The intensity of the 1997 El Niño and the 8°C sudden drop in sea surface temperature (SST) around 0°–130°W during the turn into La Niña in 1998 were a surprise to the scientific community. This succession of warm and cold events was observed from start to finish with a comprehensive set of remotely sensed and in situ observations. In this study we employ space-based observations to demonstrate, for the first time, their maturity in capturing the preconditioning, onset, evolution, and decay of the 1997 El Niño and its transition into the 1998 La Niña. An accumulation of warm water in the west and equatorial wave reflection on the western ocean boundary appeared favorable for the development of El Niño. However, the action of a series of westerly wind bursts from December 1996 to June 1997, notably in March 1997, was instrumental in setting up this huge El Niño. The westerly wind bursts excited equatorial downwelling Kelvin waves and advected the eastern edge of the warm pool eastward, which triggered a distinct warming over the central and eastern parts of the equatorial basin. Once these warmed regions joined, the coupling between the SST and surface winds was fully effective, and El Niño reached its mature phase. By that time much of the warm waters of the western equatorial Pacific was transferred toward the east by surface eastward currents. The demise of El Niño and its turn into La Niña in spring 1998 were due to the arrival in the east of various interrelated phenomena. Upwelling was brought from the west by favorable off-equatorial wind stress curl and equatorial Kelvin waves generated by easterly winds and wave reflection on the western ocean boundary. Additional upwelling was brought from the east by equatorial Rossby waves generated by westerly winds. These various upwelling signals were added to the general uplifting of the thermocline because of the slow discharge of the upper layer of the equatorial basin by diverging surface currents. A series of equatorial Kelvin and Rossby waves, characterized by upwelling and opposite surface currents, led to the breakup of the warm waters, the surfacing of the thermocline, and the drastic drop in SST in May 1998 around 0°–130°W. With the arrival of cold water in the east the easterly winds expanded from the west, and La Niña turned into a growing mode. This view of the 1997–1998 El Niño–La Niña, afforded from space, enables the testing of various El Niño theories.
 The 1982–1983 El Niño, considered then as the El Niño of the century, caught the scientific community by surprise. It was well developed when it was first detected. At that time there were scant observations in the tropical Pacific Ocean, provided mostly by merchant ships, tide gauges, and a few equatorial current meter moorings. Except for sparse sea surface temperature (SST) measurements from merchant ships none of these observations were transmitted in real time. In situ SST measurements were already blended with near-real time satellite retrievals, but the eruption of the Mexican volcano El Chichon made these space-based measurements suspect [Reynolds, 1993]. This failure to detect the 1982–1983 El Niño turned out to be a strong argument for the 1985–1994 Tropical Ocean and Global Atmosphere (TOGA) program. This international program succeeded in building an oceanic observing system with most of the data transmitting in near-real time [McPhaden et al., 1998] and in developing a collection of oceanic and coupled ocean-atmosphere models for the tropical Pacific. With the evidence that El Niño is the major disturbing factor of the Earth's climate on seasonal to interannual timescales the efforts of the international community were concentrated in the tropical Pacific. The most important element of the TOGA in situ observing system, the Tropical Atmosphere-Ocean (TAO) array [Hayes et al., 1991], is composed of 70 moorings concentrated in the 8°N–8°S equatorial band, which were all in place at the end of TOGA in 1994.
 Although the TAO array provided unprecedented in situ data coverage, these moorings could not adequately sample the entire tropical Pacific. A series of satellites that were flying during and mostly following TOGA were able to complement the in situ measurements, through their much wider coverage. The progress in SST retrievals, from the National Oceanic and Atmospheric Administration (NOAA) polar orbiting satellites, and their blending with in situ observations resulted in improved SST fields [Reynolds and Smith, 1994]. In 1984, with the launch of the U.S. Navy Geodetic Satellite (Geosat), began an era of consistent altimetry measurements. ERS-1 was launched in 1991, and the dedicated TOPEX/Poseidon altimetry satellite, launched in 1992, enabled the measurement of sea level variations with an accuracy as low as 2–4 cm [Busalacchi et al., 1994; Picaut et al., 1995]. Beginning in 1987, an accurate surface wind field was calculated from space-based observations using the Special Sensor Microwave Imager (SSM/I) data [Busalacchi et al., 1993; Atlas et al., 1996]. These winds were deduced using microwave sensors onboard the Defense Meteorological Satellite Program platforms. Later, space-borne scatterometers such as ERS, NASA scatterometer (NSCAT), and QuikScat provided high-quality global vector winds.
 The TOGA program resulted in a better description and understanding of El Niño, and the TAO array provides crucial subsurface information for El Niño–Southern Oscillation (ENSO) forecast models [Ji and Leetmaa, 1997]. While the lead time for ENSO forecasts is anywhere from 3 months to a year, it will take many more El Niño events before the skill of the current models can be rigorously ascertained. As longer lead times are envisioned, the global view provided by the multiple remotely sensed variables is likely to become increasingly important, not only in constraining these models but also in understanding the various processes. The statistical techniques for seasonal to interannual climate prediction will also be advanced by the availability of these satellite data. Thus it is important to synthesize various remotely sensed variables for understanding how well the climatic processes are represented by satellite data. This is the first such study, albeit focused on the 1997–1998 ENSO, which attempts to do this.
