The Nature and Causes for the Delayed Atmospheric Response ...

1 MAY 2003 1391K U M A R A N D H O E R L I N G

The Nature and Causes for the Delayed Atmospheric Response to El Nino

ARUN KUMAR

NOAA/NCEP Climate Prediction Center, Camp Springs, Maryland

MARTIN P. HOERLING

NOAA Climate Diagnostics Center, Boulder, Colorado

(Manuscript received 15 April 2002, in final form 13 November 2002)

ABSTRACT

Remarkable among the atmospheric phenomena associated with El Nino–Southern Oscillation (ENSO) is thelag in the zonal mean tropical thermal anomalies relative to equatorial east Pacific sea surface temperatures(SSTs). For the period 1950–99, the maximum correlation between observed zonal mean tropical 200-mb heightsand a Nino-3.4 (58N–58S, 1208–1708W) SST index occurs when the atmosphere lags by 1–3 months, consistentwith numerous previous studies. Results from atmospheric general circulation model (GCM) simulations forcedby the monthly SST variations of the last half-century confirm and establish the robustness of this observedlag.

An additional feature of the delay in atmospheric response that involves an apparent memory or lingering ofthe tropical thermal anomalies several seasons beyond the Nino-3.4 SST index peak is documented in this study.It is characterized by a strong asymmetry in the strength of the zonal mean tropical 200-mb height responserelative to that peak, being threefold stronger in the summer following the peak compared to the precedingsummer. This occurs despite weaker Nino-3.4 SST forcing in the following summer compared to the precedingsummer.

The 1–3-month lag in maximum correlation is reconciled by the fact that the rainfall evolution in the tropicalPacific associated with the ENSO SST anomalies itself lags one season, with the latter acting as the immediateforcing for the 200-mb heights. This aspect of the lagged behavior in the tropical atmospheric response occursindependent of any changes in SSTs outside of the tropical east Pacific core region of SST variability relatedto ENSO. The lingering of the tropical atmospheric thermal signal cannot, however, be reconciled with theENSO-related SST variability in the tropical eastern Pacific. This part of the tropical atmospheric response isinstead intimately tied to the tropical ocean’s lagged response to the equatorial east Pacific SST variability,including a warming of the tropical Indian and Atlantic SSTs that peak several seasons after the Nino-3.4warming peak.

1. Introduction

The zonally homogeneous response of tropical tem-peratures is the most robust impact of El Nino–SouthernOscillation (ENSO). Atmospheric warming occurs fromthe surface to the tropopause during ENSO’s warmphase (e.g., Newell and Weare 1976; Angell and Kor-shover 1978; Pan and Oort 1983; Reid et al. 1989),resulting in an elevation of pressure surfaces in the up-per troposphere. Bjerknes (1972) and Horel and Wallace(1981) highlighted the spatial coherence of the geopo-tential height response throughout the Tropics, a patternthat is consistent with the horizontal redistribution ofheat by tropical thermally direct mass circulations (e.g.,Wallace 1992).

Corresponding author address: Dr. Arun Kumar, NCEP/NOAAClimate Prediction Center, 5200 Auth Road, Room 806-H, CampSprings, MD 20746.E-mail: [email protected]

A remarkable feature of this tropical tropospheric re-sponse is its lag relative to the tropical eastern Pacificsea surface temperature (SST) forcing. This was an earlydiscovery in ENSO research, being documented in New-ell and Weare’s (1976) analysis of the relation betweensea surface temperatures of the equatorial east Pacificand 700–300-mb thicknesses at 14 tropical stations dur-ing 1958–73. In numerous other accounts, the maximumcorrelation between the zonal mean tropical tropospher-ic temperatures and the SST in the equatorial east Pacificoccurs when the SST leads the atmosphere by one totwo seasons (Newell and Weare 1976; Angell 1981; Panand Oort 1983; Reid et al. 1989; Yulaeva and Wallace1994). This feature has been confirmed from the analysisof different subperiods within the historical archives ofobservational data, within case studies of individualENSO cycles, and within climate model responses toSST boundary forcing related to ENSO. Yet, its origincontinues to defy explanation. The purpose of this study

Unauthenticated | Downloaded 01/12/22 08:19 AM UTC

1392 VOLUME 16J O U R N A L O F C L I M A T E

is to revisit the nature of the lagged behavior of thetropical response to east Pacific sea surface temperaturevariations, establish its robust features, and offer phys-ical and dynamical explanations for its cause.

A recent example is offered in Fig. 1, based on resultsof Hoerling et al. (2001), that illustrates observed andsimulated 200-mb heights during the life cycle of the1997/98 strong El Nino. East equatorial Pacific SSTsreached their peak intensity during the fall of 1997 (Fig.1, right), whereas the zonally averaged tropical heightanomalies peaked in late winter (Fig. 1, left), a lag con-sistent with many earlier studies. For this particular case,an additional feature is the prolonged warm tropical tro-posphere throughout 1998 despite the reversal in sign ofthe tropical east Pacific SST anomalies. Many of theseobserved atmospheric anomalies during 1997/98 werereproduced in an ensemble of climate simulations (Fig.1, middle), confirming the SST forcing’s strong control-ling influence.

By comparison to the extensive documentation for adelay in the tropical atmospheric response, there hasbeen a dearth of explanations for its source. It is gen-erally believed that the tropical tropospheric warmingduring El Nino results from the increase in precipitationover anomalously warm SSTs in the central to easternequatorial Pacific. Reid et al. (1989) argued for a slow(1 m s21) zonal propagation of the tropical thermal re-sponse from a point of origin in the eastern Pacific whereconvection is enhanced. But it appears to be an unlikelycause for the one season lag in view of the theoreticalresults on the rapidity with which warming can spreadacross the Tropics in response to isolated heating (e.g.,Heckley and Gill 1984; Bantzer and Wallace 1996).

