The Tropical Ocean-Global Atmosphere observing system: A decade of progress

2.1  El Niño: A Primary Focus of TOGA

We begin with a brief overview of El Niño, which was the primary phenomenological focus of TOGA, in order to highlight physical principles that helped to guide development of the TOGA observing system. This overview parallels what was known at the start of TOGA with the caveat that, as a conceptual model, many of its key mechanisms were poorly understood or not yet critically tested from observations. Progress beyond this simple description is taken up in sections 3 and 4.

In the tropical Pacific, net heat gain from the atmosphere tends to create a warmer surface layer near the equator than at higher latitudes. Under normal conditions (, top), easterly trade wind forcing drives near-equatorial surface flow westward in the South Equatorial Current (SEC), piling up this warm surface layer in the western Pacific to create a deep warm pool. Conversely, warm water is drained from the eastern Pacific, leading to an upward tilt of the thermocline to the east. The relative shallowness of the thermocline in the eastern Pacific increases the efficiency of local trade-wind-driven equatorial upwelling to cool the surface, creating a cold tongue in SST that extends from the coast of South America to near the international date line. The easterly trade winds are reinforced by the east-west SST contrast, which is associated with low atmospheric surface pressure over the warm pool in the west and high surface pressure over the cooler waters of the eastern Pacific. Atmospheric circulation on the equatorial plane (that is, the Walker circulation) is closed by ascent of warm moist air over the warm pool (associated with deep convection and precipitation), westerly winds aloft, and subsidence in the high-pressure zone of the eastern Pacific. In the ocean, westward flow in the surface SEC is in part compensated by a return flow to the east in the thermocline, i.e., the Equatorial Undercurrent (EUC). This current flows down the zonal pressure gradient associated with the east-west tilt of the thermocline and provides a source of water for upwelling in the east [Bryden and Brady, 1985].

During El Niño (Figure 1, bottom), the trade winds weaken in the central and western Pacific, leading to a local eastward acceleration of the surface currents. In addition, weakening of the trade winds excites downwelling equatorial Kelvin waves, which propagate into the eastern equatorial Pacific, where they depress the thermocline, and upwelling equatorial Rossby waves, which propagate into the western Pacific, where they elevate the thermocline [Wyrtki, 1975; McCreary, 1976; Hurlburt et al., 1976]. Anomalously warm sea surface temperatures appear from the coast of South America to west of the international date line, and the pattern of deep convection and precipitation shifts eastward with the warmest SSTs [Gill and Rasmusson, 1983]. Deep convection is the principal driving force for atmospheric circulation through the release of latent heat at midtropospheric levels, and these shifts in the centers of deep convection during El Niño affect the atmospheric circulation on a global basis [Horel and Wallace, 1981].

2.2  Key Variables and Sampling Requirements

The physical basis for ENSO and related phenomena provided a rationale for the development of an observing system to measure key oceanographic and meteorological variables. Prioritization of these variables was based on the need not only to better document and understand but also to predict short-term climate variability. Foremost were fields of surface wind stress and sea surface temperature since, as evident from discussion in the preceding section, it is these two variables by which the ocean and atmosphere most immediately interact in the tropics.

Of next highest priority was the upper ocean thermal field. The basic periodicity of ENSO is controlled in part by the vast thermal inertia of the upper ocean through the propagation of planetary-scale equatorial waves. These waves mediate coupling to the atmosphere on interannual timescales by redistributing upper ocean heat not only zonally along the equator, as evident in Figure 1, but also meridionally [Wyrtki, 1985a]. Thus the "memory" for the ENSO cycle is to be found in the ocean rather than in the atmosphere, where (excluding the mean seasonal cycle, which is externally forced by variations in insolation) intrinsic timescales are much shorter and are primarily associated with 3-5-day weather variability. Also, the slow evolution of upper ocean heat content on seasonal-to-interannual timescales suggested a logic for initializing ocean models used in climate prediction with subsurface temperature data.

Sea level variability was likewise deemed to be a crucial variable because it is a proxy for upper ocean heat content. The tropical oceans behave in many ways as a two-layer fluid, with thermocline variations reflected in sea level heights [Rebert et al., 1985]. For example, during ENSO, sea level is elevated when the thermocline deepens in the eastern Pacific, and it is depressed when the thermocline shoals in the western Pacific. Sea level thus provides a convenient measure of the vertically integrated oceanic response to atmospheric forcing.

Measurement of ocean currents was deemed to be essential for meeting the goals of TOGA because of the strong control ocean dynamics plays in creating ENSO SST anomalies. In most parts of the world ocean, seasonal-to-interannual changes in SST are controlled simply by variations in heat flux across the air-sea interface. In the equatorial Pacific, on the other hand, changes in three-dimensional ocean circulation play a crucial role in generating ENSO SST anomalies through horizontal advection and through changes in intensity of upwelling in the cold tongue region. To a certain extent, the need for information on the horizontal flow field could be met via estimates from the thermal field via geostrophy. However, it was also considered essential to directly measure horizontal currents along the equator, where pure geostrophy breaks down, and in the surface mixed layer, where frictional Ekman flows were expected to be large and likewise inaccessible via the geostrophic approximation.

Surface winds, SST, upper ocean thermal structure, sea level, and ocean currents, though of central importance in motivating the development of an observing system for TOGA, were of course not the only variables considered to be of value for studies of ENSO and related phenomena. It was appreciated that a quantitative understanding of SST variability required improved estimates of surface heat fluxes, that salinity variability needed to be better documented in the tropical oceans for a variety of reasons (e.g., its contribution to static stability and dynamic height, and its potential role in the surface heat balance in regions of heavy rainfall), and that studies of atmospheric circulation would benefit from an improved definition of precipitation (an integral measure of latent heat release) over the ocean. TOGA-sponsored research activities thus addressed measurement issues aimed at variables other than winds, SST, upper ocean thermal structure, sea level, and currents. However, these five key variables were viewed as a sine qua non both for improved understanding of short-term climate variability (section 3) and for the development of climate forecast models with significant predictive skill (section 4).

