Madden-Julian Oscillation

2.1. Intraseasonal Period

[9] The dominant period of the MJO spreads over a range of roughly 30–100 days (Figure 3). Its power peak is highly variable within this range. These reflect a fundamental but often neglected nature of the MJO. Although referred to as an “oscillation,” the MJO by no means oscillates regularly. It is highly episodic or discrete [Salby and Hendon, 1994]. The range of its local period (30–100 days) suggests the interval between two consecutive events is irregular and their propagation speeds may vary (4). The MJO undergoes a seasonal cycle (9) and interannual fluctuations (10) with its occurrence frequency higher in certain times than others. All these result in a spread of its spectral peak.

2.2. Planetary Zonal Scale

[10] The typical zonal extent of an MJO event, measured by its regions of positive and negative anomalies in cloud covers, is roughly 12,000–20,000 km (see the composite of Rui and Wang [1990]). Only one fully developed MJO event exists in the tropics at a given time. Occasionally, two weak convective centers of the MJO with weak circulations may coexist, one being just initiated in the Indian Ocean and the other decaying in the central Pacific [e.g., Wheeler and Hendon, 2004]. The zonal scale of the convective component is much less than that of the circulation because of the nature of atmospheric response to localized heating [Salby et al., 1994]. The spectral peak is thus at zonal wave number 1 for the zonal wind but spreads over zonal wave numbers 1–3 for precipitation (Figure 3). Meanwhile, the zonal extent of active phase of the MJO is much smaller than that of inactive phase. The MJO is therefore more an isolated or discrete pulse-like event than a sinusoidal wave [Salby and Hendon, 1994; Yano et al., 2004].

2.3. Eastward Propagation

[11] The slow eastward propagation at an averaged speed of 5 m s⁻¹ [e.g., Weickmann et al., 1985; Knutson et al., 1986] is one of the most fundamental features that distinguishes the MJO from other phenomena in the tropical atmosphere, especially convectively coupled Kelvin waves, which propagate eastward at greater speeds of 15–17 m s⁻¹ [e.g., Wheeler and Kiladis, 1999; Straub and Kiladis, 2002] (Figure 2b). The phase speed of the MJO varies slightly among individual events and during different stages of the life cycle of a given event [Hendon and Salby, 1994] (Figure 2). While the convective signals of the MJO normally vanish in the eastern Pacific (Figure 2b), its signals in wind and surface pressure continue to propagate farther east as free (uncoupled with convection) waves at much higher speeds of about 30–35 m s⁻¹ [Milliff and Madden, 1996; Matthews, 2000] (Figure 4). A continuous, global circumferential propagation of the MJO along the equator exists only in its upper level fields [e.g., Krishnamurti et al., 1985; Knutson and Weickmann, 1987] as atmospheric responses to the convective perturbations. Issues related to the stationary intraseasonal oscillation in the tropics are discussed in 12.

2.4. Convection-Wind Coupling

[12] The large-scale wind structure (Figure 1) is often described in terms of equatorial waves coupled to deep convection. East of the convective center, the low-level easterlies and upper level westerlies resemble the equatorial Kelvin wave. To the west, low-level westerlies (upper level easterlies) and the associated pair of cyclonic (anticyclonic) circulation or gyres straddling the equator are the characteristics of the equatorial Rossby wave [Madden, 1986; Nogués-Paegle et al., 1989] (Figure 5). Both Kelvin and Rossby wave structures have been considered dynamically essential to the MJO (16).

[13] Observations have revealed that the relative phase between the large-scale surface zonal wind and convective center varies during the life cycle of the MJO [e.g., Knutson and Weickmann, 1987; Rui and Wang, 1990; Hendon and Salby, 1994; Sperber, 2003]. When the MJO is in the Indian Ocean, its convective center is more likely to be in between the surface westerlies to the west and easterlies to the east, as depicted in Figure 1, which is synthesized as MJO model I in Figure 6. When the MJO moves to the Pacific, surface westerlies are likely to prevail through the convective center (Figure 4). This structure is referred to as MJO model II in Figure 6. In reality, the structure of an MJO event may be of either model I or II or anything in between.

2.5. Large-Scale Vertical Structure

[14] Many fundamental large-scale features of the MJO are not included in the MJO schematics in Figures 1, 5, and 6. Water vapor, temperature, divergence, and diabatic heating all show large-scale patterns coherent with the wind and deep convection of the MJO [e.g., Chen and Yen, 1991; Hendon and Salby, 1994; Lin and Johnson, 1996a, 1996b; Mote et al., 2000; Kemball-Cook and Weare, 2001; Myers and Waliser, 2003; Weare, 2003; Lin et al., 2004]. These fields, as well as the winds, also show marked zonal asymmetry and westward tilt in the vertical with respect to the convective center [Sperber, 2003; Kiladis et al., 2005] (Figure 7). Immediately ahead (to the east) of the convective center are low-level convergence, ascending motions, and positive anomalies in humidity; low-level divergence, descending motions, and negative anomalies in humidity occur immediately behind (to the west). Such zonal asymmetry provides favorable large-scale conditions for the development of new convective systems east of the existing ones and discourages such development to the west, resulting in the eastward propagation of the convective center. Some of these features have been taken into consideration in MJO theories and hypotheses (12). However, it is not clear whether the zonal asymmetry is due to the large-scale circulation (Kelvin versus Rossby waves) or embedded mesoscale systems (7 and 22).

