The potential predictability associated with the remote influence of midlatitude tropospheric anomalies over the North Pacific or the North Atlantic, via a seesawlike interannual oscillation between the surface Aleutian and Icelandic lows (AL and IL, respectively) is investigated. Data from a 24-member ensemble of 50-yr atmospheric general circulation model simulation forced with observed sea surface temperature (SST) conditions are analyzed by separating the total simulated fluctuations into the external component forced by the prescribed SST and the internal component generated by atmospheric internal dynamics. The AL–IL seesaw can be identified in both the external and internal components of the variability. In the external variability, determined through the ensemble mean, the seesaw is gradually formed from December to March through the development of a Pacific–North American (PNA) pattern–like wave train, remotely forced by the El Niño–Southern Oscillation. The amplitudes of the externally forced North Atlantic anomalies are only about half as large as the North Pacific anomalies. The potential predictability of the Atlantic anomalies, defined as the ratio of the SST-forced variance to the total variance, does not exceed the 20% level. In the internal component of the variability, determined from the deviations of each ensemble member from the ensemble mean, the negative correlation between the AL and IL anomalies is modest but persistent through winter. It is confirmed that, regardless of the polarity of the AL–IL seesaw, the IL anomalies are formed through eastward wave activity propagation of the stationary Rossby wave train emanating from the AL region in the form of what may be called a “PNAA pattern,” the extension of the PNA-like wave train into the Atlantic. Thus, the midwinter development of North Pacific anomalies is found to be a necessary, though not sufficient, condition for the seesaw formation. The persistence of the North Pacific anomalies beyond a 1-month time span appears to augment the probability of the seesaw formation by sustaining eastward wave activity propagation to the North Atlantic.
During the Northern Hemisphere winter, the tropospheric low-frequency variability in the North Pacific tends to be as strong as its North Atlantic counterpart (Blackmon et al. 1977, 1984). Furthermore, it has been found that on interannual time scales the wintertime tropospheric perturbations in these two regions vary more or less coherently, in what is known as the Aleutian low (AL) and Icelandic low (IL) seesaw (e.g., Bjerknes 1966; van Loon and Rogers 1978; Wallace and Gutzler 1981; van Loon and Madden 1983). Recently, Honda et al. (2001, hereafter referred to as Part I) showed that the seesaw is most apparent in late winter (February through mid-March), when it forms as a remote atmospheric influence of the North Pacific variability associated with the AL. This “atmospheric bridge” is apparent particularly in the upper troposphere, where the North Pacific influence extends across North America in the form of a stationary Rossby wave train, corresponding to the Pacific–North American (PNA) teleconnection pattern (Wallace and Gutzler 1981). The IL anomalies, which are associated with the North Atlantic Oscillation (NAO), form as part of another wave train that appears to emanate from the Atlantic edge of the PNA-like wave train. The feedback from synoptic-scale baroclinic eddies migrating along the Atlantic and Pacific storm tracks is important in the amplification and maintenance of stationary anomalies that comprise those wave trains (Part I). The significance of the AL–IL seesaw in the interannual tropospheric variability over the wintertime Northern Hemisphere is apparent in the fact that the seesaw can be identified as the first empirical orthogonal function (EOF) of 250-hPa height anomalies (Honda and Nakamura 2001) and also of the 850-hPa streamfunction anomalies (Ambaum et al. 2001). Hence, in most of the troposphere, the variability associated with the seesaw dominates over that associated with the Arctic Oscillation or Northern Hemisphere annular mode (NAM; Thompson and Wallace 1998, 2000), especially in late winter (Honda and Nakamura 2001). The seesaw influence on the zonal wavenumber-2 component in the stratospheric circulation has been demonstrated by Nakamura and Honda (2002).
As shown in Part I, the downstream influence of the North Pacific variability via propagation of stationary Rossby wave trains, namely the PNA pattern and its extension into the Atlantic: hereafter, the PNA–Atlantic (PNAA) pattern is a crucial dynamical component of the AL–IL seesaw formation. The presence of the North Pacific influence on the North Atlantic is also suggested by other recent studies (Ambaum et al. 2001; Lau and Nath 2001; Wallace and Thompson 2002). Thus, the strength of the IL, which exerts significant impacts on weather conditions in the Euro-Atlantic sector, is likely to come under a remote influence from the North Pacific, especially in late winter. Therefore, a successful long-range forecast of climate conditions in the Euro-Atlantic sector may depend on a North Pacific influence, specifically in the form of the AL–IL seesaw.
