A 21st century shift in the relationship between ENSO SST and warm water volume anomalies

Authors


Abstract

[1] This paper documents changes in the relationship between warm water volume (WWV), which is an index for upper ocean heat content, and El Niño/Southern Oscillation (ENSO) SST anomalies during the period 1980–2010. Upper ocean heat content represents a major source of predictability for ENSO, with WWV integrated along the equator leading ENSO SST anomalies by 2–3 seasons during the 1980s and 1990s. For the first decade of the 21st century however, WWV variations decreased and lead time was reduced to only one season, mainly due to the diminished persistence of WWV anomalies early in the calendar year. These changes are linked to a shift towards more central Pacific (CP) versus eastern Pacific (EP) El Niños in the past decade. The results are consistent with a reduced impact of thermocline feedbacks on ENSO SST development and potentially imply reduced seasonal time scale predictability during periods dominated by CP El Niños.

1. Introduction

[2] The El Niño/Southern Oscillation (ENSO) phenomenon is the most prominent year-to-year climate fluctuation on the planet, affecting patterns of weather variability worldwide [McPhaden et al., 2006]. It is predictable up to three seasons in advance, with predictability based on slow evolution of upper ocean heat content [Latif et al., 1998]. The theory for the role of heat content in affecting the ENSO cycle is elegantly summarized in Jin's [1997] Recharge Oscillator theory. Meinen and McPhaden [2000, hereinafter MM2000] substantiated the basic principles of this theory through analyses of upper ocean heat content and its relationship to sea surface temperature (SST). Consistent with theory, they found that an index for heat content averaged across the basin, namely warm water volume (WWV), led ENSO SST anomalies in the eastern Pacific by 6–7 months on average for the period 1980–99.

[3] However, the relationship between SST and heat content anomalies in the equatorial Pacific has subtly shifted since 1999. McPhaden [2008] for example pointed out that WWV lead time had shortened to only a season after 2000 based on inspection of index time series. This change coincides with other changes taking place in the tropical Pacific. Specifically, over the past few decades, there has also been a trend towards the increasing occurrence of central Pacific (CP) or Modoki type El Ninos [Ashok et al., 2007; Kao and Yu, 2009], which have there greatest warming in the central Pacific (Figure 1). The first decade of the 21st century exhibited the greatest frequency of CP El Niños over the past 30 years [Lee and McPhaden, 2010; McPhaden et al., 2011] and El Niños during this period tended to be weaker on average than more classical eastern Pacific (EP) El Niños of the previous 20 years. The predictability of recent El Niños appears to have been reduced as well. As Wang et al. [2010] reported for NOAA Coupled Forecast System (CFS) forecasts for the period 2005–08, “For the tropical eastern Pacific… the forecast skill at lead times longer than 2 months is less than the average…due to the relatively weaker ENSO variability…” compared to 1981–2004.

Figure 1.

Composites of (a) SST (in °C) and (b) and Z20 (in m) for December-February (DJF) of El Niño years during 1980–99 with zonal wind stress (in N m−2) overplotted on both. (c and d) Same as Figures 1a and 1b but for 2000–10. Reproduced from McPhaden et al. [2011].

[4] In this study, we examine the relationship between Warm Water Volume (WWV) and ENSO time scale SST anomalies for the period 2000–2010 and contrast these results with those from MM2000 for the period 1980–99. We also interpret differences between the two periods in the context of the changing character of El Niño during this time.

2. Data

[5] We use two data sets, each covering January 1980 to December 2010 in the tropical Pacific. One is for SST anomalies in the NINO3.4 region (averaged over 5°N–5°S, 120°W–170°W) and NINO4 region (averaged over 5°N–5°S, 150°W–160°E) available on the web at http://www.cpc.ncep.noaa.gov/data/indices/. These indices are derived from the Reynolds et al. [2002] blended in situ and satellite SST analysis. The other is for 20°C isotherm depth data from the Australian Bureau of Meteorology Research Centre (BMRC). This product is based on data from moored buoys, expendable bathythermographs, and Argo floats [Smith, 1995]. It is available with monthly resolution on a 1° latitude ° by 2° longitude horizontal grid. These data were integrated between 5°N-5°S, 120°E-80°W to estimate the volume of water above the 20°C isotherm, namely WWV, which is a measure of upper ocean heat content. Monthly WWV and NINO index region anomalies are computed relative to a mean seasonal cycle based on a 30-year (1981–2010) climatology; or when subsets of the data are used, relative to the record length mean climatology.