2.6. Multiscale Structure

[15] The eastward moving convective center or active phase of the MJO can be viewed as a large-scale ensemble of myriad higher-frequency, small-scale convective systems moving in all directions [e.g., Nakazawa, 1988; Hendon and Liebmann, 1994] (Figure 8). The apparent eastward propagation of the large-scale convective center of the MJO is due to consecutive development of new convective systems, each on average slightly to the east of the previous one. Among others the most noticeable high-frequency variability within the large-scale ensemble of cloud clusters are eastward propagating synoptic-scale disturbances [Dunkerton and Crum, 1995] at the speed of the coupled Kelvin wave, which has been referred to as the supercluster or supercloud cluster [Nakazawa, 1988; Lau et al., 1989], and westward moving disturbances with apparent periods of 2 and 5 days. The 2-day disturbances are likely manifestations of the diurnal cycle of deep convection [Chen and Houze, 1997] and inertiogravity waves [Takayabu, 1994; Liebmann et al., 1997; Haertel and Kiladis, 2004]. The westward propagating 5-day waves might be related to equatorial Rossby waves and mixed Rossby-gravity waves [Wheeler and Kiladis, 1999]. In addition, other types of disturbances, such as easterly waves and tropical storms, also propagate westward. However, all can exist independently of the MJO [Clayson et al., 2002] (Figures 2b and 8).

[16] The large-scale enhancement of convection during active phases of the MJO is mostly a manifestation of an increase in the occurrence of large (>90,000 km²) and deep (cloud top temperature <208 K) cloud clusters [Mapes and Houze, 1993; Chen et al., 1996]. During inactive phases, there are plenty of deep but small (<3000 km²) convective systems and far fewer deep and large ones. Convective clouds whose tops reach only to the middle troposphere because of an inversion at the melting level are also common during inactive phases [Johnson et al., 1999].

[17] The diurnal cycle in deep convection is modulated by the MJO. Over the Maritime Continent the diurnal cycle is the strongest during the inactive phase of the MJO and becomes diminished during the active phase [Sui and Lau, 1992], possibly because of the interruption of the local sea breeze circulation by the large-scale circulation of the MJO (see also 8). Over the open ocean, deep convective signals exhibit a nocturnal peak during active phases of the MJO when long-lived, organized mesoscale systems prevail; during inactive phases the cloud population is dominated by isolated, short-lived small convective systems that tend to peak in the afternoon [Chen and Houze, 1997].

[18] The low-level westerly wind component of the MJO is composed of synoptic-scale westerly wind bursts (WWB) [Luther et al., 1983]. The WWB might be stationary [Hartten, 1996; Harrison and Vecchi, 1997], but within an MJO envelope each on average tends to occur slightly east of the previous one (August–September and November–December 2000 and February–March 2001 in Figure 2a). See 34 for more discussion on WWB.

2.7. Geographical Preference

[19] MJO signals in convection are normally confined to the Indian and western Pacific oceans (Figures 2, 4, and 9), because convective instability can be sustained only over the warm sea surface known as the “warm pool.” MJO signals in some other fields can be detected in the rest of the tropics [e.g., Knutson and Weickmann, 1987; Chen and Chen, 1997; Sperber, 2003]. The effect of the warm sea surface in determining the location of the MJO can be further illustrated by two examples. One is the zonal displacement of the MJO in concert with ENSO (10). The other is the MJO signal in the eastern Pacific north of the equatorial cold tongue and adjacent to the Central American coast, which is nontrivial only in boreal summer when the sea surface temperature there is sufficiently high [Maloney and Kiehl, 2002] (Figures 9c and 9d).

[20] The convective component of the MJO over the Maritime Continent is generally much weaker than over the surrounding oceans (Figure 9). Possible explanations [Salby and Hendon, 1994; Wang and Li, 1994; Zhang and Hendon, 1997; Maloney and Sobel, 2004] for this are the following: (1) The strong diurnal cycle in convection due to diurnal heating over land tends to compete with the MJO for moisture and energy. (2) Topography interferes with low-level moisture convergence believed to be crucial to the MJO (16). (3) Surface evaporation, another possible crucial factor for the MJO (17), is severely reduced over land. These possible explanations may also be applied to tropical South America, where deep convection in local summer is almost as strong as in the western Pacific but MJO signals are intriguingly weak.

2.8. Seasonal Cycle

[21] The MJO undergoes a strong seasonal cycle in both its strength and latitudinal locations [Madden, 1986; Gutzler and Madden, 1989; Salby and Hendon, 1994; Zhang and Dong, 2004], as illustrated in Figures 9 and 10. Its primary peak season is austral summer/fall when the strongest MJO signals are immediately south of the equator; the second peak season is boreal summer when its strongest signals are north of the equator. The primary peak season in austral summer is related to the Australian summer monsoon [Hendon and Liebmann, 1990], while the secondary peak season in boreal summer is related to the Asian summer monsoon [e.g., Lawrence and Webster, 2002]. The seasonal migration of the MJO in latitude is much stronger in the western Pacific Ocean than in the Indian Ocean [Zhang and Dong, 2004]. In a narrow latitudinal band (e.g., 5°N–5°S) the seasonal cycle of the MJO shows a single peak in austral summer/fall. In the eastern Pacific a separated MJO peak exists only in boreal summer [Kayano and Kousky, 1998; Maloney and Kiehl, 2002].

2.9. Interannual Variability

[22] In the Pacific, interannual variability in zonal wind variance of the MJO is more prominent in the lower than upper troposphere [Gutzler, 1991]. During a warm event of ENSO (El Niño), as the eastern edge of the warm pool extends eastward [Picaut et al., 1996], so does MJO activity [e.g., Anyamba and Weare, 1995; Fink and Speth, 1997; Gualdi et al., 1999; Hendon et al., 1999; Woolnough et al., 2000; Bergman et al., 2001]. The MJO in the Pacific appears to be extraordinarily vigorous prior to the peak of an ENSO warm event and anomalously weak after the peak and during a cold event [Zhang and Gottschalck, 2002]. Simultaneous relationship between the level of global MJO activity and sea surface temperature (SST) indices representing ENSO has been found to be very weak [Slingo et al., 1999; Hendon et al., 1999]. This suggests that globally the interannual variability of the MJO might be driven more by the atmospheric internal dynamics than surface conditions. Possible connections between the MJO and ENSO are further discussed in 40.

