Changes in the linear relationship of ENSO–PDO under the global warming

Authors


Abstract

We examine changes in El Niño and Southern Oscillation (ENSO)/Pacific Decadal Oscillation (PDO) relationship under the global warming using coupled climate models participated in the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4). The temporal structure for the ENSO–PDO relationship is changed remarkably. The relationship between ENSO and PDO during the boreal winter (December, January and February) becomes stronger so that there would be more frequent in phase occurrence of ENSO and PDO (i.e. El Niño—a positive phase of PDO or La Niña—a negative phase of PDO). As PDO could constructively interfere with the ENSO-related climate when ENSO and PDO are in phase, in the future one may expect stronger climate signal because of ENSO in the midlatitude. The IPCC AR4 model also shows that the Pacific North America-like pattern is slightly shifted eastward and much stronger. We also discuss the possible reason for these changes. Copyright © 2012 Royal Meteorological Society

1. Introduction

El Niño and Southern Oscillation (ENSO) could affect the weather and climate variability worldwide beyond the tropical Pacific through atmospheric teleconnections (Lau and Nath, 1996; Alexander et al., 2002). However, such global influence because of ENSO teleconnections could be significantly modulated by the state of large-scale oceanic and atmospheric variability such as Pacific Decadal Oscillation (PDO), the Aleutian low pressure, the Arctic Oscillation, and so on. Among them, there is an evidence both in the observations and coupled general circulation model (CGCM) showing that the intensity of the atmospheric response to ENSO depends on the state of the North Pacific Ocean, i.e. the PDO (Gershunov and Barnett, 1998; Bond and Harrison, 2000; Higgins et al., 2000; Barlow et al., 2001; Lee et al., 2002; Lau et al., 2004; Yeh and Kirtman, 2004; Pavia et al., 2006; Yu and Zwiers, 2007; Yu et al., 2007). In other words, PDO plays a role to modulate the ENSO teleconnections from the tropics to the midlatitudes, therefore, a correct simulation of PDO is able to improve the skill of seasonal climate predictions on the ENSO's effect in the midlatitudes (Pierce, 2002).

There exist a number of previous studies to discover a physical mechanism generating the PDO. For example, Newman et al. (2003) and Pavia (2009) applied an idea of the stochastic forcing model (Hasselman, 1976) to the PDO forcing mechanism. They concluded that the PDO is largely dependent upon ENSO on from interannual to interdecadal timescales. As a first approximation of the autoregressive model, that is, the PDO can be understood as a reddened response to ENSO and atmospheric white noise. Furthermore, Schneider and Cornuelle (2005) also suggested that the PDO can be reconstructed by ENSO using the Aleutian low index, and ocean circulation anomalies in the Kuroshio–Oyashio Extension region based on a stochastic model. They argued that the variability of the Aleutian low and ENSO is essential for PDO forcing process on interannual timescales, while the ocean circulation associated with Kurosio–Oyashio Extension is relatively important on decadal timescale. Similarly, some previous studies also suggested that tropical forcings, which are associated with ENSO, are transported to the North Pacific through atmospheric teleconenctions, and then, this tropical-related atmospheric circulation is able to induce the PDO (Trenberth and Hurrell, 1994; Diaz et al., 2001; Deser et al., 2004).

In addition to a physical mechanism, there are many studies which examine how the ENSO–PDO relationship is associated with climate variability in the Pacific. The PDO and its associated atmospheric variability are correlated with sea surface temperature (SST) and precipitation anomalies of Asian and North Pacific regions as well as a modulation of ENSO teleconnection (Gershunov and Barnett, 1998; Mantua and Hare, 2002). For example, the PDO could interfere constructively and destructively with ENSO-related climate over North America. Hu and Huang (2009) demonstrated that the relationship between ENSO/PDO and climate anomalies in the Great Plains is intensified when ENSO and PDO are in phase (i.e. El Niño and a positive phase of PDO or La Niña and a negative phase of PDO). On the other hand, when ENSO and PDO are out of phase, the relationship is weakened and the climate anomalies over the Great Plains tend towards neutral. In addition, the phase of PDO is highly connected to the amount of ENSO-related rainfall over south China and the summer monsoon rainfall variability over Northeast Asia (Chan and Zhou, 2005; Yoon and Yeh, 2010). Therefore, it is important to examine characteristic changes in the ENSO–PDO relationship to understand the climate system in the pan-Pacific basin and the extended range prediction of weather and climate.

