El Niño (e.g., Venzke et al. 2000). This confirms the role of ocean dynamics in the dipole when the dipole is not strongly forced from the Pacific. These strong negative feedbacks on SSTAs also show why Indian Ocean variability is weak and why the dipoles are rare events. Both models have insufficient cooling fluxes in the west, which will contribute to the warm biases modeled in the west. HadCEM even shows a strong warming flux in the southwest in MAM, consistent with the warm SSTs modeled in JJA. HadCM3 has insufficient warming fluxes in the southeast in SON, consistent with the incomplete HadCM3 dipole termination. This is likely to be due to the erroneous meridional SST gradient in this region, as shown in Fig. 2, coupled with the realistic southeasterly wind anomalies.

So far, the models have been found to reproduce some aspects of the thermocline, SST, and wind variability of the dipoles. The downwelling in the southwest is weak, which is likely to be related to the climatological mean state in the Maritime Continent and Indian Ocean. The models incorrectly locate the large (negative) wind stress curl, SST gradients, and wind stress curl variability in the southeast rather than farther west in the Indian Ocean. The other serious error appears to be the persistence of the dipoles in HadCM3 through the austral summer and into the following year when they reinvigorate as the seasonal cycle returns to dipole favorable conditions. This means that there is no quasi-biennial nature of the HadCM3 dipoles. This is found to be consistent with the warm Maritime Continent bias leading to errors in the surface fluxes, but there are also mechanisms in the subsurface ocean variability and mean state that influence the dipole persistence in HadCM3.

d. Ocean temperature composites

The ocean temperature on a cross section of the Indian Ocean through 10°S between depths of 5 and 200 m is plotted in Fig. 12 for the dipole composites of all datasets for SON during the dipole peak and DJF during the termination. The 95% significant anomalies are plotted in color and the contours show the total composite temperature. This latitude intersects both poles of the dipole composite (Fig. 8) and also the unrealistic persistent SSTAs of HadCM3.

Fig. 12.

SODA and modeled total and anomalous composite ocean temperature averaged between 8° and 12°S in the Indian Ocean. The total temperature is shown at contour intervals of 1°C. Contours of 24°C and warmer: solid, 23°C and cooler: dashed-Anomalies significant at 95% are colored.

Fig. 12.

SODA and modeled total and anomalous composite ocean temperature averaged between 8° and 12°S in the Indian Ocean. The total temperature is shown at contour intervals of 1°C. Contours of 24°C and warmer: solid, 23°C and cooler: dashed-Anomalies significant at 95% are colored.

Figure 12 shows a deep, barotropic structure in the ocean temperature anomalies accompanying the surface dipole for all datasets. Between SON and DJF westward propagation is evident due to off-equatorial Rossby waves. Rossby waves in the southern tropical Indian Ocean have been described by Masumoto and Meyers (1998), Rao et al. (2002), and Xie et al. (2002), and the propagation in this composite has a similar speed to that described. In the total temperature contours there are steep zonal gradients between 110° and 120°E in SODA and HadCEM, off Java, and sharp gradients with depth between 20° and 24°C in the thermocline. In HadCM3, Java is not represented and the thermocline is more diffuse, which is likely to be due to the coarser vertical resolution in HadCM3.

In the models, some of the warming in the west is shallow, whereas all of the warm surface anomalies in SODA are associated with warm anomalies at depth. The shallow modeled warm anomalies are therefore to be forced by surface heat fluxes rather than by wind stress curl, which forces the ocean downwelling. This explains why the SSTAs in the southwest in the models are significant, even though the modeled thermocline is deep and hence does not interact much with the surface.

The subsurface temperature anomalies in both the east and west persist beyond the termination of the surface anomalies and into DJF in all datasets. However, in DJF the subsurface anomalies in the east become disconnected with the surface anomalies in SODA, allowing the surface signal to terminate while the subsurface anomalies persist. By contrast, in HadCM3, with the diffuse thermocline, the surface and the subsurface remain connected by anomalies of the same sign. There is no decoupling between the surface and subsurface in HadCM3, so memory of the deep ocean is transmitted to the surface, maintaining the surface anomalies though the period of weak or negative surface feedbacks in DJF. The pattern of temperature anomalies in DJF modeled by HadCEM is between those of SODA and HadCM3. Some cold anomalies do persist at 5-m depth in the east. However, these are weaker and less widespread than in HadCM3 and, in fact, do not extend to the surface as shown in Fig. 8. This is also shown in the correlation between the dipole index and the equatorial 24°C depth in the east in Fig. 7, which remains high throughout the year for HadCM3 but falls sharply between January and April in the reference data and, to a lesser extent, HadCEM.

