Breakdown of ENSO predictors in the 2000s: Decadal changes of recharge/discharge-SST phase relation and atmospheric intraseasonal forcing



[1] Variations in the warm water volume (WWV) of the equatorial Pacific and atmospheric forcing from intraseasonal variation (ISV) in the western equatorial Pacific are regarded as two good predictors of the subsequent El Niño/Southern Oscillation (ENSO), with a lead time of two to three seasons. Here we report that the robust predictability of these predictors for ENSO has changed in the 2000s. During 1981–2000, the recharge (discharge) of the WWV and strong (weak) ISV forcing preceded El Niño (La Niña) by two to three seasons. However, in the 2000s, the interrelationship between the WWV/ISV and following ENSO became weak, especially for the El Niño/La Niña events after 2005. Notably, the discharged phases of WWV that led to subsequent La Niña events were less observed since 2001. These changes may be caused by frequent occurrences of the “warm-pool El Niño,” which is characterized by SST anomalies centered in the central equatorial Pacific.

1. Introduction

[2] Forecasting of El Niño/Southern Oscillation (ENSO) and estimation of its global influence are important topics in climate research. In recent years, advancing forecast systems have been able to predict ENSO with about a 1-year lead time. Successful predictions are based on accurate initializations of ocean temperature [e.g.,Balmaseda and Anderson, 2009] from the ENSO observing system, which includes the TAO/TRITON buoy array, satellite altimetry, and Argo floats. Representation of the ocean-atmosphere coupling dynamics in models is also a key for interannual-scale numerical prediction [Luo et al., 2008]. Besides the prediction models, some ENSO precursors have also been found through simple analysis of observation data. For instance, subsurface observation has shown that variation in the equatorial warm water volume (WWV) precedes ENSO by three seasons and can be a useful ENSO predictor [McPhaden, 2003], in line with the recharge/discharge paradigm for ENSO [Jin, 1997].

[3] A factor that fundamentally limits the predictability of ENSO is the effect of atmospheric stochastic forcing [e.g., Kleeman and Moore, 1997]. The primary component of stochastic forcing is tropical intraseasonal variation (ISV), such as the Madden–Julian Oscillation (MJO) [Madden and Julian, 1972], tropical cyclones, and cold surges from the mid-latitudes. A possible linkage between ISV and ENSO has been proposed [e.g.,Bergman et al., 2001]. Recent studies have suggested that the low-frequency tail of the ISV in the western equatorial Pacific, through the excitation of equatorial Kelvin waves and triggered air–sea feedbacks, has a significant effect on ENSO [e.g.,Zavala-Garay et al., 2005]. Stochastic processes have caused prediction failure in many ENSO forecast models. One such case was for the wind bursts from November 1996 to May 1997 and the subsequent large 1997/98 El Niño [Landsea and Knaff, 2000]. Strong ISVs in boreal spring in the western equatorial Pacific have caused failures of ENSO forecast for lead times longer than one year.

[4] To understand the predictability of ENSO, we must assess both the large-scale slow ocean variability and the atmospheric stochastic forcing for ENSO. Using a simple two-predictor regression model,McPhaden et al. [2006] estimated the relative influence of the WWV and ISV on ENSO from observation data. They suggested that these two factors had approximately equal influence on ENSO during 1980–2005 at a lead time of two to three seasons. A similar scheme was presented by Clarke and Van Gorder [2003], who alternatively used zonal wind stress over the Indo-Pacific tropics.

[5] El Niño events in the 2000s were discerned as those with the warmest SST anomalies (SSTA) in the central equatorial Pacific [Lee and McPhaden, 2010], compared to the conventional El Niño composite. The recent behavior of ENSO has motivated studies focusing on differences in its spatial structure and teleconnection patterns from the canonical ENSO [Yeh et al., 2009, and references therein]. In the different kind of ENSO, SSTA were confined to the central Pacific. It was also reported that prediction of the major differences in SSTA between different El Niño types is limited to less than a one-season lead time [Hendon et al., 2009].

[6] Since ENSO behavior have changed, it is not clear if existing statistical prediction schemes still work in the 2000s. In this paper, we investigate ENSO in the 2000s, focusing on (1) the recharge/discharge of WWV and atmospheric stochastic forcing, (2) changes in the statistical predictability of ENSO, and (3) a possible linkage between the change and frequent occurrences of the so-called warm-pool El Niño. Here, referring to a plausible evolution process [Kug et al., 2009], we adopt the terms “warm-pool El Niño” for the central Pacific warming type and “cold-tongue El Niño” for the eastern Pacific warming type. We primarily report that the skills of two ENSO predictors (WWV and ISV) have significantly weakened in the 2000s.

