1. Introduction

The gross warming of tropical surface ocean and land temperatures outside the tropical Pacific (hereafter referred to as the remote Tropics) to El Niño is a well-known and robust response (e.g., Yulaeva and Wallace 1994; Klein et al. 1999). As an example, we show ocean (Fig. 1a) and land (Fig. 1b) surface temperature anomalies during March–May 1998, one season after the boreal winter peak of the strong El Niño event of 1997/98. A gross Tropics-wide warming of around 1 K is evident over most of the remote region. This warming is neither uniform nor entirely ubiquitous over the remote Tropics: in particular, sea surface temperatures (SSTs) directly west of Angola tend to cool slightly during El Niño (as noted by Enfield and Mayer 1997; cooling in this region is also apparent in Fig. 1). However, the gross sense is one of warmer remote tropical surface temperatures after an El Niño event.

What causes this gross warming? Previous studies have focused on teleconnections mediated by the atmosphere [the “atmospheric bridge” concept of Lau and Nath (1996)], although the detailed mechanisms underlying this communication are not well understood. Previous observational studies of the El Niño impact over the remote tropical oceans [e.g., Curtis and Hastenrath (1995) and Enfield and Mayer (1997) for the tropical Atlantic; Yu and Rienecker (1999) and Alexander et al. (2003) for the Indian Ocean] support the atmospheric bridge concept by arguing that atmospheric circulation changes acting on surface fluxes (in particular, wind speed effects on latent heat flux and radiative effects arising from changes in cloudiness) or through ocean dynamical processes (in particular, Ekman transports and mixing and large-scale dynamical processes near the equator) are largely responsible for the SST anomalies. Klein et al. (1999) solidified these atmospheric bridge arguments by showing that SST variations over the tropical Atlantic, Indian, and southeast Asian Oceans during El Niño are consistent with inferred variations in the surface flux. A theme that emerges from these studies is the region-specific nature in the causes of surface temperature change and their characteristic evolutions over the ENSO cycle. In particular, latent heat flux changes associated with variations in the northern trade wind intensity during the boreal winter appear to be of primary importance for the warming in the tropical North Atlantic (e.g., Curtis and Hastenrath 1995; Enfield and Mayer 1997; Klein et al. 1999), whereas the causes of Indian Ocean SST variability appear mixed, with shortwave radiation (e.g., Klein et al. 1999), wind speed (e.g., Shinoda et al. 2004), and ocean dynamics (e.g., Murtugudde and Busalacchi 1999; Yu and Rienecker 1999) all appearing to play a significant role. The regional variations are likely associated with processes that determine the mean climate state and seasonal cycle as well as the internal variability of the climate.

We take a different tack in this study by focusing on the similarity in response to El Niño over the remote Tropics: that is, why do (almost) all remote tropical regions, both land and ocean, warm to El Niño? Region-specific explanations tend to be unsatisfactory since, in the spirit of Okham’s razor, there likely exists a more general explanation for the gross warming effect of El Niño over all remote tropical regions. The argument against a general explanation is that surface flux budgets are known to vary substantially from region to region (as demonstrated in the above cited studies), with different components contributing to the warming in different regions. A problem that we foresee with this argument (as elaborated in section 2) is that the individual flux components are large terms compared to the net flux that often compensate each other, leaving a small residual as the net surface flux anomaly. Because of this compensation, it may not be sufficient to examine subsets of the individual flux components to infer the causes of SST variations. This poses a quandary for observational studies as surface flux datasets are nowhere near complete in either space or time—in particular, we do not have adequate global observations of surface longwave and shortwave fluxes for interannual studies. Furthermore, observational errors may mask the signal: for example, Chiang and Sobel (2002) noted that a typical 10 W m⁻² anomaly in latent heat flux can be driven by an air–sea humidity difference anomaly of around 0.4 g kg⁻¹, which is smaller than the mean random error of ∼1.1 g kg⁻¹ for marine 10-m specific humidity measurements from voluntary observing ships (Kent et al. 1999).

This paper explores the case for tropical tropospheric temperature warming caused by El Niño as the explanation for remote tropical surface warming. Our starting point is the observational paper by Yulaeva and Wallace (1994), which documented tropospheric temperature (TT) warming throughout the tropical belt to El Niño, with an amplitude of almost 1 K for strong El Niño years and lag relative to the peak El Niño phase by around 3 months. An example of the warming response to El Niño is shown (Fig. 1c), namely, the March–May 1998 air temperature anomalies at 400 mb. The TT warming is understood as a dynamical response to atmospheric heating caused by El Niño; because the tropical atmosphere cannot maintain strong temperature gradients, the large-scale dynamics acts to spread the heating across the entire tropical belt (Wallace 1992; Sobel et al. 2002). Yulaeva and Wallace noted furthermore that the mean TT anomalies are highly coherent with fluctuations in surface air temperatures over tropical land masses and SST anomalies over the tropical Indian and North Atlantic Oceans, which led them to propose that the tropical TT and surface temperature fluctuations represent a “thermodynamic response to the perturbations in the surface heat balance induced by the ENSO cycle in the eastern equatorial Pacific” (Yulaeva and Wallace 1994). Brown and Bretherton (1997) proposed that moist convection links the remote tropical TT to the surface, invoking the strict quasi-equilibrium hypothesis which argues that, over time scales sufficiently longer than the convective time scale, convection homogenizes moist static energy perturbations in the vertical. A previous study by one of us (Chiang and Sobel 2002, hereafter CS02) generalized these concepts by arguing that ENSO communicates to the remote Tropics through the propagation of TT and that the remote ENSO impact can be thought of as the adjustment of the remote tropical climate to the TT perturbation—hereafter, we refer to this mechanism as the “TT mechanism.”