 Many of the theories of El Niño were designed within the TOGA framework. From sea level analyses, Wyrtki  suggested that an accumulation of warm water in the equatorial basin, and in particular in the west, is a necessary precondition for the initiation of El Niño. According to Wyrtki , a warm event appears first through the spreading of the warm water toward the east. Then the upper layer of the equatorial basin is slowly depleted of its water by equatorial and coastal Kelvin waves. The end result is the uplifting of the thermocline in the east and favorable conditions for the development of a La Niña. On the basis of Wyrtki's idea and the positive feedback of the tropical ocean-atmosphere interaction, Jin [1997a, 1997b] constructed a conceptual model for ENSO. His ocean recharge/discharge paradigm relies on the non-equilibrium between the zonal mean equatorial thermocline and wind stress. Schopf and Suarez  and Battisti  proposed a theory, known as the delayed action oscillator. This theory states equatorial upwelling Rossby waves created by the growing wind/SST El Niño interaction in the east are fundamental to the demise of El Niño. These planetary waves reflect on the western boundary as equatorial upwelling Kelvin waves that erode the growing interaction in the east and turn it into La Niña. A revised theory was proposed by Picaut et al. [1996, 1997] and corroborated by Clarke et al. . It was based on the finding of an oceanic zone of convergence on the eastern edge of the western Pacific warm pool, which moves over thousands of kilometers along the equator in phase with ENSO. The simple models used in these studies suggest that zonal advection and equatorial wave reflections on the two ocean boundaries are fundamental to the oscillatory nature of ENSO, in particular, to its manifestation in the central equatorial Pacific. Coupled models often exhibit slow propagation of SST, and this incited Neelin  to propose a SST mode theory. ENSO growth is due to the vertical and horizontal advection by currents adjusted instantly to the wind anomalies (the fast wave limit). In this limit it is the timescale of these advective processes (and not of fast equatorial wave propagation) that sets the growth rate of ENSO. In view of the appearance of easterly winds in the west during El Niño, Weisberg and Wang  analyzed SST and sea level pressure in the western Pacific and came up with a western Pacific oscillator paradigm for ENSO. According to Weisberg and Wang, during El Niño the off equatorial cold SST anomaly in the west induces a high sea level pressure anomaly that initiates easterly wind anomaly. This provides the negative feedback that stops El Niño and turns it into La Niña. Recently, Harrison and Vecchi  suggested an additional mechanism for shifting El Niño into La Niña. At the peak of El Niño during austral summer the warm pool is spread eastward but also southward because of the regular seasonal heating. Hence the associated atmospheric convection and surface zonal winds are also shifted southward. As a result, the zonal wind stress is no longer symmetric about the equator; its oceanic equatorial response diminishes and this could lead to a La Niña. These various mechanisms are thought to explain the underlying cyclic nature of the ENSO events, and several authors [Wang, 2001; Jin and An, 1999] succeed in unifying most of them. However, ENSO does not have a regular cycle, and several mechanisms have been proposed for its irregularity. These include interaction with the seasonal cycle, chaotic behavior, natural and anthropogenic long-term variations, and high-frequency forcing such as westerly wind bursts [Neelin et al., 1998].
 The end of the TOGA decade coincided with a series of El Niño, which were well observed but not forecasted well [e.g., Barnston, 1994]. In addition, it seemed that El Niño events were becoming more frequent [Trenberth and Hoar, 1997]. Despite the prediction by several models of a moderate warm event as early as September 1996 [Barnston et al., 1999] the enormous intensity of the 1997–1998 El Niño was a reminder to the scientific community that many unknowns remain in the understanding of the El Niño mechanisms. In particular, the 1997–1998 El Niño, now quoted as the event of the past century, was followed by a La Niña rapid in its SST signature, contrary to the 1982–1983 El Niño, which was followed by an aborted La Niña.
 Several authors have studied the 1997–1998 El Niño–La Niña using either in situ data provided mostly by the TAO array [McPhaden, 1999] or satellite altimetry data provided by TOPEX/Poseidon [Boulanger and Menkes, 1999; Delcroix et al., 2000]. The purpose of the present study is to take advantage of the broad spatial coverage provided by an abundance of space-based data fields available during the 1996–1998 period and thereby describe and understand the preconditioning, onset, and growth of the 1997–1998 El Niño and its rapid SST change into La Niña over the tropical Pacific. This is the first time that the synoptic evolution of a complete sequence of El Niño and La Niña is described in great detail because of the coverage afforded by the high-resolution space-based observations. Given the unusual character of the 1997–1998 El Niño–La Niña events, it will be instructive to test the various theories of El Niño in view of the present analysis of satellite observations and of previous works. In section 2 the data and their processing are presented. In section 3, a general view of El Niño and La Niña is completed by a sequential presentation of these events going from the preconditioning of the 1997 El Niño to the development of the 1998 La Niña. In section 4 the major results are discussed in view of recent investigations and the various theories of ENSO, and in section 5 a conclusion is presented.
2. Data and Analysis Procedures
2.1. Sea Surface Temperature Data
 The blended SST analysis of Reynolds and Smith  is used to assess the spatial and temporal variability of the observed SST signal. The analysis is an optimal interpolation of all in situ temperature reports from ships, buoys, and satellite SST. It benefits from the spatial coverage of the advanced very high resolution radiometer installed onboard the NOAA polar orbiting satellites and takes into account any satellite biases relative to the in situ measurements. The product used in this study is a monthly mean SST gridded field on a 1° × 1° grid since 1982.
2.2. TOPEX/Poseidon Altimetric Data
 The TOPEX/Poseidon near-real time altimeter analyses of the National Centers for Environmental Prediction [Cheney et al., 1998] are used in this study to represent the observed sea level signal. Each track of the 2 day Intermediate Geophysical Data Record within the area 20°N–20°S and 120°E–76°W is corrected using standard geophysical corrections including the tide correction from the Grenoble tide model [Le Provost et al., 1994]. For each track the average over the 1993–1995 period is removed to eliminate the error in the geoid. Poseidon data are merged with TOPEX data when these data become available. The anomalies are assembled into 1° × 1° × 10 day bins. The resulting data are optimally interpolated in space onto 1° × 1° grids. These 10 day grids are then averaged monthly to provide the sea level data utilized in this study.
2.3. Wind Data
 In this study the wind data resulting from the variational analysis method (VAM) of Atlas et al.  are used. This method combines the SSM/I wind speed data of the Defense Meteorological Satellite Program with other conventional wind data. The wind speed is inferred from the differences in surface microwave emissivity at two wavelengths, and the VAM employs the European Centre for Medium-Range Weather Forecasts analyses as the first guess for assignment of the direction (see Wentz  for more details on the wind speed determination technique). The along-track zonal and meridional components of the wind speed are first converted to wind stress using the bulk formula, then any missing data are filled using a bilinear filter. The SSM/I winds were chosen over other products for several reasons. First of all, the SSM/I product does a very good job of reproducing the observed sea level and subsurface temperature structure when forcing an ocean general circulation model [Hackert et al., 2001]. It was shown that satellite winds, including SSM/I, accurately represent the meridional gradient of the zonal winds, which is the key to modeling the equatorial sea level and thermocline depth. In addition, the surface currents of Lagerloef et al.  utilize the same SSM/I winds for the Ekman currents. Therefore we will use wind stress and curl anomalies that are consistent with those surface currents presented in section 2.4.