Significant precursory signals of the seesaw are identified over the eastern North Pacific one month before its mature phase (Part I). Forecasting the seesaw long enough before its formation, however, may not be straightforward, because of its strong seasonal dependence and yet undetermined origin. Moreover, the presence of a significant decadal-scale modulation, at least during the recent half century, which consists of alternating long intervals of inactivity and activity (Part I), adds to the confusion regarding the nature and predictability of this phenomenon.
To understand the mechanisms causing the seesaw and to assess its potential predictability, we investigate in this study the dynamical characteristics of the AL–IL seesaw, in a set of ensemble hindcast climate experiments using an atmospheric general circulation model (AGCM) forced with global sea surface temperature (SST) anomalies observed over the last 50 yr or so. We diagnose the fluctuations associated with the model-simulated AL–IL seesaw by separating them into the component forced by the prescribed SST conditions and the component arising from atmospheric internal dynamics, as explained in section 2.
2. Experimental design and statistical approach
The AGCM used in this study is known as the fourth version of ECHAM (Roeckner et al. 1996). This AGCM was originally derived from a spectral weather forecast model of the European Centre for Medium-Range Forecasts (ECMWF) and has been modified and upgraded at the Max Planck Institute for Meteorology in Hamburg, Germany. We used a particular version of the model (ECHAM4.5) with a triangular truncation at wavenumber 42 and 19 levels in the vertical (T42L19). The horizontal resolution corresponds approximately to a 2.8° grid in both latitude and longitude. The above version of the model is operationally used for seasonal climate forecast at the International Research Institute for Climate Prediction. In the present study, an ensemble of 24 simulations, forced by observed SST and climatological sea ice distributions for the period 1951–2000, was utilized. We will hereafter refer to those simulations as hindcast simulations. Each ensemble member was integrated with different atmospheric initial conditions, thus providing 24 independent realizations of the model’s internal variability plus the underlying SST-forced component.
Our prime statistical tool is the so-called “analysis of variance” (ANOVA; e.g., Zwiers et al. 2000), which has been applied to the ensemble hindcast climate simulations to separate the total atmospheric variance (σ2TOT) of a monthly averaged quantity into the component (σ2EXT) arising from the SST and sea ice distribution specified in the lower boundary (hereafter external forcing) and another component (σ2INT) driven solely by internal atmospheric dynamics (e.g., Harzallah and Sadourny 1995; Rowell 1998). In the extratropical atmosphere, σ2INT is generally larger than σ2EXT (e.g., Zwiers et al. 2000; Zheng et al. 2000). The relative importance of internally and externally forced variability can be measured by calculating the “potential predictability,” defined as the ratio of the external variance to the total variance (σ2EXT/σ2TOT).
Following Zwiers et al. (2000) and Zheng et al. (2000), the externally forced variability is defined as monthly anomalies in the 24-member ensemble average, because fluctuations arising from the internal dynamics of the atmosphere should mostly cancel out given the sufficiently large number of integrations. The internally varying part of the variability is obtained as the ensemble of deviations of each individual member from the ensemble mean. For each calendar month, externally forced variability of a given variable is represented by a sample of 50 anomaly maps, whereas the internal variability is represented by a sample of 24 × 50, or 1200 maps. Throughout this paper, we use the words “external” or “forced” variability to represent the ensemble mean state, “internal” variability to represent the deviation of each ensemble output from the ensemble mean, and “total” variability to mean the sum of the internal and external variability.
3. AL–IL seesaw in external variability and its potential predictability
The model-simulated 50-yr mean wintertime SLP field closely reproduces the observed climatological-mean SLP pattern (not shown). Though not identical, the magnitudes and distribution of the interannual standard deviations of SLP are also similar to those observed (not shown).