It was also recognized at the start of TOGA that, although ENSO is predominantly a large-scale, interannual perturbation of the climate system, it could not be effectively observed without taking into account smaller-scale, higher-frequency fluctuations. There is a broad spectrum of variability in both the ocean and the atmosphere that represents a potential source of geophysical noise in estimates of climate signals. Noise contamination can arise because of inadequate sampling in space and/or time, which will alias energy from high-frequency, small-scale fluctuations into the lower frequencies and larger scales of climatic interest. The existence of this broad spectrum of variability imposes stringent sampling requirements for climate observations. As an example, Halpern [1988a] and Mangum et al. [1992] determined that about one sample per day would be required at a given location in the equatorial Pacific to estimate monthly mean winds with an accuracy of 0.5-1.0 m s^-1. Much of the equatorial Pacific was significantly undersampled relative to this criterion by volunteer observing ships (VOS), the main source of information on surface winds prior to and during the early stages of TOGA. Furthermore, some high-frequency variations were likely to be dynamically relevant in the evolution of El Niño. Potential scale interactions result from nonlinearities in the ocean-atmosphere system through processes such as atmospheric convection, ocean mixing, heat and momentum advection, etc. Considerable debate, for example, revolved around the role of episodic 1-10-day westerly wind bursts and the 30-60-day intraseasonal Madden and Julian Oscillation [Madden and Julian, 1971, 1972] in either triggering or sustaining El Niño events or in accounting for the irregular periodicity of El Niño [e.g., Keen, 1982; Luther et al., 1983; Harrison and Schopf, 1984; Lau and Chan, 1986].

Resolution and accuracy requirements established by TOGA for the measurements discussed in this study are presented in , as excerpted from the fourth edition of the "TOGA International Implementation Plan" [International TOGA Project Office, 1992]. These requirements evolved during the program as understanding of the climate system and technical capabilities improved. Table 1 represents the final assessment of the TOGA community, taking into account developments up to 1992. No specific requirements were set for subsurface temperature. For this variable it was felt that available observational techniques would fall short of expectations, especially in terms of resolution, except in certain well-sampled regions. Note that as a practical matter, surface wind velocity rather than wind stress is measured over the oceans, with stress estimated from wind velocity using bulk turbulent transfer formulae. As specified in Table 1, an accuracy of 0.01 Pa (1 Pa = 1 N m^-2) for surface stress translates roughly into an accuracy requirement of 0.5 m s^-1 for surface winds in regions of trade wind forcing.

The requirements in Table 1 were generally cast in terms of mapping and/or documenting variability, rather than in terms of requirements for initialization of climate forecast models. These latter requirements are still a matter of ongoing research. Nonetheless, by the standards of Table 1, it could be claimed that by the end of TOGA the observing system met many of the data requirements in the equatorial Pacific Ocean between 8°N and 8°S. This is partly because that was where most in situ resources were concentrated and partly because TOGA data requirements in some instances (e.g., subsurface temperature and sea level) were based on what was considered technically feasible. Outside the latitude band 8°N-8°S in the tropical Pacific, and in the tropical Atlantic and Indian oceans, the observing system fell short of specific requirements in Table 1.

In the following subsection we provide a brief summary of the observing system, element by element. Additional technical details such as instrumental design and instrumental accuracies are elaborated on in Appendices B, C, and D.

2.3  TOGA Observing System Components

2.3.1  In situ oceanographic measurements

In situ elements of the oceanographic observing system developed and implemented in support of TOGA objectives are illustrated in and summarized in . These elements include an island and coastal tide gauge network to provide sea level measurements; drifting buoy arrays to provide mixed layer velocity and SST measurements; the TOGA Tropical Atmosphere-Ocean (TAO) array of moored buoys to provide surface wind, SST, upper ocean temperature, and current measurements; and a volunteer observing ship (VOS) expendable bathythermograph (XBT) program for upper ocean temperature profiles. The XBT program was embedded in the ongoing program of VOS surface marine meteorological measurements, which provided wind, SST, and other surface data. TOGA also inherited a decade-long VOS sea surface salinity network in 1985. In addition, repeat hydrographic sections from regularly scheduled research cruises, most notably along 110°W [McPhaden and Hayes, 1990b; Hayes et al., 1991c], 165°E [Delcroix et al., 1992], and 137°E [Shuto, 1996], provided valuable information on upper ocean water mass structures in the Pacific during TOGA.

A key feature of the array elements shown in Figures 2 and 3 was that by the end of TOGA most of the data were transmitted to shore via satellite relay in real time. In addition, each array element had unique measurement capabilities that were advantageous for TOGA (Table 2). However, none of these elements by themselves would have been adequate for TOGA purposes, because each has certain disadvantages in terms of cost and/or sampling characteristics that limit its applicability for short-term climate studies. These array components were therefore viewed as complementary to one another, providing a synergistic framework in which to document and analyze climate fluctuations in the tropical oceans.

Design of the observing system was guided by general circulation model simulations of wind-forced oceanic variability and by empirical studies of space-time correlation scales. Model design studies indicated, for example, that basin-scale wind measurements were required within at least ~ 7° of the equator to simulate accurately the seasonal-to-interannual evolution of SST variability in the cold tongue region of the equatorial Pacific, and that the ocean responds most sensitively to zonal wind rather than meridional wind forcing on these timescales [Harrison, 1989]. Empirical studies indicated that zonal wind field variations are minimally coherent over 2°-3° latitude and 10°-15° longitude [Harrison and Luther, 1990], and that approximately one sample per day would be required to meet TOGA accuracy requirements [Halpern, 1988a; Mangum et al., 1992]. The space scales and timescales of upper ocean thermal structure are depth dependent and nonstationary in time. However, the most stringent thermal field sampling requirements (for thermocline temperature during non-ENSO periods) are comparable to those for zonal winds [e.g., Meyers et al., 1991; Hayes and McPhaden, 1992; Kessler et al., 1996]. Scales of variability and sampling requirements for velocity were described in work by Hansen and Herman [1989], World Climate Research Program [1990b], and McPhaden et al. [1991].