2.4. Surface Current Data
 Total near-surface currents are computed as a sum of geostrophic and Ekman parts [Lagerloef et al., 1999]. They are created using a statistical model estimated using observed 15 m drifter measurements. Geostrophic currents are computed from the slope of TOPEX/Poseidon sea level data using an equatorial formulation with a Gaussian weight function. Away from the equator the standard f plane approximation is used. The current near the equator is estimated using a polynomial expansion of the sea level gradient. Just off the equator, where the f plane approximation breaks down, a weight function derived from an inverse error variance serves to provide smoothly varying meridional gradients of geostrophic estimates. The net result of this process allows for a smooth transition from geostrophic approximation f off the equator to an effective estimate β of the currents near the equator.The Ekman term is derived using SSM/I winds with a two-parameter model whose coefficients are given by Lagerloef et al. . The two coefficients represent the amplitude and the turning to the wind and are determined from empirical relationships with the velocity of the 15 m Pacific drifter data set. The winds used for calculation of the Ekman velocity are consistent with those described in section 2.3. Total currents come from the combination of the Ekman and geostrophic components.
 However, the addition of the Ekman term seems to degrade the comparison with the zonal currents at the equator observed at several TAO sites, in particular, in the east [see Lagerloef et al., 1999, Figure 8]. Hence we have used another altimetric-derived geostrophic field for a specific study in the equatorial band. As described by Picaut et al. , a hypothetical drifter moved by the zonal currents averaged in the equatorial band is an efficient construct for highlighting the location of the zone of oceanic convergence in this band and on the eastern edge of the warm pool. The TOPEX/Poseidon currents are derived using the method of Delcroix et al. . Off the equator the first derivative of the sea level slope is used to calculate geostrophic flow. At the equator the second derivative of the sea level slope is used. These currents are averaged between 2°N and 2°S for each degree of longitude. With the removal of the geoid the altimetry and derived current products are relative to some mean, in this case over the 1992–1998 period. The anomalous geostrophic currents are thus added to the long-term mean currents observed within 2°N–2°S from hundreds of near-surface drifters in 1988–1994. The resulting currents are interpolated to fill values every degree along the equator. A hypothetical drifter is then released into these currents and allowed to travel in the equatorial band with the 10 day time step of the TOPEX/Poseidon cycle.
2.5. Additional Procedures
 An instructive way to analyze the mechanisms of ENSO is to consider equatorial Kelvin and Rossby waves. These waves are separated using the TOPEX/Poseidon data and the technique developed by Delcroix et al. . Only the Kelvin and first meridional Rossby modes of the first baroclinic mode will be presented.
 The seasonal cycle blurs the ENSO signal, especially when looking at SST and currents. Therefore the mean seasonal cycle is removed from the data covering the period 1996–1998. The seasonal cycle is formulated using data over the 1993–1995 period, which was imposed by the relatively short record of TOPEX/Poseidon altimetric data. Comparison of mean seasonal cycles of SST over the 1993–1995 period and over a longer period, shows that the 1993–1995 seasonal cycle is relatively close to climatology.
 In order to assess the displacements of the warm pool during the 1997–1998 El Niño–La Niña, figures that superimpose SST anomalies and surface current anomalies were considered. However, the representation of anomalous SST is misleading for identifying the areas of notable air-sea interaction since atmospheric convection is far more vigorous over SST warmer than 28°C [Gadgil et al., 1984; Graham and Barnett, 1987]. Hence we will present several figures that superimpose anomalous surface currents and the sum of anomalous SST and the absolute mean SST. This procedure will permit joint analysis of ENSO signals in SST and surface current that are consistent regarding their separation from the seasonal cycle. To avoid any confusion with SST, the sum of anomalous SST and the absolute mean SST will be named pseudo SST. Note that there will be little difference between SST and pseudo SST in the western equatorial Pacific because of weakness of the seasonal signal as compared to the ENSO signal [Picaut and Delcroix, 1995].
3. Sequential Analysis of the 1997–1998 El Niño–La Niña
3.1. General Presentation
 In this section the global perspective of the evolution of the series of ENSO events over the 1996–1998 period is presented in two ways. A multivariate empirical orthogonal function (MEOF) analysis summarizes the common space-time variation over the tropical basin of the main parameters used in this study, while time-longitude diagrams highlight the main space-based data evolving along the equatorial band. Note that such an MEOF analysis that combines a complete set of remotely sensed products (SST, sea level, wind stress, and surface currents) is presented here for the first time. The spatial structure of the first MEOF (Figure 1a1), which accounts for 19% of the total variance, corresponds to the pattern of the mature phase of ENSO, especially in terms of the SST and surface wind stress interaction. The SST maximum is trapped near the equator in the east, and westerly winds converge into the SST anomaly during the warm phase. Sea level exhibits the characteristics signatures of equatorial Kelvin waves in the east and equatorial Rossby waves in the west. In the center western part of the basin the sea level patterns extend well beyond 8° from the equator and thus cannot be detected by the limited range of the in situ TAO array. The surface currents are dominated by the strong signature of the equatorial Kelvin and Rossby waves, are clearly trapped near the equator, and cover most of the equatorial basin. This emphasizes the strong surface flow and exchange of mass from the west to the east in the equatorial band during the mature phase of El Niño (and from the east to west during La Niña). The time series of the first MEOF resembles the inverse of the Southern Oscillation Index and highlights the 1995–1996 weak La Niña and the drastic changes due to the 1997 El Niño and the 1998 La Niña. The second MEOF (Figure 1b) accounts for 17% of the total variance. With its time series resembling that of the first MEOF with a lead of about 7 months it describes, in part, the onset of El Niño and, most of all, the transition into La Niña. In particular, the spatial patterns for MEOF2 resemble the set up of the 1997 El Niño in March–April 1997 (for the time series >0), and they track the transition to the 1998 La Niña in May–June 1998 (for the time series <0). During the set up of the cold phase of ENSO, these patterns are characterized by easterly wind anomalies in the west and greater than normal trades off the equator, some cooling at the equator just east of the dateline and depressed sea level in the central and southwest basin. For surface currents, this mode exhibits the equatorial jet associated with the onset of ENSO west of 140°W. This feature extends east of 140°W in a weaker pattern shifted north of the equator at about 2°N and to a lesser degree shifted south at about 7°S. Interestingly, this second mode also reveals the overall meridional convergence (divergence) of flow into (from) the equator between 12°N and 10°S and 180° and 150°W during the onset of El Niño (La Niña).
