a. Data

The data used in this section come from various sources. We concentrate our analysis on March 1993 when the CEPEX was conducted in the tropical Pacific. The data collected during this experiment are used in this paper. In addition, many TOGA-COARE-related platforms supplied data. Lagrangian drifting buoys released in the Western Pacific in February and March 1993 as part of TOGA COARE effort (Ralph et al. 1997) are used to obtain SST and oceanic surface currents in the equatorial region of the Western Pacific. The sea temperatures are measured by a thermistor on a surface float, 10 cm below the surface. The buoy’s drogue is centered at 15 m. In most of the calculations we are using buoys located within 3 degrees of the equator, between 150°E and the date line. The position of the drifting buoys, SST and ocean current measurements were available every 6 hours. TAO ATLAS buoys were used to look at surface winds and the temperature structure of the mixed layer in the vicinity of the drifting buoys.

Atmospheric data were also gathered from a variety of sources. The cloud-top temperatures and albedo were obtained from GMS data converted to a 0.5° grid. To assess the synoptic situation in the area of interest, we used the analyzed fields from NOGAPS, a system that has been shown to have considerable skill in the tropical Pacific (Goerss and Jeffries 1994; Kindle and Phoebus 1995). NOGAPS consists of a multivariate optimum interpolation analysis (Goerss and Phoebus 1992), a nonlinear normal mode initialization, and a global spectral forecast model (Hogan and Rosmond 1991). For this study, NOGAPS was executed using a T79 version of the global spectral model, which has a physical grid resolution of approximately 1.5°. The NOGAPS surface wind is valid at 10 m, and surface wind speed observations from the SSM/I are routinely assimilated into this system, a particularly attractive feature for our purposes.

To examine the relationship between wind fields, cloudiness, surface currents, and SST the NOGAPS surface (10 m) winds and the GMS data are interpolated to the drifting buoy positions.

b. The influence of equatorial convection on SST changes during March 1993

During our analysis period (March 1993) two episodes of equatorial convection were observed. The first began on 5 March, in the vicinity of the local SST maximum at 170°E, and ended in the formation of twin tropical cyclones Irma and Roger on 13 March 1993. The second episode developed on 18 March near 165°E and lasted for 3 days. In agreement with Gill’s (1980) arguments on the circulation response to an equatorial heat source, both of these episodes were associated with westerly wind anomalies on the equator, with the maximum surface westerlies reaching 10 m s⁻¹ in the case of the 5 March episode, but only 5 m s⁻¹ for the late March event.

The wind and cloud-top temperatures, along with the 24-h SST changes observed during the stronger March event, are shown in Figs. 1 and 2. Initially (Fig. 1) the convective activity was concentrated on the equator. A narrow strip of strong westerly winds developed below the convective heat source, while to the east of the heat source weak easterlies prevailed. During this stage, the SSTs decreased in the region of westerly winds. As can be seen in Fig. 1, both large surface fluxes related to strong westerly winds and the presence of high clouds with their large albedos could contribute to the observed SST drop. Even though we have less drifter data in the convergence zone between easterly and westerly winds, the existing buoys indicate either no significant change or an increase of SST in this region.

Figure 2 depicts the situation on 13 March 1993, right after the tropical depressions on both sides of the equator reached tropical cyclone strength (Irma at 5.9°N, 160°E and Roger at 13°S, 165°E). Even though the equatorial winds remained strong west of 160°E, the convection shifted off the equator and became centered in developing storms. The SST began to rise—especially in the region of the low-level divergence between the storms. Near the center of Irma, a SST drop can be observed. As the storms moved poleward and westward, the equatorial westerly winds propagated westward and weakened and the equatorial SST continued to rise.

The response of equatorial surface currents and SSTs to the March 1993 atmospheric events, as measured by drifting buoys, is shown in Fig. 3. The increase in equatorial westerlies was followed by a change in the direction of the ocean equatorial surface current. A comparison of the current velocities measured by each buoy with the wind speeds interpolated to the buoy’s location showed that the lag in response was about 5 days. The eastward acceleration of the drifters was accompanied by a SST decrease, while for the westward moving buoys the temperatures generally increased. This behavior can be explained in two ways.

