1. Introduction

The tropical atmosphere over the western equatorial Pacific (warm pool) manifests continuous disturbances together with highly active air–sea interaction. Yet, the warm pool SST-(a list of acronyms appears in Table 1) as compared to that of the other sections of the tropical Pacific displays very small spatial and temporal variations. To illustrate the temporal variations, the area-averaged SST is plotted for six consecutive years (1993–99) over the eastern, western, and entire equatorial Pacific in Fig. 1. These time series are calculated from the NCEP SST analysis (more information on this data will be given in section 1). As seen from this figure, the time variation of the area-averaged SST is much greater in the eastern sector. Furthermore, the time series for the eastern sector and the entire tropical Pacific are closely related in terms of their annual variations. The western counterpart, however, displays very different variations.

Another interesting aspect of the western equatorial Pacific is that it is not largely affected by the extreme anomalies such as the ENSO cycle (see year 1997 in Fig. 1). Furthermore, the western Pacific has the warmest SSTs over the entire equatorial Pacific. Even though this is true, observations have shown that the maximum SST in the western Pacific never exceeds around 32°C. All these suggest that the western Pacific SST is driven by a unique mechanism as compared to the rest of the tropical Pacific. This persistent behavior of the western equatorial Pacific has prompted a lot of interesting debates during the last decade on whether or not the SST over this region is regulated. One of the most interesting conclusions from these debates was the relation between convection and SST. Using observational data, earlier studies (see, e.g., Krueger and Gray 1969; Gadgil et al. 1984; Graham and Barnett 1987; Waliser and Graham 1993; Waliser 1996; Lau et al. 1997) have demonstrated that scatterplots of convection (represented by OLR or cloudiness) versus SST display three different regimes. The first regime occurs when the SST is lower than about 27°C; it is characterized by little or no convection. The second regime is when SST is roughly between 27° and 29°C. This area is characterized by vigorous convection. This area is followed by the third regime when SST is over 29°C. In this regimes, the convective measures display a sharp decline.

Although the relationship between tropical SST and large-scale/local convective processes implies a clear role of convection in SST regulation, it does not explain the way this mechanism operates. The first idea toward this came from Graham and Barnett (1987) who argued that the cloud cover associated with deep convection may block some incoming solar radiation and thereby limit the SST. Their analysis was based on the SST–convection relation and an idealized surface heat balance model. Then, Ramanathan and Collins (1991) proposed the so-called thermostat hypothesis to explain the mechanism regulating maximum SST in the warm pool. This theory works in two stages; first, in the presence of convective clouds and high SST, the greenhouse effect results in uncontrolled warming. Second, the highly reflective clouds, as a result of the convection, shield the surface against solar radiation and inhibit the further warming. Wallace (1992) on the other hand, has proposed the following explanation; high SSTs destabilize the atmosphere by heating the planetary boundary layer through surface heat fluxes. In response, the atmosphere stabilizes itself by convection that carries the heat away from the boundary layer. Throughout this process, the latent heat flux is the major factor transporting heat from the ocean to the atmospheric boundary layer. According to Wallace, this stabilization process, with other elements of atmosphere–ocean interaction, is enough to control maximum SST over hot spots such as the warm pool region. A similar explanation was later proposed by Waliser and Graham (1993). They suggested that the convective systems over the Tropics act to stabilize the tropical troposphere that is heated and supplied with the moisture from the ocean surface below by cooling and drying the lower levels and warming and moistening the upper levels.

These important findings have strengthened the hypothesis of a regulated SST in the western tropical Pacific. However, it was realized that the available observations were not sufficient to unveil the regulation mechanism. A complete budget analysis of the warm pool SST is necessary to construct the dynamical and physical aspects of this regulation. Hartman and Michelsen (1993) took the first step toward this when they studied each term in the surface energy balance and their sensitivities to SST using observations and a simple numerical model. Their result displayed that the solar heating and latent heat cooling of the surface are the major contributors to SST. They also indicated that shortwave cloud forcing is more sensitive to the SST gradient rather than the absolute value of SST, implying a large-scale signature in the regulation of SST over the warm pool. This was also the point where the thermostat hypothesis was criticized by Lau et al. (1994). They argued that the cloud changes strongly depend on the large-scale processes that are derived by the complex large-scale atmosphere–ocean circulation as opposed to the local convection.

The studies mentioned above suggest a timescale of the variability of SST that is closely related to that of convection. However, Arking and Ziskin (1994) used three different datasets to show that the variability of the warm pool SST is season- and latitude-dominated. Another result from this study was that cloud effect on longwave radiation is as important as that on solar radiation, which means it is unlikely that clouds have a regulatory effect on SST.