Figure 1 presents the statistical relationship between the model-simulated AL and IL intensities, which are defined as the area-mean SLP over the North Pacific (40.5°–54.5°N, 163.2°–135°W) and North Atlantic (57.2°–62.8°N, 33.8°–14.1°W), respectively. The figure shows the 31-yr moving correlation coefficients between the AL and IL intensities for each winter month, based on the ensemble mean output (heavy solid line) and the total variability for each of the individual ensemble members (thin lines). These can be compared to the same statistics based on observed SLP (heavy dashed line; data taken from Trenberth and Paolino 1981). For each of the winter months, the correlation based on the individual simulations tends to be negative (Fig. 1), indicative of the potential existence of the AL–IL seesaw in the internal variability of the model through winter.
The seesaw relationship in the ensemble mean (or forced) variability, in contrast, exhibits a strong seasonal dependence, akin to but not identical to the observed variability. In early winter, the AL–IL correlation coefficient in the ensemble mean is nearly zero or weakly positive. It gradually becomes negative toward late winter until it reaches −0.8 in March. The correlation remains strongly negative in April. The observed seesaw relation also exhibits a seasonal dependence, but the observed negative correlation peaks in February and tends to be weaker than in the ensemble mean except during January. The stability of the seesaw, as measured by the change in the strength of the correlation over the years, is more pronounced in the model ensemble mean than in observations, especially in late winter.
Similar correlation tendencies have also been found in 50-yr hindcast ensemble simulations using an AGCM for the Earth Simulator (AFES; Ohfuchi et al. 2004).1 In the T42 version for the AFES, each of the simulation members exhibits negative correlation between the AL and IL intensities throughout the winter, and their negative correlation in the ensemble mean is strongest in March (not shown). This coherent signature suggests that this AL–IL seesaw relationship of internal and external variability is a robust phenomenon.
Figure 2a shows 50-yr ensemble-mean time series of the AL and IL intensities for March. Consistent with Fig. 1, negative correlation between the normalized AL and IL intensities is apparent (Fig. 2b). We define the AL–IL seesaw index (AII), which represents the strength and polarity of the seesaw, as the normalized IL intensity subtracted from the normalized AL intensity. The model-simulated AL–IL seesaw index is highly inversely correlated with ENSO. Figure 2c shows the AII and sign-reversed normalized time series of SST anomalies averaged over the equatorial eastern Pacific region (the area referred to as Niño-3.4, 5°S–5°N, 170°–120°W) for the 3-month mean of December, January, and February. Their high correlation coefficient (−0.77 for the 50 yr) strongly suggests that the late-winter AL–IL seesaw signal in the externally forced variability is a manifestation of the remote influence of midwinter ENSO.
The subseasonal evolution of the atmospheric teleconnection associated with the seesaw is shown in composite difference maps of ensemble-mean 200-hPa geopotential height (Z200) anomalies between the strongest positive and strongest negative AII years calculated separately for each calendar month from December to March (Fig. 3; the years used for the composites are indicated in the figure caption). In December (Fig. 3a), a PNA-like wave train forms over the North Pacific. Its leading edge already reaches the southern United States in the form of a stationary Rossby wave train, as evident in the associated wave activity flux vectors.2 This PNA-like pattern develops further in January (Fig. 3b). By February (Fig. 3c), there is indication for another wave train that emanates as an extension of the PNA-like pattern and stretches across the North Atlantic. As the latter appears, both negative anomalies over Canada and positive anomalies over the United States extend eastward into the North Atlantic. The negative anomalies correspond to the incipient IL anomalies at the surface. As the IL anomalies further develop by March, the AL–IL seesaw reaches its mature phase (Fig. 3d).
Dynamical processes of the seesaw formation in response to external forcing in the AGCM integration are essentially the same as in observations (Part I; not shown), but the time scale of the seesaw formation in the model is about 3 months, about twice as long as that observed. Furthermore, unlike in the observations (Part I), the strength of the IL anomalies in the model ensemble mean is only about half of that of the AL counterpart, as evident in the unnormalized time series of the AL and IL intensities (Fig. 2a). To further explore the implications of ensemble mean response, we evaluated potential predictability (σ2EXT/σ2TOT), following the example of Harzallah and Sadourny (1995) and Rowell (1998). As explained above (section 2), the external variance (σ2EXT) has been calculated based on the deviations of monthly means from the 50-yr climatology, all in the 24-member ensemble mean. Internal variance (σ2INT) has been obtained based on the deviation of monthly means in individual ensemble members from the 24-member ensemble mean. The total variance (σ2TOT) is the sum of the external and internal variances.