3.1. Atmospheric Response to Independent Forcing

3.2. Atmospheric Instability

[28] Instability theories of the MJO seek mechanisms for growing modes that bear characteristics similar to the observed. Nonlinear advection has been shown to be nonessential to the MJO [Van Tuyl, 1987], permitting analytical solutions for linear unstable modes. Sources of the instability inevitably involve deep convective processes, which are highly nonlinear and demand simplifications in their mathematical representations to a degree that solutions are dictated by specific assumptions. A common malady of an instability theory is that its maximum growth rate occurs at smallest scales (the “instability catastrophe” [Crum and Dunkerton, 1992]). Special tricks are needed to remedy this. Among the popular ones are positive-only heating [Lau and Peng, 1987; Wang and Xue, 1992] and time lags between the energy input (from surface evaporation or moist convergence) and convective heating [e.g., Goswami and Rao, 1994]. The justification of positive-only heating is that precipitation occurs solely in regions of large-scale ascents, which is inconsistent with observed precipitation during inactive phases of the MJO (Figure 2b). The numerical value of the time lag cannot always be chosen independently of desired results.

3.2.1. Moisture Convergence

[29] On the basis of the premise of CISK [Charney and Eliassen, 1964; Ooyama, 1964] and wave-CISK [Lindzen, 1974] the equatorial Kelvin wave becomes unstable when its convective heating interacts with its low-level convergence in “mobile wave-CISK” [Lau and Peng, 1987] or “Kelvin wave-CISK” [Chang and Lim, 1988] theories. Without additional assumptions (e.g., positive-only heating), unstable wave-CISK Kelvin modes propagate at speeds (16–19 m s⁻¹) comparable to the observed speed of the convectively coupled Kelvin waves, not the MJO. The growth rates are the greatest on smallest scales. The direct dependence of convective heating on moisture convergence in the wave-CISK type of theories has been criticized as unphysical [Emanuel et al., 1994; Raymond, 1994]. In global models the wave-CISK mechanism alone appears to be insufficient to generate MJO-like signals [Hayashi and Golder, 1997].

[30] The instability catastrophe of wave-CISK theories has been treated by including the frictional effect on moisture convergence in the atmospheric boundary layer in a “moist Kelvin wave” or “frictional-convergence” theory of the MJO [Wang, 1988a; Salby et al., 1994]. In this theory, frictional convergence in the boundary layer leads wave convergence in the lower troposphere and damps small-scale wave-CISK modes. The instability of moisture convergence due to boundary layer viscosity results in slowly eastward moving, planetary-scale, unstable modes in a parameter regime stable to inviscid wave-CISK. The instability depends crucially on the vertical distribution of moist static energy of the basic state.

[31] When the equatorial Rossby wave is included, boundary layer dynamics can generate an unstable mode consisting of coupled Rossby-Kelvin waves [Wang and Rui, 1990a; Wang and Li, 1994] in contrast to other wave-CISK theories where the Rossby waves are stable [e.g., Chang and Lim, 1988]. The inclusion of the Rossby wave helps suppress unrealistic fast growth of pure wave-CISK Kelvin mode (Figure 11). The resulting eastward propagating speed comes from a group velocity of a multiscale structure (7) that is slower than the propagation speed of the unstable, convectively coupled Kelvin wave [also see Chao, 1987]. The boundary layer convergence leading (east of) the convective center of the MJO, a key feature of this theory, has been confirmed by global data assimilation products (Figure 7a) [Hendon and Salby, 1994; Jones and Weare, 1996; Maloney and Hartmann, 1998; Kiladis et al., 2005]. The frictional-convergence mechanism has been shown to be sensitive to surface drag but not to cumulus parameterization [Moskowitz and Bretherton, 2000]. It may also be responsible for the initiation of the MJO in the Indian Ocean [Seo and Kim, 2003].

3.3. Other Factors

3.3.4. Scale Interaction

[37] Possible scale interactions involved with the MJO have recently been discussed by Slingo et al. [2003] and Moncrieff [2004]. The observed rich multiscale structure of the MJO (7) raises the question as to whether it represents a dynamical hierarchy essential to the large-scale features of the MJO [Majda and Biello, 2004] or simply a manifestation of large-scale modulation of existing smaller-scale systems [Hendon and Liebmann, 1994]. It is obvious in Figure 2 that the westward moving, 2- to 5-day signals exist in both active and inactive phases of the MJO but are modified by the MJO. In a simple wave-CISK model [Lau et al., 1989] the intraseasonal period of the MJO depends on the time the large-scale atmosphere needs to adjust to finer-scale heating and to make the planetary-scale Kelvin waves dominantly unstable. Of the models that produce MJO-like signals some simulate the multiscale structure [e.g., Hayashi and Sumi, 1986; Itoh, 1989; Wang and Li, 1994]; others do not [e.g., Gustafson and Weare, 2004a, 2004b]. In a two-dimensional (zonal and vertical) large-scale model without the Coriolis force [Chao and Lin, 1994] the eastward moving speed of supercloud clusters embedding westward propagating convective systems is much less than that of convectively coupled Kelvin waves produced in a two-dimensional cloud-resolving model without any multiscale structure [Grabowski and Moncrieff, 2001].

