The North Pacific Oscillation–ENSO and internal atmospheric variability

Authors


Abstract

[1] The relationship between the North Pacific Oscillation (NPO) and El Niño and Southern Oscillation (ENSO) is reexamined using three runs of a coupled general circulation model. The difference in the three simulations is only in the amplitude of internal atmospheric variability felt by the ocean component at the air-sea interface. In observations and a standard coupled model (one AGCM coupled to a single OGCM), the NPO and ENSO may be explained by correlated white noise time series or by two independent red noise time series processes. However, there is the possibility the influence of ENSO could be felt depending on the amplitude of the noise in the North Pacific. When the amplitude of the atmospheric noise is reduced, stronger ENSO variability goes with a stronger ENSO-NPO correlation. The analysis suggests that local North Pacific internal atmospheric dynamics overwhelms the remote influence from ENSO in the coupled model.

1. Introduction

[2] Understanding tropical-extratropical teleconnections is important in terms of extended-range prediction of weather and climate. Although variations in midlatitude circulation anomalies associated with tropical sea surface temperature (SST) forcing has been studied extensively [e.g., Hoskins and Karoly, 1981], the issue of the physical origin of the correlation between SST changes in the tropical and North Pacific remains an outstanding problem.

[3] Previous studies have suggested that there is a component of North Pacific SST variability that is linearly independent of tropical SST variability and there is variability that is linked to tropical SST forcing [Deser and Blackmon, 1995; Alexander et al., 2002]. There is still difficulty determining how much of North Pacific SST variability is due to local stochastic forcing and how much is due to remote tropical SST forcing. Part of the difficulty in coming to a consensus concerns testing the effects of stochastic forcing on the North Pacific SST variability within the context of uncoupled forced feedback versus coupled ocean-atmosphere feedbacks [Barsugli and Battisti, 1998].

[4] In this study we examine how atmospheric stochastic forcing impacts North Pacific variability in a coupled GCM. In order to control the SST variability that is forced by atmospheric noise at the air-sea interface we use the interactive ensemble approach [Kirtman and Shukla, 2002]. This approach is to extend this notion of ensemble averaging to a coupled model with the expressed purpose of reducing the variability that is forced by the internal atmospheric dynamics at the air-sea interface. As the interactive ensemble evolves, each AGCM realization experiences the same SST predicted by the OGCM. The OGCM, on the other hand, experiences surface fluxes that are the ensemble average of the six or twelve AGCM realizations. The control simulation consists of one AGCM coupled to one OGCM.

[5] Here we focus on the relationship between the leading mode of North Pacific SST variability (i.e., NPO) and the leading mode of tropical SST variability (i.e., ENSO). When the SST variability that is forced by atmospheric noise is reduced, our analysis indicates that NPO is linearly related to ENSO, although the strength of this relationship undergoes decadal variations which are related to low frequency modulation of the amplitude of ENSO. Moreover, it is found that the dominant spatial pattern of NPO changes with the variations of ENSO amplitude.

2. Model and Data

[6] The atmospheric component of the standard and interactive ensemble anomaly coupled model is the Center for Ocean-Land-Atmosphere Studies (COLA) AGCM with triangular truncation at wave number 42 and 18 vertical levels. The ocean model is adapted from the Geophysical Fluid Dynamical Laboratory (GFDL) modular ocean model [Pacanowski et al., 1993] version 3 (MOM3). The results from interactive ensemble model with the six (hereafter, IE-6) and twelve (hereafter, IE-12) AGCM realizations are compared to those of the standard anomaly coupled model (hereafter, SC). The choice of six and twelve ensemble members is based on the analysis of variance simulated in COLA AGCM. The simulated atmospheric signal in the COLA AGCM is reasonably well isolated with six ensemble members [Kirtman and Shukla, 2002]. All of the analysis shown here is based on the data for the 300 years. We also used monthly mean observed SST data taken from January 1950 to December 2000 which were analyzed by the National Centers for Environmental Prediction (NCEP) [Reynolds and Smith, 1994].