We explore the mechanisms behind the remote tropical surface warming to El Niño with a focus on the TT mechanism. CS02 explored the TT mechanism in the limited framework of a single-column model (a summary of CS02 will be given in section 3); here, we examine the more realistic context of an atmospheric general circulation model (AGCM) as well as an intermediate-level-complexity tropical circulation model, with both coupled to a fixed-depth thermodynamic slab ocean outside of the tropical Pacific. Because of the limitations of our model setup, we do not examine the role of ocean dynamical feedback. Our main conclusion is that the remote tropical ocean surface temperature warming during El Niño in the models used here is consistent with the TT mechanism proposed by CS02; in fact, it appears to be the dominant mechanism. Tropospheric temperature is also fundamental for the remote tropical land warming to El Niño, although we do not yet fully understand the details of its operation. Given that the climate processes modeled here are relatively complete, it suggests that the TT influence on remote surface temperature has some basis in reality.

2. AGCM simulations of the remote tropical surface warming to the 1997/98 El Niño

We use the Community Climate Model version 3.10 (CCM3; Kiehl et al. 1998) coupled to a constant-depth 50-m slab ocean model outside the tropical Pacific. The model is widely used; we refer the reader to Kiehl et al. (1998) and Bonan (1998) for descriptions of the atmospheric and land surface components, respectively. The atmospheric model resolution used corresponds to a triangular truncation of 31 basis functions in the meridional and 15 in the zonal, equivalent to 48 grid points in both latitude and longitude. The CCM3 standard 18 levels are retained for the vertical. Although the resolution is relatively coarse, previous studies using the same configuration have yielded reliable simulations of climate (e.g., Yin and Battisti 2001; Chiang et al. 2003). Moreover, for the scales under consideration here, the resolution is adequate. The land surface component (Land Surface Model 1.0) is a one-dimensional model for energy, momentum, water, and carbon dioxide exchange and allows for hydrological and vegetation differences as well as specification of lakes and wetlands; land surface and subsurface temperatures are determined on the basis of energy constraints. The soil is divided into six layers of 0.1, 0.2, 0.4, 0.8, 1.6, and 3.2 m from top to bottom, connected by standard thermal diffusion with a diffusion coefficient depending on the soil type. For the slab ocean, a specified spatially varying, climatological monthly flux correction, a proxy for ocean heat transport convergence, is applied so that the simulated SST seasonal climatology closely matches the observed; a description of this procedure can be found in Chiang et al. (2003). We specify SST in the region of the tropical Pacific from 20°S to 20°N and 110°E to the west coast of the Americas and merge the SST field linearly with the slab ocean SST over a 10° interval, wherever appropriate. A 57-yr control run is obtained by integrating the model with the observed SST seasonal cycle climatology, with the first 15 years discarded as spinup.

We simulated an ensemble of 1997/98 ENSO conditions by specifying anomalies of the January 1997–December 1998 period, with a linear ramp up of 3 months (i.e., no anomalies were applied for January 1997, 33% of the observed amplitude for February 1997, and so on), and a 3-month ramp down between October 1998 and December 1998 (note: Table 1 lists and briefly describes all simulations done for this paper). The simulations were continued for another 2 years past the end of 1998 but with SST climatology specified in the tropical Pacific; this allows for the memory of the 1997/98 ENSO warm event in the model climate system to play itself out. The anomalous SSTs were derived from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996). Twelve ensemble members were simulated, with each starting from different initial conditions for 1 February 1997, and ensemble-averaged anomalies were computed. The initial conditions were taken from a long-term climatological seasonal cycle simulation integrated with climatological seasonal cycle SSTs imposed in the tropical Pacific and a slab ocean elsewhere.