A combination of observational, statistical analysis,and general circulation model (GCM) experimentationis employed to identify the mechanisms for a laggedatmospheric response to east Pacific sea surface tem-perature variations. The paper first builds upon previousstudies to document the tropical atmospheric thermalanomalies associated with ENSO by examining thelead–lag relationship between the interannual variabilityof Nino-3.4 (58N–58S, 1208–1708W) SST anomalies andthe 200-mb height anomalies during 1950–99. In lightof the realism of climate simulations as shown in Fig.1, we perform a parallel analysis of such relationshipsusing the ensemble mean output of GCM experimentsforced by the global, observed SSTs during the samerecord. These allow us to establish the robust, charac-teristic features of the delay in atmospheric responses,and test hypotheses for their origin. An additional suiteof GCM experiments with different specifications ofSST anomalies is also performed in order to understandthe role of ENSO-related SST anomalies in differentocean basins on the tropical atmosphere. The datasetsand the experiments are described in section 2. Resultsare presented in section 3, the first section of whichdocuments the lagged behavior, and the second section

of which proposes and tests various causal mechanisms.A summary is provided in section 4.

2. Data and analysis procedure

a. Observational data

Our analysis focuses on anomalies in 200-mb heightsrelated to the ENSO cycle. This variable is a usefulmeasure for the vertically averaged temperature betweenroughly the surface and the tropopause, and thus servesas a proxy for the tropospheric thermal response toENSO. We analyze the observed monthly mean 200-mb heights for 1950–99 provided by the National Cen-ters for Environmental Prediction–National Center forAtmospheric Research (NCEP–NCAR) reanalysis thatare available on a 2.58 latitude–longitude grid (Kalnayet al. 1996).

Monthly mean observed sea surface temperature forthe same period is obtained on a 28 latitude–longitudegrid from a blended dataset based on two sources. Thefirst constructs global fields during 1950–80 by pro-jecting the sparse, monthly in situ SSTs onto empiricalorthogonal functions (Smith et al. 1996). The second isa global SST analysis constructed by combining in situand satellite observations using optimum interpolation(Reynolds and Smith 1994) during 1981–99. Observedmonthly anomalies are calculated for both variables rel-ative to a 1950–99 climatology.

We employ the interannual variability of SSTs in theso-called Nino-3.4 region that encompasses the area58N–58S, 1208–1708W as an index for the tropical eastPacific forcing associated with ENSO. This monthlyindex for 1950–99 is regressed against monthly 200-mb height anomalies at each grid point to obtain theglobal atmospheric linear response to the east PacificSST variations. We perform two different regressionanalyses. The first is a simultaneous regression betweenthe Nino-3.4 SST index and 200-mb heights. These arecomputed for each calendar month separately, and thespatial maps provide information about the simulta-neous atmospheric response. The second is a monthlylead–lag regression between the Nino-3.4 index and200-mb heights for up to 12 months preceding and 12months after the index. We calculate the lead–lag re-lations with respect to January values of the Nino-3.4index, corresponding roughly to the month of maximumeast Pacific SSTs. We thus seek to describe the characterof 200-mb heights that precede (lead) this SST peak byup to 12 months, and the height behavior that follows(lags) this peak for up to 12 months.

To complement the simultaneous regression analyses,ENSO composite maps for 200-mb heights are also an-alyzed. The warm and cold ENSO events are definedaccording to a one standard deviation exceedence of theNino-3.4 SST index. Composite maps are constructedbased on both simultaneous as well as lead–lag analysis.We should point out that for the simultaneous compos-



FIG. 1. Zonally averaged 200-mb height anomalies (m) during Jan 1997–Jan 1999 for (left) observations and (middle) CCM3 ensemblemean simulations. (right) The 108N–108S meridionally averaged SST anomalies (8C). Negative contours are dashed.

ites, there is no constraint that the monthly sequence ofSSTs be reflective of ENSO’s temporal evolution (e.g.,see the definition of ENSO events for different monthsin Livezey et al. 1997).

b. Atmospheric GCM experiments

We also apply the methodologies described above toan ensemble of atmospheric GCM simulations. Onemodel used is NCAR’s Community Climate Model(CCM3; Kiehl et al. 1998) run at T42 resolution andforced with the observed monthly variations of globalSSTs during 1950–99. A 12-member ensemble has beenperformed in which the individual realizations start fromdifferent atmospheric initial conditions, but are forcedwith the same prescribed SST and sea ice boundaryconditions. Our analysis is based on the ensemble meanatmospheric anomalies of CCM3, which are computedwith respect to the model’s 1950–99 climatology.

We also analyzed two sets of GCM simulations withdifferent configurations of SSTs. These simulations areperformed using the Geophysical Fluid Dynamics Lab-oratory (GFDL) R30 atmospheric GCM coupled to anocean mixed layer model. The horizontal resolution ofthe R30 GCM is approximately 2.258 latitude by 3.758longitude, and the GCM has 14 vertical sigma levels.In the first set of experiments, the observed SSTs forthe 1950–99 period are specified only over the easterntropical Pacific (158S–158N, 1728E–South Americancoast). SSTs at all the other ocean points are specifiedas a climatologically varying seasonal cycle. In this pa-per these simulations are referred to as POGA (the Pa-

cific Ocean and Global Atmosphere). The atmosphericresponses in the POGA experiment are solely due to theENSO-related SST variability in the tropical eastern Pa-cific.