[7] The NPO and ENSO mode is defined by the leading empirical orthogonal function (EOF) of the North Pacific (20°N–60°N, 120°E–120°W) and the equatorial Pacific (20°N–20°S, 120°E–90°W), respectively. Because the dominant mode of SSTA variability over the North Pacific, as inferred from EOF analysis, exhibits a rather similar spatial structure year-round with comparable magnitude [Zhang et al., 1998] our analysis period includes the entire calendar year. Note that the analysis of NPO and ENSO mode using only November–March results is repeatedly tested and has little qualitative impact on the results presented here.

3. The NPO-ENSO Association in the Observations

[8] We first examine the relationship between the NPO and ENSO for the observed period from 1950 to 2000. Figures 1a and 1b show the leading SSTA EOFs for the NPO and ENSO. The spatial pattern of the NPO mode (Figure 1a) shows an elliptical shape located in the western and central North Pacific, accompanied by anomalies of the opposite sign to the eastern Pacific coast, north and south. This EOF accounts for 24% of the total variance. The ENSO mode shown in Figure 1b is the leading EOF of tropical Pacific SST, and accounts for 53% of the total variance.

Figure 1.

The leading EOFs in the North Pacific (a) and tropical Pacific (b) for the period of 1950–2000. Contour interval is 0.1. Units are arbitrary and shading is for positive. Panel (c) shows the 20-year sliding correlation coefficient of the time series of principal components (PCs) of (a) and (b). The straight solid line (0.63) indicates that the correlation value having a 95% confidence level.

[9] The simultaneous correlations between the NPO index with various indices of ENSO are low [Mantua et al., 1997]. However, it is unknown whether the relationship of the NPO-ENSO is stationary or not. Here we attempt to diagnose changes in relationship between the NPO and ENSO. To show changes in the connection of NPO-ENSO, the 20-yr sliding correlation coefficient of the time series of the principal component (PCs) of the NPO and ENSO is presented in Figure 1c. Although there are periods where the correlations are relatively high (1950–1960) or low (1980–1990), there is no statistically significant relationship of the NPO-ENSO during the entire period.

[10] Based on a bootstrap [Efron, 1982] method, we examined whether low-frequency sliding correlations of the NPO-ENSO may be due to noise in the time series [Gershunov et al., 2001]. The standard deviation of the time series shown in Figure 1c is 0.11 which is not significant at the 95% confidence level in a one-tailed test according to Gershunov et al. (The confidence limit of 95% and 5% percentiles of the time series shown in Figure 1c is 0.25 and 0.11, respectively. The standard deviations must be outside these limits to be considered significant.) Indeed, we cannot eliminate the null hypothesis, i.e., the NPO-ENSO relationship may be explained by correlated white noise time series or by two independent red noise time series processes. In the next section we examine this relationship with much larger time series from coupled model simulations. The coupled model simulations differ in terms of the amplitude of the atmospheric stochastic variability at the air-sea interface via the interactive ensemble technique.

4. Model Results

[11] In this section we examine the NPO-ENSO relationship in the coupled model simulations. The SC has one AGCM coupled to one OGCM. The second and third simulations use the interactive ensemble technique to couple six or twelve atmospheric realizations to a single ocean realization, respectively.

[12] Figures 2a and 2b are the same as in Figures 1a and 1b except we have used the SSTA simulated by the SC model for a period of 300 years. The geographical pattern of the model's NPO and ENSO mode has some similarities and differences with observations. The simulated NPO (Figure 2a) stretches from the coast of Asia into the central North Pacific, which seems to be in agreement with observations. The center of action, which is oriented to the northeast-southwest in the central North Pacific, has some differences from observations. The structure of the ENSO mode in the SC model (Figure 2b) is similar to the observed, although the SST variability is weak in the eastern tropical Pacific, is too narrowly confined to the equator and extends to far to the west.

Figure 2.

The same as in Figure 1 except for the simulated SSTA for the period of 300 years in the standard coupled model.