Design of the observing system was constrained by logistical considerations, such as the availability of islands suitable for tide gauge installation and the availability of commercial shipping routes. It was also constrained by the practicalities of cost, since financial resources were limited. Implementation was based on existing technologies, although measurement capabilities and cost efficiencies were greatly enhanced by two significant technological breakthroughs. One was the development of a low-cost Autonomous Temperature Line Acquisition System (ATLAS) wind and thermistor chain mooring capable of telemetering its data in real time [Hayes et al., 1991a]. The second was the development of a low-cost, long-lived drifting buoy with accurate water-following characteristics [Niiler et al., 1995].

The in situ observing system was much better developed in the Pacific than in the Atlantic and Indian Oceans, as evident in Figure 3 and Table 3. In the Atlantic and Indian Oceans, fewer VOS XBT tracks and tide gauge stations were instrumented, and no long-term moorings were deployed for TOGA purposes. Drifter deployments were occasionally made in the tropical Atlantic and Indian Oceans during TOGA [e.g., Integrated Global Ocean Services System (IGOSS), 1992], but there was no program of sustained drifter deployments undertaken in either basin specifically by TOGA investigators until near the end of the program.

The full TAO array of ~ 70 moorings is situated between 8°N and 8°S, 95°W and 137°E and spans over one third the circumference of the globe at the equator (Figure 2). The backbone of the array is the low-cost ATLAS wind and thermistor chain mooring [Hayes et al., 1991a]. Five long-term current meter mooring sites are also maintained along the equator [World Climate Research Program, 1990a]. The array was built up primarily during the second half of TOGA (Figure 2 and Table 3) and was completed only at the very end of TOGA in December 1994 [McPhaden, 1995]. A major advantage of the TAO array was its finely resolved (daily or higher temporal resolution) time series data of key variables, particularly winds, which significantly reduced the amount of aliased high-frequency energy in the climate signals of interest. Data were transmitted in real time to shore via Service Argos then retransmitted on the Global Telecommunications System (GTS). Financial support was derived mainly from the United States, France, Japan, Taiwan, and Korea.

A TOGA/World Ocean Circulation Experiment (WOCE) Surface Velocity Program (SVP) was organized at the beginning of TOGA to seek broad international support for drifter acquisitions and deployments. At the time, there were several competing designs of unknown water-following characteristics. Several years of engineering and design work led to the Global Lagrangian Drifter with a mean lifetime (defined in terms of drogue retention) of roughly 300-400 days. Position information, SST, and other drifter data were telemetered to shore in real time via Service Argos then retransmitted on the GTS. In TOGA, drifters were deployed from research vessels, VOS, and airplanes. The objective was to maintain drifter arrays with enough samples in 2° latitude × 8° longitude areas to define the mean 15-m circulation, the seasonal cycle [Reverdin et al., 1994], and ENSO-related anomalies [Frankignoul et al., 1996]. SST data from the drifters have also proven to be critical for operational SST analyses (see Appendix C). By the end of TOGA, over 700 drifters were operational in the global oceans, over one third of which were deployed in the tropical Pacific. The SVP emerged from TOGA as the Global Drifter Program, maintained with resources from 16 countries.

TOGA inherited a substantial Pacific tide gauge network that was largely installed during NORPAX. Though design of the tide gauge network was constrained by the availability of islands where gauges could be placed (Figures 2 and 3), efforts in the Pacific during TOGA were focused on expanding and refining this network, under the direction of the University of Hawaii Sea Level Center. By the end of TOGA the number of stations in the Pacific had more than doubled (Table 3). Relative growth was equally impressive in the Atlantic and Indian Oceans, although the number of sites instrumented in those oceans was fewer than in the Pacific. Many sites were linked to the Hawaii Center via data channels on geostationary satellites. In addition, many of the TOGA tide gauges contributed to the Integrated Global Ocean Services System (IGOSS) Sea Level Project in the Pacific, for which data were made available via GTS with a delay of 1 month.

There are currently around 7000 VOS worldwide, operated by about 50 countries. They collect observations on sea surface pressure, wind velocity, sea state, humidity, and SST as part of the World Weather Watch (WWW). On a few routes, surface salinity is also sampled. Each month, typically 100,000 or more surface observations are collected and transmitted in real time to national meteorological centers via satellite communication systems or via coastal radio stations, then entered onto the GTS for general use. Prior to the establishment of TAO and other dedicated TOGA observing systems, data from VOS marine reports and from island weather stations constituted the bulk of the available information on seasonal and interannual variability in tropical surface marine meteorological fields. Important data sets and products such as the Florida State University (FSU) wind analysis [Stricherz et al., 1992] and Comprehensive Ocean-Atmosphere Data Set (COADS) [Woodruff et al., 1987] derive largely from VOS surface marine observations.

A subset of VOS ships also collect XBT data, and ~ 150,000 temperature profiles to a depth of 400 m or more were added to the climatological database during TOGA in the tropical Pacific. Design of the VOS XBT array for TOGA was based on a strategy of low-density sampling to provide broad-scale, widely dispersed coverage in areas of routine merchant shipping on a monthly-to-quarterly cycle for description of large-scale thermal field signals. Recommended low-density XBT sampling was prescribed as one XBT drop per 1.5° latitude by 7.5° longitude per month. TOGA also recognized a need to observe seasonal and interannual variations of major geostrophic currents in the tropical oceans. A strategy of frequently repeated sampling with higher along-track resolution was devised for a few transequatorial VOS lines to meet this need [Meyers et al., 1991]. On some routes, expendable conductivity-temperature-depth (XCTD) data were also collected [Roemmich et al., 1994]. By the end of TOGA most VOS XBT data were telemetered to shore in real time via Service Argos or via geostationary satellites, then retransmitted on the GTS.

2.3.2  Satellite measurements

Complementing in situ oceanographic observations were satellite missions to measure SST, sea level, and winds (Table 4). Sea level measurements were provided from altimeters flown on the Geosat mission, the ERS-1 mission, and the joint National Aeronautics and Space Administration (NASA)/Centre National d'Études Spatiales (CNES) TOPEX/POSEIDON mission. SST measurements were derived principally from multichannel advanced very high resolution radiometers (AVHRR) carried aboard the National Oceanic and Atmospheric Administration (NOAA) series of polar orbiting weather satellites. Wind speeds were measured by the special sensor microwave imager (SSM/I) deployed on the Defense Meteorological Satellite Program (DMSP) sponsored by the U.S. Department of Defense. Remotely sensed wind velocities were first available during TOGA beginning in 1991 from a scatterometer aboard the ERS-1 satellite. Note that Table 4 does not list all the wind speed and SST data available during TOGA from satellite platforms. For example, SST information was available from the along-track scanning radiometer on ERS-1, and wind speed was available from altimeter missions. The emphasis in Table 4 is on those satellite data sets which for technical reasons were most widely applied in TOGA studies.