First, since the magnitude of the westerlies associated with the equatorial convection is usually larger then the magnitude of the easterly winds developing east of the heat source, evaporative fluxes are larger during westerly wind episodes. The analysis of evaporative fluxes from TAO buoys as well as from NOGAPS model field data shows that this was indeed the case, especially during the early March westerly burst. In NOGAPS, the surface latent flux on the equator near 160°E reached 200 W m⁻² during the burst, compared to about 80 W m⁻² after the cyclones moved poleward. Similarly, the evaporation calculated by bulk formula from the TAO buoy data indicates that for sea surface temperatures of about 29°C the largest fluxes (200 W m⁻²) were associated with the strongest westerlies (10 m s⁻¹), while easterlies in this region seldom exceed 5 m s⁻¹ and had associated evaporative fluxes of about 100 W m⁻² or less.

Another explanation involves reduction of the solar flux by convective clouds associated with the diabatic heat source, emphasized by Ramanathan and Collins (1991). To look closer at how these two mechanisms contribute to the SST changes observed by drifting buoys, we calculate the average 24-h SST change as a function of the zonal wind velocity from NOGAPS analyses and the GMS cloud-top temperatures T_B collocated with the position of the buoys. Since SST time derivatives are calculated following Lagrangian drifters, they reflect vertical fluxes at the position of the drifter.

Figure 4 shows the 24-h SST change as a function of zonal wind, for drifting buoys located within 3° of the equator. The solid line shows the SST change averaged for zonal winds from u + Δu to u − Δu, where Δu = 0.5 m s⁻¹.

As can be expected from evaporative fluxes, the largest heating occurs for small zonal wind velocities. However, there is a significant asymmetry in SST change between easterly and westerly winds, which cannot be explained by the effects of evaporation alone. While the averaged SST change is positive for all easterly winds, with the maximum at u = −3 m s⁻¹, cooling dominates for westerlies stronger than 2 m s⁻¹. The average SST change for easterlies decreases with the magnitude of the wind velocity. This relationship can be explained by the increasing role of the evaporative cooling competing with the clear-sky solar fluxes.

The large scatter of the data can be explained by combined effects of various surface fluxes and changing depth of the mixed layer. The influence of the mixed layer depth on the magnitude of the SST change for March 1993 convective events is illustrated in Figs. 5 and 6, representing the temperature and the 24-h change of SST for one of the drifting buoys, and wind and ocean temperatures at the TAO ATLAS (Autonomous Temperature Line Acquisition System) buoy located near the drifter. Since only temperature data are available, the mixed layer in this case is defined by temperature profiles. The density mixed layer can be quite different because of the salinity effects and presence of the “barrier layer.” Comparison of Figs. 5 and 6 suggests that very large SST tendencies were associated with a shallow mixed layer in the ocean. For example, at the beginning of the westerly burst (5 March) the SST drop was sudden and, as wind magnitude was still relatively small, the temperature decrease was probably related to the reduction of the solar flux. Figure 6 indicates that the initial large SST drop was confined to a very shallow layer. As the mixed layer depth increased with growing wind speed, the effect of surface fluxes was spread over the deeper layer and SST changes were smaller. After the convection moved off the equator (after 12 March), the increasing solar fluxes began to heat and stabilize the surface layer, even though the equatorial westerlies were still relatively strong. When the equatorial winds weakened, the SST rapidly increased. The high SSTs provided favorable conditions for the development of convection, and on 18 March the SST again rapidly decreased.

The relationship between SST change and GMS-derived cloud-top temperature is more straightforward. In Fig. 7 we show the average 24-h change of SST as a function of cloud-top temperatures T_B averaged over six hours. Although there is large scatter in the data, the average SST change is positive for T_B greater than 255 K and negative for colder cloud tops. This relationship corresponds well with the linear relationship between T_B and net cloud effect on surface forcing observed in CEPEX (Collins et al. 1996). As shown by Collins et al. (1996), on the surface, the “shading” effect is greater than longwave heating, and the net radiative effect of clouds is to cool the surface. The dependence between the net cloud surface forcing C_net(0) and cloud-top temperatures T_B can be approximated by a linear relation, C_net(0) = 2.8 − (290 − T_B) − 7.