In a companion experiment, SSTs in the tropical east-ern Pacific are again specified as in the POGA exper-iment, but SSTs elsewhere in the World Ocean are freeto evolve according to coupled interactions with a mixedlayer model. In the mixed layer model a vertical columnof specified ocean depth is coupled to the atmosphere,and SST in that column evolves according to the ex-change of air–sea fluxes. Details for these so-calledPOGA–mixed layer (ML) experiments are outlined byAlexander et al. (2002). The oceanic response outsidethe tropical eastern Pacific in the POGA–ML experi-ments is due to the ‘‘atmospheric bridge’’ mechanismlinking ENSO SST variability in the tropical easternPacific with the rest of the global oceans (see Alexanderet al. 2002; Lau and Nath 1996). An ensemble of 8POGA and 16 POGA–ML realizations were availablefor the 1950–99 period.

Regression and composite analysis techniques de-scribed above are also applied to the ensemble meanatmospheric responses in the POGA and POGA–MLexperiments. The difference in the lead–lag atmosphericresponses between the POGA and POGA–ML is indic-ative of the atmospheric feedback of SST variability inthe oceans outside the tropical eastern Pacific. At thesame time, this SST variability itself is because of theENSO SST variability in the tropical eastern Pacific,which are the only SSTs specified externally in theseexperiments.



FIG. 2. Zonally averaged correlations (with a contour interval of 0.1) between the Nino-3.4 SST index and 200-mb heights (left) observedand (middle) CCM3-simulated. (right) The 108N–108S meridionally averaged correlations between the Nino-3.4 SST index and SSTs. Thelead–lag correlations are calculated relative to a Jan base month for the Nino-3.4 index. Negative (positive) y-axis values indicate monthsfor which the target variable leads (lags) the Jan Nino-3.4 index. The analysis is for the 1950–99 period, and negative correlations are dashed.

3. Results

a. Evidence for a delayed atmospheric response

Figure 2 shows the lead–lag correlations of Januaryvalues of the Nino-3.4 SST index with the tropical SSTsand 200-mb height. The correlations are calculated ateach grid point. For 200-mb heights, these are zonallyaveraged to produce the time–latitude sections in Fig.2 (left and middle). For SSTs, correlations are averagedfor 108N–108S to produce the time–longitude sectionsin Fig. 2 (right). Simultaneous correlations during Jan-uary are plotted for ordinate value of ‘‘0,’’ whereascorrelations that precede (follow) the January Nino-3.4index are specified by the leading (lagging) months onthe negative (positive) ordinate.

The SST correlations (Fig. 2, right) illustrate the trop-ics-wide oceanic evolution during ENSO’s life cycle,having a characteristic timescale of about 18 months.Beginning with initial SST signals at 9-month lead inthe west-central equatorial Pacific, the positive corre-lations indicate that the SST anomalies associated withENSO typically appear during the spring prior to theirboreal winter peak, a result consistent with earlier stud-ies (e.g., Rasmusson and Carpenter 1982). By contrast,the termination of the ENSO event occurs in the tropicalIndian and Atlantic Oceans at a 9-month lag. The pos-itive correlations between Nino-3.4 SSTs and those overthe Indian and the Atlantic Oceans are largest at a 1–2-season lag, and are in phase with the equatorial Pacific

SST anomalies as noted in Pan and Oort (1983) andstudied recently in Klein et al. (1999).

The zonal mean 200-mb heights lag the January Nino-3.4 SST index by 1–3 months in both observations (Fig.2, left) and GCM simulations (Fig. 2, middle). Duringthe evolution of ENSO, positive correlations betweenNino-3.4 SSTs and 200-mb heights initially appear onthe equator in late summer preceding the Nino-3.4 indexpeak, consistent with a tropical tropospheric warming(cooling) during the warm (cold) phase of ENSO. Thecorrelations increase with time, and also broaden me-ridionally indicative of a tropospheric warming thatspreads into the subtropics by the following spring andsummer seasons.

A remarkable feature in Fig. 2 is the appreciable dif-ference in evolution of the oceanic and atmosphericanomalies during the ENSO cycle. In particular, whilethe correlation between SSTs and the Nino-3.4 indexpeaks during December in the east equatorial Pacific,indicative of the mature phase of the tropical Pacificocean expression of ENSO, the tropical height corre-lations do not peak until the following March.

The lagged relationship between the Nino-3.4 SSTsand the atmospheric tropical thermal response has fur-ther expression in terms of the prolonged positive heightcorrelations that exist throughout the second summerseason. Note that both observed and simulated zonalmean 200-mb height correlations exceed 0.4 at an 8-month lag, despite the fact that the Nino-3.4 SST anom-



FIG. 3. Same as in Fig. 2 but for the zonally averaged warm ENSO event lead–lag composite anomalies. Warm events are defined to bethe years when the Jan Nino-3.4 index exceeds a 11 std dev departure. Units are m for the 200-mb heights (contoured every 5 m), and 8Cfor the SSTs (contoured every 0.58C).

alies have almost vanished by that time. By contrastduring the spring/summer seasons preceding the Nino-3.4 index peak, tropical Pacific SSTs are already estab-lished with large positive correlations spanning 1708E–808W, whereas only a weak atmospheric signal is evi-dent in 200-mb heights. It is worth stressing that thesimilarity between the correlations using observed dataand GCM data confirms that these features are robust,and not an artifact of sampling errors due to brevity ofthe observed record.