Satellite measurements have the advantage of being global, or nearly so, in coverage and quasi-synoptic in time, and they often have better spatial and/or temporal resolution than in situ data. The increased use of satellite data did not diminish the need for in situ oceanographic measurements, however. In situ techniques are required for measurements of variability below the surface of the ocean. Also, satellite systems rely on complicated algorithms to convert measurements of electromagnetic radiation into geophysically meaningful variables. To be useful, satellite data must be calibrated and validated against in situ observations in order to detect and remove potential biases induced by orbital errors, instrumental errors, and/or atmospheric effects (e.g., water vapor, clouds, and aerosols).

Considerable effort was devoted to calibration and validation during TOGA for satellite-derived estimates of SST [e.g., Liu, 1988; Allen et al., 1995], SSM/I surface wind speed [e.g., Bates, 1991; Halpern et al., 1993; Boutin and Etcheto, 1996], surface wind velocity from the ERS-1 scatterometer [Bentamy et al., 1996; Rufenach, 1995], sea level from Geosat and TOPEX/POSEIDON [Cheney et al., 1989, 1994; Busalacchi et al., 1994; Delcroix et al., 1991, 1994; Katz et al., 1995a; Picaut et al., 1995], and surface zonal geostrophic currents derived from satellite altimetry [Picaut et al., 1990; Menkes et al., 1995]. The accuracies achieved depended on the particular satellite sensor and the level of data processing (Appendix C). Also, blended satellite/in situ products were developed during TOGA to take advantage of the strengths of both types of data. These products include the SSM/I-based wind analysis merged with in situ data and European Center for Medium-Range Weather Forecasts (ECMWF) model output [Atlas et al., 1991, 1996] and the National Centers for Environmental Prediction (NCEP) blended satellite/in situ SST analysis, an example of which is shown in for the last week of TOGA [Reynolds and Smith, 1994, 1995] (see also Appendix C, section C1).

2.3.3  In situ meteorological measurements

Most long atmospheric time series available for climate research derive from the operational activities of the WWW. At the start of TOGA, there were about 400 upper air reporting stations between 30°N and 30°S as part of the WWW, of which TOGA identified 150 as a minimal network for documenting planetary-scale variations in atmospheric circulation. Thus the basic elements of an upper air observing system existed at the outset of TOGA. Even so, this WWW network of stations was not adequate for TOGA purposes. As a consequence, initial planning for TOGA by the various scientific bodies noted the strong desirability of expanding the network of WWW rawinsonde sites in the tropics, especially in the Pacific and Indian Ocean sectors. Sites eventually instrumented under TOGA auspices included Tarawa, Kanton, Penrhyn, and San Cristóbal (in the Galápagos Islands) in the Pacific (Figure 5) and the island of Gan in the Indian Ocean. Unfortunately, the WWW network in the tropics in general underwent significant declines in data collection and exchange through the GTS during the TOGA decade for a variety of technological, political, and economic reasons [National Research Council, 1994a].

TOGA also supported the establishment of wind profilers at several sites throughout the Pacific Basin (Figure 5), beginning with the 50-MHz very high frequency (VHF) wind profiler that commenced operation at Christmas Island in April 1986 [Gage et al., 1990, 1991a]. This Transpacific Profiler Network provides measurements of tropospheric winds between altitudes of 1.8 and 18 km height. Four times per day, hourly averaged VHF profiler data are telemetered via geostationary satellite and incorporated into the GTS for worldwide distribution. In addition, 915-MHz ultrahigh frequency (UHF) wind profilers were installed at Biak, Indonesia; Tarawa, Kiribati; and San Cristóbal, in the Galápagos Islands of Ecuador to provide more information on boundary layer wind variability.

3  Scientific Progress: Improved Description and Understanding

3.1  Long-Term Mean and Mean Seasonal Cycle

The long-term mean and mean seasonal cycle are crucial for understanding interannual variations in the coupled system. Background stratification, for example, affects the length scales, timescales, and phase speeds of planetary equatorial waves thought to be important in the ENSO cycle. Likewise, zonal asymmetries in the background state of the equatorial ocean due to mean trade wind forcing, e.g., the mean zonal slope of the equatorial thermocline and zonal SST gradient associated with it (shown schematically in Figure 1), establish conditions necessary for the growth of ENSO-related SST anomalies [e.g., Battisti and Hirst, 1989]. El Niño anomalies also tend to be phase locked to the seasonal cycle, with warmest El Niño SST anomalies often occurring in boreal winter in the equatorial cold tongue, when SST is seasonally at its coldest [Rasmusson and Carpenter, 1982]. Empirical and modeling studies have indicated that persistence and predictability of ENSO anomalies is seasonally modulated, being highest in boreal summer and winter and falling off through the boreal spring [Latif and Graham, 1992; Webster and Yang, 1992; Latif et al., 1994; Balmaseda et al., 1995]. Some theories also suggest that the mean seasonal cycle determines the basic periodicity and irregularity of the ENSO cycle via chaotic nonlinear self-interaction [e.g., Jin et al., 1994; Tziperman et al., 1994; Chang et al., 1995]. However, few, if any, coupled ocean general circulation models (GCMs) are capable of simulating both the mean seasonal cycle and interannual ENSO-like variability with equal degrees of veracity [Mechoso et al., 1995]. Finally, seasonal variations for some variables (e.g., SST in the eastern Pacific) are as large as, or larger than, ENSO-related interannual anomalies. Therefore, at minimum, one requires a clear definition of the climatological mean seasonal cycle for model validation and in order to accurately define interannual climate anomalies. Climatologies existed prior to TOGA, but in some cases, especially for subsurface oceanographic variables, they were of poor quality because of the sparsity of data on which they were based.