Dominance of seasonal variability of SST was also shown by Zhang and McPhaden (1995) in the western Pacific. They indicated that the seasonal variabilities of SST and the latent heat flux are out of phase and dominated by the surface wind changes, mainly derived by the large-scale atmospheric circulation in this timescale. According to Zhang and McPhaden, the seasonal variability of surface wind overcomes that of the humidity difference between the surface and a reference level, keeps the latent heat flux small, and lets the SST increase with solar radiation. These results have been confirmed by Zhang et al. (1995) where they pointed out that on the climate timescale, the variation of the latent heat flux at high temperatures is dominated by a variation in wind speed and, therefore, decrease in wind speed leads to a relatively suppressed latent heat flux at high SSTs.

These findings indicate that SST regulation over the western Pacific is an intricate temporal-and spatial-dependent phenomenon. Zhang et al. suggested a simple explanation stating that the climatological mechanisms are the dominant factors maintaining capped SST over the western Pacific. A later study by Lau and Sui (1997) suggested that on timescales of weeks to months over the warm pool, the SST is mainly regulated by the MJO through induced surface fluxes and upper-ocean processes. However, they also argued that the MJO itself is subject to the climatological changes. Sud et al. (1999) claimed that the downdraft during the convective phase of the MJO is a major factor in determining the warm pool SST. They claimed that the descending warm and dry air associated with the downdraft causes the abrupt increase in the surface latent and sensible heat fluxes and thereby begins to cool the ocean surface. According to them, the sudden increase in the surface heat fluxes act as a thermostat analogous to the one proposed by Ramanathan and Collins (1991) except that here, the thermostat is the surface flux and not the shortwave cloud forcing. A recent study by Tian et al. (2001) presents very detailed analysis of the warm pool atmosphere during TOGA COARE and CEPEX. They were able to show that the cloud radiative forcing has a dipole effect on the SST very much resembling the thermostat hypothesis.

In all of these studies mentioned above, the elements that are thought to be responsible in regulating the warm pool SST are usually the atmospheric parameters that are not unilaterally agreed to be the prime factor. The effect of oceanic processes in the regulation of the warm pool SST is not clearly understood and therefore, neglected in most of the studies. An important consensus was that this regulation mechanism could be explained better if each term in the balance equation is examined at each individual timescale with the use of a detailed coupled atmosphere–ocean model. In fact, Lau et al. (1997) stressed the necessity of such detailed work in order to understand the real nature of this regulation phenomenon.

Even though some of the past studies have benefited from simple numerical models (see, e.g., Hartman and Michelsen 1993; Zhang et al. 1995), lack of coupling and simplistic approaches used have made their results hardly convincing. Probably the first coupled numerical study for studying the warm pool SST came by Schneider et al. (1996) who used MPI's coupled GCM. However, they only focused on seasonal variations. Observations in and around the warm pool region were the primary source for the earlier studies and revealed significant amounts of information. However, long discontinuities in the temporal and spatial distributions proved extremely hard to cope with. Lau et al. (1997) also pointed out that the limited amount of observations can sometimes be misleading and therefore, should be used with extreme caution. Remarkably, after reprocessing the OLR–SST relationship, they showed that the third region in the scatterplots of SST versus convection (as explained before) does not really exist in the warm pool region.

Our goal in this paper is to ease the problems mentioned above and perform a residue-free thermodynamic budget of the coupled system by utilizing the FSUCGSM to assess the contributions from each term in the sea surface temperature equation. Global atmospheric analysis from ECMWF and global SST analysis from NCEP (R. Reynolds, D. Stokes, and T. Smith, personal communication) are used as initial conditions for the atmospheric model. The initial conditions for the ocean model are obtained at the end of a 10-yr spinup period. Levitus climatology (Levitus 1982) is used to initiate the spinup processes. The resulting initial state for the coupled model reflects the actual atmospheric and oceanic conditions for the initial time (we have chosen 3 March 2001 as the starting date). More information on ocean spinup processes can be found in LaRow and Krishnamurti (1998). To obtain a residue-free SST simulation, the coupled model is redesigned in a way that all the numerical interventions such as filtering, nudging, etc., are reduced to a minimal level.