The model surface temperature anomaly for March–May 1998 is shown in Fig. 2a. (Note: by “surface temperature” in CCM3 we mean the temperature of the slab ocean for ocean surfaces, and the temperature of the topmost soil layer over land.) There are spatial details of the CCM3 Atlantic and Indian SST response that do not match the observations, in particular the cooling over the southeast tropical Indian Ocean. The existence of these differences is not surprising given that ocean heat transport changes are not modeled. For the Indian Ocean in particular, Murtugudde and Busalacchi (1999) and Yu and Rienecker (1999) showed that ocean dynamics plays a significant role in the SST warming during El Niño. However, the gross warming over the remote Tropics is clearly present—even with the deficiencies in the model ocean—with approximately similar magnitude. To view temporal features of the warming, we examine spatial averages of the surface temperature anomaly over the remote tropical region spanning 20°S–20°N from the west coast of the Americas to 110°E, marked by the dashed lines on Fig. 2a (hereafter, we more precisely define this region to be the remote Tropics; ocean and land subsets of this region are the remote tropical ocean and remote tropical land, respectively). The top left and top center panels in Fig. 3 show the surface temperature averaged over the remote Tropics from the NCEP–NCAR reanalysis and the mean of the ensemble of CCM3 model runs, respectively. The temporal features of the warming are well captured by the model, including the phasing of the peak anomaly around boreal spring of 1998, although the peak amplitude of the surface warming is too low (by about 0.3 K) in the model simulation. However, we note from sensitivity experiments [where we repeated the 1997/98 El Niño simulations but varied the mixed layer depth (MLD)] that the amplitude is in part a function of the mixed layer depth chosen, with a shallower mixed layer depth associated with an increased peak remote tropical surface temperature anomaly. Since the mixed layer depth of the standard setup (50 m) is deeper than the observed 20°S–20°N average mixed layer depth of around 33 m according to the World Ocean Atlas 1994 (Levitus and Boyer 1994), the underestimate is to some extent attributable to the mixed layer depth that we chose. Additionally, some underestimation may arise from ocean dynamical feedbacks to the ENSO signal, as discussed in Murtugudde and Busalacchi (1999) for the Indian Ocean, although Venzke et al. (2000) found in a coupled GCM that surface fluxes still dominate the Indian Ocean SST response to ENSO. While ocean dynamical feedbacks are beyond the scope of our study, there is enough similarity between the model and observational responses to make it useful to understand how this model response comes about.

We first examine the surface fluxes for clues to the remote surface warming. The net flux anomaly into the surface (G′) is given by

View Expanded

where LH′, SH′, CSR′, and CRF′ are the latent heat, sensible heat, clear-sky radiation, and cloud radiative forcing flux anomalies, respectively. We partition the radiation into its clear and cloud forcing components, rather than the more typical shortwave and longwave components because we wish to gauge the impact of clouds separately from the impact of tropospheric temperature (and associated water vapor) increase on the radiative fluxes. Over the slab ocean, the net flux anomaly equals

View Expanded

where T′_s is the slab ocean temperature anomaly and c = c_p ρ H, where c_p = 3.93 × 10³ J kg⁻¹K⁻¹ is the mass heat capacity of ocean, ρ = 1.026 × 10³ kg m⁻³ is the density of ocean water, and H is the mixed layer depth. Over land, G′ is generally considerably smaller than each term on the rhs of (1) for monthly and longer term means because of its small heat capacity. Figures 4a–d show the individual components of the surface flux anomalies summed over February 1997 through April 1998—in other words, the fluxes that approximately produce the March–May 1998 surface temperature anomalies shown in Fig. 2. All surface flux components contribute significantly over the remote tropical land, whereas over the remote tropical ocean the dominant terms are the latent flux and cloud radiative forcing, with the clear-sky radiation and sensible heat flux contributing to a weak warming. Examination of the shortwave and longwave contributions to the radiation (figures not shown) demonstrates that the clear-sky radiation is dominated by the longwave component with almost no contribution by the shortwave component, whereas the cloud radiative forcing is dominated by the shortwave component, with some degree of cancellation by the longwave component. We also note (figure not shown) that the all sky (sum of clear sky and cloud forcing) longwave componment is largely determined by the clear sky component, that is, the cloud contribution to the longwave component is less important.

Interpretation of the radiative fluxes aside, the point that we underscore here is, unlike the surface temperature response to El Niño, no simple story emerges from the component fluxes: the flux components vary considerably from region to region, and sizable compensation occurs between the various flux components.

3. Role of tropospheric temperature

The complex behavior of the surface fluxes in the previous section begs the following question: Why should the net response be a relatively simple remote Tropics-wide surface warming? The development of a widespread remote tropical surface warming response suggests a more fundamental control that is not readily inferred from the surface flux budget. We investigate tropospheric temperature for that role, following the mechanism proposed by CS02 briefly introduced in section 1. We highlight the main points of the CS02 study in the next few paragraphs; the reader is referred to the original reference for details.

CS02 explored the potential impact of TT on the remote tropical climate by imposing TT anomalies above the boundary layer in a single-column model (with radiative and convective parameterizations incorporated) coupled to a simple fixed-depth slab ocean and examining their impacts. The vertical profile of the TT anomaly chosen was derived from NCEP–NCAR reanalysis (Kalnay et al. 1996) and used because it is typical of variations occurring during El Niño; essentially, the anomaly was characterized by a “first baroclinic mode” structure dictated by variations in deep convective forcing, with the maximum amplitude in the midtroposphere (see Fig. 3 of CS02). It was found that, when the model representing deep convective states was forced by observed monthly mean TT anomalies over 1979–99, the model surface temperature response qualitatively resembled the observed remote tropical ocean temperature interannual variations, namely, warming of around 1 K during El Niño years with a lag of a few months relative to the TT forcing. Diagnostics of the model response showed that the variations in the single-column model were caused primarily by latent heat flux anomalies resulting from changes to the boundary layer humidity and the air–sea specific humidity difference. The boundary layer humidity (and hence the moist static energy) perturbation varied synchronously with the imposed tropospheric temperature variations. Also, moist convection was essential to the downward communication of the free tropospheric signal: the same TT perturbations applied to the model in a nonconvecting mean state did not produce significant surface temperature variations.