[6] In this study we will use NINO3.4 SST anomalies rather than NINO3 anomalies (averaged in the region 5°N-5°S, 90°W-150°W) as inMM2000. Our results change little if we use NINO3 since the two indices are highly correlated. NINO3.4 has become a standard operational index for ENSO evolution at NOAA's Climate Prediction Center, so is a more relevant for our purposes.

3. Results

[7] El Niño warming has occurred more frequently in the past 11 years than during the previous 20 years (Figure 2a), with four El Niños between 2000–2010 (one every 2.8 yrs) versus five between 1980–99 (one every 4 years). Three of the four El Niños in the past decade were CP El Niños. Both NINO3.4 and WWV variations weakened in the 21st century as evident from their reduced standard deviations compared to the 1980–1999 period (Figure 2a).

Figure 2.

(a) Monthly anomalies of WWV and NINO3.4 SST (5°N–5°S, 120°W–170°W). Time series have been smoothed with a 5-month running mean filter. The numbers indicate standard deviations for NINO3.4 SST and WWV (×1014) for 1980–99 and 2000–2010. (b) Cross correlation of monthly WWV and NINO3.4 SST anomalies for 1980–99 (red) and 2000–2010 (blue). One standard error for peak correlations are shown using the method of Davis [1976] to determine degrees of freedom.

[8] For most El Niño and La Niña events between 1980–1999, WWV led NINO3.4 by 2–3 seasons (Figure 2a) as was found in MM2000. However, this lead time shortened to only about one season after 2000 as can be seen in the cross correlation computed between the two time series (Figure 2b). For the period 1980–1999, the peak correlation (0.68) occurs with WWV leading NINO3.4 by 5–7 months, while for the 2000–2010 the peak correlation (0.75) occurs with WWV leading NINO3.4 by 2–3 months. Thus, the heat content recharge process prior to the development ENSO events operates throughout the period 1980–2010, but at shorter lead times for the 21st century. We also see that the discharge process is evident during both periods (where WWV lags NINO3.4), but is reduced in efficacy for the period 2000–10 based lower peak cross correlation with NINO3.4 compared to 1980–1999. This feature reflects the fact that some of the El Niño events after 2000 were not associated with a significant discharge of excess equatorial heat content and thus were not followed by a subsequent La Niña, as for example in 2002–03 and 2004–05 [see also Kug et al., 2010].

[9] Repeating the analysis of McPhaden [2003], who examined the seasonality of the relationship between SST anomalies and WWV for the 1980s and 1990s, we find that the boreal spring persistence barrier that is so prominent in SST does not exist for WWV (Figures 3a and 3b); and that boreal winter to early-spring anomalies in WWV can provide a reliable indicator peak SST anomalies nine months later (Figure 3c). Thus, the lack of a boreal spring heat content persistence barrier helps to explain why accurate initialization of upper ocean heat content in ENSO forecast models can reduce the prominence of the spring SST predictability barrier [e.g., Balmaseda et al., 1995; Smith et al., 1995].

Figure 3.

Correlations of (a) NINO3.4 SST with itself, (b) WWV with itself, and (c) WWV and NINO3.4 as a function of start month and lead-time for 1980–99. Positive values in Figure 3c imply that a WWV anomaly leads a NINO3.4 anomaly of the same sign. (d–f) Same as Figures 3a–3c but for 2000–2010.

[10] For 2000–2010, the SST persistence barrier is relatively unchanged (Figure 3d), but the correlation at lead times up to 9 months of WWV with itself (Figure 3e) and up to 12 months of WWV with Nino3.4 SST (Figure 3f) is greatly diminished early in the calendar year. Thus, unlike for the period 1980–99, WWV in the early part of the year is of little practical value as a predictor of SST anomalies occurring in the second half of the year. There is still high correlation (values >0.9) at 2–4 month lead times between WWV and SST in the latter part of the calendar year (Figure 3f), but these correlations do not exceed the autocorrelation of SST with itself at the same time. Thus, the unique value of WWV as predictor of SST is diminished for the period 2000–2010 compared to 1980–1999. The large negative WWV correlations in mid-to late year at lags of 6-12 months may be related to a higher frequency ENSO cycle during 2000-10 and the appearance of prominent La Niña variations after 2006.

[11] While variability WWV and equatorial cold tongue SST decreased in the most recent decade (Figure 2a), the standard deviation of SST in the NINO4 region increased from a of 0.60°C for 1980–1999 to 0.68°C for 2000–2010. This increase resulted in NINO4 variations almost as large as NINO3.4 variations and corresponded to the increased in frequency and amplitude of CP El Niños [Lee and McPhaden, 2010]. In the central Pacific, zonal advection is the dominant source of ENSO SST variability since interannual variations in thermocline depth, and thermocline feedbacks on SST, are relatively weak compared to the eastern equatorial Pacific [An and Jin, 2001].

