The global ocean imprint of ENSO

Authors


Abstract

[1] The ENSO-related spatial patterns and global averages of ocean temperature, salinity, and steric height are estimated from over 7 years of Argo data, 2004–2011. Substantial extratropical variability is seen in all variables in addition to familiar tropical ENSO signals. Surface layer (0–100 dbar) and subsurface (100–500 dbar) temperature variations are both important in determining steric height and sea surface height patterns. For the two years prior to the 2009 El Niño, the upper 100 dbar of the ocean gained 3.3 × 1022 J yr−1 of heat, while the 100–500 dbar layer lost a similar amount. The ENSO-related vertical redistribution of globally-averaged heat content between surface and subsurface layers, occurring throughout the record, is due primarily to changes in the east-west tilting of the equatorial Pacific thermocline. The large temperature changes in the individual layers mask the smaller vertically-averaged temperature change, in which the ocean loses heat when the surface layer is anomalously warm and gains heat when the surface layer is cool.

1. Introduction

[2] El Niño/Southern Oscillation (ENSO) variability in sea surface temperature (SST), air temperature, and other atmospheric variables is well-known to be global in scale [e.g., Pan and Oort, 1990; Jones, 1989; Wigley, 2000; Horel and Wallace, 1981]. Moreover, increases in globally-averaged air temperature and SST have been observed during El Niño episodes, with decreases during La Niña. For example, during the strong El Niño of 1997–1998, globally-averaged SST from the NOAA OI SST product [Reynolds et al., 2002] was more than 0.2°C warmer than in the adjacent years. In the same episode, globally-averaged sea surface height (SSH) was anomalously high by 20 mm [Nerem et al., 1999], and correlation between SST and SSH records suggested that thermosteric expansion was responsible for the large SSH increase.

[3] Prior to the implementation of the Argo Program it was not possible to describe subsurface ocean temperature and salinity variability globally, although the ENSO observing system [McPhaden et al., 2001] provided data from the tropical Pacific. Since early 2004, the global array of Argo profiling floats has grown from about 1000 instruments to more than 3000, and now provides over 100,000 temperature/salinity profiles annually [Roemmich and Gilson, 2009], spaced approximately every 3° in latitude and longitude and 10 days in time. The Argo record is still very brief for describing interannual variability, and it contains no strong El Niños comparable to the 1997–1998 episode. Nevertheless, global patterns of recent ENSO variability are seen in Argo data. ENSO teleconnections impact the subsurface ocean, for example through anomalous wind-forcing, and we can now begin to explore the subsurface global impacts.

[4] In the following section it is shown that the short Argo record has ENSO variability in SST with a very similar global pattern to the much longer SST record. Then, the global patterns of ENSO variability are described for subsurface temperature and for sea surface salinity (SSS). Section 3 examines global averages of these same quantities. An adiabatic oscillation is seen in the global averages, consisting of a very large vertical redistribution of heat between the surface layer (0–100 dbar) and the layer beneath it (100–500 dbar). The diabatic elements of ENSO are enabled by this adiabatic oscillation. That is, anomalous heat is brought into the surface layer during El Niño, where it can interact with the atmosphere. During La Niña the surface layer is anomalously cool as much of the “missing” heat is sequestered in the subsurface layer.