The close variation of the boundary layer moist static energy perturbation with the free tropospheric temperature variations, and the requirement that moist convection be present for this downward communication, suggested that quasi-equilibrium arguments (as proposed by Brown and Bretherton 1997) were likely applicable. However, in violation of strict quasi equilibrium, CS02 noted significant convective available potential energy (CAPE) variations associated with the TT variations, giving rise to variations in convection. CS02 argued that the CAPE variations resulted from surface thermal inertia that also influenced the boundary layer moist static energy variations through surface fluxes, effectively pulling the boundary layer out of a rigid lockstep with moist static energy variations in the free troposphere.

Despite the inapplicability of rigid strict quasi-equilibrium arguments, we argue that quasi equilibrium is still a useful concept for understanding the El Niño influence on the remote surface temperature, given the relatively long time scales (i.e., months) of the El Niño influence and the link between boundary layer and free tropospheric moist static energy perturbations, a concept that has heretofore not been seriously considered in understanding the El Niño influence on the remote tropical climate. The CS02 prediction of surface warming mediated through boundary layer humidity is yet to be supported from observational analysis, though CS02 pointed out the difficulty of detecting the atmospheric humidity signal given the errors involved in the measurements of specific humidity. We now examine the CCM3 simulations as a more realistic framework in which to evaluate the TT mechanism.

How well is TT modeled in the CCM3 simulation? Figure 2b shows the March–May 1998 TT anomalies in the CCM3 1997/98 simulation, which should be compared to the observations shown in Fig. 1c. While regional features of the TT anomaly field differ between the model and observations, a Tropics-wide warming of approximately correct magnitude is clearly evident in both. The model vertical profile of March–May 1998 TT anomalies averaged over the tropical belt 20°S–20°N (Fig. 2c, solid line) shows TT anomalies increasing from around 0.5 K at the surface and peaking at around 1.8 K near 200 mb, before dropping to around –0.4 K around 50 mb. It resembles a positive perturbation of the moist adiabatic vertical structure, a result also reported in CS02 using NCEP–NCAR reanalysis. Comparison with the same TT anomaly but from NCEP–NCAR reanalysis (Fig. 2c, dashed) shows a good match up to around 300 mb, although below 300 mb the observed TT anomaly cools more dramatically in the stratosphere. However, since the TT mechanism is mainly concerned with temperature anomalies below the tropopause, the model–data divergence above the tropopause should not unduly affect our study.

The lower-left and lower-center panels in Fig. 3 show time series of the 200–800-mb averaged air temperature for the remote tropical region for the NCEP–NCAR reanalysis and the CCM3 simulation, respectively. The model realistically simulates TT although, as with SST, the amplitude is underestimated (to reconcile this result with Fig. 2c, note that Fig. 2c is for the entire tropical belt, not just the remote Tropics—it implies that the model Pacific TT is warmer than in observations). There is a close correspondence of the time evolution of the SST and TT anomalies that suggests close association between the two. To demonstrate more concretely the role of TT, we repeat the 1997/98 ENSO simulations but with a change in the specified tropical Pacific SST designed to reduce the ENSO impact on TT. Following Sobel et al. (2002), who suggested and found observational evidence for a linkage between interannual variations in tropical TT and interannual variations in the warmest SST regions of the tropical Pacific, we uniformly subtract a constant value from the imposed tropical SST anomalies in the tropical Pacific such that the monthly SST anomaly area averaged over the imposed SST region (20°S–20°N and 110°E to the west coast of the Americas) is zero. In other words, the spatial gradients of the SST anomalies are preserved, but the anomalous SST field has been reduced by a uniform amount. This adjustment has the effect of reducing the warmest SST regions in the tropical Pacific. The results for remote tropical surface temperature and TT, illustrated in the third column in Fig. 3, show that this modified SST forcing reduces the amplitude of both TT and remote tropical SST by around 50%.

An interesting feature of this simulation is that the “anomalous Walker circulation”—the zonal circulation change associated with anomalous zonal displacements of equatorial Pacific convection—often implicated for the tropical ENSO teleconnection (e.g., Saravanan and Chang 2000; Chiang et al. 2000) is apparently not responsible for the changed teleconnection in this case. Figures 5a–b show the precipitation anomalies in the tropical Pacific from the “standard” 1997/98 ENSO simulation and from the uniform Pacific SST reduction case, for November 1997–January 1998 period when the El Niño was at its peak. The striking feature of this comparison is that the Pacific precipitation anomalies are not very different, although there is some reduction in the amount of tropical Pacific precipitation in the uniform reduction simulation, mostly around 10°N and 10°S in the western part of the basin. Since the shift in convection associated with El Niño is still present, the reduction to the remote tropical warming in this perturbation experiment is unlikely to be attributable to changes in the anomalous Walker circulation—it comes, rather, from the relatively small decrease in total tropical Pacific precipitation.