2. Methodology

SST equation as defined in FSUCGSM can be written as

View Expanded

where V is the horizontal velocity, w is the vertical velocity, and D_H and D_υ are horizontal and vertical diffusion coefficients, respectively. The surface forcing F can actually be decomposed into the following components:

View Expanded

where C is the specific heat for the ocean and ρ₀ is the density of the water. The terms in the parentheses from left to right represents solar, terrestrial, evaporative, and sensible heat fluxes, respectively. Now, it can be seen that the SST tendency in Eq. (1) is made up of eight partial tendencies, namely, shortwave, longwave, latent heat, sensible heat, horizontal advection, vertical advection, horizontal diffusion, and vertical diffusion. Since all of these appear in Eq. (1), thereafter, they will be referred as direct contributions. The first four of these contributions are the direct outcome of air–sea interaction, whereas all the others are involved with the oceanic processes that may indirectly be linked to the air–sea interaction due to momentum transport into the ocean. There is one additional widely used term, which is the convective adjustment resulting from numerical representations. This term is considered as a direct contribution since it is applied to the vertical temperature profile in the ocean at each time integration.

Any variable that is used in calculation of the direct contribution will be referred as indirect contribution. Surface wind speed, cloud cover, etc., are, for example, some of the indirect contributions. To obtain a residue-free analysis, it is important that Eq. (1), after adding convective adjustment, leave very small residue during the time integration of the coupled model. For this reason, the FSUCGSM is redesigned to produce a very small residue up to order of 10⁻¹³°C day⁻¹ as shown in Cubukcu (2001).

The entire context of this project is based on the 1-yr numerical integration of the redesigned FSUCGSM, from March 1996 to March 1997 (the date is chosen arbitrarily). An area-averaged time series for each direct contribution is derived from the numerical integration. Area averaging for the warm pool domain is done on an area between ±10° latitude and between 125° and 150°E longitude. Unless otherwise stated, this area will be called the warm pool, or, interchangeably, as the western equatorial Pacific. The land points in the warm pool region are excluded from the calculations. When preparing the time series, each term is taken as it appears in Eq. (1). Since we will mostly be working with the anomaly fields, it should be noted that for a term that is always negative (such as the latent heat term) negative/positive anomalies denote more/less cooling.

As part of the residue-free analysis, we first obtain the power spectrum of a 1-yr-long record for the warm pool SST. Based on this spectrum, the most important harmonics are identified. Then, using the bandpass-filtering technique, the time series of each term is filtered for each important timescale. Next, the filtered time series are used to explore the primary forcing(s) on the warm pool SST for each timescale. This part is accomplished using PCA analysis. All this information is then used in order to explore the regulation mechanism of the warm pool SST.

3. Numerical model

4. Verification

In this section, the validation and verification of the predicted SST with NCEP SST are discussed. The coupled model has previously been applied to numerous climatological problems (see, e.g., LaRow and Krishnamurti 1998; Krishnamurti et al. 2000). The results have proven that the FSUCGSM is an excellent tool for climatological studies. Based on previous successes of this model, we assume that all direct and indirect contributions to SST are reasonably simulated provided that the SST agrees well with the observations. It is hard to perform verification of each forcing in the SST equation due to the lack of available observational data consisting of the same spatial and temporal characteristics. Figure 2 shows the differences (model − observed) in the seasonal mean SST fields for one year of model integration.

Overall, this figure shows that the comparison of the model results to NCEP analysis displays good agreement in most of the equatorial Pacific. There are some discrepancies in the eastern equatorial Pacific in the region of the cold tongue and also in the downstream parts of the California and Peru Currents as also noted in Latif et al. (1994). These discrepancies are due to inadequate simulation of the narrow canal coastal currents such as those off the coasts of California and Peru and thereby the strong coastal upwelling in these regions. Also, the resolution and the interpolation routines used in the model and those used in the NCEP analysis often cause large differences during the comparison. The comparison in the western Pacific, which is the main focus of this study, is very encouraging.

Time variations of the weekly mean model and NCEP SSTs are shown in Fig. 3 for the western, eastern, and entire equatorial Pacific regions. The sampling period is taken as one week to be consistent with NCEP's weekly analysis. It is again clear that the model performs well especially in the warm pool region for the entire period of the integration. The warm pool region has marked intraseasonal variations in both the observed and the predicted time series. These variations were also noted by Lau and Sui (1997) using observations during TOGA COARE. In the longer timescale, both the model-predicted SST and the NCEP analysis display semiannual variation in the western equatorial Pacific. Although the model simulation displays similar low-frequency variation in the eastern Pacific, the analysis shows a seasonal type of variation. Our simulation did not capture this shift at low frequencies as one goes from west to east. Both the semiannual and annual variations are associated with the annual march of the ITCZ. Observations shows that the ITCZ usually stays around 10°N in the eastern Pacific (Waliser and Gautier 1993). In fact, if that was the case in the model simulation, the time series of the SST would have had seasonal variations instead of semiannual type of variations. The model fails to capture this, probably due to same reasons we mentioned above for the spatial comparison.