2. The Global Ocean Imprint of ENSO

[5] The global pattern of ENSO variability (Figure 1) is estimated for SST, sea surface salinity (SSS), subsurface temperature, SSH, and steric height (SH). In each case, we take the ENSO state to be characterized by the spatial mean SST of the Niño 3.4 region (170°W to 120°W, 5°S to 5°N, hereafter N34 [e.g., Trenberth, 1997]). After removing the mean annual cycle from N34, a linear regression is calculated at each spatial grid point, of the 3-month running mean of N34 anomaly onto anomalies of each of the other variables. For SST this is done using both the NOAA OI SST product [Reynolds et al., 2002] from 1981–2011 (Figure 1a), and Argo near-SST from January 2004–March 2011 (Figure 1b) based on gridded Argo data [Roemmich and Gilson, 2009]. The former time-series has a substantial trend in globally-averaged SST over the 30-year period, and so is de-trended prior to calculating the linear regression. For the shorter Argo era no trend is removed. The pattern of ENSO SST variability in Figures 1a and 1b is familiar [e.g., Pan and Oort, 1990] – strong positive anomalies along the equatorial Pacific during El Niño, with a band of negative anomalies extending from the equator in the far western Pacific poleward to 40°N [Deser and Blackmon, 1995] and 40°S in the central and eastern Pacific respectively. The long-term SST pattern (Figure 1a) is effectively captured in the 2004–2011 Argo record (Figure 1b). Even regions with reduced amplitude appear to have similar ENSO variability in the Argo 7-year period and the 30-year period, including the tropical Indian Ocean [Reason et al., 2000], the southern central Pacific (50°S, 150°W) [Qiu and Jin, 1997], the North Atlantic and the coastal wave guide extending poleward from the equator in the eastern Pacific. A notable difference between Figures 1a and 1b is the westward shift of the equatorial Pacific maximum in 1b, reflecting the central Pacific “Modoki” El Niño pattern [Ashok et al., 2007] characteristic of the last decade.

Figure 1.

Color shading indicates the slope of the linear regression of N34 anomaly on the given variable, (a) NOAA OI SST, 1981–2011 (de-trended), (b) Argo SST, 2004–2011, (c) Delcroix et al. [2011] SSS, 1981–2008, (d) Argo SSS, 2004–2011, (e) SODA T160, 1981–2007, (f) Argo T160, 2004–2011, (g) AVISO SSH, 1993–2010, Ducet et al. [2000], “reference” product (de-trended), (h) Argo SH, 0/2000 dbar, 2004–2011. Units are: variable's unit (°C, psu, or cm) per °C of N34. All variables are smoothed with a 3-month (12-month for SODA) and 5° longitude × 3° latitude running mean. Black contours outline regions where the variance explained is greater than 20% (thin lines) or 50% (thick lines).

[6] SSS anomalies during ENSO (Figures 1c and 1d) are due to a combination of physical processes. These include anomalous rainfall associated with shifts in the tropical atmospheric convergence zones [Rasmusson and Carpenter, 1982; Folland et al., 2002] and mid-latitude storm tracks, changes in horizontal advection, and entrainment of shallow thermocline waters into the surface layer. Throughout the Pacific islands and the Pacific rim, there are strong ENSO impacts on regional rainfall [Ropelewski and Halpert, 1987]. The pattern of ENSO variability in Argo-era SSS (Figure 1d) is similar to that estimated from a tropical Pacific historical SSS dataset (Figure 1c) [Delcroix et al., 2011], but with the amplitude and location of maxima differing due to the time periods represented and to the sparse sampling of the pre-Argo SSS.

[7] The spatial pattern of ENSO temperature variability at 160 dbar (T160, Figures 1e and 1f) is quite different from that at the sea surface. It reflects the known ENSO-related tilting of the thermocline in the equatorial Pacific, characterized by deepening (warming of T160) in the central and eastern Pacific and shoaling in the west during El Niño. The western Pacific branch has two off-equatorial lobes extending over a broader latitude range, about 15°S to 15°N, than the signal in the central Pacific. The scales of subsurface temperature variability are shorter than for SST, so the pattern of T160 is relatively noisier. The Argo-era record (Figure 1f) is very similar in the tropical Pacific and Indian oceans to the longer record from the Simple Ocean Data Assimilation (SODA, version 2.2.4) model (Figure 1e) [Carton et al., 2000].

[8] SH is an integral over pressure of density anomaly, and in most cases SH changes are dominated by changes in temperature. In turn, SSH anomalies include both a steric and a mass-related component, though the former dominates in the tropics on seasonal and interannual time-scales. We therefore expect the pattern of ENSO variability in SSH (Figure 1g) and SH (Figure 1h) to be similar and to reflect vertically-averaged temperature. The SSH pattern [see also Merrifield et al., 1999] is smoother than SH because of the longer satellite altimetry time-series, but otherwise the two are very similar. SSH and SH resemble the SST pattern in regions where the thermocline is shallow, such as the eastern tropical Pacific. In other regions, where the thermocline is deeper, such as the western tropical Pacific and the tropical Indian Ocean, they are similar to the T160 pattern.