4. Adjustment of the remote Tropics to an abrupt introduction of “El Niño” forcing

We now examine transients of a CCM3 50-m slab ocean simulation in which El Niño–like forcing is suddenly switched on in the model by abruptly introducing tropical Pacific SST anomalies representative of a peak El Niño (we chose December 1997 in this case). To reduce the noise level, we ran an ensemble of 20 members of the perturbation and control simulations, each starting on 1 February but with different initial conditions. Five-day ensemble-averaged perturbation fields were computed, and 5-day-averaged control fields were subtracted from the perturbation fields to obtain pentad anomalies. Figure 6 shows the tropospheric temperature and surface temperature anomalies averaged over days 1–5, 6–10, and 81–85 following the introduction of anomalous Pacific SST forcing. The fast propagation of tropospheric temperature anomalies originating from the eastern Pacific is notable, with a TT wavefront traveling eastward along the tropical waveguide. The structure of the TT anomalies appears consistent with the propagation of a transient Kelvin wave forced from the anomalous convection in the tropical Pacific. A rapid onset of remote tropical land surface warming is evident in the first two panels in Figs. 6a–b. The remote tropical oceans warm more slowly, presumably because of the ocean thermal inertia, but warming over the remote tropical oceans especially near the equator is quite apparent by days 81–85 (Fig. 6c).

The spatial congruence of the TT and the surface warming anomalies is not perfect: in particular, there are large sections of the southern tropical Atlantic and Indian Oceans that have not warmed by days 81–85. The CS02 hypothesis may offer some explanation for the lack of warming in the southern Atlantic and Indian Oceans in terms of the mean state, as moist convection is required to communicate the TT anomaly to the ocean surface. Since the southern tropical oceans are generally cooler than their northern counterparts and a sizable fraction of these oceans are covered with stratus decks, the troposphere–surface convective link in these regions may be quite weak. There are indeed maxima in the CCM3 climatological low clouds over the regions in question (not shown), although they tend to be somewhat farther south of the southern tropical ocean regions that have not appreciably warmed by days 81–85 (Fig. 6c); thus, the cool SST–thick stratus cloud regions are not entirely collocated. In examining the causes of the SST structure, we found increased surface southeasterly trades collocated with the southern Atlantic and Indian Oceans that did not show appreciable warming. It suggests that regional circulation effects play a significant role in the SST response over the southern tropical oceans—in particular, anomalous SST gradients that form over the tropical Atlantic and Indian Oceans and drive surface circulations, which in turn increase evaporation over the southern tropical oceans. Another possibility is that the strong southeasterly trades are very efficient at removing moisture out of these regions, promoting stronger evaporation. Although such regional circulations are quite important, our main objective is to point out the general collocation of the TT and SST warming in the deep Tropics as a whole.

The temporal evolution of spatially averaged anomalies in the remote tropical TT and surface temperature fields further demonstrates the interdependence between the two. Figure 7 shows the TT and surface temperature anomalies averaged over 20°S–20°N for South America, the Atlantic Ocean, Africa, and the Indian Ocean, respectively, from day 1 to day 115 when the anomalies have approximately equilibrated. In all regions, the TT and surface temperature anomalies increase more or less in concert, although land surface temperatures exhibit greater variability than the ocean surface temperatures. Note that the equilibrated TT anomalies decrease from South America eastward, consistent with a damped propagation of TT anomalies from the eastern Pacific. More interesting is the ratio of TT to surface temperature anomalies over land and oceans, around 1:1 for both land regions, and around 1:0.3 for both ocean regions. We interpret the existence of fixed land and ocean ratios of equilibrium TT and surface temperature anomalies as suggestive of an intimate linkage between the two, presumably through deep convection. But how exactly is this linkage realized in the surface fluxes, given the apparent complexity in the surface flux response shown in section 2? We address this issue in the next two sections for remote ocean and land regions.

5. Mechanisms for the remote ocean response

a. Mechanisms that initiate the remote surface ocean warming

We examine the fluxes that initiate the warming over the remote ocean surface to El Niño. We start with a model configuration in which the remote ocean SST is not allowed to respond to or feed back on the El Niño forcing, thereby allowing us to isolate the mechanisms that initiate the remote ocean warming. To do this, we repeat the CCM3 1997/98 simulation but with the remote SST fixed to climatology throughout the simulation, equivalent to an infinitely deep mixed layer.

The remote ocean areally averaged net surface flux for this ensemble simulation (Fig. 8, first column) shows clearly net surface flux heating into the remote ocean surface. (Note: all fluxes shown in Fig. 8 are cumulatively summed from the beginning of the El Niño simulation on 1 February 1997 to the time in question; for example, the cumulative flux for January 1998 is the flux sum from 1 February 1997 through January 1998. This allows us to evaluate the net contribution of the flux over the entire time history of the El Niño forcing.) Using the peak cumulative net surface flux month of June 1998 as a baseline, the heating is clearly dominated by latent heat flux, with measurable contributions by the clear-sky radiation and (to a lesser extent) sensible heat flux. Cloud radiative forcing does not contribute.

What causes this change to the latent heat flux? Using a linearization of the latent heat flux bulk formula following Saravanan and Chang (2000), we decompose the latent heat flux change (Fig. 9, second panel from top) into its wind speed and air–sea specific humidity difference contributions:

View Expanded

where F is the mean latent heat flux, W the wind speed (overbars denote monthly mean climatological values and primes are anomalies about them), Δq the air–sea specific humidity difference, and F′_W and F′_q are the estimated changes to the latent heat flux due to wind speed and air–sea humidity difference changes, respectively. Figure 9 (third panel) shows W ′/W averaged over the remote tropical oceans, and Fig. 9 (fourth panel) shows Δq′/Δq also averaged over the remote oceans. It is clear from this decomposition that it is the air–sea specific humidity difference that is primarily causing the reduced latent heat fluxes over the El Niño forcing period from early 1997 to mid-1998 (there is a brief period of time in January–February 1998 where the reduction in wind speed makes a significant contribution). The change to the air–sea specific humidity difference, in this case, has to come from the change to the boundary layer humidity since the remote ocean surface temperature is fixed in this simulation.