We have thus shown that subsurface signatures of the dipole persist past the termination of the surface dipole in SODA and in both models. However, in HadCM3, with reduced horizontal and vertical ocean resolution, the surface signal also persists through DJF. The next question to address is how the persistent subsurface signal influences the Indian Ocean the following year and, conversely, how the Indian Ocean conditions the year before a dipole influence the dipole.

Hovmöller diagrams of the 24°C depth anomalies along the equator and along 10°S for the dipole composites in Fig. 13 show the evolution of the subsurface Indian Ocean variability over a longer time scale. The center of these composites is in October year zero, when there are upwelling negative anomalies in the east and downwelling positive anomalies farther west at both latitudes. On the equator there are eastward propagating Kelvin waves and at 10°S there are westward propagating Rossby waves. The Rossby waves of both models and the HadCEM Kelvin waves have realistic speeds. The HadCM3 Kelvin waves are not obviously present in the dipole composite but equatorial anomalies during individual years do show eastward propagation of approximately the correct speed (not shown). The wave propagations in HadCM3 are contaminated by the reoccurrence of the dipole for more than one yr.

Fig. 13.

Hovmöller diagrams of longitude vs time of 24°C isotherm depth dipole composite anomalies in the Indian Ocean. Contour interval: 4 m, negative countours: dashed and start at −2 m. Anomalies significantly less than zero at the 95% confidence level: light shaded, more than zero: dark shaded.

Fig. 13.

Hovmöller diagrams of longitude vs time of 24°C isotherm depth dipole composite anomalies in the Indian Ocean. Contour interval: 4 m, negative countours: dashed and start at −2 m. Anomalies significantly less than zero at the 95% confidence level: light shaded, more than zero: dark shaded.

The downwelling during the dipole peak in SODA is stronger than in the models in Fig. 13, as was also noted for Figs. 10 and 12, due to the larger local wind stress curl anomaly in the reanalysis than modeled. In SODA, the downwelling signal starts in the southeast, a year before the peak of the dipole. It propagates westward and intensifies when it reaches the dome in the southwest, which corresponds with the peak phase of the dipole. Therefore the downwelling in the east during the year before may be a dipole precursor that relies upon propagation to the thermocline dome. This precursor is not evident in either model, which may be because the thermocline dome is misplaced, so interaction between the thermocline and the surface is weaker in the models.

It has been suggested by Webster et al. (1999) that downwelling equatorial Kelvin waves arriving in the east are instrumental in the termination of the surface dipole. However, the comparison of the subsurface with surface composites (Figs. 8) shows that the surface anomalies terminate before the subsurface anomalies. Instead, it appears that these downwelling signals can only propagate from west to east once the positive feedbacks between the SST, surface winds, and thermocline depth have subsided, during the austral summer. This explains the strong seasonality of the termination. Once this positive feedback has subsided, the Kelvin waves from the west act to remove the subsurface anomalies in the east, influencing conditions the following year.

6. The role of ocean vertical resolution in dipole termination

A number of important physical processes have been identified for the simulation of the termination of the dipole. First, the negative feedback from the surface fluxes, which are influenced by the SST gradients. Second, it must be possible for the thermocline and surface anomalies to decouple in DJF in the southeast so that the surface anomalies can terminate while the deeper anomalies persist. This occurs in HadCEM but not HadCM3 and is hypothesized to be due to the enhanced vertical resolution of the thermocline rather than the horizontal resolution or the resolution of the coastlines.

To test the hypothesis concerning the role of vertical resolution in the termination of the dipole, results of a third coupled model are presented, with ocean resolution in between that of HadCM3 and HadCEM. A low-resolution version of HadCEM (HadCEML) was developed and run as part of the development of HadCEM (Roberts et al. 2004). HadCEML has the same 40 vertical levels with 10-m resolution in the top 135 m but 1.25° horizontal resolution. This horizontal resolution is the same as HadCM3 but HadCM3 has 3.75° × 2.5° resolution of the coastlines whereas HadCEML has 1.25° resolution of coastlines and hence a representation of Java. The resolutions are summarized in Table 1. The last 100 years of a 150-yr integration is studied, which contains 14 dipoles (as defined in section 5a), 8 of which are not accompanied by strong El Niños and are hence included in the dipole composite. The dipole years included and not included in the composite are given in Table 2.