The comparison for the entire tropical Pacific, however, indicates that excluding the coastal features in the eastern Pacific, the FSUCGSM captures a great portion of the temporal and spatial variations of the equatorial Pacific SST on an annual timescale. Errors associated with the coastal features in the eastern Pacific can be assumed to have little effect on the entire equatorial Pacific SST, especially in the western end. Thus, FSUCGSM can be a very useful tool for studying the mechanism of the warm pool SST.

5. Residue-free budget analysis of the warm pool SST

In this section, the types of variations existing in the warm pool SST are investigated using area-averaged time series. For this goal, we have first computed the power spectrum of the warm pool SST for the time period defined in the introduction. The power spectrum is displayed in Fig. 4. Instead of using the conventional frequency–power relationship, we have made an area-conserving transformation in order to have a better look. The new sketch of the variables [frequency × power vs log(frequency)] provides consistency between the variations in the time and frequency domains. This technique was first used by Zangvil (1975) and is still being used in numerous meteorological applications. As shown in Fig. 4 it suggests that by the nature of the spectrum and earlier observational studies (see, e.g., Lau and Sui 1997), it may be possible to partition the spectrum into three distinct regions. The first region is where the spectrum has a peak at around 180 days. This suggests the existence of a semiannual variation of the warm pool SST. The semiannual variation is associated with the annual variation of the solar radiation. The second region in the spectrum marked as MJO falls roughly into the 30–60-day band. The variation of the warm pool SST on this timescale contains a great portion of the total variance, as it was the case for the semiannual variation. MJO can be regarded as a baroclinic wave disturbance originating over the Indian Ocean and traveling eastward around the globe (Madden and Julian 1971, 1972; Krishnamurti and Gadgil 1985). The most intense phase of these oscillations occurs over the tropical Indian Ocean and over the warm pool region. Atmospheric conditions associated with MJO change drastically. The wet phases of these oscillations are characterized by heavy convection, often with westerly wind anomalies, whereas the dry phases of these oscillations are characterized by clear and dry air with easterly wind anomalies (Knox and Halpern 1982; McPhaden and Taft 1988; Fasullo and Webster 2000). The SST responds to these atmospheric variations through air–sea interaction. The third region is the 10–25-day band. Lau et al. (1997) described these as biweekly oscillations created by the local synoptic events regardless of the wet/dry phases of MJO. In a more recent paper by Fasullo and Webster (2000), it was argued that these oscillations are associated with the Kelvin waves that are believed to be ignited by the westerly wind bursts. The dynamical aspect of these oscillations is quite similar to that of MJO, except that their impact on atmospheric and oceanic variables is smaller compared to that of MJO. These disturbances also comprise wet/dry phases associated with the westerly/easterly wind anomalies. Even though there are notable similarities between MJO and the 10–25-day oscillations, there is no clear indication that these variations are related.

The possible regulation mechanism for the warm pool SST needs to be explored at each of these individual timescales. This isolation is necessary because there is no apparent relation among these distinct variations. After separating each timescale, the direct contributor(s) acting on the SST can be evaluated without facing the contamination problem from other types of variations. Once these contributors for each type of variation are determined, we can then seek what determines their variations using related atmospheric and oceanic variables previously named as indirect contributions. The use of a bandpass filter combined with the principal component analysis may be an excellent tool to reach this goal. The layout for this plan is as follows.

1. Use a bandpass filter for the timescales described above to obtain the filtered time series of the direct contributions.

2. Apply PCA (EOF analysis) to the time series obtained in the previous step.

3. Obtain the most representative contributor(s) using the information gathered from the principal components and corresponding variances.

For the first step, BBPF is used. This filter has been employed extensively in meteorological applications (see, e.g., Murakami 1979; Krishnamurti and Subrahmanyan 1982). Only the time series of shortwave, longwave, latent heat, sensible heat, total advection, vertical diffusion, and convective adjustment are filtered, along with the SST tendency. The covariance matrix used in the EOF analysis is constructed by placing these contributors at each row in the order listed above following the columns corresponding to the time levels.