3.1.1  Long-term mean

Key features important in characterizing the coupled ocean-atmosphere system in the equatorial Pacific include the western Pacific warm pool with SSTs > 28°C and the equatorial cold tongue of the eastern and central equatorial Pacific (Figure 4). These structures, evident in all long-term mean SST climatologies, are modulated in intensity and areal coverage on seasonal, interannual, and decadal timescales. Understanding how these features relate to surface winds and subsurface ocean hydrodynamics is critical to understanding climate variability related to ENSO.

An example of the improved definition from the TOGA observing system of mean upper ocean temperature, surface dynamic height, and wind stress along the equator is shown in . The mean temperature section, on the basis of all available TAO data between 2°N and 2°S, is similar to that presented by Kessler et al. [1996]. It shows the increase in SST from east to west, the warm pool of 28°C water in the upper 100 m of the western Pacific, the downward sloping thermocline in the upper 300 m, and the existence of a weakly stratified "thermostad" of 13°C water in the eastern Pacific [Stroup, 1969]. Situated in the middle of the highly stratified upper thermocline is the 20°C isotherm; for this reason this isotherm is often used as an index for the depth of the thermocline in the tropical Pacific. The mean surface dynamic height associated with the temperature field rises by 40 dynamic centimeters (dyn. cm) between 95°W and 170°E, after which it decreases slightly to the west. Zonal variations in dynamic height and thermocline depth along the equator are a response to steady easterly trade wind forcing in the eastern and central Pacific [McPhaden and Taft, 1988]; reversal of these gradients in the western Pacific is associated with local westerly winds [see also Wyrtki, 1984; Mangum et al., 1990; McPhaden et al., 1990a]. The zonal section in Figure 6 has many features in common with sections composited from different individual cruises prior to TOGA [e.g., Philander, 1973; Halpern, 1980] but is more representative of long-term mean conditions.

The mean thermal structure of the Pacific along quasi-meridionally oriented VOS XBT lines (Figure 7) also shows the downward slope of the thermocline toward the west in response to mean trade wind forcing. In addition, the meridional structure of ridges and troughs in the thermocline, which are related to major zonal currents [e.g., Donguy and Meyers, 1996a], is also clearly delineated. Evidence of trade-wind-driven equatorial upwelling (local minima in temperatures near the equator in the surface layer) is apparent in the central and eastern Pacific sections.

Methods to estimate the volume transport of the major equatorial currents from monthly, synoptic VOS XBT sections, as in Figure 7, were developed by Kessler and Taft [1987], Taft and Kessler [1991], Picaut and Tournier [1991], and Donguy and Meyers [1996a]. A comparison of transports from VOS XBT data to research vessel data (Table 5) shows that all of the geostrophic current transports can be reasonably well monitored by the VOS program. Differences between means based on research vessel and VOS data are of the order of only 7-20% (Tables 5a and 5b). The temporal variation inferred from research cruise data is highly correlated to the VOS estimates [Picaut and Tournier, 1991]. Although somewhat different methods were used to calculate XBT transports by Kessler and Taft [1987] and Picaut and Tournier [1991], the mean and standard deviation of transports over a 7-year period are only slightly different (Table 5c).

Drifter data allow for a definition of the surface circulation (combined Ekman and geostrophic components) across the entire basin, rather than just along prevailing shipping routes. The average velocity at 15-m depth from the drifter data for 1988-1994 (Figure 8) shows the persistent and well-documented surface current systems of the tropical Pacific: the North Equatorial Current (NEC), South Equatorial Current (SEC), North Equatorial Countercurrent (NECC), and a vestigial South Equatorial Countercurrent (SECC) (in the region 6°-10°S, 160°-176°E). The standard error of the velocity shows that the general circulation of the tropical Pacific is well defined everywhere, even to the extent that divergence and relative vorticity fields can be computed from this data with a high degree of confidence.

Significant departures from the patterns that have been reported by ship drift charts, or from interpretation of the gradients of dynamic height as an index of the surface current, emerge from the drifter data. For example, dynamic height maps show that there should be a geostrophic flow toward the equator nearly everywhere, while drifter data indicate that there is a flow toward the pole nearly everywhere. Thus the meridional Ekman flows are strong enough not only to cancel the near-surface geostrophic currents but also to transport surface layer water in the opposite direction. Surface layer Ekman divergence near the equator in particular is important in determining the equatorial upwelling circulation [Wyrtki, 1981]. Also, compared to ship drift charts, the drifter data show a splitting and divergence of the South Equatorial Current between 110° and 136°W, with maxima in westward flow to the north and south of the equator.

3.1.2  Mean seasonal cycle

The seasonal cycle of SST in the equatorial Pacific has been well documented from COADS and other VOS-based analyses [e.g., Reynolds and Smith, 1995]. Warmest SSTs in the cold tongue occur in boreal spring, and coolest SSTs occur in boreal autumn. The amplitude of these annual period variations diminishes from east to west as the thermocline deepens (Figure 9); similarly, the timing of maximum temperatures occurs later in the boreal spring progressing from west to east [e.g., Horel, 1981; Enfield, 1986; Chao and Philander, 1991]. The westward progression of the annual cycle of SST along the equator in the Pacific is related to the westward progression in the zonal winds [Chang, 1994; Xie, 1994]. Annual variations in SST in turn set up atmospheric boundary layer pressure gradients which drive annual period zonal wind variations [Nigam and Chao, 1996].

Although solar forcing near the equator is predominantly at semiannual periods, SST in the equatorial cold tongue of the eastern and central Pacific is dominated by annual period variations because of the importance of ocean dynamics and the influence of land masses bordering the Pacific [Li and Philander, 1996]. Recent diagnostic studies and model results illustrate the complex mix of ocean processes in accounting for the amplitude and phase of seasonal SST variations in this region [Hayes et al., 1991b; Köberle and Philander, 1994; Chang, 1993, 1994; Chen et al., 1994a]. The shallow mean thermocline depth in the eastern Pacific, which is due to large-scale wind forcing (Figure 6), is important in facilitating upwelling and vertical mixing to cool the surface. Zonal advection associated with seasonally varying currents is also important, particularly in the central Pacific [Chen et al., 1994a; Minobe and Takeuchi, 1995]. Variations in surface heat fluxes (mainly solar irradiance and latent heat flux) are significant at all locations. These fluxes assume a dominant role as ocean dynamical processes diminish poleward away from the equator and in the western equatorial Pacific where the thermocline is deep. In this latter region the semiannual period in solar irradiance forcing leads to the dominant semiannual period in SST (Figure 9).