Considering the dominance of the AL–IL seesaw in late winter, we compare potential predictability in Z200 in November–December–January (NDJ) and February–March–April (FMA; Fig. 4). In early winter (Fig. 4a), the potential predictability is relatively high (35%–45) over the eastern North Pacific and the southern U.S. regions, which correspond to the centers of action of the PNA-like pattern. This PNA-like signal becomes more apparent in late winter (Fig. 4b), as potential predictability reaches 50% or even higher in those two regions. In addition, potential predictability also increases over Canada and the western portion of the North Atlantic. In particular, the predictability increases from 5% to 20% in late winter around the upper-tropospheric center of the IL anomalies. Correspondingly, a slight increase in the potential predictability is also seen in SLP anomalies over the North Atlantic in late winter (not shown). The increase of potential predictability toward late winter over North America and the North Atlantic is consistent with the remote influence of ENSO reaching into these regions in the form of stationary Rossby wave trains in the upper troposphere (PNAA pattern) and the associated AL–IL seesaw near the surface.
4. Dynamics of AL–IL seesaw in internal variability
We have thus far confirmed that the ENSO remote influence reaches in the North Atlantic through the atmospheric bridge between the North Pacific and North Atlantic in the form of the PNAA pattern. However, the low potential predictability over the midlatitude North Atlantic relative to the North Pacific indicates the dominance of internal variability, part of which may be associated with the seesaw. In this section, we examine the AL–IL seesaw signatures identified in the internal variability. As in the total variability (Fig. 1), a tendency for negative correlation between the AL and IL intensities is also evident in the model’s wintertime internal variability (as hinted in Fig. 1). Here, we focus on the simulated internal variability in February, when the observed AL–IL seesaw is most apparent (Part I). Qualitatively, the results are similar in January and March.
Figure 5a shows a scatter diagram of the normalized AL and IL intensities associated with the internal variability in February for all members. The second and fourth quadrants of the scatter diagram (Q2 and Q4, respectively) correspond to what we will refer to as “negative AL–IL seesaw (AIS)” (weak IL and strong AL) and “positive AIS” (strong IL and weak AL), respectively. The first and third quadrants (Q1 and Q3, respectively) correspond to in-phase changes in the intensity of the AL and IL. For each of the quadrants, the number of the months in which magnitudes of the normalized AL and IL anomalies both exceed an arbitrary threshold (0.7 or −0.7) is given in parenthesis. The dominance of the number of out-of-phase cases over the number of in-phase cases is evident. In fact, a null hypothesis that the anomalous AL and IL intensities in the internal variability are independent can be rejected at the 99.9% confidence level based on a chi-squared test.
Figures 6b,d show composite February Z200 fields of the negative and positive phases of the AL–IL seesaw, which exceed the threshold value (scatterplot quadrant Q2 and Q4), respectively. The seesaw is evident in each of the composites. Unlike in the external variability (Fig. 3), the North Atlantic anomalies are nearly as strong as the North Pacific ones, a situation similar to the observed seesaw, which suggests that the AL–IL seesaw is most likely a mode of internal variability. One can notice that height anomalies over the Rockies associated with the internally generated seesaw are oriented more meridionally than their counterpart associated with the external variability. This particular tendency is consistent with the more zonally oriented PNA-like wave train associated with the internally generated seesaw, as evident in the eastward-pointing wave activity flux vectors over midlatitude North America (Figs. 6b,d). A similar zonally oriented wave train can be observed over North America (over the United States) in the AL–IL seesaw signal in the anomaly field from which the remote ENSO influence has been statistically removed (Part I). Straus and Shukla (2002) argued that the PNA pattern is essentially a mode of internal variability but is not necessarily forced by ENSO. One may thus argue that the AL–IL seesaw can be triggered by the PNA pattern formed through internal midlatitude dynamics.
Figures 6a,c show the composites for corresponding January (one month earlier). In good agreement with the observations (Part I), significant precursory signals over the North Pacific sector are found in each phase of the seesaw. This suggests that circulation anomalies over the North Pacific may become a trigger of the seesaw formation a month later, even in the internal variability. Corresponding maps two months earlier do not show clear precursory signals over the North Pacific (not shown), indicating that the internally generated seesaw tends to be formed within a two-month time frame.