 The longitude-time diagrams of the zonal wind stress, sea level anomalies and SST along 2°N–2°S (Figure 2) and of the equatorial Kelvin and first meridional Rossby waves (Figure 3) illustrate the dynamical response of the equatorial ocean to wind stress forcing. Superimposed is the trajectory of a hypothetical drifter advected by the derived geostrophic currents averaged within 2°N–2°S. Such a trajectory emphasizes the horizontal advection of the eastern edge of the warm pool and its role in the mechanisms of ENSO [Picaut et al., 1996, 1997]. The weak 1996 La Niña was marked by easterly winds stronger than normal over most of the equatorial basin. This increased the equatorial upwelling in the east and pushed the warm pool west of the dateline. In the central and eastern basin the sea level dropped in response to the arrival of equatorial upwelling Kelvin waves (EUKWs) and rose in the west with the arrival of equatorial downwelling Rossby waves (EDRWs). The 1997 El Niño began in the west with a series of westerly wind bursts (WWBs) in December 1996 and March and June 1997, each of which had ever increasing fetch as the eastern edge of the warm pool was displaced eastward. These WWBs excited a series of equatorial downwelling Kelvin waves (EDKWs) that propagated toward the east, where they stopped the seasonal equatorial upwelling and resulted in positive SST anomalies. Equatorial upwelling Rossby waves (EURWs) were excited with the migration of the westerly winds with the warm pool over the central equator region in mid-1997. They propagated toward the west where they decreased the sea level. The reversal of the 1997 El Niño into the 1998 La Niña was due to anomalous easterlies in the west that excited EUKWs toward the east and the return of the warm pool toward the west.
3.2. Preconditioning of the 1997–1998 El Niño
 During the 1995–1996 weak La Niña, SSTs were characterized by waters colder than normal in the central equatorial Pacific. Coupled to the cold SSTs in the central equatorial Pacific were stronger than normal easterlies, most notable during the first 4 months of 1996 (Figure 4, February 1996). This increased the equatorial upwelling in the east and pushed the warm pool west of the dateline (Figure 2). The anomalous easterlies generated EUKWs and EDRWs over most of 1996 (Figure 3). The generation of the downwelling waves can also been seen through the wind stress curl maps (not shown), which were favorable for driving EDRW during most of 1996. As a consequence of this wind forcing and oceanic wave propagation, the westward South Equatorial Current (SEC) was stronger than normal, at first in the western equatorial Pacific during the first half of 1996. In fall–winter 1996 the westward surface current anomaly covered most of the equatorial band (Figure 5, December 1996). EDRWs and the anomalously strong SEC led to an accumulation of warm water in the west during 1996. This can be seen through the horseshoe pattern of positive sea level anomaly of Figure 1a (with the time series <0 in 1996). As noted by Wyrtki , the accumulation of warm water in the west was instrumental for the development of the 1997–1998 El Niño.
3.3. Onset of the 1997–1998 El Niño
 Superimposed upon this buildup of warm water in the west and its transmission from the western boundary into continuous EDKWs (Figure 3) was the action of a series of WWBs. It has been hypothesized that the 1997 El Niño needed the action of the WWBs to occur with such strength. The first significant WWB came in December 1996 in the extreme west (Figures 2a and 4). It generated an EDKW that manifested itself as a positive sea level anomaly that propagated eastward (Figures 2b and 3). Along its characteristic this wave encountered the diminishing effect of the stronger than normal easterly wind in the center part of the basin (Figures 2a and 4). Nevertheless, this EDKW was still capable of reaching the eastern side of the basin in March 1997, where it produced a slight warming (Figures 2c and 4). With the help of the reflected equatorial waves on the western ocean boundary, this wind-forced EDKW induced an anomalous eastward surface flow along the equator, which was particularly important in the eastern equatorial basin in February–March 1997 (Figure 5). In the west the surface eastward jet induced by the December WWB, broadened by a slightly stronger than normal North Equatorial Countercurrent (NECC) from December 1996 to February 1997, resulted in a small eastward displacement (about 10° longitude) of the eastern edge of the warm pool (Figures 2c and 5). This set up favorable SST conditions for increasing the fetch of the subsequent WWBs.
 In early March 1997 a strong and most effective WWB of >0.4 dynes cm−2 over a fetch of 40° (Figures 2a and 4) occurred in the western equatorial Pacific. Its amplitude over the ocean equatorial waveguide was not much stronger than that of the December burst, but its longer fetch resulted in a stronger oceanic response. In addition, the EDKW generated by this forcing was not damped in the central Pacific since the easterly wind anomalies in the central equatorial Pacific were weaker as compared to the December 1996 WWB (Figures 2a and 4). The EDKW excited by the March 1997 WWB was seen in the TOPEX/Poseidon data with a clear equatorially trapped sea level signal propagating eastward at about 2.5 m s−1 (Figure 2b). Given the basin-wide wavelength of the equatorial Kelvin wave, this resulted in a nearly continuous eastward flow anomaly all along the equatorial band starting April 1997 (Figure 5). The strong eastward flow was the main reason for moving the eastern edge of the warm pool by about 2000 km in 2 months (Figures 2c and 5, April 1997). This notable displacement of the warm pool set the condition for an increased westerly wind fetch.
 In June 1997 a third WWB was clearly detectable (Figure 2a). Its fetch along the equator was about 4 times longer than the corresponding December WWB and 2 times longer than the March WWB. Because of the successive eastward displacements of the warm pool, the central equatorial Pacific wind fetch in June 1997 was well situated for El Niño coupling between surface wind and SST (Figures 2 and 6, June 1997). The equatorial Kelvin wave signature of this last WWB and its propagation was still discernible in TOPEX/Poseidon data (Figures 2b and 3). As the westerly wind forcing moved toward the central part of the basin, EURWs were generated to the west.
 The reflections of equatorial waves on the ocean boundaries paved the way for the spreading of an eastward flow all along the equator. As noted above and seen in Figure 3, the EDRWs associated with the 1996 La Niña reflected at the western boundary as EDKWs, which helped the EDKWs forced by the WWBs in December 1996 and March 1997. On the other side of the basin the EUKWs associated with the 1996 La Niña reflected at the eastern boundary as equatorial EURWs, which brought along the equator a westward current of opposite sign to the original EUKWs (i.e., eastward). As seen in Figure 3b, this relatively weak eastward current joined the wind-forced and wave-reflected currents issued from the west around April–May 1997. Altogether, an eastward anomalous flow appeared all over the equatorial band during most of 1997 (Figure 5, April 1997). However, the direct effects of the WWBs in the west were more fundamental to the growth of El Niño. They were the main reason for moving the eastern edge of the warm pool eastward, thus increasing the wind fetch and finally setting up the ocean-atmosphere system into a fully coupled mode after spring 1997.