[38] A possible scale interaction may occur between the MJO and the diurnal cycle through the water vapor feedback due to low and midlevel clouds (20). Midlevel clouds in an inactive phase of the MJO [Johnson et al., 1999] may moisten the lower to middle troposphere and create a favorable condition for the upcoming active phase [Inness et al., 2001]. These midlevel clouds undergo a strong diurnal cycle, which may be related to the diurnal cycle in SST [Slingo et al., 2003]. The diurnal cycle could be important to the MJO also through air-sea interaction. The diurnal cycle in surface heating is critical to intraseasonal fluctuations in SST (35) and therefore to the MJO if SST feedback to the MJO is important (30 and 38).

[39] Mesoscale convective systems dominating active phases of the MJO feature stratiform precipitation [Houze, 1989] and midlevel rear inflows [Zipser, 1969]. During active phases of the MJO such a rear inflow comes from the west into mesoscale convective systems in the large-scale convective center of the MJO [Kingsmill and Houze, 1999]. There, it descends because of diabatic cooling due to evaporation and melting of stratiform precipitation, and, when reaching into the boundary layer, it enhances large-scale surface westerlies [Houze et al., 2000; Mechem et al., 2005] (Figure 12). This mesoscale downward momentum transport, supplementary to the large-scale momentum transport by the Rossby gyres [Moncrieff, 2004], is another example of scale interaction associated with the MJO. Whether the scale interactions are essential for the scale selection of the MJO is unknown.

4.1. Realism of Current Simulations

[44] The MJO has been such an elusive modeling trophy that it is hard to resist the temptation of claiming it whenever intraseasonal signals propagating eastward are discerned in numerical simulations. The realism of simulated MJO should, however, be evaluated at the highest possible standard. Otherwise, the risk of misrepresenting the physics and dynamics of the MJO is too high. Only when model simulations are diagnosed following the same procedures as for observations can the simulated MJO be evaluated fairly and model improvement judged objectively. The following progressive procedure, synthesized from previous observational analysis of the MJO, can be used in evaluation of model simulations against observations.

[45] 1. Examine time-space power spectra [Hayashi, 1979]. Intraseasonal and planetary-scale eastward propagating power in convection and zonal wind much larger than the corresponding westward propagating power signifies the existence of the MJO [e.g., Zhang and Hendon, 1997].

[46] 2. Objectively extract MJO signals using methods such as the empirical orthogonal function [e.g., Lau and Chan, 1985] or singular vector decomposition [e.g., Weare, 2003]. The leading modes that are supposed to represent the MJO should be separable from the others based on a priori criteria [e.g., North et al., 1982].

[47] 3. Reconstruct the MJO from the leading modes and examine its primary features, such as the zonal scale, propagation speed, and structure [e.g., Sperber, 2003; Kiladis et al., 2005].

[48] 4. Examine the spatial distribution and seasonal cycle of the MJO [e.g., Zhang and Dong, 2004].

[49] When used to validate simulations against observations, each of the above steps progressively demands a higher degree of realism from the simulated MJO. The first three should be considered the minimum validation. If simulated MJO signals are too weak for some of these steps, for example, steps 2 and 3, to be applicable, other methods can be adapted to analyze only selected data with identifiable signals [e.g., Duffe et al., 2003; Sperber, 2004].

[50] Examples are given in Figures 13 and 14 to illustrate the current stage of MJO simulations. All selected global weather forecast and climate models produce some MJO signals, as evidenced by their eastward propagating power that is greater than the westward propagating power in zonal wind (Figure 13). Even though the intraseasonal peaks are not always distinguishable from the lower-frequency power and the MJO signals in precipitation are much weaker, these power spectra are much more realistic than those produced by many other global models.

[51] A realistic spectrum does not guarantee a realistic structure of the simulated MJO. Figure 14 demonstrates a common problem [e.g., Itoh, 1989; Wang and Li, 1994]: Positive anomalies in precipitation tend to be in regions of low-level easterly anomalies, contrary to the observations (Figures 4 and 5). This simulated structure of the MJO is synthesized as MJO model IV in Figure 6. Few models can produce the MJO with a structure similar to the observed [e.g., Sperber et al., 1997; Inness and Slingo, 2003]. Unrealistic vertical distributions of heating might be a reason. MJO signals in models are often too weak in the Indian Ocean or too strong in the tropical northeastern Pacific [Maloney, 2002; Liess and Bengtsson, 2004]. The seasonal cycle of the simulated MJO often misses the observed latitudinal migration or peaks at a wrong time [Slingo et al., 1996; Liess et al., 2004]. In short, models that produce dominant eastward propagating power with reasonable zonal scales, periods, and phase speeds may still suffer from unrealistic structures, spatial distributions, and seasonal cycles of their simulated MJO.

4.2. Sensitivity to Model Configurations

[52] Simulations of the MJO can be affected by four factors among others.

4.3. Prediction

[59] The ability to simulate the intraseasonal fluctuations associated with the MJO by some models makes it tempting to explore the possibility of extending numerical weather prediction in the tropics beyond the known predictability limit for synoptic-scale systems (7–14 days). Investigations on this subject [Lau and Chang, 1992; Jones et al., 2000; Hendon et al., 2000; Waliser et al., 2003a; Jones et al., 2004] have showed that useful prediction skill for filtered intraseasonal perturbations can last up to 15–20 days, provided models can reproduce the MJO reasonably well. Better prediction skill is likely to be obtained when an MJO event is in its active phase at the initial time and remains strong during the forecasting period. The extended 15- to 20-day prediction skill is, however, only a quarter of the life cycle of the MJO. In comparison, weather prediction at midlatitudes is limited by the full life cycle (up to 14 days) of the dominant extratropical weather systems. Meanwhile, dynamic models do not outperform statistical models in MJO forecasts [Waliser et al., 1999a; Lo and Hendon, 2000]. The prediction limit demonstrated by these investigations therefore illustrates more the limitation of the numerical model's capability of simulating the MJO than the predictability of the MJO itself. Without further improvement of model simulations of the MJO, empirical methods are more feasible for practical intraseasonal prediction in the tropics [e.g., Wheeler and Weickmann, 2001; Newman et al., 2003].