Considering this one-month lag relationship, we prepared another scatter diagram based on the North Pacific Z200 anomalies in January, as an upper-tropospheric manifestation of the anomalous surface AL, and the anomalous surface IL for February (Fig. 5b). The Pacific anomalies in the upper troposphere are defined as the area mean Z200 for the domain (37.7°–48.4°N, 168.8°–140.6°W). As in Fig. 5a, negative correlation is apparent also in the lag relationship between the Pacific and Atlantic anomalies (Fig. 5b). In Figs. 7a, b, composite upper-tropospheric anomalies for January and February, respectively, are shown for the case of the positive Z200 anomalies over the North Pacific for January and negative surface IL anomalies for February (Q4 in Fig. 5b), corresponding to the positive AIS. The composite maps display a typical AL–IL seesaw formation, in association with wave activity propagation in the form of stationary Rossby wave trains, as an extension of the PNA pattern and across the North Atlantic (i.e., the PNAA pattern; cf. Part I). The corresponding anomaly composites for the negative AIS (Q2 in Fig. 5b) exhibit an anomaly evolution very similar to that for the positive phase but with the reverse polarity (not shown).
Though less frequently realized, there exist some cases in which the Pacific anomalies in January and the Atlantic anomalies in February have the same sign. Figures 7c,d display sets of anomalous composite Z200 maps for January and February, respectively, based on those cases with positive Z200 anomalies over the North Pacific in January and the positive surface IL anomalies in February (Q1 in Fig. 5b). In the “in-phase” case, the North Pacific influence on the North Atlantic is not apparent. Rather, the spatial patterns are characterized by a PNA-like pattern in January and an NAO-like pattern in February, with no indication that the former pattern influences the latter. Unlike the seesaw formation cases (Figs. 7a,b), the Rossby wave activity carried with the PNA-like wave train across North America is mostly dispersed into the subtropics (Fig. 7c). It should be stressed that the North Pacific anomalies in January for the in-phase composite (Fig. 7c) bear striking similarities to the corresponding anomalies of the out-of-phase composite (Fig. 7a; pattern correlation is 0.87). This result suggests that prediction of the AL–IL seesaw formation based solely on the AL anomalies one month earlier may be difficult. Thus, in the framework of our AGCM simulations, anomaly development over the North Pacific in midwinter is a necessary but not sufficient condition for the AL–IL seesaw formation.
We can estimate the probability of the seesaw formation based on the overall number (24 members × 50 yr) of North Pacific anomalies identified in the internal variability. The probability that a significant circulation anomaly over the North Pacific in January whose normalized amplitude is greater than 0.7 will form an AL–IL seesaw in February (with an IL anomaly whose normalized magnitude exceeds 0.7) is about 35% (86 out of 269 cases and 116 out of 319 cases of large positive and large negative Z200 anomalies over the Pacific in January, respectively). It is higher than the expectation (25%) that the amplitude of a given normalized IL anomaly exceeds 0.7 assuming a normal distribution. In contrast, the probability that a significant anomaly observed over the Pacific in January will lead to an in-phase relationship in February with an significant IL anomaly is about 15% (38 out of 269 cases and 52 out of 319 cases of strong positive and negative Z200 anomalies over the Pacific in January, respectively). In fact, as shown in Fig. 5b, the probability density distribution of February IL anomalies is significantly skewed (above the 95% confidence level; e.g., White 1980) negatively and positively if sampled only for the cases in which the normalized January Z200 anomalies over the Pacific are greater than 0.7 and less than −0.7, respectively.
In this AGCM experiment therefore, the development of the North Pacific anomalous circulation does not necessarily guarantee the AL–IL seesaw formation. Nevertheless, the development appears to be an important trigger for the seesaw formation. Based on further investigation, we found that a persistent anomalous AL event, lasting more than a month, is one of the characteristics of seesaw events. In fact, the composite evolution of Z200 anomalies in Fig. 7 indicates more apparent persistence of the North Pacific anomalies in cases of a seesaw formation (Figs. 7a,b) than in cases of in-phase relationship (Figs. 7c,d) between the AL and IL intensities. This is despite the fact that all composites use the same criterion for a North Pacific anomaly in January, and the requirement for persistence of a North Pacific anomaly is not included in the composite criteria. Persistent anomalies effectively sustain eastward wave activity emanation into the North Atlantic, contributing to the development of the IL anomalies (not shown). The probability for the seesaw formation becomes even higher (about 45%) when the midwinter North Pacific anomalies persist for two months from January to February (35 out of 81 and 47 out of 106 for the positive and negative phases of the seesaw, respectively). In contrast, the probability for the development into the in-phase relation (no seesaw) is less than 10% (7 out of 81 cases and 10 out of 106 cases for the positive and negative phases, respectively). Thus, if we observe a significant anomaly over the North Pacific in a particular month in winter and can predict that the anomaly persists for a month, the probability of the AL–IL seesaw formation in the following month is high.