[9] The pattern of ENSO temperature variability versus pressure along the Pacific equator (Figure 2a) is that of a dipole, with extrema in the thermocline, but extending below it. This pattern signals the familiar ENSO fluctuation in the thermocline's zonal slope.

Figure 2.

Slope of the linear regression of Niño 3.4 SST anomaly onto Argo temperature anomaly (color shading), for vertical sections (a) along the Pacific equator and (b) along 160°E. Black contours indicate mean temperature.

[10] ENSO-related changes in the location or strength of zonal geostrophic currents are indicated by three subsurface minima in the regression slope along 160°E (Figure 2b). Changes in zonal currents during the strong ENSO cycle of 1996–1998 were described by Johnson et al. [2000]. The minimum in regression slope at 3°S likely indicates a southward shift of the South Equatorial Current during El Niño. The minimum at 7.5°N, at the thermocline ridge, indicates an El Niño strengthening of vertical shear in the eastward North Equatorial Countercurrent [Meyers and Donguy, 1984] and in the westward North Equatorial Current.

3. ENSO and Globally-Averaged Temperature and Salinity

[11] ENSO-related variability in temperature and salinity is seen not only regionally (Figures 1 and 2), but also in global averages (Figure 3). The familiar warming of globally-averaged SST during El Niño increases with depth to a maximum at 50 dbar (Figure 3a) and then decreases to zero at about 100 dbar. Deepening of the shallow thermocline of the eastern Pacific during El Niño dominates the global mean. Below 100 dbar, shoaling of the deeper thermocline of the western Pacific results in net cooling in the global mean during El Niño, with maximum cooling around 150–200 dbar. A caveat for the global averages is that Argo does not cover most of the marginal seas, including those of the Indonesian archipelago where there is a strong ENSO signal.

Figure 3.

(a) Time-series of globally-averaged (60°S to 60°N) temperature anomaly from the monthly mean, versus pressure (dbar). The contour interval is 0.02°C and values are smoothed by 3-month running mean. (b) Time-series of globally-averaged SST (black, °C), T160 (blue), and the N34 regression estimate for SST (red). (c) Same as Figure 3a except for salinity anomaly. The contour interval is 0.004. (d) Time-series of globally-averaged SSS (black, psu) and S160 (blue). The vertical ranges of Figures 3d and 3b represent equivalent effects of T and S anomaly on density.

[12] The correlation of globally-averaged temperature with N34 is high for both the 0–100 dbar and 100–500 dbar layers (Figure 3b), with opposite phasing. The N34 linear regression accounts for 80% of the variance in 3-monthly smoothed SST anomaly and 60% of the variance in T160.

[13] As with temperature, globally-averaged salinity (Figure 3c) and the time-series of both SSS and salinity at 160 dbar (S160, Figure 3d) show a tendency for the surface layer and subsurface layer anomalies to have opposite sign. The global mean salinity anomalies have only about 1/3 as much effect on density and steric height as do the temperature anomalies, but their density effects tend to reinforce one another (fresh anomaly corresponding to warm anomaly). Time-series of globally-averaged steric height, and separately of the thermosteric and halosteric contributions show a similarly dominant role of temperature. In addition to the ENSO-like fluctuations in salinity, an equally large but longer time-scale signal is seen in salinity but not in temperature, with overall fresh anomalies for the first 5 years followed by a 0.03 increase in surface salinity in 2009–2010. This increase in globally-averaged surface layer salinity is balanced by decreasing salinity in the interval from 100–1000 dbar.