We repeat the analysis of Fig. 2 but using lead–lagcomposites for both the warm and cold phases of ENSO.These provide amplitude information about the oceanicand atmospheric signals, and also serve to test the lin-earity of responses that is implicit in the correlationapproach. Results for the warm event composites areshown in Fig. 3, whereas cold event composites areshown in Fig. 4. The temporal evolution of the atmo-spheric and oceanic anomalies in the composites arevery similar to those seen in the correlations, and thedelay in the atmospheric response relative to the Nino-3.4 index is evident for both phases of ENSO.

The composite results reiterate the divergence be-tween the tropical oceanic and atmospheric responsesduring the evolution of ENSO’s extreme phases. First,the peak amplitude of the tropical tropospheric warming(cooling) during warm (cold) events occurs in the Feb-ruary–April period, whereas the tropical east PacificSST anomalies peak in the previous December. Second,the atmospheric response lingers with appreciable am-

plitude into the following summer and fall season de-spite the sharp decline in tropical east Pacific SST anom-alies. Note, for example, that the zonally averaged 200-mb height anomalies at the 6-month lag (July) are equalto their values at the 1-month lead (December), whereasthe tropical east Pacific SST anomalies decline by morethan half their value.

To further illustrate the nature of the delay in atmo-spheric responses, we present in Figs. 5 and 6 the globalpatterns of the composite warm and cold event SST (leftpanels), observed 200-mb height (center panels), andGCM 200-mb height (right panels) anomalies. Theseare shown for three select months, one for the Januarybase month of the Nino-3.4 index (middle rows), andthe others for the Julys that lead (top rows) and lag(bottom rows) the January base, respectively. The twoJuly composites have been selected to emphasize theasymmetry in the atmospheric response with respect toENSO evolution, and we will subsequently refer to theprevious summer as July(2) and the following summeras July(1).

Consistent with many previous studies, the compositemaps show that the tropical Pacific SST anomalies dur-ing ENSO are already well established in July(2), withfurther strengthening by January. By July(1), theseanomalies decay considerably, and a point to note is thatthe amplitude of SST anomalies in the tropical easternPacific during July(2) is larger than the amplitude dur-ing July(1). Elsewhere in the Tropics, SST anomaliesare generally not strong in the Indian and the Atlantic



FIG. 4. The same as in Fig. 3, but for cold ENSO event lead–lag composites.

FIG. 5. The spatial maps of warm ENSO event composite anomalies for (left column) SST (8C), (middle column) observed 200-mb height(m), and (right column) CCM3 simulated 200-mb height (m). Warm events defined as in Fig. 3. The top (bottom) row shows Jul compositespreceding (following) the Jan base month of the Nino-3.4 index. The middle row shows simultaneous composites for Jan. Negative anomaliesare dashed.



FIG. 6. The same as in Fig. 5, but for cold ENSO event composites.

Oceans during July(2), but these grow by winter andassume the same sign as the east Pacific SST anomalies.As such, a tropicalwide pattern of like-signed SSTanomalies exists by July(1).

The tropical 200-mb height response is larger in Jan-uary than either in the previous or following July. Like-wise, a prominent response in the Northern Hemisphereextratropics can be seen in January, whereas a muchweaker signal occurs in summer. With regard to thedelay in atmospheric 200-mb height responses, it is ev-ident from Figs. 5 and 6 that a stronger, and larger-scaleheight response exists in July(1) compared to July(2).This asymmetry across the consecutive summers of theENSO cycle occurs despite the fact that SST anomaliesin the tropical Pacific are actually weaker in July(1)compared to July(2). In many ways, the increase inhomogeneity of the tropospheric thermal response byJuly(1) is consistent with the increased homogeneitySST anomalies throughout the Tropics.

To what extent does the temporal evolution of thezonal mean tropical response differ from the monthlyvarying simultaneous atmospheric response to Nino-3.4SSTs? In other words, how important is the evolutionaryaspect of tropical SSTs in determining the evolving zon-al mean response? To answer these questions, we com-pare the lead–lag analysis with the monthly varying,simultaneous sensitivity to Nino-3.4 SSTs. The latter iscalculated by regressing the 200-mb heights onto the

Nino-3.4 index for each calendar month. The regres-sions are then scaled by the actual phase and amplitudeof the Nino-3.4 SST index following the ENSO cycle.As such, identical Nino-3.4 SST forcing is employed inboth approaches, but the temporal evolution of the trop-ical SSTs are preserved only in the lead–lag analysis.Results are shown for the (warm minus cold event) SSTdifferences, the evolution of which is shown in Fig. 7(right).

Comparing results from the simultaneous (Fig. 7, left)and lead–lag (Fig. 7, middle) analyses reveals very sim-ilar atmospheric thermal responses during the growingphase of Nino-3.4 SST anomalies. Furthermore, the 1–3-month lag in the maximum tropical atmospheric re-sponse is reproduced in the simultaneous regressions.By contrast, there is a large difference between the twoanalyses during the decay phase of Nino-3.4 SST anom-alies, and it is clear that specifying the seasonally vary-ing tropical response solely from knowledge of the si-multaneous sensitivity to Nino-3.4 SSTs severely un-derestimates the atmospheric signal during the latter halfof the ENSO life cycle.

b. Possible origins for the delayed atmosphericresponse

Our analysis of the origins for the delayed atmo-spheric response deals with two separate issues: 1) a lag