The composite anomalies of SST and the 10-m wind are shown in Fig. 14 from JJA before the peak to MAM after the termination of the dipole and of ocean temperature through 10°S from SON to DJF. This shows that the surface dipole terminates completely in DJF in HadCEML (cf. with SSTAs in Fig. 8 for the other datasets). This supports the hypothesis regarding the importance of vertical ocean resolution in the dipole termination. If the upper ocean and thermocline are well resolved, then they can decouple so that the memory of the deep ocean does not influence the surface when the surface feedbacks terminate.

Fig. 14.

HadCEML SST, 10-m wind, and Indian Ocean temperature through 10°S for the Indian Ocean dipole composites. Anomalies significant at 95% are plotted.

Fig. 14.

HadCEML SST, 10-m wind, and Indian Ocean temperature through 10°S for the Indian Ocean dipole composites. Anomalies significant at 95% are plotted.

The thermocline at 10°S, shown in the temperature cross sections in Fig. 14, is well resolved in comparison with HadCM3, shown in Fig. 12. Hence the thermocline and the surface separate in DJF and the surface anomalies terminate. However, the thermocline in the southwest is even deeper than in HadCEM and HadCM3, which means the thermocline–surface coupling will be even weaker.

7. Conclusions

A number of conclusions can be drawn regarding the simulation of the Indian Ocean mean climate and dipole variability, the relationship between climate mean errors, and the variability and the ocean resolution required to model aspects of the dipole.

a. The Indian Ocean mean state

  1. The coupled models capture the broad features and timings of the large seasonal wind reversals that occur in the Indian Ocean region.

  2. The modeled mean climate of the Indian Ocean is dominated by a warm SST bias in the Maritime Continent, leading to overly strong zonal and meridional SST gradients. This bias is reduced only slightly with enhanced ocean resolution.

  3. The warm Maritime Continent SST bias forces overactive cyclonic circulation in the southeast tropical Indian Ocean. Hence the minimum of the wind stress curl is in the southeast rather than the southwest, as observed.

  4. Consequently, the modeled thermocline dome is in the southeast tropical Indian Ocean rather than in the southwest.

  5. To model sufficient ocean upwelling south of Java, the coastline must be represented. This is possible with horizontal resolutions of ⅓° and 1.25° but not with 3.75° × 2.5° resolution.

  6. The thermocline is too diffuse in HadCM3, which has a low vertical resolution of 20 vertical levels and approximately 20-m vertical spacing at a depth of 50 m. This is improved in HadCEML and HadCEM, which have 40 vertical levels and 10-m vertical spacing down to 135 m.

b. The Indian Ocean dipole

  1. The coupled models accurately reproduce (i) dipole variability peaking in October; (ii) the timing and speed of onset; (iii) the main features of the SST, wind, and thermocline behavior during the peak; and (iv) the timing and speed of the termination with the reversal of the climatological winds in the southeast during the austral summer. This degree of model accuracy is also achieved by many coupled models of varying complexity.

  2. All of the models, with all the ocean resolutions, have a long memory of the dipole in the ocean. South of the equator, shoaling of the thermocline underneath the cool SSTA and deepening underneath the warm SSTAs, contemporaneous with the peak of the surface dipole, persists for at least a season after the termination of the surface dipole.

  3. The misplacement of the thermocline dome by the models and the consequent lack of surface–thermocline interaction in the southwest means that, when downwelling anomalies in the southeast propagate to the southwest, they do not trigger dipoles when they reach the southwest. This dipole precursor of deep thermocline anomalies in the southeast the year before a dipole is therefore not present in the models. However, although this dipole precursor is a 95% significant anomaly of a composite of four observed events, this does not prove that the propagation of this anomaly influences the dipole. To prove or disprove this hypothesis, more events need to be studied and modeling experiments performed. Therefore, coupled models must be able to reproduce the wave dynamics associated with the surface dipole more accurately.

  4. The warming in the southwest tropical Indian Ocean is stronger in the models than observed during weak El Niño dipoles. This is despite the lack of interaction between the surface and the thermocline modeled in the southwest. The overly strong warming in the southwest is due to the lack of negative feedback provided by cooling surface heat fluxes present in the reanalysis. The warming flux anomaly errors are consistent with the strong meridional SST gradients modeled.