[13] Figure 2c shows the 20-yr sliding correlation coefficient of the PC time series of the NPO and ENSO modes for the period of 300 years. Note that a value of 0.52 is the 95% confidence level. The temporal relationship of the NPO-ENSO is characterized by larger decadal-to-multidecadal swings in the sliding correlations from −0.2 to 0.4, however, there is no significant relationship of the NPO-ENSO during the entire data record. We have also applied the statistical test based on Table 1 in Gershunov et al. to the sliding correlation shown in Figure 2c. The confidence limit of 95% and 5% percentiles of the time series is 0.29 and 0.12, respectively. The standard deviation of the time series shown in Figure 2c is 0.14, indicating that the low-frequency variability in the relationships between the NPO-ENSO modes may be explained by correlated white noise time series.

[14] Based on the results of IE-6 and IE-12 simulation we test the effects of stochastic forcing on the NPO-ENSO relationship. When the SST variability forced by internal atmospheric dynamics is reduced, it is found that there is a significant linear relationship between the NPO and ENSO. Figure 3 is the same as in Figure 2c except for the IE-6 (thin) and IE-12 (thick). The correlation remains positive for the entire period in both the IE-6 and IE-12 with low frequency fluctuations in the relationship between the two modes. For the relationship of the NPO-ENSO modes in the IE-12, there are more than 250 years where the correlation exceeds the 95% confidence level. Based on the same statistical test applied to the observations and the SC, the low-frequency sliding correlations of the NPO-ENSO in the IE-12 are also significant at the 95% confidence level. This indicates that the low-frequency variability in the relationship between the NPO-ENSO modes cannot simply be explained by correlated white noise time series when the forcing due to internal atmospheric dynamic has been significantly reduced.

Figure 3.

The same as in Figure 2c except for the IE-6 (thin line) and IE-12 (thick line). The dashed straight line (0.52) indicates that the correlation value having a 95% confidence level. The two solid straight lines (0.61 and 0.53) determine periods for the high and low NPO-ENSO periods, respectively. (See the text for the details).

[15] One question of interest here is to diagnose whether there are any structural changes in the NPO associated with the low frequency changes in the correlation seen above. In order to address this question we examine the spatial pattern of the NPO mode for each NPO-ENSO correlation period. The high NPO-ENSO period is based on periods when the sliding correlation (thick in Figure 3) exceeds 0.61. Similarly, the low NPO-ENSO period is based on periods when the same time series is below 0.53. Note that the grand mean of sliding correlation is 0.57 and a half standard deviation is 0.04. Figures 4a–4c are the same as in Figure 2a except we have used the SSTA simulated by the IE-12 for a period of 300 years (a), high NPO-ENSO period (b) and low NPO-ENSO period (c). The NPO mode for the high NPO-ENSO period (Figure 4b) is similar to that for a whole period (Figure 4a). However, the NPO mode for the low NPO-ENSO period is distinctly different, i.e., the tilting structure to the northeast-southwest is changed into the east-west extended pattern in the central North Pacific. In order to test whether these differences in the dominant patterns seen in Figures 4b and 4c are significant we applied a Monte Carlo approach. We randomly choose two pairs of time chunks from the data such that the length of the first (second) chunk is the same as the length of the high (low) ENSO-NPO period. We calculated a pattern correlation of NPO modes based on randomly selected two pairs of time chunks. We repeated this procedure to get the distribution of pattern correlation between two pairs of randomly chosen time chunks. If the pattern correlation is less than 0.5, then the patterns are different at the 95% level based on this Monte Carlo test. The pattern correlation between Figure 4b and Figure 4c is 0.42, which is significant at 95% confidence level. We argue that the changes in the dominant spatial pattern of NPO are caused by, rather than due to, the low-frequency variations of the NPO-ENSO relationship (which themselves are forced by the amplitude of the ENSO signal). This change can only be detected when the forcing due to internal atmospheric dynamic has been significantly reduced. Note that there is no particular difference in ENSO modes for each NPO-ENSO period (not shown).

Figure 4.