The increase in the boundary layer humidity is reflected in the temporal variations of the remote ocean moist static energy variations in the boundary layer (over 826–1000 mb: Fig. 9, fifth panel), which in turn closely follow moist static energy variations in the free troposphere (over 220–826 mb: Fig. 9, bottom panel). The magnitudes of the free tropospheric and boundary layer moist static energy variations are comparable, suggesting that quasi equilibrium is a good assumption for this comparison. About 60% of the free tropospheric moist static energy increase comes from the increase in tropospheric temperature, and the rest from free tropospheric humidity. The remote ocean latent flux response and the linkages to free tropospheric moist static energy variations are therefore consistent with the TT mechanism proposed by CS02, as highlighted in section 3.

Our interpretation is that equatorial wave dynamics (shown in section 4) communicates the tropospheric warming to the remote Tropics, increasing the moist static energy of the free troposphere. A rapid increase in the water vapor content as a consequence of the warmer tropospheric temperature—Su et al. (2004) estimate the lag between TT and water vapor to be around a week—substantially amplifies the moist static energy increase. Convective linkages compel the boundary layer moist static energy to vary commensurately with the free tropospheric value; in turn, evaporation is reduced because of the increased boundary layer specific humidity. The increased temperature and moisture in the atmospheric column is also consistent with the increase of clear sky downwelling longwave radiation and with the small decrease in sensible heat fluxes out of the remote ocean.

6. Mechanisms for the remote land response

The individual and net cumulative remote tropical land flux responses to the 1997/98 El Niño forcing and their variation with ocean mixed layer depth are shown in Fig. 10 (this is the land region equivalent of Fig. 8). An immediately notable feature is that the net surface flux is a small residual of substantially larger individual flux components that typically cancel: in particular, latent heat flux tends to be a warming influence, whereas sensible heat flux is a strong cooling influence; the role of radiation varies with mixed layer depth. Another notable feature is the substantial variation in all individual flux components as the mixed layer depth is changed.

However, the larger point we make from this analysis is that, despite the significant variations in the individual flux responses, the net cumulative heat flux remains essentially the same for all mixed layer depths as does the remote land surface temperature, suggesting a regulatory mechanism for the surface warming. However, this mechanism is not easily inferred for the tropical land region, and, while we do not make any firm conclusions in this regard, in the next section we will show results of simulations using an idealized tropical circulation model that suggests sensible heat flux as the regulator. We note that the remote tropical land boundary layer moist static energy increase during the boreal spring of 1998 (this is when the remote tropical temperature response to El Niño approximately peaks) is substantially larger than the corresponding free tropospheric temperature variations (figure not shown), so we cannot readily apply quasi-equilibrium arguments here.

7. Idealized simulations of the 1997/98 El Niño

We now demonstrate directly the role of TT on the remote surface temperature warming through idealized simulations of the Quasi-Equilibrium Tropical Circulation Model (QTCM). The model description can be found in Neelin and Zeng (2000). Briefly, the QTCM is a tropical beta-plane model employing analytically derived vertical structure functions of temperature and moisture (one each) and two velocity structure functions. As in a GCM, the QTCM contains parameterizations of microphysical processes like radiation, convection, and clouds. The QTCM has been extensively used for tropical ENSO teleconnection studies (e.g., Su et al. 2001, 2003).

Application of the same 1997/98 ENSO forcing scenario to the QTCM with similarly imposed SST in the tropical Pacific and a 50-m slab ocean everywhere else yields a qualitatively similar remote tropical surface temperature and TT warming (Fig. 11, first column). However, the QTCM response is significantly attenuated compared to CCM3 (in particular, the remote ocean surface temperature response is severely underestimated), and consequently the timing of the peak surface temperature anomaly over the entire Tropics occurs a few months earlier than observed. We think part of the reason why the QTCM response is significantly underestimated is that the QTCM anomalous convective response in the tropical Pacific to the El Niño SST is significantly less than observed (e.g., see Su et al. 2003; Figs. 2b and 4a compare observed and QTCM anomalous precipitation response during January–March 1998). As a consequence, anomalous convective heating in the QTCM tropical atmosphere is reduced during El Niño, with consequently reduced tropical tropospheric temperature anomalies. However, as the ENSO–surface warming teleconnection appears qualitatively correct in the QTCM, we take advantage of the flexibility of this model to execute idealized experiments to elucidate the mechanisms for remote tropical surface warming to El Niño. The QTCM simulations are performed in similar fashion as the CCM3 simulations except that, since the QTCM exhibits substantially less weather “noise” than CCM3, 10 ensemble members are sufficient.