2. Data

[7] The target ENSOs are those that occurred between 1980 and 2011, during which reliable observation data sets were obtained. Niño-3.4 index (SSTA averaged over 5°S-5°N, 170°W-120°W) was obtained from the Climate Prediction Center website ( El Niño/La Niña was defined as the period when the Niño-3.4 index exceeded 0.5°C in magnitude more than three consecutive months. The time-series of WWV was taken from the Pacific Marine Environmental Laboratory website ( The WWV was computed as the volume integral of water above the 20°C isotherm over the entire equatorial Pacific (5°S-5°N, 120°E-80°W) [Meinen and McPhaden, 2000]. To investigate the behavior of ocean subsurface temperature, we also used the ocean temperature dataset provided by the National Centers for Environmental Prediction (NCEP) Global Ocean Data Assimilation System (GODAS) [Saha et al., 2006]. For these data, monthly anomalies were calculated by removing the mean seasonal cycle from 1981 to 2010.

[8] The index of high-frequency surface wind forcing in the western equatorial Pacific was computed in the almost same way as byMcPhaden et al. [2006] and Hendon et al. [2007]. Using daily surface wind data from the NCEP/National Center for Atmospheric Research (NCAR) reanalysis dataset [Kalnay et al., 1996], we extracted the 20–100-day signal of the surface zonal wind by a Lanczos filter to capture the variance associated with the ISV. The major component of ISV was due to the effect of the MJO and its low-frequency tail [Hendon et al., 2007]. Then we squared the filtered time series to obtain the variance or “power.” Finally, monthly averages were calculated over the western equatorial Pacific (5°S-5°N, 120°E-180°). The wide area was chosen to capture the zonal shifts of ISV in the different kinds of ENSO. The results were insensitive to minor modification of the area. The extracted strong ISV period was consistent with the period of westerly wind bursts [e.g.,Seiki and Takayabu, 2007].

3. Decadal Changes in ENSO Predictors

[9] In the 1980s and 1990s, the WWV variations preceded ENSO with about a three-season lead time (Figure 1a). Here, the increase of WWV (recharge) corresponds to a deepening of the equatorial thermocline in the Pacific, and vice versa. It demonstrates that the WWV variation and Niño-3.4 SSTA are close to quadrature, in line with the recharge/discharge paradigm for ENSO [Jin, 1997]. The WWV anomaly in the preceding 7 to 11 months (boreal spring) was well correlated with the subsequent November–December–January (NDJ) Niño-3.4 (Figure 2a). These results are in agreement with previous findings [Meinen and McPhaden, 2000; Hasegawa and Hanawa, 2003] in which the linear relationship between the amplitude of the WWV anomaly and subsequent eastern SSTA was presented. Although there were several exceptions such as 1990 (no El Niño event with recharged WWV) and 1994 (onset of El Niño without recharged WWV), the WWV was a significant predictor of ENSO during 1980–2000.

Figure 1.

(a) Monthly anomalies of Niño-3.4 SST (black) and warm water volume (WWV) (blue). The Niño-3.4 was smoothed using a 5-month running mean filter. (b) Niño-3.4 (black) and 3-month averaged ISV over the western equatorial Pacific (120°E-180°E, 2.5°S-2.5°N) (red). (c) Composite time series of Niño-3.4 (bold black) and ISV (bold red) for five El Niño events during 1980–2000. The time axis is from January of the onset year (0) to December of the following year (+1). The thin red line shows the ISV in each year. (d) As in Figure 1c, but for four El Niño events during 2001–2011.

Figure 2.

Scatter plots of the (a) WWV (b) ISV and Niño-3.4 for the period 1980–2000. Niño-3.4 is the average for the boreal winter (NDJ), and numeral pairs denote the last two digits of the year (e.g., ‘97’ means from November 1997 to January 1998). WWV (ISV) is the 3-month (4-month) average from February to April (March to June) in each year. All values are anomalies relative to the mean seasonal cycle. The red (blue) characters represent El Niño (La Niña). (c and d) Same as Figures 2a and 2b, respectively, except for the period 2001–2011.r indicates the correlation coefficient. The number shown in parentheses is the correlation excluding the strong events (82, 88, and 97).

[10] The frequency and strength of ISV events were also associated with ENSO during 1980–2000 (Figures 1b and 1c). In 1986, 1991, and 1997, strong ISV occurred just before the onset of El Niño, indicating triggering activity of high-frequency atmospheric forcing for El Niño. The preceding ISV was significantly correlated with the subsequent ENSO on a two-season lead time during 1980–2000 (Figure 2b), in line with Hendon et al. [2007]. There were also inactive periods with La Niña episodes, such as in 1983 and from 1998 to 2000 (Figure 1b), suggesting that the forcing is not purely stochastic [Eisenman et al., 2005; Seiki and Takayabu, 2007]. On the other hand, some ISV peaks were found in non-El Niño years, such as in 1981 and 1990.

[11] Turning to ENSO in the 2000s, the relationship between WWV and Niño-3.4 is found to be no longer one of quadrature (Figure 1a). The change can be also observed in the phase diagram (Figure S1 in the auxiliary material). WWV preceded ENSO by only a few months, and both time series oscillated at almost the same phase, as in 2003–2004. Surprisingly, the robust relationship found in 1980–2000 broke down in the 2000s (Figure 2c). There was no marked recharged phase of the WWV preceding El Niño. Discharged phase in boreal spring occurred only in 2007 during these 11 years. Furthermore, La Niña events originated from recharged phases, such as in 2005, 2008, and 2011.