In spite of the diversity of the coupled model results, a majority of the Coupled Model Intercomparison Phase 3 (CMIP3) multimodel analysis indicated that the global warming forced by CO2 can alter properties of SST variability in the tropical Pacific (van Oldenborgh et al., 2005; Meehl et al., 2007a; Yeh and Kirtman, 2007; Guilyardi et al., 2009). Therefore, one may expect that the tropical–extratropical SST teleconnections including the ENSO–PDO relationship could be different in the future climate. In fact, there already is a study showing that the atmospheric teleconnections of both El Niño and La Niña over the North Pacific and American regions are changed under the global warming using the CMIP3 CGCM simulations (Meehl and Teng, 2007). However, there is little investigation on the changes of SST teleconnections. In this study, we, thus, show how the linear relationship between the ENSO and PDO changes under the global warming and explain the processes governing the changes using the CMIP3 CGCM simulations.

2. Model and methodology

The CMIP3 CGCM simulations are used in order to asses ENSO–PDO relationship under the global warming. The CGCM simulations are made available by the Program for Climate Model Diagnosis and Intercomparison from the website http://www-pcmdi.llnl.gov/ipcc/about_ipcc.php (Meehl et al., 2007b). We examine 12 sets of CGCMs, each of which consisted of a control run and a climate change run, due to the availability of the climate change run data (Table I). The control run is the 20th Century Climate Change Modelling (20C3M) simulation to year 2000 with anthropogenic and natural forcing, and hereafter the term ‘20C3M run’ refers to data from the last 100-year simulation period (i.e. 1900–1999) from the control run. The climate change run corresponds to the future climate change simulation with Special Report for Emission Scenario A1B (SRESA1B) in which the atmospheric CO2 concentration increases to 720 ppm by the end of 21st century and then held fixed. The term ‘SRESA1B run’ refers to the model results from 100 years long period between 2100 and 2199 during which the concentration of CO2 is fixed at 720 ppm. The extended reconstructed sea surface temperature provided by the National Centers for Environmental Prediction–National Center for Atmospheric Research for the period of 1900–1999 (Smith and Reynolds, 2004) is also utilized to compare with the 20C3M run.

Table I. Summary of the AR4-IPCC coupled models used in this study
Originating groupsCMIP3 IDAGCM resolutionOGCM resolution
Canadian Center for Climate Modelling and AnalysisCGCM3.1(T47)T47 L31192 × 96 L29
Meteo-France/Centre National de Recherches MeteorologiquesCNRM-CM3T42 L45180 × 170 L33
Geophysical Fluid Dynamics LaboratoryGFDL-CM2.0N45 L241 × 0.33–1 L50
Geophysical Fluid Dynamics LaboratoryGFDL-CM2.1N45 L241 × 0.33–1 L50
NASA/Goddard Institute for Space StudiesGISS-ER72 × 46 L1772 × 46 L13
LASG/Institute of Atmospheric PhysicsFGOALS-g1.0128 × 60 L26360 × 170 L33
Institute for Numerical MathematicsINM-CM3.04 × 5 L212 × 2.5 L33
Center for Climate System Research, National Institute for Environmental Studies, and Frontier Research Center for Global Change (JAMSTEC)MIROC3.2 (medres)T42 L20256 × 192 L44
Meteorological Research InstituteMRI-CGCM2.3.2T42 L302 × 0.5–2.5 L23
National Center for Atmospheric ResearchPCMT42 L18384 × 288 L32
Hadley Centre for Climate Prediction and Research, Met OfficeUKMO-HadCM32.5 × 3.751.25 × 1.25 L20
Hadley Centre for Climate Prediction and Research, Met OfficeUKMO-HadGEM1N96 L381 × 0.33–1 L40