We now show directly the dependence of the remote tropical surface temperature anomalies on TT in the QTCM through a simple manipulation experiment: we repeat the standard 1997/98 ENSO simulation, but at each time step we remove the zonally averaged tropospheric temperature anomaly from each atmospheric grid point between 20°S and 20°N; in other words, the zonally averaged tropospheric temperature in the Tropics is constrained to follow its climatology. However, anomalous zonal gradients in TT (and the circulations associated with these gradients) are preserved. This can be viewed as a more precise version of the 1997/98 CCM3 simulation in section 3 in which the SST anomaly field over the tropical Pacific forcing region is reduced by a uniform amount to eliminate the basinwide SST anomaly. The result (Fig. 11, second column) shows that the remote tropical zonal mean TT anomaly is more or less removed and the remote tropical surface temperature anomaly (Fig. 11d, thick line) is essentially zero. There are zonal asymmetries involved, however: decomposition of the remote response into remote land and ocean contributions shows that the land warms by about half its original value, whereas the ocean cools slightly. Closer examination of the QTCM simulation shows in particular that the equatorial South American response to the 1997/98 El Niño forcing is substantially stronger than in CCM3, and it is this regional effect that likely contributes to the asymmetric remote land response in the above simulation.

We now directly address the role of individual flux components by altering each individually and directly in the 1997/98 simulations. For altering the sensible and shortwave fluxes (hereafter the “altered sensible flux” and “altered shortwave” runs, respectively), we imposed the climatological sensible and shortwave surface fluxes, respectively, on the model simulations over the regions outside the tropical Pacific where SST was imposed. For the “altered longwave” simulation, we fixed the tropospheric temperature and specific humidity as seen by the atmospheric longwave calculation to climatology, also over the regions outside the tropical Pacific. Finally, for the “altered latent flux” simulation, we fixed the atmospheric specific humidity as seen by the latent heat flux calculation to climatology. We did the latter two cases differently from the first two cases as we were unable to produce reasonable 1997/98 simulations with imposed climatological latent or longwave fluxes; fixing those fluxes considerably alters the simulated climate, for reasons we do not yet understand. Figure 12 shows the individual surface fluxes averaged over the remote ocean for each of these idealized perturbation cases, with the altered flux component highlighted with a thicker panel border. For all of the cases, the altered flux component increases the cooling over the remote oceans. Despite this, the ocean surface temperature response (as evidenced by the net cumulative flux anomaly) for each of the simulations, except for the altered latent flux simulation, undergoes a warming comparable to the standard 1997/98 simulation. For the altered sensible and shortwave experiments, the loss of the individual flux component is largely compensated for by reduction in the latent heat flux cooling, with a noticeable but lesser contribution by the longwave. For the altered longwave experiment, the strong cooling effect now seen with the net longwave radiation should have resulted in a net cooling of the remote oceans had all else been the same. However, because of strong compensation by the latent heat flux (its cooling influence was drastically reduced), the net remote tropical ocean response is still a warming. With the altered latent flux simulation, however, while the net flux still warms the ocean, the magnitude is substantially less than in the other perturbation cases and the standard 1997/98 simulation. This behavior is consistent with our notions of latent heat flux, acting through the boundary layer specific humidity, as the regulatory mechanism for the El Niño–related remote tropical ocean warming.

A similar examination over the remote tropical land was done to see if we could identify a regulatory mechanism for the remote land regions. The results (Fig. 13) reveal that, for all but the altered sensible flux simulation, the net flux anomalies in each of those cases show a warming qualitatively similar to the standard 1997/98 simulation. This is quite remarkable given that the net flux over land regions comes from a very small residual of the individual flux components. The altered sensible flux simulation, however, does not show a warming like the standard 1997/98 simulation but rather a fast-varying oscillation of warming and cooling. We interpret this to suggest that sensible heat flux acts as the primary regulator of remote tropical land warming during El Niño, although we do not yet understand the details of this regulation.

We caution that these results should be taken only as suggestive because of the idealized nature of these simulations; additionally, the QTCM is a simplified model that may not faithfully simulate the observed ENSO teleconnection. However, we are struck by how the well the QTCM simulates the remote surface warming to El Niño, suggesting that the model contains the fundamental physics necessary to simulate this behavior.

8. Summary and discussion

We examined mechanisms causing the observed remote tropical surface warming that occurs during El Niño using ensemble simulations of the 1997/98 El Niño with an atmospheric general circulation model (CCM3) coupled to a 50-m slab ocean outside the tropical Pacific. Our line of reasoning can be summarized as follows.

1. The surface fluxes responsible for the remote warming showed large and spatially variable contributions by all components, with significant compensation between them. Because the response is a relatively simple remote Tropics-wide warming, the existence of a more fundamental mechanism was suggested. We examined the “tropospheric temperature mechanism” proposed by Chiang and Sobel (2002) as this mechanism.

2. A simple modification to the 1997/98 El Niño SST forcing allowed us to reduce the El Niño impact on tropical tropospheric temperature. The remote surface temperature response was similarly reduced, demonstrating the close link between TT and SST. The “anomalous Walker circulation,” often implicated in the tropical ENSO teleconnection, is apparently not involved in this reduction in the remote tropical surface warming since the zonal shift in Pacific convection associated with the anomalous Walker circulation was still very much evident in this simulation.

3. Examination of transients in a simulation with an abrupt switch-on of El Niño forcing showed clearly the rapid eastward propagation of warm TT anomalies along the tropical waveguide and the almost instantaneous response of land surface warming collocated with the TT anomalies, followed by the remote tropical ocean warming peaking a few months later. An apparent fixed ratio between TT and surface temperature warming (approximately 1:1 for land regions and 1:0.3 for ocean regions) suggested a regulatory mechanism that ties TT to surface temperature.

