1 Introduction

Many studies have documented the interdecadal climate fluctuations in both the atmosphere and ocean in the North Pacific region during the winter half of the year in the late 1970s (e.g.,Trenberth, 1990;Trenberth and Hurrell, 1994). These interdecadal climate fluctuations are characterized by significant changes in atmospheric circulation that resulted in a more intense and eastward-shifted Aleutian low-pressure system, advecting warmer and moister air along the west coast of North America and into Alaska and directing colder air over the North Pacific, after approximately 1976. Consequently, surface air temperatures and sea surface temperatures (SSTs) increased along the west coast of North America and Alaska, whereas SSTs decreased over the North Pacific (Trenberth, 1990;Trenberth and Hurrell, 1994). Simultaneously, changes emerged in the form of southward-shifted Pacific storm tracks and associated synoptic eddy activity (Lau, 1988;Rogers and Raphael, 1992;Trenberth and Hurrell, 1994), shifts in the surface ocean sensible and latent heat fluxes (Cayan, 1992), and a weakening of intraseasonal fluctuations such as blocking activity (Nakamura, 1996). Also, accompanying these changes were distinct weather/climate anomalies in the regions over the North Pacific and North America. Cold outbreaks occurred more frequently in the eastern United States (Downton and Miller, 1993), the Northwest Pacific experienced less winter precipitation (Chen et al., 1996), the temperatures in Alaska and western Canada tended to be much warmer (Trenberth and Hurrell, 1994), and the amount of sea ice in the Bering Sea appeared to decrease (Manak and Mysak, 1987). Moreover, the epipelagic ecosystems in the North Pacific were also affected (Venrick et al., 1987;Mantua et al., 1997;Hare and Mantua, 2000).

This climatic regime shift was evident not only in the North Pacific and its surrounding areas but also in global atmospheric circulation patterns. The Walker circulation exhibited weaker action centers, which were located farther to the west in the western Pacific (Garcia and Kayano, 2008); the global and Northern Hemispheric mean surface temperatures increased abruptly (Nitta and Yamada, 1989;Trenberth, 1990;Lau and Weng, 1999); and obvious changes in precipitation and temperature were observed over China (Yu and Lin, 1997;Li et al., 2010).

In terms of much longer timescales, it is suggested that the climatic regime shift of the late 1970s is not unique and that analogous climatic transitions also occurred in the early 1920s, the mid 1940s, and the late 1990s (Kondo, 1988;Mantua et al., 1997;Minobe, 1997;Mantua and Hare, 2002;Deser et al., 2004). This recurring pattern of interdecadal climate variability is referred to as the Pacific Decadal Oscillation (PDO) and is charac-terized by a pan-Pacific phenomenon that also includes interdecadal climate variability in the tropical Pacific that is similar to the ENSO pattern (Mantua et al., 1997). In addition to the large influence of the PDO on regional climate variations over the Pacific and its surrounding areas (Latif and Barnett, 1994;Mantua et al., 1997;Zhang et al., 1997;Bond and Harrison, 2000;Frauenfeld and Davis, 2002;Zhou et al., 2007;Lu et al., 2013), the PDO also exerts obvious modulating effects on the interannual variability of other climate signals as an interdecadal background, among which the modulations exerted by the PDO on the interannual variability of the ENSO cycle have drawn considerable attention. Corresponding to different PDO phases, there are distinct differences in both the temporal and spatial features of the interannual ENSO cycle and the interannual relationship between ENSO and the global climate (Gershunov, 1998;Gershunov and Barnett, 1998;Power et al., 1999;Zhu and Yang, 2003;Yang et al., 2004;Pavia et al., 2006;Wang et al., 2008;Hu and Huang, 2009;Chen et al., 2014;McAfee, 2014;Wang et al., 2014;Jin et al., 2016). For example, the influence of ENSO on climate/weather variations and regional predictability in North America is clearly regulated by the PDO (Gershunov, 1998;Hu and Huang, 2009), and the effect of ENSO on dry–wet changes also varies with the PDO's phase (Wang et al., 2014).