A similar analysis for the internal variability was applied to the total (external plus internal) variability. There are no substantial differences in the AL–IL seesaw statistics between the internal and total variability. Furthermore, we separated the total variability into ENSO years (the 20 yr shown in the caption for Fig. 3) and non-ENSO years (the other 30 yr) and separately applied to them the same analysis as above. The probability of the anomaly persistence over the North Pacific into the next month is generally higher in the ENSO years (50%–60%) than that in the non-ENSO years (30%–40%) through winter, based on 0.7 sigma criterion. Furthermore, precursory signals over the North Pacific one month earlier are also always stronger in the ENSO years, and these signals can be identified two months earlier. The probability for seesaw formation in March when the North Pacific anomalies persist from February to March is 49% during ENSO years (as an average for the positive and negative cases), which is just slightly higher than that during non-ENSO years (43%). Consistent with the results, the probability of the occurrence for the AL–IL seesaw to all samples in March is significantly higher in the ENSO years (28.1%: 135 out of 480) compared to the non-ENSO years (16.9%: 122 out of 720). The latter one is nearly identical to that for the internal variability (18.3%: 219 out of 1200), In other months, however, there are no significant differences in the probability for the seesaw formation between the ENSO and non-ENSO years when the North Pacific anomalies persist from one month to the next. This result indicates that ENSO-related persistent anomalies over the North Pacific do not necessarily guarantee the formation of the AL–IL seesaw, except for in March. Finally, as in the results for the internal variability, it is confirmed that an anomalous signal over the North Pacific one and/or two months earlier is a necessary but is not a sufficient condition for the seesaw formation.
5. Summary and discussion
In the present study, we have investigated the evolution and potential predictability of the AL–IL seesaw in an ensemble of AGCM simulations forced by the observed history of global SST, by separating the variability into its external and internal components using an ANOVA approach. The AL–IL seesaw, namely the North Pacific influence on the North Atlantic, via the upper-tropospheric PNAA teleconnection, is unambiguously identified in both the external and internal components of the variability. The externally forced seesaw exhibits a strong seasonal dependence, which is most apparent in March. It is characterized by its slow formation as a midwinter ENSO remote influence extending into the North Atlantic, and also by the substantially smaller amplitude of the IL anomaly compared with its AL counterpart. The latter characteristic reflects the low potential predictability (i.e., predictability associated with surface forcing) over the North Atlantic relative to the North Pacific, indicative of the dominance of the internal variability in the model-simulated AL–IL seesaw. In the internal variability, the seasonality in the seesaw formation is less distinct than that in the external variability. Modest but significant negative correlation (about −0.3) is present between the AL and IL intensities throughout winter. The internally generated seesaw bears striking resemblance to its observed counterpart, that is, comparable magnitudes between the AL and IL anomalies and a relatively short formation time scale. This result suggests that the AL–IL seesaw is most likely a mode of internal variability in the Northern Hemisphere wintertime circulation.
Low potential predictability found over the North Atlantic sector implies difficulty in predicting the formation of the AL–IL seesaw, although a precursory signal appears over the North Pacific in either phase of the AL–IL seesaw formation. Through the analyses of the internal variability, we find that the development of a PNA-like circulation anomaly pattern over the eastern North Pacific is a necessary but not a sufficient condition for the seesaw formation one month later. Thus, the AL–IL seesaw prediction may be difficult if based only on the AL behavior. It is noteworthy that the precursory pattern of the AL–IL seesaw (e.g., Fig. 6) is similar to a midlatitude circum-global wave pattern studied by Branstator (2002). The AL–IL seesaw formation might be interpreted as a regional amplification of the particular pattern identified by Branstator (2002). We will pursue this issue in a future work.