Studies using XBT and conductivity-tempera- ture-depth (CTD) data have described the seasonal cycle of upper ocean thermal structure based on the dynamics of Ekman pumping and Rossby waves [Delcroix and Henin, 1989; Kessler, 1990; Kessler and McCreary, 1993]. Seasonal variations in transports of major currents have also been documented using XBT and tide gauge data by Taft and Kessler [1991], Picaut and Tournier [1991], and Donguy and Meyers [1996a]. Mitchum and Lukas [1990] used a set of sea level data lying along the North Equatorial Countercurrent trough to show that annual variations propagate to the west as a Rossby wave resonantly forced by westward propagating components in the wind field. Recent model simulations of the seasonal cycle, validated against TOGA observations [e.g., Minobe and Takeuchi, 1995], confirm the results of these empirical studies on the importance of wind stress forcing and equatorial wave processes.

Reverdin et al. [1994], developed a climatology of the surface currents in the tropical Pacific from TOGA drifter and mooring data. A notable aspect of the mean seasonal cycle along the equator is the "springtime reversal" of the normally westward flowing South Equatorial Current [Halpern, 1987b]. It is most evident in the eastern Pacific where, for example, eastward flow of over 30 cm s^-1 occurs in April-May at 110°W (Figure 9). This reversal in flow propagates westward along the equator [McPhaden and Taft, 1988], as do zonal winds and SST [Horel, 1981; Lukas and Firing, 1985], with variations at 140° and 170°W lagging those farther to the east. The springtime reversal in the SEC had been known for nearly a century [Puls, 1895], though its magnitude was underestimated because of contamination of ship drift estimates by windage on ship's hulls [McPhaden et al., 1991]. Model simulations suggest that the springtime reversal results from the seasonal relaxation of the zonal component of trade winds, causing flow to accelerate eastward down the zonal pressure gradient [Chao and Philander, 1991; Yu et al., 1997].

The mean seasonal cycle of the Equatorial Undercurrent along the equator has been described in several reports [Halpern, 1987b; McPhaden and McCarty, 1992; McCarty and McPhaden, 1993; Weisberg and Hayes, 1995]. Juxtaposing seasonal analyses based on these studies (Figure 9) helps to highlight some of the important characteristics of variability on this timescale. The EUC, on average, is located in the upper thermocline and is therefore found at greater depths in the west than in the east. Zonal current variations are confined principally to above the Undercurrent core, with a maximum eastward flow in the thermocline occurring in boreal spring at all longitudes.

Recent analyses suggest that the seasonal cycle is nonstationary in the eastern equatorial Pacific [Gu et al., 1997]. Specifically, at 110°W the annual period in thermocline depth variations was much more pronounced in the 1990s than in the 1980s, presumably because of changes in the annual cycle of zonal wind forcing farther to the west. Interestingly, amplification of thermocline depth variations was not reflected in amplified annual SST variations at 110°W. The mean depth of the thermocline remained sufficiently shallow in the eastern Pacific that, consistent with the theories of Köberle and Philander [1994] and Xie [1994], the efficiency of ocean-atmosphere interactions and ocean dynamical processes to cool the surface would not have been significantly impacted.

3.2  ENSO Variability

Some of the hallmark manifestations of the ENSO cycle are illustrated in , which shows time series of the Southern Oscillation Index (SOI) and of surface zonal wind stress anomalies and sea surface temperature anomalies along the equator. The period shown (1982-1995) encompasses the 1982-1983 El Niño and interannual variability during the TOGA decade (1985-1994). Each warm episode (1982-1983, 1986-1987, 1991-1992, 1993, and 1994-1995) is associated with negative SOI values and weaker than normal trade winds over about 60° of longitude in the central and western Pacific. In the case of the intense 1982-1983 El Niño the trade winds weakened progressively from west to east all the way across the basin. Conversely, the 1988-1989 cold La Niña event was associated with high SOI values and a strengthening of the trade winds over roughly 60° of longitude. Also noteworthy in Plate 1 is the persistence of warm SST anomalies near the date line and the occurrence of three distinct warm episodes in the eastern Pacific in concert with consistently low Southern Oscillation Index values between 1991 and 1995. Although it is known that the frequency and intensity of ENSO events are modulated on decadal and longer timescales [Gu and Philander, 1995], the duration of warm phase ENSO conditions over 5 calendar years is unparalleled in this century [Trenberth and Hoar, 1996].

The relationship between surface winds and SST for December 1994 (Figure 10) illustrates another important aspect of ENSO variability. Deep atmospheric convection typically occurs over the warmest SSTs in the tropical Pacific [e.g., Graham and Barnett, 1987]. Warmest SSTs (> 30°C) in December 1994 were situated just south of the equator near the date line in a region of strongly convergent surface winds and active deep atmospheric convection [Climate Analysis Center, 1994]. Converging winds act to sustain both deep convection (via moisture convergence) and warm SSTs (via ocean dynamics) [Philander et al., 1984]. These processes tend to locally reinforce one another, and representing them properly in coupled ocean-atmosphere models has been one of the challenges of ENSO modeling [e.g., Zebiak and Cane, 1987; Battisti, 1988; Battisti and Hirst, 1989; Schopf and Suarez, 1988].

An important oceanic feature of the ENSO cycle is the zonal redistribution of warm surface layer water masses [White et al., 1985; Donguy, 1987; Donguy et al., 1989; McPhaden et al., 1990a; McPhaden and Hayes, 1990b; Kessler and McPhaden, 1995a]. In the western Pacific the thermocline (as indicated by the depth of the 20°C isotherm) shoals 20-50 m in the latitude band 15°S to 20°N during El Niño, whereas in the eastern Pacific the thermocline deepens by a comparable amount but in a narrower band of latitudes than in the west. These thermocline depth variations, illustrated along the equator in for the 1991-1993 El Niño, are correlated with changes in the strength of major currents. The westward SEC weakens significantly during El Niño episodes, while in some events the NECC intensifies [Taft and Kessler, 1991; Kessler and McPhaden, 1995a]. Thus there is an anomalous eastward mass transport of warm water by the equatorial surface currents during the onset of warm events.

