These extrema in the modes of atmosphere/ocean instability are unattractive candidates for explaining nature's ENSOs for three reasons. First, in the Pacific Ocean there is a fundamental east-west asymmetry in the basic state (e.g., a colder eastern basin). Second, the observed SST anomalies do not seem to propagate systematically. Finally, from observations, the timescales associated with the local air-sea interaction are comparable to the dynamical adjustment time of the tropical ocean. For the destabilized coupled Rossby mode to exist the anomalies in upwelling must be a small term in the heat budget of the ocean mixed layer. However, the existing buoy data and the analysis of hindcast simulations of the ocean response during ENSO using ocean general circulation models indicate that throughout the eastern and central equatorial basin, subsurface temperature anomalies and upwelling anomalies are an important process in changing SST. For the destabilized eastward propagating Kelvin mode to be favored, horizontal SST advection must be unimportant and the adjustment (or mechanical damping) timescales of the ocean must be long compared with the timescales associated with local SST changes, neither of which is true for the Pacific Ocean. Clearly, the pure destabilized propagating modes noted above are not relevant to nature.
When the entire observational record is considered, the robust tropical
features of the observed ENSO are as follows: (1) quasi-stationary SST
anomalies in the eastern and central Pacific; (2) at event onset, a
relaxation of the trade winds associated with positive SST anomalies; (3)
there is a deepening in the east and a shoaling in the west of the
thermocline along the equator; (4) prior to the peak of the ENSO event, the
anomalously deep thermocline in the eastern/central Pacific begins to return
to climatological values. In addition, there is usually an increase in
the trades in the far western Pacific one or two seasons prior to the onset
of the ENSO event.
Presently, the theory for ESNO events that best fits with the robust
observations (1--4) noted above is the delayed-oscillator theory of ENSO,
put forward by Suarez and Schopf [1988] and Battisti and Hirst
[1989]. The delayed oscillator mechanism for ENSO incorporates the important
observational constraint that the timescale of the basin adjustment is
comparable to the timescale associated with the local air-sea interactions.
In addition, the delayed oscillator mechanism postulates that the dynamical
adjustments in the eastern half of the ocean basin affect the SST (and
consequently the atmospheric circulation) vis a vis changing the
temperature of the water that is subsequently entrained into the surface
mixed layer. Both the Suarez and Schopf [1988] and the Battisti
and Hirst [1989] delayed oscillator models were formulated by assuming that
the coupled atmosphere/ocean mode was stationary and trapped to the eastern
basin. It is
important to note that if nature's ENSO events are captured by the delayed
oscillator mechanism, the latter theory also provides a theoretical basis
for the role of the ocean memory in ENSO and, thus, the expectation that the
state of the tropical Pacific climate system (i.e., ENSO) should be
predictable for lead times of (at least) one year.
In a series of papers, Jin and Neelin [1993a,b] and Neelin and Jin [1993] explored the atmosphere/ocean instabilities that developed in a hierarchy of coupled models, including an ocean general circulation model (GCM) coupled to a statistical atmosphere model, and a simplified equatorial strip atmosphere/ocean model in which only the important ocean thermodynamics that are identified from the Zebiak and Cane [1987] coupled model are retained. In these studies, they explore the continuum of coupled atmosphere/ocean modes that exist in the parameter space mapped out by varying the coupling strength, the ratio of the ocean dynamical adjustment time to the timescale associated with the SST changes, and the relevant strength of the upwelling vs. horizontal advection terms in the ocean thermodynamic equation. By varying the parameters to more realistic values, introducing increasingly complete ocean (SST) thermodynamics, and tracking the eigenmodes, Jin and Neelin demonstrated how the extreme (idealized) eigenmodes that arise from the homogeneous basic state problem (see, e.g., Hirst, [1988]) give way to an increasing realistic stationary dominant eigenmode, the delayed oscillator. It is clear from these papers, and the summary review by Neelin et al. [1994], that unless some important physics is being neglected (e.g., cloud feedbacks) the delayed-oscillator mode is rather robust and is likely to be the dominant unstable atmosphere/ocean mode in nature.