FIG. 7. Simultaneous and lead–lag composites of zonally averaged 200-mb heights for (warm 2 cold event) ENSO composites. (left) Themonthly varying, simultaneous regressions between 200-mb height and the Nino-3.4 SST index. These are scaled according to the amplitudeand phase of the Nino-3.4 SST index following the 24-month ENSO cycle. (middle) The lead–lag (warm 2 cold event) difference calculatedby subtracting results of Fig. 4 from results of Fig. 3. (right) The 108N–108S meridionally averaged (warm 2 cold) lead–lag SST composites.The top (bottom) row shows observations (CCM3 simulations). Units are m for the 200-mb heights (contoured every 10 m), and 8C for theSSTs (contoured every 0.58C). Negative values are dashed. See text for further details.

of 1–3 months between the maximum in 200-mb tro-pospheric height response and the maximum in theNino-3.4 SST anomalies in the tropical eastern Pacific,and 2) a marked asymmetry between the temporal evo-lution of Nino-3.4 SST anomalies and tropically aver-aged 200-mb heights.

Regarding the origin of the lagged atmospheric re-sponse, one hypothesis is that it may reflect a seasonallyvarying sensitivity of the tropical troposphere to Nino-3.4 forcing. Indeed, the results of Fig. 7 indicate thatthe 1–3-month lag in the zonal mean 200-mb heightresponse could originate from such a sensitivity. It isalready well established that the extratropical responseto ENSO varies strongly with the seasonal cycle, andthat this results from changes in the background at-mospheric state rather than changes in tropical east Pa-cific forcing (e.g., Webster 1982). The seasonal cycleof the tropical atmospheric circulation is considerablyweaker, and it remains an open question if these vari-ations modulate the tropical atmospheric response toENSO as implied by Fig. 7.

In fact, further diagnosis indicates that the reproduc-tion of a 1–3-month lag in atmospheric responses fromthe simultaneous sensitivity analysis is due to such sea-sonally varying sensitivity in the atmospheric responseto Nino-3.4 forcing. It specifically involves a seasonalchange in the tropical rainfall response, rather than achange in the atmospheric sensitivity due purely to theseasonal cycle of the tropical atmospheric circulation.The temporal evolution of zonal mean precipitation isshown in Fig. 8 (bottom left) based on the CCM3 sim-ulations, and it too possesses a 1–3-month lag relativeto the Nino-3.4 index. It thus maximizes simultaneouslywith the zonal mean 200-mb height response (Fig. 7,bottom middle). Calculations using a linear baroclinicmodel driven by time-evolving zonal mean diabaticheating that mimics this behavior of the zonal meanrainfall confirms that the 200-mb heights are indeedresponding to such forcing (not shown).

A fundamental question regarding the 1–3-monthlagged atmospheric thermal response is thus the causefor the lag in the tropical rainfall. It should first be noted



FIG. 8. Lead–lag (warm 2 cold event) composites of (top) SST (8C), and (bottom) CCM3-simulatedprecipitation (mm day21). (left) The zonally averaged anomalies, and (right) the 108N–108S meridonallyaveraged anomalies. Negative values are dashed.

that the tendency of zonal mean tropical rainfall to max-imize in late winter is not due to an increase in zonalmean SST anomalies, since the latter exhibit no lagrelative to the Nino-3.4 index (see Fig. 8). The laggedzonal mean rainfall is instead related to a sensitive phas-ing of interannual SST anomalies with the climatolog-ical seasonal cycle. Of importance are the total, ratherthan anomalous SSTs, in forcing the tropical convectionin the eastern Pacific. To illustrate, Fig. 9 superimposesequatorially averaged SST anomalies during ENSO’slife cycle (shading) onto the seasonal cycle of clima-tological SSTs (contours). When east equatorial PacificSST anomalies approach their peak during fall, the sea-sonal cycle of SSTs is near a minimum. The climato-logical SSTs warm in subsequent seasons, and this oc-curs at a rate much faster than the decay of ENSO SSTs.As such, the total SSTs achieve peak amplitude in thatregion by early spring. Since the convective rainfalldepends on the total SSTs (Graham and Barnett 1987;Zhang 1993), these weaker east Pacific SST anomalies

following the Nino-3.4 peak are in fact more effectivein exciting rainfall. The equatorial average of CCM3rainfall following ENSO’s evolution (Fig. 8, bottomright) confirms the existence of such an increase in theeast equatorial Pacific rainfall, in phase with the increasein zonal mean rainfall. It follows that the zonal meandiabatic forcing associated with the temporal evolutionof ENSO does not decrease at the same pace as impliedby the evolution of the Nino-3.4 index, but is phaselocked with the seasonal cycle of climatological tropicalSSTs.

Another factor contributing to a lag in the zonal meanrainfall relative to Nino-3.4 SSTs is the evolution ofrainfall anomalies in the western equatorial Pacific. Dur-ing the growth phase of Nino-3.4 SST anomalies in fall,these are out of phase with the east-central equatorialPacific rainfall anomalies (Fig. 8, bottom left). By thefollowing spring, these rainfall anomalies weaken andbecome in phase with the rainfall anomalies over theeastern Pacific, thereby adding to the amplitude of the



FIG. 9. Lead–lag composite of the 108N–108S meridionally aver-aged (warm 2 cold event) SSTs (shading, 8C). Contours (every 0.58C)are of the climatological seasonal cycle of SSTs meridionally aver-aged 108N–108S, and are repeated twice. Positive (negative) valuesare shown in solid (dashed) contours, and denote when the clima-tological SSTs are above (below) their annual mean. For clarity, neg-ative values of the (warm 2 cold event) SST difference are notplotted.