3. Analysis

3.1. Changes in the ENSO–PDO relationship

We focus on the ENSO–PDO relationship during the boreal winter (December–January–February) because both ENSO and PDO have strong peaks during the season. Hereafter, the ENSO and PDO refers to the ENSO and PDO during winter, respectively, for convenience. To obtain the PDO and ENSO in the 20C3M and SRESA1B runs, respectively, firstly, we perform an empirical orthogonal function (EOF) analysis using detrended SST from the 12 CGCMs. The PDO is defined as the first EOF mode of Pacific SST anomalies (SSTAs) north of 20°N (Mantua et al., 1997). Similarly, ENSO is defined by the first EOF mode of SSTAs over the equatorial bands of Pacific Ocean (120°E–80°W, 20°N–20°S). The first principal component time series of the PDO and ENSO is defined as the PDO index and ENSO index, respectively.

Figure 1(a) and (b) shows the ENSO and PDO in the observations, respectively, for the period of 1900–1999. The positive phase of PDO (Figure 1(a)) is characterized by an elliptical shape of cool temperature anomalies over in the western and central North Pacific and warm temperature anomalies to the east, north, and south. On the other hand, the ENSO (Figure 1(b)) is characterized by a large warming over the central and eastern equatorial Pacific. The ensemble means of PDO and ENSO simulated in the 12 20C3M runs (Figure 1(c) and (d)) deviate from the observation to some degree. As compared with the observation, for instance, in the central North Pacific the centre of the simulated PDO (Figure 1(c)) is slightly shifted southward and the maximum SST variance is located in the western North Pacific. On the other hand, the maximum variance of the simulated ENSO (Figure 1(d)) extends further to the west, and the SST variance over the far eastern tropical Pacific is relatively weaker than the observations. In spite of such discrepancies, the pattern correlation of the PDO between the observation and the 20C3M run is 0.83, and that of the ENSO is 0.92. Both correlation coefficients are significantly high, and we can conclude that the 20C3M run reasonably captures the main structure of the observed ENSO and PDO regardless of the aforementioned discrepancies.

Figure 1.

The PDO (a) and ENSO (b) in observations. (c)–(d) are the same as in (a)–(b) except but the 20C3M run. (e)–(f) are the same as in (a)–(b) except but the SRESA1B run. Unit is non-dimensional. The percent variance explained by the PDO and ENSO is noted at the top of panel. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Figure 1(e) and (f) is the same as in Figure 1(c) and (d) except for the SRESA1B run. The spatial pattern of the ENSO and PDO simulated in the SRESA1B run is similar compared with the 20C3M run so that the pattern correlation of the PDO between the 20C3M run and the SRESA1B run is significantly high, 0.97 and that of the ENSO is also high, 0.94. This result indicates that the global warming would not alter the spatial patterns of both PDO and ENSO, which are defined as the first EOF SST modes of variability, substantially as consistent with Overland and Wang (2007).

On the other hand, change in the temporal structure for the ENSO–PDO relationship under the global warming is remarkable as displayed in Figure 2 that shows the correlation coefficients between the ENSO and PDO indices in each 12 CGCM and the ensemble mean correlation. The significance of the relationship was estimated using the bootstrap method as follows. In order to construct the probability distribution function for the correlation coefficient between the ENSO and PDO, we randomly choose a 100 years log period from the 20C3M simulation, and then calculated the correlation coefficient. By repeating this process 10 000 times, the probability density function is constructed and the coefficient from the SRESA1B run is compared with this probability density function.

Figure 2.