BBPF is first applied to the time series of the warm pool SST for the frequency bands mentioned in the previous section. The resulting filtered time series are displayed in Fig. 5. In relation to the spectrum of the warm pool SST in Fig. 4, this figure clearly displays that the variation of the warm pool SST is inversely proportional to the frequency, that is, the largest deviations from the annual mean occurs at the low frequencies and the smallest variations occur at the high frequencies. The dominance of the low-frequency variations was also pointed out by many authors (see, e.g., Arking and Ziskin 1994; Zhang and McPhaden 1995) using various kinds of observational data over the warm pool domain. Furthermore, the comparison between the 30–60- and the 10–25-day variations is also consistent with the observational findings (Fasullo and Webster 2000). We now feel very confident that the FSUCGSM not only successfully simulates the types of variations in the warm pool SST, but also captures the relative importance of each timescale.

Considering that area-averaged quantities are used for this analysis, it is realized that the warm pool SST undergoes substantial changes. Therefore, in the following discussion, considerable effort is made to investigate all these variations. Figure 6 displays the filtered time series of the direct contributions and the tendency of the warm pool SST for 10–25-, 30–60-, and 120–360-day variations. This figure is the basis of the entire EOF analysis.

The results of EOF analysis for the direct contributions are displayed in Fig. 7 and Table 2. Starting with the 10–25-day variations, the results can be interpreted as follows. The first (largest) eigenvector corresponding to the first (largest) eigenvalue belongs to the shortwave contribution. All other first eigenvectors are less than around one-third of the eigenvector that belongs to the shortwave contribution. The first principal component contains 66% of the total variance. This is to say that the shortwave contribution is the quantity most heavily altered by the 10–25-day oscillations. The second largest eigenvalue in Table 2 corresponds to the second largest eigenvector associated with the latent heat term. The second principal component contains roughly 24% of total variance. Together, the first two principal components contain about 90% of the total variance. With this information at hand, one can easily assume that the shortwave and, secondarily, the latent heat term, can explain the 10–25-day variations seen in the evolution of the warm pool SST. Since the temporal variation is fixed using the bandpass filter, the PCs should imitate the dominant terms in SST equation.

Similar analysis is also applied to the 30–60-day timescale. It is seen that the shortwave term contains most of the variance (about 67% of total variance). The first principal component shows a lot of similarity to the filtered SST tendency in Fig. 6. Nevertheless, the 28% variance in the second PC is important to consider. By comparing the second EOFs, one can easily say that the variations of the latent heat contributions are responsible for this portion of the total variance. Together, the first and second PCs contain about 97% of the total variance and thus, they undoubtedly dominate the MJO oscillations in the variation of the warm pool SST.

The analysis of the semiannual cycle is also displayed in Fig. 7 and Table 2. For this case, the bandpass filter covers the range between 120- and 360-day periods. The PCA results reveal that the main driving force behind the semiannual cycle of SST is now the latent heat term as can be inferred from the EOF plots. The first principal component contains about 67% of the total variance. This is followed by the shortwave contribution with a partial variance of 30.6. Similar to the 10–25- and the 30–60-day oscillations, both evaporative and shortwave fluxes contain almost the entire variance of the semiannual oscillations. The dominance of the latent heat term in the seasonal timescale was also pointed out by Zhang and McPhaden (1995) using observations.

In summary, it is concluded that the variations of the warm pool SST discussed above, appear to arise primarily as a result of air–sea interaction. Among the surface forcing terms, the shortwave and latent heat terms are the primary forcings on the warm pool SST. For the timescales extending from 10–25 to 30–60 days, the shortwave contribution is found to be the primarily important, and the evaporative contribution is found to be a secondarily important factor. However, as the timescale becomes longer, the evaporative contribution becomes more important than the shortwave contribution. In relation with these findings, one can argue that there is a mutual cancellation between shortwave and evaporative flux contributions in timescales roughly between 60 and 120 days. As seen in the spectrum of the warm pool SST in Fig. 4, this results in very small variations on such timescales.

Knowing the importance of these terms is the first important step to unveil the possible regulation of the warm pool SST. Yet, we still have to establish physical and dynamical reasoning of the variations. For this, each contribution to the shortwave and latent heat term ought to be examined. These were previously defined as indirect contributions. In the next two sections, this issue will be investigated thoroughly. The outcome of this investigation and the results discussed above will be used to construct the regulation mechanism of the warm pool SST.