3.4. Mature Phase of the 1997–1998 El Niño
 Once the El Niño coupled system was developed, its center of action was moved into the central eastern equatorial Pacific. This can be seen from the wind/SST interaction depicted in Figures 1a and 6. The eastward anomalous current was established all along the equator and induced an eastward transport of mass of more than 50 Sv within the top 120 m along the equator [Johnson et al., 2000]. Because of the deepening of the thermocline by EDKWs, the normal seasonal equatorial upwelling was greatly diminished in boreal summer, and this sustained the warming in the east.
 By the end of 1997 the advection of the warm pool, as represented by the hypothetical drifter trajectory, reached its maximum eastward location at about 125°W (Figure 2c). The warm water to the west of the eastern edge of the warm pool joined the warming in the east resulting from the EDKWs and the regular seasonal heating off the coast of Central America (Figures 2c and 7, December 1997). This resulted in surface temperatures exceeding 28°C throughout the equatorial basin and a maximum of westerly wind anomalies in the center of the basin (Figures 2a and 6, January 1998). The relative eastward location of the strongest westerlies provided a long fetch for EURWs to develop. Evidence of strong EURW signals can be seen west of 170°W from June through December 1997 in Figure 3b. These EURWs, together with the strong eastward flow, depleted the warm water of the western basin and uplifted the thermocline. The thermocline in the west started to shoal compared to its mean position as early as August 1997 (Figure 2b), and this slowly set up favorable conditions for the development of the 1998 La Niña.
3.5. The Turn Into the 1998 La Niña
 The shoaling of the thermocline in the western tropical Pacific was more pronounced in the north and south away from the equator because of the specific meridional pattern of the EURWs excited by the westerly winds in the central equatorial basin. However, starting in June 1997, the upwelling favorable wind stress curl in the northwest made this shoaling occur first in this region, as seen with the negative sea level anomaly of Figure 8 (October 1997). Then in early 1998, wind stress curl of the opposite sign resulted in a similar shoaling of the thermocline in the southwest, which culminated in April 1998 (Figure 8). The shoaling of the subsurface thermal structure in the northwest and then in the southwest, manifested as EURWs, reflected on the western boundary in the form of EUKWs (Figure 3). These reflected waves enhanced the EUKWs, forced by easterly wind anomalies that showed up in the extreme west of the basin in boreal winter 1997–1998 (Figures 2a, 6, and 1b). The continuous arrival of these EUKWs since September 1997 (Figure 3) slowly reduced the El Niño tendency for a deeper thermocline in the east and eventually led to a surfacing of the thermocline in spring 1998. Note that the impact of the wind-forced EUKWs was gradually strengthened by the increasing fetch of the easterly wind anomalies from December 1997 to July 1998 (Figure 2a). Over this same period the associated westward and meridionally divergent surface currents (Figure 1b) depleted the surface layer of its warm water and contributed to the gradual uplifting of the thermocline in the east. Another phenomenon that contributed to this uplifting was the remaining westerly wind anomalies that were associated with the warm SST in the east in January–April 1998 (Figures 2a, 6, and 7). These westerlies spawned EURWs that propagated westward of 110°W, inducing a broad uplifting of the equatorial thermocline.
 The combined impact of the EURWs and EUKWs was fundamental in the generation of the sudden drop of SST in May 1998 around 0°–130°W. In order to highlight the interaction between SST and the propagation of the waves associated with this cooling, a complex multivariate empirical orthogonal function (CMEOF) analysis was done combining the fields of SST and zonal surface currents associated with the equatorial Kelvin and first meridional Rossby waves. This technique is more appropriate for application to propagating features than using standard MEOF as in Figure 1. In order to simplify the interpretation of the temporal evolution each mode was reconstructed separately for each variable. Then the gridded results were formulated into longitude versus time plots along the equator for the period 1996–1998.
 The first (second) mode accounting for 66% (20%) of the total variance corresponds to the broad westward (eastward) propagation associated with the development of El Niño (transition to La Niña), confirming those features as described in the text in sections 3.3'Mature Phase of the 1997–1998 El Niño'–3.5. These two modes add no new information, so they are not shown. However, the third CMEOF is particularly useful for detailed analysis of the warm pool separation around 0°–130°W. The third CMEOF (5% of the total variance) depicts the specific sequence of constructive interference of the equatorial waves in relation to this unique SST drop. In this manner we have used the CMEOFs to isolate the propagating signals giving rise to the abrupt La Niña transition. As seen in Figures 9a9 and 3, the EUKWs and EURWs intersected around 130°W, starting May 1998. Most important was the opposite sign of the surface currents associated with the arrival of these equatorial upwelling waves (i.e., westward for EUKW and eastward for EURW). As early as March 1998, the zonal currents in the equatorial band were of opposite sign on each side of 130°W. The arrival of EDRWs between 150°W and the dateline in March–June 1998 (probably issued from the reflection of EDKWs on the eastern boundary, Figure 3) enhanced the westward currents west of 130°W. The opposing surface zonal currents led to the breakup of the warm waters at this location (Figure 9b). This breakup not only shows up in the CMEOF analysis, but it is also illustrated in the two drifter trajectories of Figure 2c, which clearly diverge in March and May 1998 between 110°W and 150°W. In <2 months the uplifting of the thermocline, associated with the arrival of equatorial upwelling waves in this area, resulted in a surge of cold water at the surface and the 8°C drop in SST around 0°–130°W.
 These features were not symmetric to the equator. As noted by Murtugudde et al.  and in Figure 10, anomalous northerly winds generated an Ekman upwelling, which shifted the EUKWs to the north by about 2°. Between March and June 1998, cold water between 120° and 140°W slowly propagated from 5°N to the equator (Figure 9b). In the same way the breakup of the warm waters by divergent surface zonal currents was not symmetric to the equator. The northern shift of the EUKWs resulted in a similar shift in the westward surface currents, as can been seen in Figures 7 (April 1998) and 1b. With the collapse of the northerly winds in April 1998, the breakup of the warm waters, and the displacement of upwelling toward the equator, in May 1998 a surge of cold water manifested itself as the beginning of the La Niña centered at 130°W and the equator.