FIG. 10. Lead–lag composites of 200-mb zonally averaged heights (warm 2 cold events) for (left) POGA simulations and (middle) POGA–ML simulations. (right) The meridional average of lead–lag composites for (warm 2 cold) SST anomalies in the POGA–ML experiments.Units are m for the 200-mb heights (contoured every 10 m) and 8C for the SSTs (contoured every 0.58C). Negative values are dashed.

zonal mean rainfall forcing at that time. The origin forthis variation in the rainfall anomalies over the westernPacific is not clear, and cannot be determined conclu-sively from these simulations.

It is through the analysis of additional GCM exper-iments that the 1–3-month lag between the tropicalheights and the ENSO SST variability can be attributedto the SST variability in the tropical eastern Pacificalone, and its interaction with the seasonal cycle. Thezonal mean of the lead–lag composites for 200-mbheights based on the POGA experiment is shown in Fig.10. The height response in these simulations (Fig. 10,left) also has a 1–3-month lag relative to the Nino-3.4SST index. Recall that the observed SSTs variability arespecified only in the tropical eastern Pacific in thePOGA simulations, and SSTs outside this region evolvethrough a climatological seasonal cycle. This 1–3-monthlag in the tropical height response is thus entirely dueto the interannual variability in the tropical eastern Pa-cific.

The behavior of the equatorial zonal mean tropicalrainfall cannot, however, be used to explain the lingeringzonal mean 200-mb response throughout spring andsummer following the peak of the Nino-3.4 index [here-after denoted by year(1)], nor the striking asymmetryof the tropical atmospheric response between July(2)and July(1). Note in Fig. 8 that the equatorial zonalmean rainfall during July(2) is almost identical toJuly(1). One might thereby be tempted to interpret theslow decay time of atmospheric zonal mean thermal



FIG. 11. Temporal evolution of the POGA minus POGA–ML dif-ferences for (top) 200-mb heights, (middle) SSTs, and (bottom) pre-cipitation, area averaged over the entire tropical belt between 308Sand 308N. Differences are for the temporal evolution of lead–lagcomposites for warm 2 cold events as in Fig. 10. For all the panels‘‘0’’ on the x axis refers to the month of Jan. Units are 8C for SSTs,mm month21 for precipitation, and m for heights.

anomalies in the Hovmoller results of Fig. 7 as reflectingan atmospheric memory of Nino-3.4 forcing. However,there is ample evidence to dismiss the notion of suchan atmospheric thermal inertia. Theoretical and mod-eling considerations indicate that the tropical tropo-sphere equilibrates rapidly to anomalous tropical heatsources (e.g., Heckley and Gill 1984; Jin and Hoskins1995; Bantzer and Wallace 1996). The mechanism forthis equilibration involves convectively driven Kelvinwaves, which effectively communicate heating anom-alies within the longitudinal expanse of the near-equa-torial troposphere on the timescale of a week to 15 days.In the absence of Nino-3.4 SST and related tropicaldiabatic heating forcing, the atmospheric signal wouldbe expected to return to climatological conditions on asimilar timescale. Thus, while it is reasonable to seeka link between the lingering aspects of delayed thermalresponse and the steady tropical forcing, neither Nino-3.4 SSTs nor the equatorial zonal mean diabatic heatingappear to be useful candidates.

Another possibility is the impact of the slow evolutionof SST anomalies in the Indian and the Atlantic Oceanbasins, which are linked to SST variability in the tropicaleastern Pacific (see Fig. 2, right). Evidence for thiscomes from the comparison of the zonal mean lead–lag200-mb height composites for the POGA and POGA–ML experiments in Fig. 10, where the meridional av-erage of the lead–lag SST composites for the POGA–ML experiment are also shown.1 The POGA–ML runrealistically simulates the lagged warming of the Indianand the Atlantic Oceans. It is evident from the differ-ences in the atmospheric responses between the POGA–ML and POGA runs that this lagged ocean warmingexerts a strong influence on the tropical atmosphere. Theevolution of the tropospheric height response in thePOGA–ML simulation is highly asymmetric relative tothe January Nino-3.4 SST peak, as also seen in theobservations, whereas such a marked asymmetry islacking in the POGA run. The lingering response in thetropical tropospheric warming following the ENSO evo-lution is thus maintained by the lagged evolution ofSSTs in the tropical ocean basins outside the Nino-3.4index region.

To further illustrate this, Fig. 11 shows the temporalevolution of the POGA–ML minus POGA differencesfor select variables that have been area averaged overthe entire tropical belt between 308S and 308N. Byyear(1) 200-mb heights for the POGA–ML are higherthan for the POGA run, reflecting the warmer tropicaltroposphere in the POGA–ML. This elevated tropicaltropospheric warmth in the POGA–ML is accompaniedby an elevated warmth in the tropical SSTs (due to the

1 SST composites for the POGA are same as for the POGA–MLexcept that the POGA simulation SST composites do not have anySST evolution east of 1728E, which is the easternmost boundary upto which interannual SST variations in these experiments is specified.

coupled model’s response to Nino-3.4 SSTs), and alsoby an increase in tropically averaged rainfall.