6. The elements of the evaporative flux contribution

As described in the previous section, the evaporative fluxes are the main factor determining the warm pool SST at the semiannual timescale and is secondarily important at shorter timescales. One major result that may be derived from this is that the direct response to the annual cycle of the incident solar radiation does not entirely work over the western Pacific. Since the evaporative flux term was found to be the primary contributor in this timescale, it becomes natural to seek similar semiannual variations in its contributors. In this way, it is possible to find the very first element that carries the semiannual oscillation. The evaporative flux in FSUCGSM is defined as

Q_eρLC_Eu_aq,(3)

where u_a is the surface wind, Δq = q₀ − q_a is the moisture deficit where q₀, q_a are the surface and reference level specific humidity, respectively, and C_E is the stability-dependent bulk coefficient. The variations in the bulk coefficient are usually neglected (see, e.g., Zhang and McPhaden 1995). Similarly, our results indicate that the semiannual variations of bulk exchange coefficients are about 3 orders of magnitude smaller than those of surface wind speed and humidity deficit. Therefore, for the sake of simplicity, it is assumed that these coefficients remain constant throughout the time integration. Then, only the time evolution of the surface wind speed and moisture deficit should be enough to explain the semiannual variation in the evaporative fluxes and thereby in the warm pool SST. The individual contributions from the surface wind and humidity deficit, however, are not straightforward to obtain, since Eq. (3) is not a linear equation. Using simple perturbation technique, the linearized contributions of the surface wind and moisture deficit can be obtained as

View Expanded

where the bars and primes denote the mean and perturbed states, respectively. In addition to C_E, we also assumed that ρ₀ is constant. The first term on the right-hand side is the contribution from the moisture deficit and the second term is the contribution from the surface wind speed. This new equation enables us to investigate the contributions independently, yet it is only a first-order approximation to the actual relation in Eq. (3). Thus, some caution needs to be exercised when using this method, since it may be misleading in the presence of strong nonlinearity, especially at high frequencies. The following discussions are based one the terms in Eq. (4).

7. The elements of the shortwave contributions

Earlier, we had found three distinct timescales in which the warm pool SST shows marked variations. It was shown that, of these variations, in addition to the diurnal variation, 10–25- and 30–60-day oscillations were primarily functions of shortwave contribution. We now attempt to determine the processes that give rise to these variations. To do that, one has to first establish a clear understanding of the shortwave parameterization in our numerical model. We can divide the parameterization of the shortwave contribution into three different components, namely, O₃, clear sky, and cloudy sky contributions as explained in Cubukcu (2001). The influence of O₃ absorption is a function of atmospheric moisture content in addition to the model-independent O₃ content. Therefore, O₃ absorption is calculated for the clear and cloudy air separately.

The second component of the shortwave contribution is the clear-sky absorption. For this part, the effective optical path of the scaled water vapor is the most important component for clear-sky absorption. Here, the term clear sky refers to a nonscattering atmosphere. Therefore, the only source for the clear-sky upward fluxes is the ground reflection. Considering that the albedo of the surface over the tropical Pacific does not change much throughout the year, variations of the clear-sky fluxes can be attributed mostly to those of the optical path.

Using the information given above, it may be assumed that the net solar flux at the surface of the ocean originates from two main components, namely, clear-sky and cloudy-sky fluxes. The O₃ absorption may be regarded as an integral part of these components as explained above.

8. The regulation of the maximum SST

So far, the important variations in the warm pool SST have been discussed using area-averaged quantities. Recall from the introduction that the most intriguing aspect of the warm pool SST is the capping of the maximum values around 32°C. Therefore, the main topic of this section is to see whether or not we can explain the variations of the maximum SST using those of the area-averaged values. The time variation of the maximum warm pool SST is shown in Fig. 12. This figure clearly illustrates that the types of variations found in the area-averaged warm pool SST also exist in the variation of the maximum SST. All the important variations (semiannual, 30–60, and 10–25 day) can be seen easily. Moreover, the time variation of the maximum SST displays at least three peaks reaching around 32°C during the early summer and winter seasons. As claimed in the observational studies, the maximum SST appears to be capped around 32°C. However, it should be noted that all numerical simulations like ours contain adjustable coefficients that can distort the results dramatically. Therefore, it is hard to prove numerically that the warm pool SST is capped below 32°C. The promising part of our simulation is that it is in very good agreement with the observations over the warm pool area as discussed previously. Such level of accuracy encourages us to construct a regulation mechanism using the findings gathered so far from the FSUCGSM.