5.1. MJO Forcing

[61] The MJO perturbs the upper ocean through surface fluxes of momentum, latent and sensible heat, radiation, and fresh water [e.g., Krishnamurti et al., 1988; Zhang, 1996; Hendon and Glick, 1997; Lau and Sui, 1997; Jones et al., 1998]. The latter three in combination make up buoyancy flux. A schematic diagram in Figure 15, based on in situ observations in the equatorial western Pacific, illustrates different forcing components during the two phases of the MJO. Amplitudes of these fluxes in the Indian Ocean can be different [Shinoda et al., 1998]. Surface fluxes from global reanalyses suffer from large biases [e.g., Shinoda et al., 1999].

[62] Perturbations in the momentum flux (0.02–0.055 N m⁻²) are dominated by the zonal wind, whose intraseasonal amplitudes are 5–10 m s⁻¹ in daily mean observations [McPhaden et al., 1992] and 2–5 m s⁻¹ in MJO composites [Zhang and McPhaden, 2000]. The intraseasonal amplitude of surface meridional wind is negligibly weak [Zhang, 1996]. The net freshwater flux into the ocean (precipitation minus evaporation or P − E) is mainly controlled by precipitation. Strong evaporation in convective centers of the MJO compensates only slightly the freshwater input of precipitation.

[63] Perturbations in solar radiation flux (controlled by cloudiness) and latent heat flux (mainly controlled by surface winds) have similar amplitudes (25–30 W m⁻²). Intraseasonal perturbations in infrared (longwave) radiation and sensible heat fluxes are smaller than measurement uncertainties (≤5 W m⁻²). The intraseasonal amplitude of the net heat flux, composed mainly of the radiation and latent heat fluxes, depends on the relative phase among different components of the MJO [Zhang and Anderson, 2003]. It is the largest (50–60 W m⁻²) when the cooling maxima due to enhanced latent heat flux and reduced solar radiation flux collocate in convective centers of the MJO (model II in Figure 6).

[64] Buoyancy flux into the ocean increases with surface warming by solar radiation and freshening by precipitation. It decreases with cooling by latent heat flux and nighttime infrared radiation and salinizing by evaporation. If during an active phase of the MJO the precipitation center and maximum surface cooling of the MJO are collocated (model II in Figure 6), then the contribution to total buoyancy flux by the fresh water (2.5 × 10⁻⁶ kg m⁻² s⁻¹) partially cancels that by heat fluxes (−5 × 10⁻⁶ kg m⁻² s⁻¹). The net buoyancy flux (−2.5 × 10⁻⁶ kg m⁻² s⁻¹) reduces the stability of the ocean mixed layer [Zhang and McPhaden, 2000]. Fluctuations in these surface fluxes in combination constitute the MJO forcing, which propagates eastward along with the MJO from the Indian Ocean to the western Pacific Ocean (Figure 16), leaving a trail of a disturbed ocean.

5.2. Sea Responses

[65] Oceanic responses to MJO forcing can be categorized into two types: mixed layer responses and wave responses. Mixed layer responses in currents, salinity, and temperature profiles in the upper ocean, forced mainly by the perturbations in local surface fluxes, are governed by mixed layer dynamics and energy balance. They are confined where MJO forcing is strong (i.e., the Indian Ocean and western Pacific Ocean). Responses of the oceanic Kelvin wave are forced by the perturbations in surface momentum flux and governed by large-scale geophysical fluid dynamics. They propagate from the western Pacific to the eastern Pacific where MJO forcing is absent. The two are not completely separated in the western and central Pacific.

5.2.1. Mixed Layer Responses

[66] One of the most significant footprints of the MJO in the oceanic mixed layer is an intraseasonal fluctuation in its temperature. As a direct consequence of the surface cooling in convective centers of the MJO and warming outside (Figure 15), fluctuations in SST propagate eastward in tandem with the MJO; their phases are in quadrature, with the maximum SST leading convective centers of the MJO (Figure 17). This MJO-related fluctuation in SST is on average about 0.5°C from the minimum to maximum [e.g., Zhang, 1996; Jones et al., 1998] but can be greater than 2°C in individual cases [Krishnamurti et al., 1988; McPhaden et al., 1992].

[67] This seemingly simple, clear picture of the mixed layer responses to MJO forcing is sometimes blurred by other processes, such as horizontal advection by ocean currents, surface freshwater flux, cold water entrainment at the bottom of the mixed layer, the effect of the diurnal cycle, and the structure of the MJO. Strong surface wind of the MJO forces ocean currents and hence the possible effect of horizontal advection. Because the vertical component of the Coriolis force vanishes, strong surface westerly winds force eastward equatorial (3°N–3°S) currents of about 1 m s⁻¹ near the surface [Yoshida, 1959], which penetrate downward as deep as 100 m [Lindstrom et al., 1987; McPhaden et al., 1992]. Its horizontal thermal advection associated with active phases of the MJO can be strong in cases [Cronin and McPhaden, 1997; Wijesekera and Gregg, 1996; Feng et al., 2000], especially in a deeper layer [Song and Friehe, 1997]. Near the eastern edge of the warm pool, where the zonal SST gradient is strong, the eastward thermal advection can push the warm pool farther to the east, which may affect ENSO [Picaut et al., 1996] (40). Within the warm pool, horizontal thermal advection is generally incoherent on the scale of the MJO; its inclusion is not essential to the overall simulation of the intraseasonal fluctuations in SST [Shinoda and Hendon, 2001]. The role of horizontal saline advection in the mixed layer salt budget is significant [Smyth et al., 1996; Cronin and McPhaden, 1998; Feng et al., 2000], mainly because of the large saline gradient at the surface caused by localized rainfall events.