[12] The behavior of ISV with ENSO has also changed in the 2000s. El Niño onsets in the 2000s were not always accompanied by strong ISV events in boreal spring (Figures 1b and 2d). In addition, the timings of ISV peaks were quite different between the El Niño events (Figure 1d). The 2005 La Niña and 2006 El Niño were particularly extreme: despite strong ISV forcing around April–May in 2005, the system shifted to La Niña conditions by the subsequent winter. In 2006, westerly wind bursts in August, which blew relatively later compared with other El Niño years, triggered a moderate El Niño.

[13] Low-frequency zonal wind in the western Pacific is closely related to the ISV because it can provide favorable conditions for the strong ISV [Seiki and Takayabu, 2007; Kug et al., 2010]. It can also be a good ENSO predictor [Clarke and Van Gorder, 2003]. The result was essentially same using the low-frequency zonal wind instead of the ISV as the predictor: the skill also has significantly fallen in the 2000s (Figure S2).

[14] Given the idea that the WWV and ISV can represent a deterministic and stochastic components of ENSO, respectively, McPhaden et al. [2006]formed a simple multi-regression model for estimation of Niño-3.4 SSTA based on the antecedent WWV and ISV for the period of 1980–2005. It is clear from ourFigure 2 that the statistical prediction scheme succeeds for ENSO during 1980–2000 and fails during 2001–2011. Although our sample numbers demonstrated here were small (21 vs. 11) and the separation of 1980–2000 and 2001–2011 was subjective, we believe that these results have physical significance related to a possible decadal change in ENSO characteristics.

4. Discussion and Concluding Remarks

[15] What caused the decadal changes in WWV and ENSO? The distributions of ocean temperature during the El Niño peak might provide one possible explanation. For El Niño peaks during 1980–2000 (Figure 3a), temperature anomalies developed along the main thermocline, with negative (positive) anomalies to the west (east). This feature was consistent with the cold-tongue El Niño, indicating the efficacy of the thermocline feedback in the eastern Pacific. Then, divergence of the interior water mass to off-equator and eastward propagation of negative temperature anomaly can cause the equatorial thermocline to become shallower during a discharge phase. On the other hand, for El Niño events during 2001–2011, positive temperature anomalies were mainly stayed in the central Pacific at the peak, compared with that during 1980–2000 (Figures 3b and 3c). This is a typical feature of the warm-pool El Niño [Kug et al., 2009], in which the zonal advective feedback at the eastern side of the warm pool [An and Jin, 2001] is more effective. Under the conditions, WWV and ENSO SSTA can develop nearly in phase [Kug et al., 2010]. After the peak, both water-mass divergence and eastward expansion of negative temperature anomalies are weak, with no significant discharge of WWV. It is reasonable to think that the frequent occurrence of the warm-pool El Niño in the 2000s could provide non-discharged conditions that preclude significant La Niña events.

Figure 3.

Longitude–depth section of the ocean temperature (contours) and the anomalies (shades) in the equatorial region (2°S–2°N). NDJ composite of (a) five El Niño events (82, 86, 91, 94, and 97) and (b) four events (02, 04, 06, and 09). Contour interval is 2°C. (c) Temperature difference between Figures 3b and 3a

[16] Regarding the ISV, some spatial-temporal changes were found in the 2000s. Notably, most of the El Niño/La Niña events in the 2000s were associated with an atmospheric forcing relatively later in the calendar year (boreal summer–fall). The strong equatorial easterly (westerly) wind anomaly in June–October (July–September) and subsequent La Niña in 2005 (El Niño in 2006) is a typical example, as in a moderate El Niño year of 2009. In the 1980s and 1990s, westerly wind bursts were observed less often in the western equatorial Pacific in the boreal summer–fall season [Seiki and Takayabu, 2007]. The ISV activities and westerly wind bursts in the 2000s were also confined to the western side of the date line, compared with that for 1980–2000. Jones and Carvalho [2011] noted that the clear warming trend in the tropical Indian and western Pacific in recent decades may have led to frequent MJO events and hence an increase in ISV in the warm pool region.

[17] The change in atmospheric forcing may relate to frequent occurrence of the warm-pool El Niño. Under the condition of the enhanced ISV and westerly wind anomalies in the western Pacific, eastward warm advection at the east side of the warm pool and deepening of thermocline over the central Pacific lead to the central-Pacific SSTA [Kug et al., 2009, 2010]. The resultant SSTA could produce further westerly wind anomalies. The air-sea feedback in the western-central Pacific in the boreal summer is the distinctive feature of the warm-pool El Niño [Ashok et al., 2007].

[18] This study reports a possible change in ENSO characteristics in the 2000s, which could greatly influence ENSO predictability. Interestingly, the present result of frequent occurrence of the warm-pool El Niño is consistent with coupled model simulations under global warming [Yeh et al., 2009]. Further studies with maintaining/extending appropriate observations are needed to find out what is happening in the tropical Pacific.


[19] We thank J.-S. Kug and an anonymous reviewer for the constructive comments. We also thank A. Seiki, K. Ando, and S.-P. Xie for helpful discussions.

[20] The Editor thanks an anonymous reviewer for assisting with the evaluation of this paper.