Correlation coefficients between the ENSO and PDO indices in the 20C3M run (red) and the SRESA1B run (blue), respectively, for the entire analysed period. The vertical segment in the red bar denotes the upper and lower limit associated with an increase and decrease of the correlation coefficient at 95% confidence level based on a bootstrap method, respectively. A 10 000 bootstrap resamples were used in order to get confidence limits

Among the 12 CGCMs, the correlation coefficient increases from the 20C3M run to the SRESA1B run in the eight cases and decreases in the four. In particular, in five CGCMs [CGCM 3.1 (T47), GFDL-CM2.0, GISS-ER, MRI-CGCM2.3.2, and UKMO-HadGEM1] the increase in the correlation coefficient is statistically significant. The correlation coefficient of the ensemble mean is 0.41 in the SRESA1B run, and is statistically significantly larger than that in the 20C3M run, 0.29, at 95% confidence level (Figure 2). This result indicates that under the global warming the PDO and ENSO are more tightly linked with each other in a linear sense, and in the future El Niño (La Niña) tends to accompany a positive (negative) phase of PDO more frequently than in the present.

Table II shows the number of conditional occurrence of El Niño and La Niña by the phase of PDO in the 20C3M run and the SRESA1B run, respectively. The El Niño and La Niña events are defined when the ENSO index is above and below 0.5 the normal standard deviations, respectively. Similarly, the positive and the negative phases PDO are defined when the PDO index is above and below 0.5 the normal standard deviation, respectively. As consistent with the result in the correlation analysis, the number of conditional occurrence of both El Niño—a positive phase PDO and La Niña—a negative phase PDO increases from the 20C3M run to the SRESA1B run. In contrast, the number of conditional occurrence of both El Niño—a negative phase PDO and La Niña—a positive phase PDO decreases from the 20C3M run to the SRESA1B run. This has an important implication in terms of prediction of climate around the Pacific basin under the global warming because the intensity of midlatitude response to ENSO is significantly modulated by the ENSO–PDO linear relationship. In the present climate, as mentioned earlier, the midlatitude response to ENSO is strong when ENSO and PDO are in phase (i.e. El Niño—a positive phase PDO or La Niña—a negative phase PDO), while such response is weak when ENSO and PDO are out-of-phase relationship (i.e. El Niño—a negative phase PDO or La Niña—a positive phase PDO). Therefore, one may expect that in the future an event in which the midlatitude atmospheric circulations respond strongly to ENSO will be more frequent. Note that there are El Niño and La Niña events without accompanying a positive or negative phase PDO in both the 20C3M run and the SRESA1B run. The midlatitude response to such El Niño and La Niña events would change from the 20C3M run to the SRESA1B run, which is not discussed in this study.

Table II. The number of conditional occurrence of El Niño and La Niña by a phase of PDO in the 20C3M run and the SRESA1B run, respectively
 El Niño—a positive phase of PDOEl Niño—a negative phase of PDOLa Niña—a positive phase of PDOLa Niña—a negative phase of PDO
20C3M1407195152
SRESA1B1834877194

To elucidate the effect of changes in the ENSO/PDO relationship, in Figure 3(a) and (b) we show the conditional composite of the anomalous SST (shading) and 500 hPa geopotential height (contour) for an El Niño—a positive phase PDO and a La Niña—a negative PDO in the 20C3M run, respectively. Figure 3(c) and (d) is the same as in Figure 3(a) and (b) except for the SRESA1B run. One can find that when ENSO and PDO are in phase both in the 20C3M run and the SRESA1B run anomalous 500 hPa geopotential height is characterized by a Pacific North America (PNA)-like pattern (Wallace and Gutzler, 1981). In the SRESA1B run (Figure 3(c) and (d)), the centre of PNA-like pattern over the North Pacific is slightly shifted eastward by about 10 degrees while substantially strengthening compared that from the 20C3M run. Both the intensification and shift could cause strong northward (southward) and warm (cold) advection of air along the west coast of America during El Niño (La Niña). Thus under the global warming an enhanced anomalous SST would be found along the west coast of America and the eastern subtropical Pacific as shown in Figure 3(c) and (d).