On the 10–25- and 30–60-day timescales, the wave dynamics of the equatorial atmosphere determines the nature of the air–sea interaction and thereby the variations in the SST. The equatorial wave disturbances have widely been studied (Wyrtki 1975; Emanuel 1987; Murakami and Sumathipala 1989; Takayabu 1994; Fasullo and Webster 2000). Even so, the mechanism that triggers these disturbances remains unresolved. Nevertheless, these studies have well established that the 10–25- and 30–60-day oscillations are often associated with easterly traveling waves. Many of these studies have shown that these equatorial waves often alter the atmospheric circulation causing successive westerly and easterly wind anomalies. Moreover, it was shown (Wyrtki 1975; Emanuel 1987; Fasullo and Webster 2000) that the westerly wind anomalies are usually associated with the wet phase of these disturbances and easterly anomalies are associated with the dry phase of the disturbances. Such anomaly wind fields were also noted in our simulation (not shown here).

Considering that the warm pool region is continuously influenced by the wave disturbances, it is inescapable to think that these disturbances constitute the warm pool SST. Ideally, this controlling mechanism works in a way that any increasing tendency of the warm pool SST during the dry (easterly wind anomalies) phase is followed by a decreasing tendency during the wet (westerly wind anomalies) phase. The compensation between two successive phases is often disturbed by other processes such as wave–wave interaction. Therefore, the net variations on each timescale may display a rather wide range of magnitudes.

Unlike the wave disturbances discussed above, the semiannual timescale is dictated heavily by the convective processes associated with the ITCZ. As seen in Fig. 6, the largest variations of the warm pool SST occur in this timescale. This variation is much more recurrent than those of the 10–25- and 30–60-day timescales probably because the main cause is rooted in the annual variations of the incoming solar radiation. Unless there is a sizable change in the incoming solar radiation, the variations of the warm pool SST at the semiannual timescale are strictly controlled.

In summary, the 10–25-, 30–60-day, and semiannual oscillations may impose certain restrictions on the maximum SST over the warm pool. One of the clear evidences for this is their continuous influences of these variations over this region. This may be a necessary condition for capping the maximum SST but not sufficient enough to complete the discussion. The magnitudes of these oscillations should also be bounded such a way that they will not cause the SST to shoot over the maximum limit. One way we can check the magnitude is to plot the scatter diagrams of the SST versus the most influential variables (i.e., atmospheric moisture, surface wind, and moisture deficit). These plots will provide sufficient information on the variation of these important elements when the SST gets closer to its upper limit. The diagrams are displayed in Figs. 13–15.

First, from the scatterplot of SST versus the optical path of the water vapor, it is possible to identify three distinct regimes. The first regime is when the SST is between 27° and 28°C, where the atmospheric moisture content is at its lowest level. This is followed by the convective regime where SST is between 28° and 29.5°C. In this region, atmospheric moisture content increases substantially with increasing temperature. Beyond 29.5° is the third regime where the moisture content displays a decreasing trend. These regimes are remarkably similar to those found observationally (Graham and Barnett 1987; Waliser and Graham 1993; Waliser 1996; Lau et al. 1997) as discussed in the introduction. Second, for Figs. 13–15 the great portion of the scatterplots lies roughly between 28° and 31°C and the contributions hardly lead SST beyond these values. Furthermore, all these figures clearly display that the variations beyond these threshold temperature values are very small as compared to the variations in the convective regime. This can be depicted judging from the spread of the scatterplots at high and low SSTs.

All these results point to the important conclusion that the tendencies that drive the warm pool SST have their most active state, and thereby the largest variations, when the SST is roughly in the range of 28°–31°C. Beyond this range the contributions, and thereby the SST, have very low variability. Judging from the scatterplots, it might also be argued that the warm pool SST mostly stays between 28° and 31°C. Thus, it is probably more appropriate to describe the first and third regions as transition zones rather than as 2 of 3 regimes of the warm pool SST. In the nonconvective region, where the SSTs are around the minimum values, the sea surface receives maximum amount of radiation. Although the surface winds are high, the evaporative cooling is negligible because of the low moisture deficit due to the cold temperatures. Therefore, solar heating causes the SST to increase. Consequently, convective measures such as increasing moisture content of the atmosphere, developing low-level convergence (decreasing winds), and increasing moisture deficit become prominent when the SST reaches around 29°C. In the temperature range of 29°–30°C, these convective measures clearly define the convective activities. All these features can be seen in the time series of the direct and indirect contributions shown earlier. The SST range of 30°–32°C is characterized by the diminishing convective measures. The role of the evaporative cooling and solar heating at this stage of the convective phase is to cool the SST. Thus, SSTs between 30° and 32°C are the highest values that can ever be reached. Despite the cooling of the SST, convection may continue until the SSTs decrease to a point where the atmosphere cannot sustain upward motion and all available potential moisture in the atmosphere is consumed by demoisturization processes. SSTs approximately from 29° to 32°C and then from 32° to 29°C, respectively, coincide with the intensification and weakening stages of the convective systems. Also, maximum SSTs tend to occur during the transition from intensification to weakening stages. After the passage of the convective systems, the conditions for the dry phase begin to emerge. The initial stage of the dry phase involves further cooling due to very large evaporative fluxes. As SSTs become cooler the evaporative cooling becomes smaller. Lack of convective clouds lets the surface of the ocean receive solar radiation that eventually causes the SST to increase. Thus, similar to the convective phase, the nonconvective phase coincides with SSTs approximately from 29° to 27°C and then, from 27° to 29°C, respectively, the initial and final stages of the dry phase.