  5. The most striking inaccuracy of the dipole variability of the model with the low-resolution ocean, HadCM3, is the persistence of the dipoles through the austral summer and into the following year, when they reinvigorate. This error is not present in HadCEML or HadCEM, with enhanced vertical resolution; complete termination of the surface dipole anomalies occurs by the end of the year. The enhanced vertical resolution leads to a better resolution of the thermocline, which permits the surface in the southeast Indian Ocean to become decoupled from the subsurface thermocline anomalies so that the cool surface anomalies terminate while the subsurface anomalies persist. The persistence of negative heat flux anomalies into SON also carries the HadCM3 dipoles into DJF. These heat flux errors may be related to the strong SST gradients in HadCM3 since advection by the atmosphere of erroneous surface air temperature gradients leads to erroneous temperature anomalies (the nonlinearity of advection). The atmospheric temperature anomalies will be communicated to the ocean via the surface heat fluxes.

  6. Although the Indian Ocean dipole is simulated most realistically by HadCEM, with an eddy-permitting ocean, there is no direct evidence that the eddy-scale processes are important for the dipole processes. Much of the improvement in HadCEM is likely to be due to the enhanced vertical and equatorial resolution and the consequent improved climatological SSTs in the Pacific.

In conclusion, vertical ocean resolution is found to be important for the termination of the dipole and a 1.25° resolution of coastlines is found to be important for coastal upwelling associated with the dipole. The mean state influences the spatial patterns of SST and thermocline anomalies.

Acknowledgments

H. Spencer acknowledges NERC’s support through Grant C790200 and R. Sutton acknowledges the support of the Royal Society. H. Spencer, R. Sutton, J. Slingo, and E. Black are members of the NERC Centres for Atmospheric Science (NCAS) and the Centre for Global Atmospheric Modelling (CGAM). M. Roberts was funded by the Department of the Environment, Food, and Rural Affairs, under the Climate Prediction Programme PECD/7/12/37.