4. Discussion and Conclusions

[14] The present work focuses on the global ocean's three-dimensional expression of ENSO variability, including the regional patterns and global averages. Between early 2008 and the El Niño of late 2009, the heat content of the upper 100 dbar of the global ocean increased by 3.3 × 1022 J yr−1 (about 4 W m−2), with a similar decrease in the 100–500 dbar layer (Figure 4, top). By comparison, multi-decadal ocean warming accounts for about 0.4 × 1022 J yr−1 of net heating in the top 700 m [Levitus et al., 2009]. Thus, the vertical redistribution of ocean heat content on interannual timescales (Figure 4, top) is at a rate 8 times that of multi-decadal global warming. The layered heating/cooling pattern is notable for several reasons. First, these large interannual temperature signals mask the smaller decadal variability and trends, increasing the time-series requirements for detecting global change signals. Second, the vertical redistribution of heat serves alternately to make heat available at the sea surface for air-sea exchange during El Niño and to sequester it below 100 dbar during La Niña. The net heating/cooling of the combined layers, 0–500 dbar (Figure 4, bottom), is also large, though smaller than the individual layer signals.

Figure 4.

(top) Time-series of heat content for the layers 0–100 dbar (black) and 100–500 dbar (blue). Solid lines indicate the global integrals (60°S to 60°N) and dashed lines indicate equatorial Pacific integrals (5°S to 5°N). (bottom) Time derivative of 0–500 dbar heat content anomaly, globally-averaged with a 15-month Hanning smoother (black). Regression estimate based on N34 anomaly with the same smoothing (red).

[15] Globally-averaged ocean heat content (solid lines in Figure 4, top) is distinguished from equatorial Pacific heat content (dashed lines in Figure 4, top) [e.g., Jin, 1997], or equatorial warm-water volume [Meinen and McPhaden, 2000]. Global and equatorial Pacific heat content differ from one another, especially for the 100–500 dbar layer, because of the strong subsurface off-equatorial signal (Figure 1f). A conclusion is that to describe the planetary heat balance during ENSO, it is necessary to integrate ocean heat content horizontally over the globe and vertically to the base of ENSO-related variations.

[16] As noted in the introduction, this work was motivated partly by the question of whether global ocean heat content variations are strongly tied to ENSO, as suggested by observations of the 1997–1998 El Niño maxima in SST and SSH [Nerem et al., 1999]. The ENSO variability in surface layer heat content is partially cancelled by that in the subsurface layer (Figure 4, top). However, a relationship between the time rate of change of the combined (0–500 dbar) heat content and N34 anomaly is suggested (Figure 4, bottom). Here, the smoothed N34 anomaly accounts for 56% of the variance in the smoothed time-series of heat gain. While the Argo record is short and contains only a few weak-to-moderate ENSO oscillations, a consistent interpretation – also seen in the longer but noisier SODA record - is that the ocean as a whole loses heat when the surface layer is warm. That is, the ENSO-related vertical redistribution (Figure 3) brings heat close enough to the sea surface for mixing and/or upwelling to affect SST and air-sea exchange. The ocean loses heat at a global mean rate of >1 W m−2 during El Niños (Figure 4, bottom) and similarly gains heat during La Niñas.

[17] This relationship, of net heat loss coinciding with El Niños, is consistent with a radiative balance for ocean heat content since globally-averaged SST is warm during El Niño. It differs from previous findings specific to the 1997 El Niño episode, of 1.5 W m−2 maxima in downward net planetary radiation and ocean heat gain in late 1997 [Wong et al., 2006, Figure 7]. The issue is of considerable interest for understanding interannual and longer variability in the planetary energy budget, but will need careful study of a longer instrumental record than the present Argo dataset. Argo's value will continue to grow as its global coverage is extended in time to a decade and longer.

Acknowledgments

[18] The Argo data used here were collected and are made freely available by the International Argo Program and by the national programs that contribute to it. The authors participate in Argo through NOAA grant NA17RJ1231(Scripps Institution of Oceanography). We thank Phil Sutton, Catia Domingues, and Rory Bingham for their valuable comments and suggestions.

[19] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.

Ancillary