The leading EOF in the North Pacific for the simulated SSTA for the periods of 300 years in the IE-12 (a). The same as in (a) except for the high NPO-ENSO period (b) and low NPO-ENSO period (c). (d) is the time series of standard deviation of NINO3.4 SST index with a 20-year running mean (solid). Unit is °C. Note that the amplitude of the standard deviation time series is indicated on the left panel. The straight solid lines (0.35°C, 0.31°C) determine the high and low ENSO amplitude periods (See the text for the details). The dashed line is the sliding correlation coefficients of the IE-12 as in Figure 3. The same as in (a) for the high ENSO amplitude period (e) and low ENSO amplitude period (f).

[16] In order to understand changes of the dominant pattern of the NPO mode shown in Figures 4b and 4c we note that these relationship between the NPO-ENSO modes shows a similar phase relationship with variations of ENSO amplitude. Figure 4d shows the time series of standard deviation of NINO3.4 (5°N–5°S, 170°E–240°E) SST index with a 20-yr running mean (solid line) in the IE-12 simulation. Note that the magnitude of ENSO amplitude is indicated on the left of the panel and the dashed line is the sliding correlation coefficients of the IE-12 as in Figure 3. The variations in ENSO amplitude are highly correlated to the sliding correlations of the NPO-ENSO modes (simultaneous correlation: 0.81). This indicates that the ENSO amplitude is relatively high during the high NPO-ENSO periods, conversely, the low NPO-ENSO periods are marked by relatively low ENSO amplitude. Figures 4e and 4f are the same as in Figures 4b and 4c except for the high and low ENSO amplitude periods. The periods for the high/low ENSO amplitude are determined by the same procedure as the high/low NPO-ENSO correlation period (i.e., plus or minus a half of the standard deviation). As expected, the spatial pattern of the NPO mode for the high (low) ENSO amplitude period resembles that for the high (low) NPO-ENSO correlation period. This result suggests that the tropical-North Pacific SSTA teleconnections appear to be linearly correlated to the strength of tropical forcing when the forcing due to internal atmospheric dynamics has been significantly reduced. However, it is not true of the observations and the SC. The variations of ENSO amplitude are unrelated to the NPO-ENSO correlation in the observations and the SC. Note that the correlation between the variations of ENSO amplitude and the sliding correlations of the NPO-ENSO modes is 0.12 (observation), −0.33 (SC) and 0.69 (IE-6). When the SST variability forced by internal atmospheric dynamics is reduced, the NPO-ENSO relationship is correlated to the strength of tropical SST variability, and accompanied by changes in the pattern of the NPO.

5. Concluding Remarks

[17] Recently, Newman et al. [2003] argued that it is the “re-emergence” process that reddens both “atmospheric noise” and the “ENSO-related atmospheric teleconnection”, and it is this combination that leads to the observed power spectrum of NPO SST variability. Using the interactive ensemble technique for controlling the amplitude of the atmospheric noise, we reexamine the relationship between the NPO and ENSO modes. In observations and a standard coupled model, the temporal relationship of the NPO-ENSO is characterized by decadal variations, however, we cannot eliminate the possibility that these variations are simply due to correlated white noise time series or due to two independent red noise time processes.

[18] The coupled model behavior suggests that local North Pacific internal atmospheric dynamics overwhelm the remote influence form ENSO. When the amplitude of the atmospheric noise is reduced, it is found that there is a significant linear relationship between the NPO and ENSO. Stronger ENSO variability goes with a stronger ENSO-NPO correlation. The model results show that the variations of ENSO amplitude largely determine the low-frequency variations of the NPO-ENSO relationship accompanying changes of the dominant pattern of the NPO. Our result is somewhat different from the hypothesis suggested by Newman et al., who argued that ENSO forces changes in the NPO. In other words, the NPO-ENSO relationship is based on rates of change (i.e., equation imageENSO), which would indicate the ENSO and the NPO should be in quadrant with little simultaneous correlation. However, we argue here that there is a simultaneous relationship, but in both the observations and the SC simulation the relationship is difficult to detect due to destructive interference from atmospheric stochastic forcing.

Acknowledgments

[19] Comments from two anonymous reviewers greatly improved this manuscript. This research was supported by grants from the National Science Foundation ATM-9814295 and ATM-0122859, the National Oceanic and Atmospheric Administration NA16-GP2248 and National Aeronautics and Space Administration NAG5-11656.

Ancillary