3.6. Development of the 1998 La Niña
 With the outbreak of cold water at the equator around 0°–130°W the northeast and southeast trades increased in magnitude and converged into the equator in the west, slowly extending the fetch of the easterly wind anomaly from the west (Figures 2 and 10). Stronger EUKWs were generated in June–July 1998 (Figure 3), which increased the equatorial upwelling in the east (Figure 11). The combination of the resulting negative sea level at the equator and of the positive sea level anomaly near 10°N (Figure 10) and the associated slope caused a broad band of westward current that pushed most of the warm pool well into the west (Figures 2c and 11). Because of the asymmetry of the current structure relative to the equator, some of the warm pool remained in the Southern Hemisphere for several months. Farther east, the development of the wind/SST feedback associated with the cooling in the central equatorial Pacific was also restrained for several months by the remains of warm water in the east, noted above (Figures 2 and 10). During summer 1998 a continuous stream of EUKWs helped to slide the negative sea level and coherent cold SST anomalies to the east, slowly eroding the warm water remnants along the equator in the east. By September 1998 these warm waters had completely disappeared from the equatorial band, and the coupled La Niña system was in full strength, with easterly wind anomaly over most of the central western equatorial basin (Figure 10). The cold SST and negative sea level anomalies were trapped to the equator (Figures 10 and 11, September 1998), suggesting the dominance of vertical upwelled motion in the generation of the La Niña event over other processes. Finally, it took several more months for the remnants of the warm anomaly off Peru to disappear under the reinforcement of the southeast trades.
 The aims of this section are to examine these results of the 1997–1998 El Niño–La Niña mainly on the basis of the heretofore unprecedented coverage of remotely sensed observations, taking into consideration earlier investigations. The elements that made this ENSO sequence so unusual are also discussed, with emphasis on the reasons for the strength of the 1997 El Niño and the sudden shift of SST into La Niña. Finally, the different theories of ENSO are discussed in view of these various results concerning the 1997–1998 El Niño–La Niña.
4.1. Preconditions for a Strong El Niño
 During the past few decades the frequency and intensity of the El Niño events have increased significantly [e.g., Trenberth and Hoar, 1997]. Decadal oscillations may have been the reason for such an increase [e.g., Lau and Weng, 1999], but global warming was also seen as a possible candidate [Timmermann et al., 1999]. In any case such a tendency was favorable for setting up a strong El Niño after the weak La Niña of 1995–1996.
 A buildup of warm water in the western Pacific warm pool prior to the 1997 El Niño was noted in section 3. In a study on the ENSO zonal displacements of the eastern edge of the warm pool, Picaut et al.  found that the zone of convergence, which usually marked this edge, was pushed to the western boundary during the 1995–1996 La Niña. This unusual feature occurred only twice in observations and model simulations over the past 20 years, briefly during the 1988–1989 La Niña and then in 1995–1996. In particular, an ocean general circulation model (OGCM) forced by the Florida State University (FSU) wind stress since 1979 confirmed the unusual behavior of the westward surface currents in the equatorial band during the 2 years preceding the 1997 El Niño. As noted by Potemra and Lukas , most of the buildup in 1995–1996 happened symmetrically about the equator, suggesting the importance of EDRWs as a mechanism. Despite the nearly perfect reflection of equatorial waves on the irregular western boundary suggested by Boulanger and Menkes  over the 1993–1998 period, the relative importance of the EDRWs and local wind forcing in this buildup remains an issue. For example, the variation of heat storage in the northern part of the warm pool was mainly attributed to the local wind stress curl [Wang et al., 1999].
4.2. A Strong El Niño
 The buildup of warm water in the west reached its maximum at the end of 1996. During most of 1996 the warm pool remained in the west, resulting in a reduced fetch for the WWBs. However, the WWBs in December 1996 and March 1997 were very energetic, and the last one benefited from an extended fetch. According to Yu and Rienecker  these WWBs were embedded with the Madden-Julian Oscillation (MJO) originating from the Indian Ocean. Most of all, the westerlies were reinforced by equatorial twin cyclones, which were helped by cold surges from the northwest Pacific. The intensity of the March WWB, its fetch, and nonlinear oceanic processes at the eastern edge of the warm pool were important factors for making the 1997 El Niño event so strong [McPhaden, 1999; Vialard et al., 2001; Boulanger et al., 2001]. This is confirmed by the fact that dynamical models, which succeeded in predicting the 1997 El Niño warming as early as fall of 1996, were unable to forecast its intensity [Barnston et al., 1999; Anderson and Davey, 1998; Landsea and Knaff, 2000]. In fact, these models had to wait to incorporate the effect of the March 1997 WWB to predict a strong warming for the end of 1997. This emphasizes the need for a better understanding of stochastic and high-frequency forcing such as MJO and WWBs in order to improve ENSO prediction [Moore and Kleeman, 1999; Slingo et al., 1999].
 Other concurrent factors helped this warm event to become so strong. Zhang and Busalacchi  suggest that a shallow pathway into the north branch of the warm pool may have brought warm SST anomalies in the NECC region at the end of 1996 and early 1997. As noted in section 3.3, the increase of the NECC helped the eastward migration of the eastern edge of the warm pool, and thus increased the fetch of the subsequent WWBs. Equatorial wave reflection at the western boundary also helped to set up the 1997 El Niño. The 1996 La Niña generated EDRWs that reflected on the western boundary, reinforcing the wind-forced EDKWs. These equatorial waves participated in the generation of El Niño in early 1997 by moving the eastern edge of the warm pool eastward and pushing down the thermocline in the east. According to the studies of McPhaden and Yu , Delcroix et al. , and Boulanger and Menkes  the amplitude of the EDKWs issued from the reflected EDRWs on the western boundary was smaller than the wind-forced EDKWs. Hence these authors recognized the dominance of the WWBs in setting up the strong 1997–1998 El Niño.
4.3. The Shift Into La Niña
 SST is the oceanic parameter that directly interacts with the atmosphere. With the 8°C drop in SST around 0°–130°W in May 1998 the termination of the 1997 El Niño was considered as abrupt [e.g., Takayabu et al., 1999]. Despite its geographic limitation to ±8° from the equator the TAO subsurface data indicated that the transition into La Niña was, in fact, a rather smooth progression over many months. The larger coverage of the remotely sensed observations shows that the seed of this transition can be found already in October 1997 northwest of the equatorial band (Figure 8).