Figure 3.

The conditional composite of anomalous SST (shading) and 500 hPa geopotential height (contour) for an El Niño—a positive PDO (a) and a La Niña—a negative PDO (b) in the 20C3M run. (c)–(d) are the same as in (a)–(b) except but the SRESA1B run. Unit is °C for SST and metre for geopotential height. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Furthermore, the anomalous North Pacific SST is also greater in the SRESA1B run (Figure 3(c) and (d)) than in the 20C3M run (Figure 3(a) and (b)) when ENSO and PDO are in phase. In particular, in the SRESA1B run the maximum anomalous SST is located in the central North Pacific while in the 20C3M run it is observed in the western North Pacific. In other words, the centre of the maximum anomalous North Pacific SST in response to ENSO is shifted to the eastward from the 20C3M run to the SRESA1B run. However, such a shift is not found when ENSO and PDO are out of phase (not shown) or if we just compare the PDO modes from the two periods (Figure 1(c) and (e)). It is worthwhile to note that there are two modes of SST variability in the North Pacific. One is the SST variability around the subpolar front including the Kuroshio–Oyashio Extension in the western part of the North Pacific, which has little relationship with the tropical Pacific SST variability. The other is the SST variability around the subtropical front in the central part of the North Pacific, which is associated with the tropical and subtropical SST variability (Nakamura et al., 1997; Miller and Schneider, 2000; Barlow et al., 2001; Schneider et al., 2002; Wu and Liu, 2003). Therefore, the shift of the maximum anomalous North Pacific SST as shown in Figure 3 indicates that the SST variability in the PDO, which is connected to the tropical and subtropical SST variability, is enhanced under the global warming in association with the intensification of ENSO–PDO linear relationship.

3.2. Physical links

To examine why the in-phase ENSO–PDO relationship becomes strong in the future climate, we compare the conditional composite of the in-phase ENSO–PDO with that of the out-of-phase ENSO–PDO as shown in Figure 4. In the 20CM3 runs, the magnitude of the composited El Niño is stronger in the El Niño—the positive phase PDO (Figure 4(a)) than that in the El Niño—the negative phase PDO (Figure 4(b)), while in the SRESA1B runs, the magnitude of former (Figure 4(c)) is weaker than that of the latter (Figure 4(d)). However, the centre of the maximum anomalous tropical SST is located farther to the west in the in-phase composite than that in the out-of-phase composite both the 20CM3 and the SRESA1B. In short, when the El Niño and the PDO are in phase the maximum of the tropical SST anomaly tends to occur westward irrespective of the strength of the anomaly in both the 20C3M run and the SRESA1B run. We also compared the conditional composites of the anomalous SST for the La Niña—the positive phase of PDO and the La Niña—the negative phase of PDO (not shown). It is found that there is a shift of minimum anomalous SST from the La Niña—the positive phase of PDO to the La Niña—the negative phase of PDO in the two runs although it is not much significant compared with the El Niño–PDO cases.

Figure 4.

The conditional composite SST anomalies for an El Niño—a positive PDO (a) and an El Niño—a negative PDO (b) in the 20C3M run. (c)–(d) are the same as in (a)–(b) except but the SRESA1B run. Unit is °C. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