The ultimate cause for the lingering tropical tropo-spheric warming is thus clearly the lagged warming inthe tropical SSTs as a whole. In fact, it is seen fromFig. 11 that there exists virtually no lag between trop-ically averaged SSTs and tropospheric warming. Yet, aquestion remains as to how the influence of these SSTanomalies is communicated to the free atmosphere. Ithas been previously argued that the tropical tropospherictemperature anomalies are associated with the SSTanomalies at the lower boundary through a simplemoist-adiabatic adjustment process (Emanuel et al.1994). This relationship between SSTs and the tropo-spheric temperatures is consistent with a physical modelin which deep convection renders the tropical tropo-spheric temperature profile a moist adiabat starting fromthe near-surface moist static energy. Within this para-digm, the lingering of the tropical height response willbe consistent with the above-normal SSTs in the tropicalocean basins, with deep convective rainfall anomaliescommunicating the near-surface layer warming to the



free atmosphere. It is not clear whether anomalous pre-cipitation is a necessary condition for this to occur, how-ever. Despite the strong relationship between tropicallyaveraged tropospheric warming, SST warming, and in-creased rainfall seen in Fig. 11, the role of precipitationanomalies may be coincidental to achieving the tropo-spheric warming. For example, Su et. al. (2003) find apoor relationship between tropically averaged tropo-spheric temperature and rainfall anomalies when com-paring different El Ninos, though they find a stronglinear relation between tropically averaged tropospherictemperatures and SSTs.

4. Summary

Results of our observational analysis covering the1950–99 record highlight two prominent features of thetropical atmospheric thermal response to SST variationsin the east equatorial Pacific. One, well documented inearlier studies, is that the maximum correlation relatingthe zonal mean tropical 200-mb height response toNino-3.4 SSTs occurs when the atmosphere lags by 1–3 months. The second, and hitherto less well-recognizedfeature, is a lingering of the tropical atmospheric re-sponse by up to 9 months after the winter (January)peak in Nino-3.4 SST anomalies. This latter behaviorwas a most singular and remarkable aspect of the zonalmean anomalies observed during the recent 1997–99ENSO cycle (Hoerling et al. 2001; Kumar et al. 2001),and it is here confirmed to be characteristic of the at-mosphere’s response during ENSO’s evolution of thelast half-century. It is manifested by a strong asymmetryin the tropical zonally averaged thermal response be-tween the antecedent summer when the Nino-3.4 SSTanomalies are moderate and growing, and the followingsummer when they are weak and decaying. The zonalmean signal is shown to be threefold larger in July(1)despite the decline in Nino-3.4 SSTs to less than halftheir July(2) strength. A parallel analysis of the delayin atmospheric responses is conducted using climatemodel simulations forced by the monthly SST variationsof 1950–99. These confirm the robustness of the ob-servational results, both the one season lag in the max-imum correlation between Nino-3.4 SSTs and tropical200-mb zonal mean heights, and the strong asymmetryof those heights between the consecutive summersbracketing the peak in the Nino-3.4 forcing.

Causes for a one season lag in the atmospheric re-sponse are intimately linked with the temporal evolutionof tropical rainfall during ENSO’s life cycle. The oneseason lag relative to the Nino-3.4 SSTs occurs becausethe tropical rainfall itself lags the Nino-3.4 index by oneseason. Therefore, while Nino-3.4 SST anomalies peakin early winter, the zonal mean rainfall anomalies peakin late winter. The associated zonal mean diabatic heat-ing is thus in phase with the zonal mean tropical 200-mb heights, a relation that likely reflects the forcing ofthe latter by the former. The lag in zonal mean tropical

rainfall anomalies can be attributed to the seasonalityof the total SST variability in the tropical eastern Pacific.The decay in east equatorial Pacific SST anomalies inlate winter of year(1) occurs while the climatologicalannual cycle of SSTs is approaching its warmest state.As such, total SSTs are greatest in late winter, and aremost effective in forcing local enhancement of convec-tion at that time. This attribution is further confirmedin additional sets of GCM experiments where the ob-served SST variability is specified in the tropical easternPacific alone.

The cause for the lingering of the tropical zonal meanthermal response for several seasons after the Nino-3.4peak cannot be reconciled with a lingering in the equa-torial tropical Pacific SST anomalies. In fact, the im-pressive asymmetry in strength of the zonal mean at-mospheric signal between the consecutive summers thatembrace an ENSO event has no counterpart in the be-havior of the equatorial zonal mean rainfall, which isinstead slightly weaker by July(1) compared to July(2).

Using a parallel set of coupled and uncoupled generalcirculation model experiments, the lingering of the trop-ical thermal response is shown to be due to a laggedevolution of tropical Indian and Atlantic SST anomaliesrelative to the Nino-3.4 index. In one set of GCM sim-ulations (referred to as POGA), observed interannualSST variability was specified in the tropical eastern Pa-cific alone, and SSTs over other ocean basins evolvethrough a climatological seasonal cycle. In the com-panion set of GCM simulations, referred to as POGA–ML, SSTs outside the tropical eastern Pacific were com-puted using a mixed layer ocean model. SST anomaliesoutside the tropical eastern Pacific evolve much as ob-served in the POGA–ML simulations, consistent withan atmospheric bridge mechanism (see also Alexanderet al. 2002; Lau and Nath 1996). The 308N–308S trop-ically averaged 200-mb height response retains high val-ues well beyond the Nino-3.4 peak in the POGA–MLsimulations as is seen also in the observations, but failto do so in uncoupled POGA simulations. The com-parison of these experiments reveals that the lingeringof the tropical tropospheric signal relative to the Nino-3.4 peak is due to the SST response in the remote trop-ical ocean basins that emerge during the evolution ofENSO.