 Several authors have discussed the components that resulted in the transition from El Niño to La Niña. Wang and Weisberg  pointed out that the off-equatorial western Pacific SSTs were cold during El Niño (Figure 1). This induced positive sea level pressure anomalies that in turn, initiated equatorial easterly winds over the western Pacific in fall 1997. Historically, the existence of easterlies in the west during El Niño was known [Rasmusson and Carpenter, 1982; Deser and Wallace, 1990], but not that many authors have looked at their possible role in the transition from El Niño to La Niña. Wang et al.  focused on the associated elevation of the thermocline in the western equatorial Pacific as a trigger of the eastward migration of the rising thermocline. Boulanger and Menkes [1999, 2001], McPhaden and Yu  and Delcroix et al.  stressed the role of EUKWs generated by these easterlies in the decay of El Niño and its turn into La Niña. These authors also suggested that the reflection of EDRWs at the western boundary into EUKWs was another factor, of similar or smaller importance, through the uplifting of the equatorial thermocline in the east and the retreat of the warm pool. However, these mechanisms are not sufficient to explain the sudden arrival of cold water at the surface around 0°–130°W in May 1998. Takayabu et al.  proposed an MJO with a strong easterly signature over the equatorial ocean in May 1998 as a triggering mechanism for the rapid SST shift from El Niño to La Niña. The erosion of the shallow mixed layer in the central equatorial Pacific by the rapid return of easterly winds in May 1998 was certainly another factor. The current paper stresses other concomitant factors, in particular, the breakup of the warm pool in early 1998 associated with the coincidental arrival around 0°–130°W of equatorial waves from the west and the east. McPhaden and Yu  and Delcroix et al.  noted the constructive uplifting of the thermocline by equatorial upwelling Kelvin and Rossby waves, but they did not detect the breakup of the warm waters by opposite surface currents. The constructive uplifting of the thermocline by the arrival of equatorial upwelling waves was also noted by Picaut and Delcroix  during the turn from El Niño to La Niña in 1988. The breakup of the warm waters is also perceptible in SST data in 1983–1984. Unique to the 1998 La Niña is the constructive arrival of the breakup of the warm waters and of the uplifting of the thermocline, which led to the sudden and drastic drop in SST.
4.4. Testing ENSO Theories
 As noted in section 3.2, the buildup or recharge theory of Wyrtki  seemed to be at work for the initiation of the 1997 El Niño event. During most of 1997 the equatorial band was occupied by eastward anomalous surface currents, which peaked in fall–winter when the El Niño wind/SST interaction was at its strongest (Figure 6, January 1998). This resulted in a progressive depletion of the western equatorial basin from spring 1997 to winter 1997–1998 (Figure 2b). Then the anomalous surface currents became slowly westward and divergent (Figure 7). This depleted the equatorial basin of its warm waters and led to La Niña. As noted in section 3 and in particular in Figure 8, the discharge concerned the whole tropical band, with a depletion of water occurring first in the northwest and then in the southwest. The exchanges of warm water during the ENSO cycle of 1997–1998 are summarized in the two MEOF of Figure 1. The first one, represents the exchange within the equatorial band during the mature phase of ENSO, while the second represents the exchanges within the southwest central and the northern regions during the setup of El Niño and most of all during its turn into La Niña. Several authors have tentatively interpreted EOF observational analyses in a similar way for various ENSO events: Delcroix  over the 1961–1995 period, Johnston and Merrifield  over 1975–1997, Xue et al.  over 1980–1996, and Meinen and McPhaden  over 1997–1998. However, none were able to show the converging movement in the surface currents as in the second MEOF of Figure 1b. This feature, only discernable through remotely sensed observations, indicates that the equatorial band was recharged mostly from the middle of the basin and from a significant distance away from the equator. In any case the asymmetry of the recharge/discharge was not anticipated by the paradigm of Jin [1997a, 1997b].
 The delayed action theory was shown to be active during this ENSO cycle. As noted several times in this paper and in others, EDRWs excited during the 1996 La Niña reflected as EDKWs between the fall of 1996 and spring of 1997. Some of these waves propagated to the east and pushed down the thermocline. Hence this reflection process participated in the annihilation of the equatorial seasonal upwelling and in the subsequent warming of this region. Conversely, the reflection of EURWs generated during El Niño at the western boundary reinforced a series of EUKWs that reduced the El Niño warming in the east and helped shift it into La Niña in spring 1998.
 The revised or convergence zone theory of Picaut et al. , also referred to as zonal advective feedback by An et al. , was at work during this ENSO cycle [Delcroix et al., 2000]. With an extensive displacement of the eastern edge of the warm pool during the 1997 El Niño the zone of convergence on this edge was not as well defined as it was during normal ENSO cycle [Picaut et al., 2001]. As seen in Figure 3, the reflection of EDRWs generated during the 1996 La Niña at the western boundary as EDKWs induced eastward surface currents that helped push the eastern edge of the warm pool. However, the latter contribution of the reflection of EUKWs at the eastern boundary into EURWs was not significant given the modest amplitude of the reflected waves and their dissipation over the distance they had to travel to reach the eastern edge of the warm pool in May–July 1997. On the contrary, during El Niño the reflection of EDKWs at the eastern boundary resulted in significant EDRWs and therefore in strong westward equatorial current (Figure 3), which worked to stop and eventually reverse the eastward movement of the warm pool as early as December 1997 (Figure 2). The reflection of the EURWs associated with El Niño at the western boundary as EUKWs were an additional factor for moving the warm pool westward a few months later. However, as pointed out by Picaut et al. , a difficulty in this theory of ENSO may happen for a very strong El Niño like in 1997–1998. With the SST nearly constant all along the equatorial band in boreal winter 1997–1998 (Figures 2c and 7) the return of the warm pool by the westward current will not change SST much, and a source of cold water is needed at the surface in order to transition into La Niña. As noted earlier, the depletion of the central equatorial basin by the diverging current was one of the reasons for uplifting the thermocline in winter–spring of 1998 in the eastern central equatorial basin. More uplifting came from the EUKWs forced by the easterly wind anomaly in the west and from the reflected waves on the western boundary following the original delayed action oscillator theory.