[68] Surface freshwater flux increases the stability of the mixed layer, which reduces the strength of buoyancy-driven mixing. Strong surface winds during active phases of the MJO minimize the sensitivity of the mixed layer temperature to freshwater flux [Shinoda et al., 1998]. This sensitivity can be amplified if surface wind is weak, for example, when its maximum is dislocated away from a convective center of the MJO [Zhang and Anderson, 2003].

[69] A direct consequence of buoyancy flux is the barrier layer [Godfrey and Lindstrom, 1989], which tends to prevent entrainment cooling from affecting the heat budget of the mixed layer and makes the latter more sensitive to surface heat fluxes [Lukas and Lindstrom, 1991; Sprintall and Tomczak, 1992]. A barrier layer usually forms when positive buoyancy flux due to precipitation and surface heating is accompanied by light surface wind, which is typical during inactive phases of the MJO. It can also form because of horizontal advection and vertical shear caused by strong surface wind [Cronin and McPhaden, 2002]. Extraordinarily strong surface winds during active phases of the MJO can generate sufficient mixing to erode away the barrier layer [Lukas and Lindstrom, 1991] and thus evoke entrainment cooling [Cronin and McPhaden, 1997]. The barrier layer therefore fluctuates intraseasonally: it is the thickest during inactive phases of the MJO [Sprintall and McPhaden, 1994; Zhang and McPhaden, 2000].

[70] The role of entrainment in the heat balance of the upper ocean during the life cycle of the MJO is difficult to estimate accurately. Direct measurements of microscale turbulence have been made only for few cases [Smyth et al., 1996; Wijesekera and Gregg, 1996]. These estimates suggest that entrainment heat flux at the base of the mixed layer varies between −25 W m⁻² (heat imported upward into the mixed layer) during inactive phases of the MJO and 35 W m⁻² (heat exported downward from the mixed layer) during active phases. The heat flux is normally downward as the mixed layer, warmer than the water below, is deepened by enhanced turbulence due to wind work and/or oceanic convection. Upward heat flux occurs when a large fraction of solar radiation penetrates through a very thin mixed layer during inactive phases of the MJO and creates a temperature inversion at its bottom [Anderson et al., 1996; Shinoda and Hendon, 1998].

[71] The contrast between daytime heating (dominated by solar radiation) and nighttime cooling (by longwave radiation) leads to a profound diurnal cycle in the depth of the mixed layer, which can shoal to a few meters during the day (with weak winds) and deepens to 60 m or more at night [Anderson et al., 1996]. During active phases, owing to a combination of strong surface winds and nocturnal cooling, the mixed layer deepening at night can break a weak barrier layer and evoke entrainment and thus enhance the mixed layer cooling on intraseasonal timescales. This diurnal cycle is critical to the amplitude of the intraseasonal fluctuation in SST in response to the MJO forcing [Anderson et al., 1996; Sui et al., 1997; Shinoda and Hendon, 1998]. Without it the mixed layer would become constantly thin, lose a certain amount of solar radiation because of its penetration during inactive phases, and experience no entrainment cooling (or even experience entrainment warming) during active phases. In consequence, the intraseasonal fluctuations in the mixed layer temperature would be less. This is an excellent example of scale interaction (22). Another example of scale interaction associated with the MJO is its rectification effects on lower-frequency variability of the ocean [Kessler and Kleeman, 2000; Waliser et al., 2003b], which may lead to an interaction with ENSO (40).

[72] Surface westerlies associated with the MJO also induce local responses below the mixed layer. A westward countercurrent can develop to 20–40 cm s⁻¹ between the surface jet and the equatorial undercurrent; the thermocline can be depressed by 20–30 m by a downwelling of 2–3 m d⁻¹ [McPhaden et al., 1992]. The associated temperature perturbations near the thermocline are of about 1°C and are disconnected from those in the mixed layer induced by surface heat fluxes [Zhang, 1997]. Oceanic Kelvin waves are responsible for the thermocline perturbations.

5.2.2. Wave Responses

[73] Intraseasonal equatorial Kelvin waves stand out as another distinct footprint of the MJO in the ocean. Their signals can be detected from the current [Johnson and McPhaden, 1993a], thermocline depth [Kessler et al., 1995], and sea level height [Enfield, 1987; Delcroix et al., 1991]. The pulse-like structure of the MJO, with its westerly wind much stronger than its easterly wind, forces pulses of downwelling Kelvin waves. They propagate from their birthplace in the western Pacific into the eastern Pacific (Figure 18) where the MJO is very weak or absent. The associated vertical displacement in the thermocline is typically 20–30 m but can be as great as 60 m [Lukas et al., 1984; McPhaden et al., 1988, 1992]. This may have significant consequences in its interaction with ENSO (40).

[74] The power of the intraseasonal Kelvin wave is centered at 70–90 days with zonal scale of 13,000–15,000 km [Hendon et al., 1998] and its maximum in the central Pacific (∼140°W) [Cravatte et al., 2003]. Observed propagation speeds of the Kelvin wave are about 2.1–2.8 m s⁻¹ [McPhaden and Taft, 1988; Delcroix et al., 1991; Johnson and McPhaden, 1993b; Kessler et al., 1995; Kutsuwada and McPhaden, 2002]. The phase speed expected from linear theories is about 2–3 m s⁻¹ [Cane and Sarachik, 1981], with higher values in the west than in the east because of the sloping equatorial thermocline [Giese and Harrison 1990]. The speed can also be affected by the phase of ENSO [Benestad et al., 2002] and nonlinearity [Ripa, 1985]. In addition, there may be a slight enhancement of eastward zonal phase speeds because of Doppler shifting by the equatorial undercurrent [Johnson and McPhaden, 1993b].