A possible clue to the AL–IL seesaw prediction may be persistence of the midwinter AL anomalies beyond one month. A good AL prediction, including its persistence, may possibly lead to a good AL–IL seesaw prediction. In our AGCM simulations, such a situation occurs randomly as part of the internal variability. Hence, there is no distinct seasonal dependence for the occurrence of the AL–IL seesaw in the internal component. As shown in Part I, the observed seesaw has a strong seasonal dependence, and persistent circulation anomalies observed over the North Pacific are often related to remote influence of ENSO variability. However, it is suggested from the result in the total variability of the model output that persistent anomalies over the North Pacific associated with the ENSO forcing do not necessarily contribute to the formation of the AL–IL seesaw. Actually, the observed AL–IL seesaw signal has been found to be a mixture of the ENSO-related variability and ENSO-independent variability (Part I). It is suggested in the present study that it is important to monitor midwinter circulation anomalies over the North Pacific as a precursor for anomalous weather condition over the Euro-Atlantic sector. Further analysis is needed to determine what mechanisms maintain persistent anomalies that potentially act as an effective precursor of the seesaw formation.
The slowness of the remote response to the external forcing, which may cause the strong seasonal dependence of the AL–IL seesaw in the ensemble mean field (Fig. 1), is consistent with an AGCM experiment investigating the influence of ENSO teleconnections by Lau and Nath (2001). An observational fact that a significant relationship between midwinter ENSO and late winter AL–IL seesaw has been confirmed (Part I) may suggest the existence of this seesawlike slow response to ENSO. Thus, the observed seasonal dependence of the AL–IL seesaw may be partly related to the slowness of the seesaw formation by the external forcing. However, we cannot show clear answers in the present study as to why the model remote ENSO response into the North Atlantic has such a long time scale.
Throughout our analysis, we have neglected interaction between the external and internal components of variability. No significant differences can be found through a chi-square test in the probability of the AL–IL seesaw formation in the internal component of variability between the ENSO and non-ENSO winters (not shown). Still, we cannot totally exclude a possibility that the behavior of internally generated anomalies might be modulated in the presence of persistent anomalies as the ENSO response, nor another possibility that the propagation of the remote ENSO response into the Atlantic might be modified by the internally generated anomalies. Nonlinearity as described above may be manifested as a sort of “aliasing” of the ENSO-related seesaw signal into the “internal component” in our simple framework, where the latter is based on a residual left after the removal of the ensemble mean. The issue of the nonlinearity will be addressed in our future investigation.
The main part of this work has been carried out during a stay of the first author (MH) at the International Research Institute for Climate Prediction (IRI) under a collaboration program between IRI and the Frontier Research System for Global Change (currently, Frontier Research Center for Global Change). We thank Dr. D. Dewitt of IRI for providing ECHAM ensemble data and his technical support and data handling. Our thanks are extended to Dr. L. Bengtsson for his useful comments and encouragement. The Grid Analysis and Display System (GrADS) was used for drawing the figures.
Corresponding author address: Dr. Meiji Honda, Frontier Research Center for Global Change, Japan Agency for Marine-Earth Science and Technology, 3173-25 Showa-machi, Kanazawa-Ku, Yokohama, 236-0001 Japan. Email: firstname.lastname@example.org
The original code of the AFES was adopted from the version 5.4.02 of an AGCM jointly developed by the Center for Climate System Research, University of Tokyo and the Japan National Institute for Environmental Studies (Numaguti et al. 1997). The code has been fully modified to attain the best computational efficiency on the Earth Simulator, a massive vector-parallel supercomputer (Shingu et al. 2003).
The wave activity flux vectors used in this study are defined by Takaya and Nakamura (1997, 2001) for stationary Rossby waves propagating through zonally asymmetric time-mean westerlies. The flux is a generalized form of Plumb’s (1985) flux derived for the zonally uniform westerlies. Within the Wentzel–Kramers–Brillouin limit, the flux vectors plotted in the figures are, in theory, independent of wave phase and are parallel to the local group velocity vectors. Therefore, they are useful in illustrating qualitative aspects of the three-dimensional propagation of a stationary Rossby wave train in the zonally varying climatological-mean flow.