 In a study of the 1997–1998 ENSO, Wang and Weisberg  stated that their western Pacific oscillator conceptual model [Weisberg and Wang, 1997] was consistent with the onset and decay of El Niño. Twin off-equatorial anomalous cyclones were part of the WWBs that started the 1997 El Niño. When the warm event was fully developed, the off-equatorial western Pacific SSTs changed from slightly warm to slightly cold. These cold SSTs were accompanied by a change in sea level pressure in the same region, which may have been responsible for the arrival of the easterly wind anomalies in the extreme west near the equator. As noted above, the easterlies were important in the excitation of EUKWs and thus in the turn into La Niña.
 There was a clear slow eastward propagation of SST during the 1997 El Niño, associated with the displacement of the warm pool (Figure 2). From the trajectory of the hypothetical drifter on Figure 2c and the studies of Picaut and Delcroix  and Picaut et al. , it appears that the main mechanism for this slow propagation is advection by zonal current anomaly in the equatorial band. Hence the propagation speed of SST is slower than those of equatorial waves. All these features (slow propagation, dominance of zonal advection, and fast wave limit) are consistent with the SST mode theory of Neelin .
 Finally, high-frequency forcing, such as MJO and WWBs, was definitely at work in the generation of the 1997–1998 El Niño. As seen in section 3.3, the succession of WWBs in December 1996 and March 1997 was determinant in advecting the warm pool eastward and in generating EDKWs. This high-frequency forcing was likely responsible for the abnormal strength of the 1997–1998 El Niño but also for the difficulty in its forecast prior to the March 1997 WWB [Barnston et al., 1999; Moore and Kleeman, 1999].
 The installation of an ENSO oceanic observing system was completed at the end of the international TOGA program [McPhaden et al., 1998]. In addition to this in situ observing system, a series of research satellites was flying during the 1996–1998 period. Hence the 1997–1998 El Niño–La Niña was by far the most observed ENSO. SST, sea surface topography, and surface wind observations were made from space, providing a wider and more comprehensive coverage than the TOGA observing system could provide. In particular, the accurate TOPEX/Poseidon altimeters together with satellite-derived winds enable a reliable estimate of surface currents over the tropical Pacific [Lagerloef et al., 1999]. Owing to the two-layer approximation for tropical ocean models, altimetry data are a useful indicator of the heat content above the thermocline and of the depth of the thermocline. In view of the excellent space-time coverage afforded by remotely sensed data alone the present study demonstrates the value of these data in providing comprehensive and insightful information on the mechanisms of the 1997–1998 El Niño–La Niña.
 Several factors contributed to the rapidity and strength of the El Niño event. With the 1995–1996 weak La Niña an initial buildup of warm water in the western equatorial Pacific and equatorial wave reflection on the western boundaries were favorable factors. However, the succession of the westerly wind bursts in December 1996 and March and June 1997 was fundamental to the onset time and amplitude of El Niño. In particular, they were able to push the eastern edge of the warm pool to the east, increasing their own fetch and thus their response into the ocean. This led to an unstable air-sea coupled system in summer 1997, which increased further the eastward displacement of the warm pool and the surface warming in the east produced by the arrival of EDKWs. By August 1997 the warming from the west and the east joined, and the entire equatorial band was covered with water >28°C. During the setup phase of El Niño the accumulation of warm water in the western equatorial Pacific was spread toward the eastern equatorial basin by eastward currents. During the end of the mature phase the warm water in the equatorial band was slowly depleted by westward and meridionally surface divergent flow. Hence the thermocline in the eastern part of the equatorial basin was pushed down during the end the setup of El Niño and then upwelled during the end of the mature phase. Eventually, this set up favorable conditions for the turn into La Niña.
 The turn into La Niña in spring 1998 was due to the combination of additional factors. Favorable wind stress curl in the northwest during fall 1997 and several months later in the southwest, together with the reflection of EURWs on the western boundary, generated EUKWs. The advent of easterly wind anomalies in the west at the end of 1997 also generated such waves. These waves brought an additional upwelling signal toward the central eastern part of the equatorial basin. In the east, another upwelling signal propagated toward the west in the form of EURWs generated by the remains of the westerly wind anomalies in this region. It is the coincidental arrival of a series of equatorial waves from the east and from the west near 0°–130°W that resulted in the sudden drop of SST in May 1998 and in the turn into La Niña. Opposing zonal surface currents associated with these equatorial waves resulted in a breakup of the warm waters around 0°–130°W starting in March 1998. Two months later and at the same location the superposition of the two upwelling signals associated with the equatorial Kelvin and Rossby waves resulted in a strong uplifting of the thermocline, which led to the rapid drop of SST.
 A lesson to remember from the study of the 1997–1998 El Niño is the decisive role of the westerly wind bursts. Despite increasing evidence of their relation with El Niño [e.g., Kessler et al., 1995; Moore and Kleeman, 1999] none of the current theories of ENSO includes these bursts. These theories explain the oscillatory nature of ENSO in its basic state, and the present study using exclusively space-based data indicates that these theories were all legitimized at various stages. The delayed action oscillator appeared to be at work during the initiation of El Niño and the turn into La Niña. The importance of the displacement of the eastern edge of the warm pool substantiates the convergence zone or advective feedback theory. The action of the easterly wind anomalies in the turn into La Niña supports the western Pacific paradigm. Finally, the recharge/discharge mechanism was also at work during this sequence of ENSO. In particular, during the setup of La Nina, meridionally diverging westward surface flow, located near the middle of the basin (between 180° and 150°W), depleted the equatorial basin from its warm water (Figure 1b). The present study indicates that these various theories have worked constructively to help understand the 1997–1998 El Niño–La Niña. However, more work needs to be done with the remarkable data set collected from space and in situ over the 1997–1998 period and with oceanic and coupled models in order to understand fully the specific characteristics and differences of the El Niño of the twentieth century.
 We would like to thank R. Cheney and his group for providing the TOPEX/Poseidon data, T. Delcroix for providing the TOPEX/Poseidon-derived current field, and R. Reynolds for providing the SST data, and R. Atlas for providing the SSM/I product. In addition, we acknowledge Alan Leonardi, whose expertise on CMEOF proved to be very valuable. Much of the work was done at NASA Goddard Space Flight Center during the 2 year visit of the first author. This research was supported by NASA, IRD, CNES, and PNEDC.