The contribution of the radiative effects to SST changes can be roughly estimated by comparing 24-h SST changes for −3 m s⁻¹ easterlies (corresponding to the maximum temperature increase) to those for 3 m s⁻¹ westerlies. We chose the identical wind speed for our comparisons to minimize the influence of evaporative fluxes and mixing effects on SST tendency. Therefore the difference may be attributed to radiative effects of the convective clouds in the westerlies region. We calculate the net flux F_n as

View Expanded

where dSST/dt is the SST change estimated from drifters, ρ_w and L_w are the density and specific heat of water, and d_m is the depth of mixed layer. We use a 4-m mixed layer depth for the undisturbed period (characterized by easterly winds) and a 23-m mixed layer depth for the disturbed period (westerly winds). These values are based on Webster et al. (1996) calculations of SST and mixed layer properties for different atmospheric conditions. Their results were obtained using TOGA COARE IOP measurements supplemented with a one-dimensional model of the oceanic mixed layer. If the values for the whole month of March 1993 are used, the average dSST/dt is 5.6 × 10⁻² K day⁻¹ for u = −3 m s⁻¹ and −2.7 × 10⁻² K day⁻¹ for u = 3 m s⁻¹, which gives the average net vertical flux of 11.6 W m⁻² under easterlies and −26 W m⁻² under westerlies. Therefore the difference in net surface fluxes between disturbed and undisturbed conditions, with the same average wind speeds of 3 m s⁻¹ is about 38 W m⁻². If only the data from the stronger episode (1–15 March) are considered, the net flux difference becomes 50 W m⁻², due to the fact that the equatorial radiative fluxes were unusually large right after formation of the twin cyclones.

a. Conceptual model of convection–SST feedback

The observational results presented in the previous section suggest a strong influence of equatorial convective systems on local changes of underlying sea surface temperature. These results agree with other TOGA COARE observations of the decrease in SST and heat content during MJO-related westerly bursts (Webster 1994) and the coincidence of highest SST with weakest winds (Serra et al. 1997). Figure 8 illustrates the relationships between equatorial cloud-top temperatures, zonal wind stress, and SST during MJO episodes observed during TOGA COARE IOP (from ECMWF analysis). The presence of high clouds is associated with westerly surface winds. The sea surface temperatures are highest before the onset of convective activity and drop sharply after MJO passage.

Nakazawa’s (1995) detailed analysis of intraseasonal oscillation (ISO) episodes observed during the TOGA-COARE IOP showed that high SST preceeded convective activity by 12–13 days. In addition maximum total precipitable water observed by SSM/I was leading convection by about 5–10 days. In the case of IOP convective systems, the increase in precipitable water occured in the regions of relatively small winds and therefore could not be attributed to wind-evaporation feedback. Nakazawa suggests that this moisture increase was related to enhanced evaporation associated with moisture deficit terms in regions of high SST.

Based on our observations and results from TOGA COARE we suggest a new conceptual model of intraseasonal oscillation, taking into account the interaction between SST and equatorial convection. We call it air–sea convective intraseasonal interaction (ASCII). A comparison of ASCII with the wave-CISK theory of intraseasonal oscillation (Lau et al. 1989) and the wind-induced surface heat exchange theory (Emanuel 1987) is shown in Fig. 9.

In the wave-CISK theory the region of convergence between easterlies and westerlies in an eastward propagating Kelvin wave (region C in Fig. 9) becomes the preferred site for the subsequent development of convection. The formation of the new cells in this region causes the eastward propagation of the equatorial supercluster, while the individual clusters that form on both sides of the equator move west as Rossby waves (Lau et al. 1989) This approach was criticized by Emanuel (1987), who argued that convection does not create atmospheric temperature perturbation but only redistributes the boundary layer Θ_e. Therefore, if convection is caused solely by convergence of moisture at constant Θ_e, adiabatic cooling exactly balances diabatic heating and there is no temperature perturbation in the troposphere and development of growing, eastward propagating modes is not possible. Recent numerical models of tropical wave instabilities (Brown and Bretherton 1995; Yu and Neelin 1994) indicate that earlier successes in modeling of MJO as wave-CISK instability depended on the use of Kuo-like parameterization that explicitely includes the relationship between convergence and development of convection. When other parameterizations of convection like convective adjustment scheme (Yu and Neelin 1994) or Emanuel parameterization (Brown and Bretherton 1995) are used, wave-CISK modes do not develop.