We hypothesize that the structural changes in the tropical Pacific SST variability, in particular, the change in the location of the tropical Pacific SST forcing, is associated with the changes in ENSO–PDO relationship from the 20C3M run to the SRESA1B run. Previous studies (Alexander et al., 2002; Barsugli and Sardeshmukh, 2002) suggested large sensitivity of the North Pacific pattern to the location of the tropical SST forcing. The westward shift of the anomalous warm SST forcing could enhance the magnitude of the extratropical atmospheric response to El Niño, therefore, the shift is more effective in accompanying the positive phase of PDO (Vimont, 2005). Recently, Yeh et al. (2009) showed that the central Pacific El Niño occurs more frequently in the SRESA1B run than the 20C3M run using the same CGCMs indicated in Table I. They argued that such frequent central Pacific El Niño occurrence is associated with change in the background state under anthropogenic global warming, in particular change in the thermocline structure in the equatorial Pacific. An increase of the El Niño event in which the centre of maximum SST is located in the western and central equatorial Pacific may lead an enhanced linear relationship of ENSO–PDO in the SRESA1B run compared with the 20C3M run.

On the other hand, one may raise the important question that the PNA-like pattern is enhanced in SRESA1B runs compared with the 20C3M run. One of the possible reasons is that the global warming mean state of the Pacific Ocean could be a favourable condition for producing the PNA-like wave pattern. The global warming climate state can change the tropical heating, which in turn could maintain the enhanced stationary wave pattern over the North Pacific and North America. Figure 5(a) and (b) shows the difference of mean precipitation in the El Niño—the positive phase of PDO and the La Niña—the negative phase of PDO between the two runs, respectively, in the eight CGCMs where the linear relationship of ENSO–PDO increases under global warming (Figure 2). Both figures show the stronger convection activity in the equatorial central Pacific (150°E–180), which is very sensitive to the intensity of PNA-like atmospheric anomalies (Newman, 2007). This enhanced convection is associated with mean SST increase in the tropical Pacific according to global warming (Kug et al., 2009).

Figure 5.

Maps of composite precipitation differences between 20C3M and SRESA1B runs for an El Niño—a positive PDO (a) and a La Niña—a negative PDO (b) in the eight runs in which the ENSO–PDO relationship increases under the global warming. Unit is mm d−1. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

4. Concluding remarks

By directly comparing results for the present climate simulation (i.e. 20C3M run) and those for the future climate simulation (i.e. SRESA1B run), we examined changes in the relationship between ENSO and PDO under the global warming. It is found that a linear relationship between the two becomes stronger in the future climate. Therefore, due to the constructive influence of PDO in the future one may except to observe stronger climate response to ENSO in the midlatitude. In particular, it is found that the PNA-like pattern over the North Pacific is substantially strengthened with an eastward shift of its centre in the in-phase ENSO–PDO relationship from the 20C3M run to the SRESA1B run.

An enhanced linear relationship of ENSO–PDO in the SRESA1B run could be associated with the position of maximum anomalous SST in the tropical Pacific. In both the 20C3M run and the SRESA1B run, the maximum anomalous SST is shifted to the west in the El Niño—the positive phase of PDO compared with the El Niño—the negative phase of PDO. In the SRESA1B run, the El Niño event in which the maximum anomalous SST is located in the western and central tropical Pacific occurs more frequently leading the enhanced linear relationship of ENSO–PDO. On the other hand, one of the possible reasons for the enhanced PNA-like wave pattern is changes in the tropical heating. In the SRESA1B run, the mean precipitation in the El Niño—the positive phase of PDO and the La Niña—the negative phase of PDO increased in the central tropical Pacific, which is able to amplify the PNA-like wave pattern over the North Pacific. We, thus, do not exclude the possibility that the enhanced atmospheric response to the central tropical convection anomalies, which is mainly because of changes in tropical Pacific mean state, could induce a large surface warming or cooling in North Pacific Ocean via boundary layer heat exchanges. Therefore, the enhanced atmospheric response to El Niño and La Niña may also lead to increase the ENSO–PDO linear relationship under the global warming.

Acknowledgements

This work was supported by the National Research Foundation of Korea Grant funded by the Korean Government (MEST) (NRF-2009-C1AAA001-2009-0093042). MinHo Kwon is supported by Korea Ocean Research and Development Institude (grants PE98801).

Ancillary