[75] The dominant period of the Kelvin wave (70–90 days) is at the low-frequency end of the spectrum for its forcing wind (30–90 days) [McPhaden and Taft, 1988; Hendon et al., 1998]. This can be explained in terms of the eastward propagation of the MJO. The energy received by the Kelvin wave is an integration of zonal stress along the characteristic line of the Kelvin wave [Kessler et al., 1995; Boulanger and Menkes, 1995]. Surface westerly winds propagating eastward cover a longer range of longitude (the fetch) and spend a longer “contact time” with the Kelvin wave. This would make the low-frequency (periods > 60 days) part of intraseasonal wind stress forcing, their weak amplitude notwithstanding, nearly resonant with the gravest baroclinic Kelvin wave and therefore able to force Kelvin waves of longer periods than if it were stationary [Kessler et al., 1995; Hendon et al., 1998].

[76] The Kelvin wave may affect the energy balance of the mixed layer in different ways. In the central Pacific near the eastern edge of the western Pacific warm pool, where the zonal SST gradient is large, the eastward surface current of the Kelvin wave can advect warmer water eastward [Picaut et al., 1996; McPhaden, 2002]. In the eastern Pacific, where the zonal SST gradient is weak but the mean thermocline is shallow, the displacement of the thermocline associated with the downwelling Kelvin waves weakens the cooling of equatorial upwelling that counteracts solar heating [McPhaden, 2002]. Both effects can lead positive anomalies in equatorial SST there [Johnson and McPhaden, 1993a; Zhang, 2001].

5.3. Oceanic Feedback

[77] The MJO is sometimes incorrectly labeled as a coupled mode simply because there is intraseasonal coherence between atmospheric variables and SST [Kawamura, 1991]. Whether the MJO is a coupled mode depends on whether the intraseasonal feedback from SST is essential to its dynamics. This is an issue much less certain than SST responses to the MJO. While numerical simulations have shown different degrees of improvement in MJO simulations when SST feedback is included (30), the mechanism for the improvement is unclear. It has been proposed that energizing effects of the positive anomaly in SST ahead (east of) convective centers of the MJO, through an enhancement in either surface latent flux [Flatau et al., 1997] or low-level moisture convergence [Waliser et al., 1999b; Kemball-Cook et al., 2002], help improve MJO simulations. While the destabilizing effect of the positive SST anomaly east of the convective center may enhance the eastward propagation of the MJO [Inness and Slingo, 2003], it is unclear why this reduces the eastward propagation speed in some simulations [e.g., Waliser et al., 1999b] and increases it in others [e.g., Watterson, 2002]. The atmosphere does not see SST; it only senses it through surface fluxes. The net effect of SST perturbations is to reduce the amplitude of intraseasonal perturbations in surface fluxes, because of the phase relations between SST and surface winds [Shinoda et al., 1998]. MJO simulations can be degraded by intraseasonal perturbations in surface fluxes [Colon et al., 2002]. An understanding of intraseasonal feedback of SST to the MJO needs to be achieved in tandem with an improved understanding of the sensitivity of the MJO to surface heat fluxes and SST effects on the MJO on the annual and interannual timescales (9 and 10).

6.1. Uniqueness of the MJO

[83] When stochastic forcing f(t) is assumed to be completely random (white) in time and space [e.g., Thompson and Battisti, 2000], it includes all types of high-frequency weather variability, such as tropical cyclones, waves, and westerly wind bursts as well as the MJO. None of them should be more special than others. Some studies suggested that f(t) should be spatially coherent with a special structure in order to be the most efficient; such a structure is referred to as the stochastic optimal of the coupled system [e.g., Kleeman and Moore, 1997]. In one coupled model of intermediate complexity [Moore and Kleeman, 1999a, 1999b] the stochastic optimal is a planetary-scale dipole in the surface zonal stress and surface heat flux in the equatorial Pacific (Figure 19). Even for a model whose stochastic optimal maximizes away from the equator, stochastic forcing in zonal stress that matters the most to ENSO predictability concentrates along the equator in the western and central Pacific (Figure 20). In both cases the MJO is the best candidate to explain the most effective patterns of stochastic forcing.

[84] In a coupled model that produces ENSO-like interannual fluctuations in equatorial SST only when f(t) is included, 71% of the variance induced by f(t) can be attributed to its leading modes representing only 28% of its total variance [Zavala-Garay et al., 2003]. These leading modes, with the same spatial structures of the stochastic optimals, affect the ocean mainly by exciting equatorial Kelvin waves. The MJO component of f(t) alone can reproduce most of the ENSO-like variability; the interannual variability induced solely by the non-MJO component of f(t) is much weaker [Zavala-Garay et al., 2005]. In observations the interannual variability of Kelvin wave forcing by the MJO component of wind stress is much greater than by non-MJO wind stress [Zhang and Gosttchalck, 2002]. The relative importance of the MJO compared to other types of stochastic forcing of ENSO, such as westerly wind bursts independent of the MJO [Vecchi and Harrison, 2000], is a subject under debate and needs to be further scrutinized quantitatively.

6.2. Mechanisms for the MJO Effect

[85] The MJO may affect an ENSO warm event by helping reduce the zonal gradient of SST. Three processes can be involved (34). Mean SST in the western Pacific can be reduced by net cooling due to the MJO [e.g., Kessler and Kleeman, 2000; Shinoda and Hendon, 2002]. Zonal current forced by the MJO westerly wind advects eastward the eastern edge of the western Pacific warm pool [Kessler et al., 1995; Picaut et al., 1996]. Oceanic Kelvin waves forced by the MJO propagate into the eastern Pacific, where they suppress the thermocline, reduce cooling due to the upwelling, and induce warm anomalies at the surface [Zhang, 2001].