The delayed oscillator theory for ENSO is remarkably consistent with the
ENSO events that occur in the Hamburg coupled atmosphere/ocean general
circulation model [ Latif et al., 1993], and it is qualitatively
consistent with the results from the Geophysical Dynamics Laboratory (GFDL)
coupled high resolution model discussed by Philander et al. [1992].
The ENSO events from the aforementioned models also share many similarities
with the observed canonical [ Rasmusson and Carpenter, 1982] ENSO
event. However, the interannual variability in a third coupled
atmosphere/ocean GCM, reported by Nagai et al. [1992], is rather weak
compared to that observed and fits the delayed oscillator model of ENSO only
during the evolution and decay of the ENSO event: the Rossby debris from the
cold events does not always kick off the next warm event (see also section
3.4). Finally, Lau et al. [1992] reported on the tropical
interannual variability from a coupled atmosphere/ocean GCM in which the
Kelvin waves are significantly distorted by the combination of numerics with
a coarse (4
latitude) resolution [see Ng and Hsieh, 1994].
Moreover, the latitudinal extent of the upwelling is known to be fundamental
to the nature of the interannual variability from theoretical constraints
[ Wakata and Sarachik, 1992]. In the low-resolution model reported on
by Lau et al., the upwelling is necessarily diffuse. As a result and
consistent with all the results above, the interannual variability in the
low-resolution model is well described by a slowly (westward) propagating
destabilized Rossby mode.
To what extent is the delayed oscillator theory for ENSO supported by the observations? Kessler [1990] has examined the observed wind stress data, the sea surface temperature anomalies, and the variability in the upper ocean thermal structure from the VOS XBT data. He concluded that the variability in the thermocline was consistent with that expected from the delayed oscillator theory. ( Wakata and Sarachik [1991] reached the same conclusion by examining the response of a shallow water model to the observed (Florida State University) wind stress anomalies.)
While the aforementioned robust observations of the development and termination of an ENSO event are consistent with delayed oscillator theory, there are clearly inconsistencies between the robust observations during event onset. Specifically, the strengthening of the trades that usually precedes the event is not a feature of the delayed oscillator theory or of the intermediate atmosphere/ocean models from which the theory is derived. Li and Clarke [1994] (hereafter LC) used equatorial wave theory and tide gauge data to construct a record of the ``observed'' western Pacific Kelvin amplitude. They correlated the reconstructed Kelvin signal with an index of the large scale zonal wind anomaly and argued the structure of the lag-correlation curve was inconsistent with the delayed oscillator theory of ENSO. The physics associated with the correlation structure reported in LC was examined in a complimentary analysis by Mantua and Battisti [1994a], whereby an ocean model was forced with the observed windstress anomalies to obtain the western Pacific Kelvin signal (which was highly correlated with that derived by LC using independent data and methods). They demonstrated that ``delayed oscillator theory'' did account for the termination of warm (ENSO) events but the cold events are not usually terminated by the (delayed oscillator) ocean adjustment process. Thus, the overall correlations between the ``observed'' Kelvin amplitude and the wind stress are degraded significantly from pure delayed oscillator theory: in nature, the ENSO events are not nearly periodic. Similar conclusions are reached by Nagai et al. [1992] in their diagnosis of the tropical Pacific variability in a coupled atmosphere/ocean general circulation model.
Finally, Penland and Magorian [1993] and Penland and Sardeshmukh [1995] have hypothesized that the tropical Pacific atmosphere/ocean system is stable in a global sense, and that the ENSO variability is best thought of as a response of the tropical Pacific system to stochastic forcing. Thus, without an external noise forcing the system there would be no ENSO events. This hypothesis for ENSO is, apparently, at odds with the all the aforementioned studies on unstable modes of variability. If the hypothesis is correct, the the intermediate level atmosphere/ocean coupled models and the hybrid numerical ocean/statistical atmosphere coupled models should demonstrate stable, not oscillatory, behavior.