The interpretation given above explains the main mechanism of how warm pool SST oscillates between its maximum and minimum values. It should be noted that the maximum and minimum for each cycle of the modes may differ slightly. For example, during early summer, when SSTs are close to the upper limit, the 10–25- or 30–60-day oscillations will probably let the SSTs oscillate between 32° and 29°C. In this case, SSTs are concentrated between 29° and 32°C. Conversely, during early winter, SSTs are very close to the lower limit and are concentrated between 27° and 29°C. Apart from these extreme cases, SSTs generally stay between 28° and 31°C as can also be perceived from the dark areas of Figs. 13–15.

In summary, the results from our numerical simulation suggest that the warm pool SST oscillates between certain threshold values. As implied by these variations, it is highly possible that there is an upper limit to the warm pool SST. Furthermore, our results also suggest a lower limit around 27°C that can be anticipated from Figs. 13 to 15. The main process that maintains this balance is the convection associated with these important timescales described before. These oscillations seem to be continuously reproduced in time. The continuity of the semiannual variations is straightforward since it is related to the annual variation of the solar radiation. In the presence of the equatorial waveguide the intraseasonal variations are also expected to be continuous in time. Such conditions are well represented by our simulation.

9. Summary and conclusions

The FSUCGSM is utilized to study the possible regulation of the maximum SST over the warm pool. For this goal, we first identified all terms that contribute to the warm pool SST. This procedure is done in a way that the SST balance equation becomes residue free. The results of a 1-yr time integration of the FSUCGSM are used to obtain the time evolutions of the residue-free warm pool SST and its contributors. To ensure the validity of the coupled model, the NCEP SST analysis for the period of the model integration is compared with the model-predicted SST. Overall the comparison displayed good agreement with the NCEP analysis. The best agreement was seen over the western equatorial Pacific. Such good agreement has demonstrated that the FSUCGSM is able to capture most of the dynamical and physical processes over the equatorial Pacific on an annual timescale.

It was found that the warm pool SST is derived mainly by three important types of oscillations, namely, semiannual, 30–60-, and 10–25-day oscillations. Then, the results of combined Butterworth bandpass filter and EOF analysis have revealed that the tendency of the solar radiation is the primary cause of the high-frequency oscillations (10–25 and 30–60 day) and the secondary cause for the low-frequency oscillations (semiannual). Moreover, the evaporative cooling was found to be the primary cause of the low-frequency oscillations and secondary cause of the high-frequency oscillations. Thus, it was concluded that the possible regulation mechanism of the warm pool SST can be explained by examining only the tendencies of the solar radiation and evaporative cooling.

The variations of the cloud cover in relation with the atmospheric moisture content, surface wind, and humidity deficit were found to be the most important factors relating the changes in the SST to those in the convective activities. In relation with the atmospheric moisture content, it was found that the cloud shortwave forcing plays the most crucial role on the solar radiation at the surface.

In the light of these findings, it is proposed that the warm pool SST may have an upper limit as suggested by previous authors. In addition, our findings indicated that there may also be a lower limit at around 27°C.

To our knowledge, our study is the most detailed numerical study ever done concerning the regulation of the warm pool SST. We have benefited from both observational findings and model simulation. Unlike the previous studies, we have used coupled global atmosphere–ocean models and analyzed each contribution individually. As stated by earlier authors, one of the advantages of using coupled GCMs is that the comparison among the terms in the SST equation can be done efficiently without worrying about datasets being from different sources. Also, all diagnostic and prognostic variables that are indirectly related to the SST can be analyzed instantly. However, there is more to be done. For instance, the resolution we have used in our study does not permit the narrow-canal coastal currents that seem to be very important in simulating the ocean circulation. Also, the cloud parameterization scheme used in this model does not adequately consider mesoscale convective processes abundant over the warm pool region. Thus a similar study with higher resolution and a better cumulus parameterization scheme might be the next step toward understanding the real nature of the warm pool SST.