REFERENCES

REFERENCES
Allan
,
R.
, and
Coauthors
,
2001
:
Is there an Indian Ocean dipole and is it independent of the El Niño–Southern Oscillation? CLIVAR Exchanges, No. 6, International CLIVAR Project Office, Southampton, United Kingdom, 18–22
.
Annamalai
,
H.
,
R.
Murtugudde
,
J.
Potemra
,
S.
Xie
,
P.
Liu
, and
B.
Wang
,
2003
:
Coupled dynamics over the Indian Ocean: Spring initiation of the zonal mode.
Deep-Sea Res.
,
50
,
2305
2330
.
Ansell
,
T.
,
C.
Reason
, and
G.
Meyers
,
2000
:
Variability in the tropical southeast Indian Ocean and links with southeast Australian winter rainfall.
Geophys. Res. Lett.
,
27
,
3977
3980
.
Baquero-Bernal
,
A.
,
M.
Latif
, and
S.
Legutke
,
2002
:
On dipolelike variability of sea surface temperature in the tropical Indian Ocean.
J. Climate
,
15
,
1358
1368
.
Behera
,
S K.
,
R.
Krishnan
, and
T.
Yamagata
,
1999
:
Unusual ocean–atmosphere conditions in the tropical Indian Ocean during 1994.
Geophys. Res. Lett.
,
26
,
3001
3004
.
Behera
,
S K.
,
P.
Salvekar
, and
T.
Yamagata
,
2000
:
Simulation of interannual SST variability in the tropical Indian Ocean.
J. Climate
,
13
,
3487
3499
.
Bernie
,
D J.
,
S. J.
Woolnough
,
J. M.
Slingo
, and
E.
Guilyardi
,
2005
:
Modeling diurnal and intraseasonal variability of the ocean mixed layer.
J. Climate
,
18
,
1190
1202
.
Black
,
E.
,
J.
Slingo
, and
K.
Sperber
,
2003
:
An observational study of the relationship between excessively strong short rains in coastal East Africa and Indian Ocean SST.
Mon. Wea. Rev.
,
131
,
74
94
.
Carton
,
J A.
,
G.
Chepurin
,
X.
Cao
, and
B.
Giese
,
2000
:
A simple ocean data assimilation analysis of the global upper ocean 1950–95. Part I: Methodology.
J. Phys. Oceanogr.
,
30
,
294
309
.
Feng
,
M.
, and
G.
Meyers
,
2003
:
Interannual variability in the tropical Indian Ocean: A two-year time scale of IOD.
Deep-Sea Res.
,
50
,
2263
2284
.
Feng
,
M.
,
G.
Meyers
, and
S.
Wijffels
,
2001
:
Interannual upper ocean variability in the tropical Indian Ocean.
Geophys. Res. Lett.
,
28
,
4151
4154
.
Fischer
,
A S.
,
P.
Terray
,
E.
Guilyardi
,
S.
Gualdi
, and
P.
Delecluse
,
2005
:
Two independent triggers for the Indian Ocean dipole/zonal mode in a coupled GCM.
J. Climate
,
in press
.
Goddard
,
L.
, and
N.
Graham
,
1999
:
Importance of the Indian Ocean for simulating rainfall anomalies over eastern and southern Africa.
J. Geophys. Res.
,
104
,
D16,
.
19099
19116
.
Gordon
,
C.
,
C.
Cooper
,
C.
Senior
,
H.
Banks
,
J.
Gregory
,
T.
Johns
,
J.
Mitchell
, and
R.
Wood
,
2000
:
The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments.
Climate Dyn.
,
16
,
2–3,
.
147
168
.
Gualdi
,
S.
,
E.
Guilyardi
,
A.
Navarra
,
S.
Masina
, and
P.
Delecluse
,
2003
:
The interannual variability in the tropical Indian Ocean as simulated by a CGCM.
Climate Dyn.
,
20
,
567
582
.
Hastenrath
,
S.
,
A.
Nicklis
, and
L.
Greischar
,
1993
:
Atmospheric–hydrospheric mechanisms of climate anomalies in the western equatorial Indian Ocean.
J. Geophys. Res.
,
98
,
C11,
.
20219
20235
.
Joseph
,
P V.
,
B.
Liebmann
, and
H.
Hendon
,
1991
:
Interannual variability of the Australian summer monsoon onset: Possible influence of Indian summer monsoon and El Niño.
J. Climate
,
4
,
529
538
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
.
Latif
,
M.
,
D.
Dommenget
,
M.
Dima
, and
A.
Grötzner
,
1999
:
The role of Indian Ocean sea surface temperature in forcing East African rainfall anomalies during December–January 1997/98.
J. Climate
,
12
,
3497
3504
.
Levitus
,
S.
, and
T P.
Boyer
,
1994
:
Temperature. Vol. 4, World Ocean Atlas 1994, NOAA Atlas NESDIS 4, 117 pp
.
Levitus
,
S.
,
R.
Burgett
, and
T P.
Boyer
,
1994
:
Salinity. Vol. 3, World Ocean Atlas 1994, NOAA Atlas NESDIS 3, 99 pp
.
Li
,
T.
,
B.
Wang
,
C.
Chang
, and
Y.
Zhang
,
2003
:
A theory for the Indian Ocean dipole-zonal mode.
J. Atmos. Sci.
,
60
,
2119
2135
.
Loschnigg
,
J.
,
G.
Meehl
,
P.
Webster
,
J.
Arblaster
, and
G.
Compo
,
2003
:
The Asian monsoon, the tropospheric biennial oscillation, and the Indian Ocean zonal mode in the NCAR CSM.
J. Climate
,
16
,
1617
1642
.
Masumoto
,
Y.
, and
G.
Meyers
,