Changes in the zonal distribution of upper ocean heat content are reflected in sea level variations [e.g., Rebert et al., 1985; Delcroix and Gautier, 1987] because of the vertically coherent structure of the upper ocean thermal field on seasonal-to-interannual timescales. In other words, anomalously deep thermocline tends to be associated with anomalously high sea level and vice versa. Wyrtki [1984] described the sea surface height gradient along the equator during the 1982-1983 El Niño assuming that the long-term mean sea level at tide gauges along the equator was equal to the long-term surface dynamic height relative to a deep reference level. He showed that the normal upward slope of sea level from east to west (Figure 7) was sharply reduced and at times reversed in the eastern and central Pacific during 1982-1983. Reduction and reversal of the sea surface slope also occurred in the 1986-1987 and 1991-1992 El Niño events (Figure 12). Variations were weaker at these times than in 1982-1983 though, as expected from the weaker and less zonally extensive westerly wind anomalies along the equator (Plate 1). Conversely, during the 1988-1989 cold La Niña event the sea level slope along the equator intensified, in association with stronger than normal trade winds (Figure 12).

Sea level slope along the equator is an index for the strength of the zonal pressure gradient, which is the driving force for the Equatorial Undercurrent [Philander and Pacanowski, 1980; McCreary, 1980; McPhaden, 1981]. Reduction and reversal of this sea level slope were associated with a significant weakening and disappearance of the EUC in the thermocline during the 1982-1983 El Niño [Firing et al., 1983; Halpern, 1987b] and the 1986-1987 El Niño [McPhaden et al., 1990a]. The EUC, though it did not disappear during the 1991-1993 El Niño, was greatly reduced in strength in the central Pacific for several months [Kessler and McPhaden, 1995a]. El Niño related reductions in Undercurrent strength have significant implications for the heat balance of the surface layer, since the Undercurrent is normally a source of cold water to feed equatorial upwelling [Bryden and Brady, 1985].

Near the equator, adjustment of the upper ocean heat and mass is strongly influenced by excitation and propagation of equatorial Kelvin and long Rossby waves, which are the primary mechanisms by which the winds communicate their influence to other parts of the ocean basin. The Kelvin waves most prominent in equatorial time series data are associated with forcing by westerly wind bursts and the atmospheric Madden and Julian Oscillation [Miller et al., 1988; McPhaden et al., 1988a; Kessler et al., 1995]. These waves are clearly evident in 20°C isotherm depth variations (e.g., Figure 11), as well as in time series of sea level, dynamic height, and zonal currents within 2° latitude of the equator. Using TAO data and Geosat-derived sea level data, Cheney et al. [1987], Miller et al. [1988], McPhaden et al. [1988a], McPhaden and Hayes [1990b], Delcroix et al. [1991, 1994], Johnson and McPhaden [1993a], and Picaut and Delcroix [1995] clearly documented equatorial Kelvin waves propagating eastward with first baroclinic mode phase speeds of 2-3 m s^-1 prior to and during the 1986-1987 El Niño. Similarly, analysis of TAO data and TOPEX/POSEIDON sea level data indicated prominent oceanic variability due to equatorial Kelvin waves generated by wind forcing west of the date line during 1991-1995 [Busalacchi et al., 1994; Kessler et al., 1995; Boulanger and Menkes, 1995].

Weakening of the trade winds near the equator in the central and western Pacific at the onset of warm ENSO events leads to a pattern of upwelling favorable wind stress curl which elevates the thermocline locally at extraequatorial latitudes [e.g., Kessler, 1990]. Weakening of the trade winds also excites upwelling long Rossby waves [White et al., 1985, 1987; Kessler, 1990; Boulanger and Menkes, 1995; Kessler and McPhaden, 1995b], the fastest of which propagates westward at phase speeds of one third the Kelvin wave speed. The slower propagation speed of these waves compared to equatorial Kelvin waves implies that elevation of the thermocline in the west lags depression of the thermocline in the east by several months as evident in thermal field and sea level analyses (e.g., for 20°C along the equator between late 1991 to early 1992 in Figure 11). The Geosat analysis of Delcroix et al. [1991] and subsequent modeling study of du Penhoat et al. [1992] for the 1986-1987 El Niño suggest that, in addition to wind forcing, eastern boundary reflections of equatorial Kelvin waves can generate equatorial Rossby waves that affect the evolution of ENSO.

Empirical studies of the surface layer heat balance emphasize the complex mix of processes controlling SST variability on ENSO timescales. For example, the importance of remotely forced equatorial waves in mediating SST variability in the eastern and central Pacific can be inferred from Plate 1. Largest ENSO SST anomalies during 1980-1995 were located significantly to the east of the largest zonal wind anomalies; moreover, large SST anomalies were found in the far eastern Pacific where zonal wind anomalies were weak. Waves affect SST in the cold tongue region by inducing changes in thermocline depth which affect upwelling and vertical mixing rates [e.g., Hayes et al., 1991b; Kessler and McPhaden, 1995a, b]. Waves can also advect temperature fields meridionally and, more importantly, zonally along the equator. Wave- and current-induced zonal advection of the eastern edge of the warm pool produces large interannual SST anomalies in the central Pacific [McPhaden and Picaut, 1990; Picaut and Delcroix, 1995; Picaut et al., 1996].