4. Additional 1997/98 simulations in which we modified the remote ocean thermodynamic feedback (through changing the mixed layer depth) showed that it is the latent heat flux that regulates the remote ocean surface temperature anomaly to the TT anomaly, although it need not itself be the source of the warming. In fact, clear-sky longwave radiation—presumably originating from the TT anomaly and associated water vapor increase—was found to be a significant source for warming for the remote ocean during El Niño.

5. The remote land response in the same 1997/98 simulations varying the remote ocean thermodynamic feedback also showed significant variations in the individual flux components; however, the net surface flux was essentially the same in all simulations. We view this as evidence for a similar regulator for the land surface temperature response, although we made no conclusions about the nature of this mechanism.

6. We used a simplified tropical circulation model (QTCM: Neelin and Zeng 2000) for “direct manipulation” experiments, simulating the 1997/98 El Niño but altering individual flux components (shortwave and longwave radiation and sensible and latent flux) separately to assess their relative importance. The remote tropical oceans warmed to the 1997/98 El Niño in all perturbation cases except for the “altered latent flux” case; similarly, the remote tropical land regions warmed for all perturbation cases except for the “altered sensible flux” case. These results are consistent with CCM3 results for the regulatory role of latent heat flux in the remote tropical ocean warming to El Niño, and furthermore suggest a regulatory role for sensible heat flux for the warming over remote land regions.

In one sense, the physical situation proposed here for the remote surface warming is simple: El Niño releases heat from the ocean into the atmosphere that is subsequently distributed to the rest of the tropical free troposphere through the equatorial waveguide. The linkage between the free troposphere and boundary layer by moist convection requires that the boundary layer (and hence the surface through surface fluxes) must also warm with free tropospheric warming. The heating also results in an increase in the tropical tropospheric water vapor content, thereby increasing the clear-sky downwelling longwave radiation. The remote surface (mainly the ocean) absorbs some of this heat, but eventually, the heat transferred to the remote Tropics is dissipated, either by radiative cooling or export to the extratropics.

On the other hand, we note that many studies examining SST variability in the remote oceans from either observational or modeling viewpoints simply do not take into account this basic but important physical effect. In particular, by focusing solely on surface fluxes, there is a danger of missing the bigger picture. In our view, the complexity of the individual surface flux components is in part related to how distinct mean climate regimes respond differently to tropospheric influences, but in the end the net flux has to act in such a way that the surface temperature tends toward consistency with TT.

The mechanism regulating the remote tropical land surface warming to El Niño still needs to be elucidated. We note that the remote land surface does not appear to be as strongly tied to the tropospheric temperature as the remote ocean: a possible reason for this is that, unlike the ocean, land is moisture limited so that moist convection may not be operating on a sufficient spatial scale and frequency for the land surface to be tightly tied to the free troposphere. We also wonder if dry convection may play a role in the regulation, given that it occurs if the surface gets too hot and the boundary layer is destabilized. Adding to the complexity is that land surface characteristics can modify over time, particularly soil moisture.

We emphasize that we are describing basinwide and continental responses to El Niño, and not specific regional responses. Regional differences in the remote ocean surface warming to El Niño can and do occur and they are associated with circulations that alter tropical convection patterns; of particular interest is the El Niño–induced forcing of significant anomalous meridional SST gradients in the tropical Atlantic (e.g., Chiang et al. 2002) and the Indian Ocean dipole (e.g., Baquero-Bernal et al. 2002). The spatially varying regional responses are clearly the more important aspect of tropical ENSO impacts, but are beyond the scope of this study. However, an understanding of the origins of the remote Tropics-wide warming is important for a complete picture of how ENSO affects the remote tropical climate. At the very least, it allows us distinguish the gross warming signal from other ENSO impacts and possibly aid in correctly attributing SST variability over the tropical Atlantic and Indian Oceans.

The TT mechanism can itself lead to nonuniform regional surface climate changes. One way that this may come about is through the dependence of the TT impact on the mean climate of affected regions. In particular, the TT–surface connection hypothesized in CS02 works less effectively over stably stratified regions over the oceans with no convection since the boundary layer there is decoupled from the free troposphere. This may be an explanation for the anomalous meridional surface temperature gradient occurring across the ITCZ latitudes in the Atlantic during El Niño since the warmer tropical North Atlantic would respond more effectively to warmer overlying TT anomalies than the cooler tropical South Atlantic. Another regional effect of increased tropospheric temperatures is the “upped ante” mechanism proposed by Neelin et al. (2003) whereby the increased threshold for convection introduced by the increased tropospheric temperatures ultimately leads to drying of marginally convective regions because of increased inflow of air into these regions from dry subsidence regions.

While we focus on the TT mechanism in this study, there exist many other proposed mechanisms for remote tropical ENSO impacts of possibly comparable importance: for example, the Pacific–North American stationary wave bridge linking ENSO to the tropical North Atlantic (Nobre and Shukla 1996) or an anomalous Walker circulation influence on the Atlantic resulting from a shift in the tropical Pacific convection (e.g., Saravanan and Chang 2000). The challenge for tropical ENSO teleconnection studies is to understand what the different pathways are, how important each pathway is, and how regional climate processes in the remote Tropics may act to modify these impacts.