Emanuel (1987) as well as Neelin et al. (1987) argued that, in the presence of the mean equatorial easterly wind, enhanced surface evaporation in the region of the strong easterlies (Fig. 9b) can provide surface forcing necessary for development of instabilities similar of observed oscillations. This process is usually referred to as wind-induced surface heat exchange (WISHE). Increased surface fluxes in the region of large easterlies are responsible for positive temperature perturbations that cause the eastward propagation of the wave and provide energy for growth of instability. However, the results from the TOGA-COARE IOP (Lin and Johnson 1996) suggest that, contrary to the behavior postulated by Emanuel (1987) or Neelin, in the warm pool region the maximum surface fluxes associated with the presence of strong winds tend to develop underneath and toward the west of the convective system, since equatorial westerlies (“westerly bursts”) are usually stronger than easterlies (Gill 1980). In addition, the assumption of mean easterly flow used by Emanuel and Neelin is not always true, especially during the El Niño years.

In our hypothesis (Fig. 9c) the surface forcing necessary for enhancement of the Kelvin wave is provided by the SST distribution generated by the supercluster itself. As we have shown in the previous section, in a narrow band of strong westerlies underneath the cluster, SST drops not only due to evaporation but also because of oceanic vertical mixing and cloud shielding effects. However, east of the convective source, in the region of the weak easterly winds, the conditions exist for increase in SST (Serra et al. 1997; Nakazawa 1995). This configuration favors the increase of the surface moist entropy east of the system and development of convection in the convergent region of the Kelvin wave.

Both our theory and WISHE emphasize the importance of surface fluxes in maintaining the convection. However, if the ocean is treated as an infinite energy source, the increased wind speed is necessary for raising the surface moist static energy east of a convective complex. If the ocean is allowed to respond to convection, the surface forcing related to SST changes can provide surface moist static energy anomalies, necessary for growth and propagation of eastward propagating perturbations. When the modification of SST by convective systems is allowed, the regions of weak surface winds—rather than strong winds—become the preferred sites for future convective development.

b. Model results

Results of three model runs are presented here. In the first experiment, the model is run with steady-state SST. In the second, the SST is modified according to the relationship discussed above. In a third experiment we use steady-state SST and, in addition, assume constant wind speed in calculations of latent and sensible heat fluxes to eliminate WISHE modes.

As can be seen from Fig. 10a, in the case of steady SST, the equatorial circulation is dominated by a wave propagating around the globe with a phase speed of about 20 m s⁻¹ and low wavenumber perturbations (s = 1 and s = 2) dominating in the zonal direction. When we neglect the wind–evaporation relationship (Fig. 10b), the propagating wind features are significantly reduced, suggesting that the modes observed in Fig. 10a are a combination of Kelvin and WISHE modes. The spectral analysis of equatorial wind shown in Fig. 11 indicates that WISHE modes, removed by setting the surface wind to a constant, have periods of 20–30 days, which agrees with periods of WISHE modes obtained by Emanuel (1987) or Neelin (1987).

The situation changes if we allow for wind–SST dependence in the equatorial region. The most dramatic change can be observed in rainfall (Fig. 12) where the 40–60 day modes become dominant. As shown in Fig. 10, the wind perturbations become more organized and their phase speed decreases. In the spectral analysis of the equatorial zonal wind (Fig. 12), the amplitude of the mode with a 30-day period increases, which suggests that WISHE modes were amplified by including SST perturbations. In addition, the slower mode, with the period corresponding to a dominant mode in the rainfall, begins to develop.

In the numerical experiments described above, we do not attempt to produce a realistic depiction of intraseasonal oscillations. However, we have been able to show that within the framework of our model, SST modifications with realistic magnitudes and timescales (Fig. 13) can influence convective development and circulation, leading to amplification of equatorial modes with frequencies similar to those of the observed intraseasonal oscillations.