1998
:
Forced Rossby waves in the southern tropical Indian Ocean.
J. Geophys. Res.
,
103
,
C12,
.
27589
27602
.
Meyers
,
G.
,
1996
:
Variation of Indonesian throughflow and the El Niño–Southern Oscillation.
J. Geophys. Res.
,
101
,
C5,
.
12255
12263
.
Murtugudde
,
R.
, and
A.
Busalacchi
,
1999
:
Interannual variability of the dynamics and thermodynamics of the tropical Indian Ocean.
J. Climate
,
12
,
2300
2326
.
Murtugudde
,
R.
,
A.
Busalacchi
, and
J.
Beauchamp
,
1998
:
Seasonal-to-interannual effects of the Indonesian throughflow on the tropical Indo–Pacific Basin.
J. Geophys. Res.
,
103
,
C10,
.
21425
21441
.
Murtugudde
,
R.
,
J.
McCreary
, and
A.
Busalacchi
,
2000
:
Oceanic processes associated with anomalous events in the Indian Ocean with relevance to 1997–1998.
J. Geophys. Res.
,
105
,
C2,
.
3295
3306
.
Neale
,
R B.
, and
J M.
Slingo
,
2003
:
The Maritime Continent and its role in the global climate: A GCM study.
J. Climate
,
16
,
834
848
.
Pope
,
V D.
,
M.
Gallani
,
P.
Rowntree
, and
R.
Stratton
,
2000
:
The impact of new physical parametrizations in the Hadley Centre climate model: HadAM3.
Climate Dyn.
,
16
,
2–3,
.
123
146
.
Rao
,
S A.
,
S.
Behera
,
Y.
Masumoto
, and
T.
Yamagata
,
2002
:
Interannual subsurface variability in the tropical Indian Ocean with a special emphasis on the Indian Ocean Dipole.
Deep-Sea Res.
,
49
,
1549
1572
.
Rayner
,
N A.
,
D.
Parker
,
E.
Horton
,
C.
Folland
,
L.
Alexander
,
D.
Rowell
,
E.
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
.
Roberts
,
M J.
, and
Coauthors
,
2004
:
Impact of an eddy-permitting ocean resolution on control and climate change simulations with a global coupled GCM.
J. Climate
,
17
,
3
20
.
Saji
,
N H.
, and
T.
Yamagata
,
2003a
:
Possible impacts of Indian Ocean Dipole mode events on global climate.
Climate Res.
,
25
,
151
169
.
Saji
,
N H.
, and
T.
Yamagata
,
2003b
:
Structure of SST and surface wind variability during Indian Ocean dipole mode events: COADS observations.
J. Climate
,
16
,
2735
2751
.
Saji
,
N H.
,
B.
Goswami
,
P.
Vinayachandran
, and
T.
Yamagata
,
1999
:
A dipole mode in the tropical Indian Ocean.
Nature
,
401
,
360
363
.
Schneider
,
E K.
,
D.
DeWitt
,
A.
Rosati
,
B.
Kirtman
,
L.
Ji
, and
J.
Tribbia
,
2003
:
Retrospective ENSO forecasts: Sensitivity to atmospheric model and ocean resolution.
Mon. Wea. Rev.
,
131
,
3038
3060
.
Siedler
,
G.
,
J.
Church
, and
J.
Gould
, and
Eds.
,
2001
:
Ocean Circulation and Climate: Observing and Modelling the Global Ocean. International Geophysics Series, Vol. 77, Academic Press, 715 pp
.
Song
,
Q.
, and
A L.
Gordon
,
2004
:
Significance of the vertical profile of the Indonesian Throughflow transport to the Indian Ocean.
Geophys. Res. Lett.
,
31
.
L16307, doi:10.1029/2004GL020360
.
Venzke
,
S.
,
M.
Latif
, and
A.
Villwock
,
2000
:
The coupled GCM ECHO-2. Part II: Indian Ocean response to ENSO.
J. Climate
,
13
,
1371
1383
.
Wajsowicz
,
R C.
, and
E.
Schneider
,
2001
:
The Indonesian Throughflow’s effect on global climate determined from the COLA coupled climate system.
J. Climate
,
14
,
3029
3042
.
Webster
,
P J.
,
A.
Moore
,
J.
Loschnigg
, and
R.
Leben
,
1999
:
Coupled ocean–atmosphere dynamics in the Indian Ocean during 1997–98.
Nature
,
401
,
356
360
.
Xie
,
P P.
, and
P.
Arkin
,
1996
:
Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions.
J. Climate
,
9
,
840
858
.
Xie
,
S. P.
,
H.
Annamalai
,
F A.
Schott
, and
J P.
McCreary
,
2002
:
Structure and mechanisms of South Indian Ocean climate variability.
J. Climate
,
15
,
864
878
.
Yamagata
,
T.
,
S.
Behera
,
S.
Rao
,
Z.
Guan
,
K.
Ashok
, and
H.
Saji
,
2002
:
The Indian Ocean dipole: A physical entity. CLIVAR Exchanges, No. 24, International CLIVAR Project Office, Southampton, United Kingdom, 15–18
.
Yu
,
L S.
, and
M.
Rienecker
,
1999
:
Mechanisms for the Indian Ocean warming during the 1997–98 El Niño.
Geophys. Res. Lett.
,
26
,
735
738
.

Footnotes

Corresponding author address: Dr. Hilary Spencer, CGAM, Department of Meteorology, University of Reading, Earley Gate, Reading, Berkshire RG6 6BB, United Kingdom. Email: h.spencer@reading.ac.uk