[86] From a linear point of view, the MJO contributes significantly to sustain the ENSO variability only if its low-frequency energy is strong [Syu and Neelin, 2000; Zavala-Garay et al., 2003]. Intraseasonal SST anomalies associated with individual Kelvin waves (≤0.3°C) are on average an order magnitude smaller than the typical SST anomalies associated with ENSO warm events [Johnson and McPhaden, 1993a]. The low-frequency energy of the MJO in f(t), mainly from its interannual fluctuations (10), must be amplified by air-sea coupling (L and N(Ψ)) to sustain the ENSO variability [Zavala-Garay et al., 2005].

[87] Nonlinear processes can transfer energy from intraseasonal to interannual timescales. Several nonlinear mechanisms of rectification have been proposed. One is the Ekman effect, which makes the equatorial oceanic current responses stronger to westerly wind forcing than to easterly wind forcing [Kessler and Kleeman, 2000]. Another is surface wind speed, whose mean can be enhanced by about 1 m s⁻¹ over a cycle of the MJO and therefore can promote a net increase in surface heat flux to cool the warm pool [Kessler and Kleeman, 2000; Shinoda and Hendon, 2002]. The nonlinearity can also come from the fact that the MJO is more a pulse-like phenomenon, whose westerly wind is much stronger than its easterly wind (hence westerly wind bursts not easterly wind bursts) [Yano et al., 2004], as are its forced downwelling ocean Kelvin waves [e.g., Lukas et al., 1984].

[88] A positive feedback mechanism between the MJO and ENSO has been suggested [Kessler et al., 1995; Bergman et al., 2001]: Thermal advection by the MJO-forced oceanic Kevin waves results in an eastward expansion of the western Pacific warm pool, which allows the MJO to propagate farther into the central Pacific. A longer zonal fetch of wind forcing would generate a stronger Kelvin wave. This progressive eastward penetration of the MJO is commonly observed during the onset and development stages of ENSO warm events [e.g., Anyamba and Weare, 1995; Hendon et al., 1999]. This possible feedback between the MJO and ENSO SST is manifested by the observed correlation between the local MJO in the Pacific and ENSO SST indices [Kessler, 2001], in contrast to the interannual variability of the global MJO that is independent of ENSO SST [Slingo et al., 1999; Hendon et al., 1999]. Observed correlation between ENSO and MJO forcing of the Kelvin waves in the equatorial Pacific (Figure 21) also suggests a possible role of the MJO in enhancing ENSO warming at its very early stage.

[89] If forcing the oceanic Kelvin wave is central to the MJO influence on ENSO, then only in the equatorial waveguide can the strength in zonal stress of the MJO determine the efficiency of this influence. This may explain why not all MJO events, even some strong ones, necessarily lead to a warm event of ENSO [Bergman et al., 2001]. Other possible explanations are the following: The susceptibility of ENSO to the influences of the MJO depends on the mean state of the coupled system and on the timing of the MJO relative to the phase of ENSO [Fedorov, 2000]. ENSO is perhaps influenced more effectively by the seasonal activity of the MJO than by any individual MJO event [Zhang and Gottschalck, 2002], as implied by the importance of the seasonal cycle of stochastic forcing [Penland and Sardeshmukh, 1995; Penland, 1996].

6.3. Implications for ENSO Prediction

[90] Numerical studies have demonstrated the effect of the MJO on individual ENSO warm events by prescribing through model integrations wind anomalies associated with the MJO, equivalent to f(t) with G(Ψ) = 0 in equation (1). Thermal advection by intraseasonal downwelling Kelvin waves forced in the western Pacific could have contributed substantially to simulated positive SST anomalies during the 1982 El Niño event [Harrison and Schopf, 1984]. Oceanic responses to a westerly wind event lasting for a month in the western Pacific have led to SST anomalies persisting for 12 months in a coupled model [Latif et al., 1988]. Many simulations have particularly demonstrated the roles of westerly wind events associated with the MJO in the onset and growth of the 1997–1998 El Niño [Kessler and Kleeman, 2000; Perigaud and Cassou, 2000; Lengaigne et al., 2003; Boulanger et al., 2004]. These results suggest that the timing and amplitude of an ENSO warm event might be better predicted should MJO activities be known.

[91] With its current prediction skill limited to 15–20 days (32), the MJO is unpredictable on interannual timescales [Slingo et al., 1999; Waliser et al., 2001]. This implies that any significant effect from the MJO on ENSO might, as other types of stochastic forcing, limit the ENSO predictability. Better knowledge of the characteristics of the MJO (e.g., its structure, geographic preference, and seasonal cycle) might help determine more realistic spread of the uncertainties in probabilistic predictions of ENSO subject to stochastic forcing [Fedorov et al., 2003]. If stochastic forcing of the MJO is indeed multiplicative, then it is conceivable that an ENSO prediction model with a capability of producing realistic MJO statistics and its relationship with ENSO SST would do better than a comparable model but without any MJO signals. The observed lag correlation between anomalies in seasonal MJO activities in the western Pacific and in ENSO SST (Figure 21) and some ENSO simulations with prescribed wind events (discussed earlier in this section) both suggest that to benefit 6- to 12-month ENSO prediction by models able to produce MJO signals, it is perhaps more important for the models to maintain a realistic level of seasonal activities of the MJO at the initial time than to forecast the MJO 6–12 months into the future. ENSO prediction assisted by including MJO activities remains an uncharted territory.