Acknowledgments. The support offered by NOAA’sClimate Dynamics and Experimental Prediction Pro-gram and CLIVAR—Pacific is gratefully acknowl-edged. We thank Gabriel Lau and May Jo Nath for kind-ly providing us the data for the GFDL model simula-tions. We would also like to thank two anonymous re-viewers for their constructive and insightful comments.

REFERENCES

Alexander, M. A., I. Blade, M. Newman, J. R. Lanzante, N.-C. Lau,and J. D. Scott, 2002: The atmospheric bridge: The influence of



ENSO teleconnections on air–sea interaction over the globaloceans. J. Climate, 15, 2205–2231.

Angell, J. K., 1981: Comparison of variations in atmospheric quan-tities with sea surface temperature variations in the equatorialeastern Pacific. Mon. Wea. Rev., 109, 230–242.

——, and J. Korshover, 1978: Estimate of global temperature vari-ations in the 100–30 mb layer between 1958 and 1977. Mon.Wea. Rev., 106, 1422–1432.

Bantzer, C. H., and M. Wallace, 1996: Intraseasonal variability intropical mean temperature and precipitation and their relation tothe tropical 40–50-day oscillation. J. Atmos. Sci., 53, 3032–3145.

Bjerknes, J., 1972: Large-scale atmospheric response to the 1964–65 Pacific equatorial warming. J. Phys. Oceanogr., 2, 212–217.

Emanuel, K. A., J. D. Neelin, and C. S. Bretherton, 1994: On large-scale circulations in convective atmospheres. Quart. J. Roy. Me-teor. Soc., 120, 1111–1143.

Graham, N. E., and T. P. Barnett, 1987: Sea surface temperature,surface wind divergence and convection over tropical oceans.Science, 238, 657–659.

Heckley, W. A., and A. E. Gill, 1984: Some simple analytical so-lutions to the problem of forced equatorial long waves. Quart.J. Roy. Meteor. Soc., 110, 203–2217.

Hoerling, M. P., J. Whitaker, A. Kumar, and W. Wang, 2001: Themidlatitude warming during 1998–2000. Geophys. Res. Lett., 28,755–758.

Horel, J. D., and J. M. Wallace, 1981: Planeraty-scale atmosphericphenomena associated with Southern Oscillation. Mon. Wea.Rev., 109, 813–829.

Jin, J., and B. Hoskins, 1995: The direct response to tropical heatingin a baroclinic atmosphere. J. Atmos. Sci., 52, 307–319.

Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Re-analysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.

Kiehl, J. T., J. J. Hack, G. B. Bonan, B. A. Boville, D. L. Williamson,and P. J. Rasch, 1998: The National Center For AtmosphericResearch Community Climate Model: CCM3. J. Climate, 11,1131–1149.

Klein, S., B. J. Soden, and N.-C. Lau, 1999: Remote sea surfacetemperature variations during ENSO: Evidence for a tropicalatmospheric bridge. J. Climate, 12, 917–932.

Kumar, A., W. Wang, M. P. Hoerling, A. Leetmaa, and M. Ji, 2001:

The sustained North American warming of 1997 and 1998. J.Climate, 14, 345–353.

Lau, N. C., and M. J. Nath, 1996: The role of atmospheric bridge inlinking tropical Pacific ENSO events to extratropical SST anom-alies. J. Climate, 9, 2036–2057.

Livezey, R. E., M. Masutani, A. Leetmaa, H. Rui, M. Ji, and A.Kumar, 1997: Teleconnective response of the Pacific–NorthAmerican region atmosphere to large central equatorial PacificSST anomalies. J. Climate, 10, 1787–1820.

Newell, R. E., and B. C. Weare, 1976: Factors governing troposphericmean temperature. Science, 194, 1413–1414.

Pan, Y. H., and A. H. Oort, 1983: Global climate variations connectedwith sea surface temperature anomalies in the eastern equatorialPacific ocean for the 1958–73 period. Mon. Wea. Rev., 111,1244–1258.

Rasmusson, E. M., and T. H. Carpenter, 1982: Variations in the trop-ical sea surface temperature and surface wind fields associatedwith the Southern Oscillations/El Nino. Mon. Wea. Rev., 110,354–384.

Reid, G. C., K. S. Gage, and J. R. McAfee, 1989: The thermal re-sponse of the tropical atmosphere to variations in equatorialPacific sea-surface temperatures. J. Geophys. Res., 94, 14 705–14 716.

Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surfacetemperature analyses using optimum interpolation. J. Climate,7, 929–948.

Smith, T. M., R. W. Reynolds, R. E. Livezey, and D. Stokes, 1996:Reconstruction of historical sea surface temperatures using em-pirical orthogonal functions. J. Climate, 9, 1403–1420.

Su, H., J. D. Neelin, and J. E. Meyerson, 2003: Sensitivity of tropicaltropospheric temperature to sea surface temperature forcing. J.Climate, 16, 1283–1301.

Wallace, J. M., 1992: Effect of deep convection on the regulation oftropical sea-surface temperature. Nature, 357, 230–231.

Webster, P. J., 1982: Seasonality in the local and remote atmosphericresponse to sea surface temperature anomalies. J. Atmos. Sci.,39, 41–52.

Yulaeva, E., and J. M. Wallace, 1994: The signature of ENSO inglobal temperature and precipitation fields derived from micro-wave sounding unit. J. Climate, 7, 1719–1736.

Zhang, C., 1993: Large-scale variability of atmospheric deep con-vection in relation to sea surface temperature in the Tropics. J.Climate, 6, 1898–1913.


The Nature and Causes for the Delayed Atmospheric Response ...

Documents

Transcript of The Nature and Causes for the Delayed Atmospheric Response ...