[12] Kim et al. [2009]stated that, “…there is a greater and earlier predictability of a central Pacific warming event than an eastern Pacific warming event…” in part because NINO4 SST anomalies exhibit no obvious spring predictability barrier as occurs in the eastern Pacific. However, they based their conclusion on predictability studies using the ECMWF forecast system for the period 1980–2007 inclusive. If one stratifies NINO4 data by 1980–1999 and 2000–2010 separately, we find empirically that a spring persistence barrier emerges like that further to the east along the equator in the CP El Niño-dominated 21st century (Figure 4). This result raises the question of whether predictability of ENSO SSTs is diminished all along the equator for early calendar year start times during 2000–2010 since neither SST nor WWV are of great value as predictors then at lead times beyond a season. It is not clear why a boreal spring persistence barrier develops in NINO4 SST during 2000–2010. However, the largest seasonal increase in NINO4 variance occurs during January to March during this period (Figure S1 in the auxiliary material), so that conditions antecedent to the development of ENSO events later in the year are more variable.

Figure 4.

Autocorrelation of (a) NINO4 SST and (b) NINO3.4 SST as a function of start month and lead-time for 1980–99. (c, d) Same as Figures 4a and 4b but for 2000–2010.

4. Discussion and Conclusions

[13] This paper has documented changes in the relationship between upper ocean heat content (as measured by WWV) and ENSO SST anomalies during the period 1980–2010. During the first decade of the 21st century, NINO3.4 and WWV variations weakened relative to the period 1980–1999, while NINO4 variations increased. Also, the lead time between WWV and NINO3.4 SST anomalies decreased from 2–3 seasons to only one season mainly due to the diminished persistence of WWV anomalies in boreal winter and early spring prior. Coincidentally, a spring persistence barrier also developed for NINO4 SST anomalies during 2000–2010. These changes appear to be linked to a shift in the character of El Niño, with a greater prevalence of CP versus EP El Niños in the past 10 years. The reduced magnitude of thermocline depth and WWV variations (Figures 1 and 2) and the weaker correlation of WWV with NINO3.4 at 2–3 season lead times in particular are consistent with a reduction in the effectiveness of thermocline feedbacks in governing the evolution of CP El Niño SSTs.

[14] These results may have implications for the predictability of ENSO. In particular, based on the weaker amplitude of NINO3.4 SST and WWV anomalies, the shortening of the lead time of WWV relative to NINO3.4 SST anomalies, and the appearance of a spring SST persistence barrier in NINO4 SST, we might expect reduced predictability for SST at 2–3 season leads especially from early in the calendar year during 2000–2010. These tendencies may account for some of the reduction in reported lead times for NCEP CFS forecasts made during the period 2004–2008 as reported by Wang et al. [2010]. They likewise suggest that predicting whether an event will evolve into a CP versus EP El Niño at 2–3 season lead times will be a challenge [Hendon et al., 2009].

[15] Why the changes we have observed in the first decade of the 21st century are occurring is an unresolved question. One could argue that the shorter lead time between WWV and NINO3.4 SSTs during 2000–2010 is simply the result of an apparent higher frequency ENSO cycle as suggested by the more frequent occurrence of El Niños during that period. However, that argument if correct leaves open the question of why El Niños, and CP El Niños in particular, have occurred more frequently. The tendency towards more frequent CP El Niños may be related to global warming [Yeh et al., 2009], to natural decadal variability [McPhaden et al., 2011; Yeh et al., 2011], or to some combination of the two. Resolving these questions will require longer data sets and improved coupled ocean-atmosphere climate models capable of simulating ENSO variability under the growing influence of greenhouse gas forcing.

[16] The instrumental records available to us over the past 30 years are sufficiently comprehensive and resolved in space and time to examine the relationships between ENSO time scale SSTs and WWV in detail. They are nonetheless relatively short for an examination of decadal variations. With small sample sizes, inferences based on a single statistical result may therefore be questionable. However, there is a consistency across various analytical approaches and a logical connection to dynamical concepts that bolster the reliability of our results. Thus, despite the limitations of this study, it can be a useful guide for diagnosis and interpretation of more complete and extensive ocean analyses and coupled ocean-atmosphere model outputs.

Acknowledgments

[17] This research was carried out at NOAA/Pacific Marine Environmental Laboratory. PMEL contribution 3788.

[18] The Editor thanks two anonymous reviewers for assisting with the evaluation of this paper.

Ancillary