Local air-sea heat exchanges are also important in the surface layer heat balance of the tropical Pacific on interannual time scales [Liu and Gautier, 1990; Hayes et al., 1991b; Kessler and McPhaden, 1995a]. The most strongly varying components of the surface energy balance are solar irradiance, which is modulated by changes in cloudiness, and latent heat flux which is modulated by changes in wind speed, SST, and relative humidity [Liu, 1988; Waliser et al., 1994]. East of the date line, where ocean dynamics are crucial for generating SST anomalies on interannual time scales, latent heat flux tends to increase with increasing SST, and therefore acts as a negative feedback on developing SST anomalies [Kessler and McPhaden, 1995a; Weisberg and Wang, 1997]. In the western Pacific warm pool, the thermocline is deep, mean horizontal SST gradients are weak, and ocean dynamical processes are less capable of generating large scale SST anomalies than further east. In this region air-sea turbulent heat exchange is an important generating mechanism for SST anomalies, through enhanced evaporation during periods of strong westerly winds [Meyers et al., 1986]. Variations in short wave radiation tend to damp developing SST anomalies throughout the tropical Pacific since high cloudiness, which reduces insolation, tends to occur over the warmest surface waters [Waliser et al., 1994].

Data from the TOGA observing system have been used to test various theories of El Niño and the ENSO cycle. An early theory espoused by Wyrtki [1975] suggested that prior to El Niño, the trade winds strengthened, and there was a increase in sea level (a proxy for heat content) in the western Pacific warm pool. When the trade winds weakened, the overcharged warm water pool would collapse and surge eastward in the form of a Kelvin wave to initiate a warm event. The importance of Kelvin waves in the development of El Niño has been confirmed by many studies. However, other aspects of Wyrtki's theory were undermined when prior to the 1982-1983 El Niño, the strongest of the century, there was no anomalous rise in sea level in the western Pacific or intensification of the easterly trades [Cane, 1984]. Similarly, prior to the equatorial warming in 1993, there was no buildup of heat content in the western Pacific warm pool or intensification of the easterlies [Kessler and McPhaden, 1995b].

Wyrtki [1985a] proposed another hypothesis, namely that warm water accumulated in the tropical Pacific prior to an El Niño on a zonally averaged basis between 15°N and 15°S. In this scenario, El Niño represents a mechanism whereby excess heat is purged to higher latitudes. Cane et al. [1986] interpreted the interannual oscillations in their coupled ocean-atmosphere model in terms of this mechanism. Springer et al. [1990], in a wind-forced ocean model simulation, found a buildup of heat content near the equator prior to the 1982-1983 El Niño as hypothesized by Wyrtki, but only between 5°N and 5°S. The difference in latitude bands over which the buildup was assumed to occur resulted from Wyrtki's use of tide gauge station data which had to be interpolated over great distances zonally beyond 5°N-5°S [Springer et al., 1990]. Miller and Cheney [1990], however, did not find a buildup at all prior to the 1986-1987 El Niño event using Geosat data. Thus Wyrtki's [1985a] mechanism, modified to a narrower band of longitudes, may be operative during some but not all El Niño events.

McCreary [1983] proposed a theory for ENSO in which the timescale between warm events was set by the slow westward propagation of long extraequatorial Rossby waves and their reflection off the western boundary as equatorial Kelvin waves. The reflected Kelvin waves would alter thermocline depths (and by proxy SST) in the eastern Pacific, thereby affecting the strength of the trade winds. In order to get a realistic 3-4-year periodicity for the ENSO cycle, Rossby waves with significant amplitudes at roughly 20° latitude from the equator were required. Using XBT data, Graham and White [1988] argued for the existence of extraequatorial Rossby waves along 12°N and 12°S and their reflection into equatorial Kelvin waves at the western boundary. However, Kessler [1990] offered alternative explanations for the observed variability along the equator in terms of direct wind forcing rather than Rossby wave reflection, and Kessler [1991] showed that only Rossby waves equatorward of about 8° latitude could reflect into equatorial Kelvin waves with significant amplitudes.

The delayed oscillator theory of ENSO [Battisti, 1988; Battisti and Hirst, 1989; Schopf and Suarez, 1988] also involves the reflection of Rossby waves into equatorial Kelvin waves at the western Pacific boundary. In contrast to McCreary's [1983] theory though, equatorial Rossby waves closely trapped to the equator, rather than extraequatorial Rossby waves at higher latitudes, are most relevant. Thermocline changes associated with reflected Kelvin waves lead to SST anomalies in the eastern Pacific cold tongue by altering upwelling rates. The SST anomalies affect the atmospheric convection and circulation, giving rise to local positive feedbacks that reinforce the SST and wind anomalies (e.g., Figure 10). The anomalous surface winds in turn excite equatorial oceanic waves of opposite sign to those that generated the original SST anomalies. The timescale for the ENSO cycle in this theory is set by the competition between the local positive feedbacks and delayed negative feedbacks associated with remotely forced equatorial waves and their western boundary wave reflections.

Tests of the delayed oscillator have focused primarily on the question of whether equatorial Rossby waves can reflect from the irregular and gappy coastal geometry of the western Pacific. Theories suggest coastal irregularities should not be a fundamental limitation to this reflection process [Clarke, 1991; du Penhoat and Cane, 1991]. However, although in principle western boundary reflections should work equally well to both initiate and terminate El Niño events, it appears that they are most effective in terminating events [Li and Clarke, 1994; Mantua and Battisti, 1994]. In this situation, reflection of an upwelling Rossby wave at the western boundary excites an upwelling equatorial Kelvin wave train which erodes the warm SST anomaly in the cold tongue, eventually leading to cool La Niña SST anomalies. Even so, not all warm events appear to be terminated by western boundary reflections. Boulanger and Menkes [1995], for example, found that wind-forced upwelling Kelvin waves, rather than boundary-reflected Kelvin waves, led to cooling along the equator in the eastern Pacific in late 1993. Also, Picaut and Delcroix [1995] argued that the 1986-1987 El Niño was terminated by Rossby waves emanating from the eastern boundary, rather than Kelvin waves emanating from the western boundary.

Few, if any, El Niño events of the TOGA decade appear to have been initiated by delayed oscillator physics. Through extended empirical orthogonal function (EOF) analysis of Geosat data during the 1986-1989 El Niño-La Niña cycle, White and Tai [1992] suggested that an equatorial Rossby wave reflected into an equatorial Kelvin wave at the western boundary, consistent with delayed oscillator theory. However, a detailed projection of Geosat sea level and derived surface currents on individual equatorial wave modes indicated very little evidence of first meridional Rossby wave reflection into Kelvin waves during this time [