Atmospheric mass (AM) is a good indicator for describing atmospheric circulation (Lorenz, 1951;Christy and Trenberth, 1985;Christy et al., 1989). Fluctuations in AM are induced not only by dynamic atmospheric processes such as vertical circulations and Rossby waves, but also by thermal forcings due to the thermal contrasts between different types of surfaces. The striking thermal differences between oceans and continents lead to the redistribution of AM over these surfaces in seasonal cycles (Van Den Dool and Saha, 1993;Lu et al., 2008;Hu et al., 2014). Because the thermal states of liquid water oceans, continents, and sea ice display considerable discrepancies and vary from year to year, AM is driven to migrate between sea and land surfaces and between sea and ice-covered surfaces, in part because of the thermal contrasts between them (Guan et al., 2015). In the yearly variations of AM over liquid water oceans, significant anomalous AM accumulations (losses) occur over the North Pacific; of these AM anomalies, those over the midlatitudes and northeastern regions are compensated by anomalies over continents at mid and high latitudes and sea-ice-covered Arctic regions, respectively. It is the North Pacific that provides the connections for AM exchanges among these regions. Therefore, a climatic shift involving the atmosphere and ocean in the North Pacific will inevitably influence the AM variations over liquid water oceans, continents, and sea ice, further affecting their relationships with climate anomalies.

Therefore, as the boreal winter climate shifted remarkably around 1976 and the PDO is one of the strongest signal in the Northern Hemisphere, we investigate in the present paper the interdecadal differences in relationships of the interannual AM variations over the Eurasian continent and North Pacific in association with the PDO between the epochs before and after 1976.

The paper is organized as follows. In Section 2, the datasets and methods used in the study are briefly introduced. In Section 3, investigations are carried out on the differences in the interannual AM variations over the three types of surfaces in the two epochs. In Section 4, the possible roles of the PDO in modulating the interannual AM variations are explored, and the relationships between AM variations and climate anomalies in the different epochs are explored in Section 5. Finally, conclusions and further discussion are presented in Section 6.

2 Data and methodology

The data used in the present study originate from the ERA-40 dataset (Uppala et al., 2005) for the period 1958–2002, including monthly mean surface pressure, geopotential height, horizontal wind, air temperature, specific humidity, surface air temperature, and sea-ice concentration. All variables are defined on a 2.5° × 2.5° lon./lat. grid mesh. All physical quantities are scattered over 23 levels from 1000 to 1 hPa, with the exception of surface variables. The surface pressure data for a longer period of time (1871–2012) are from the 20th century reanalysis V2 dataset (Compo et al., 2011).

The precipitation data are from the GPCC (Global Precipitation Climatology Centre) dataset (Schneider et al., 2014), at a horizontal resolution of 1° × 1° lat./lon. and available at ftp://ftp.dwd.de/pub/data/gpcc/.

Observational data, including surface air temperature and rainfall, from 160 meteorological stations in China, are also employed. These observational data, covering the period from January 1951 to January 2016, are available from the National Climate Center of China.

The PDO index is defined as the first empirical orthogonal function (EOF) of the SST in the Pacific Basin north of 20°N, and the PDO index data are available at http://jisao.washington.edu/pdo/PDO.latest. Calculations are performed after subtracting the global mean SST from the Pacific Ocean SST to eliminate the impact of “global warming” (Zhang et al., 1997).

The period considered for all the data listed above spans from September 1957 to August 2002. The winter of a particular year is defined as the period from the December of the previous year to the February of that year (DJF); hence, the period of 1959–2002 corresponds to 45 winters in total. The winter means are averaged over the DJF period.

The surface air pressure (p_s; SAP) is a suitable parameter for describing the variations in AM over the surface of earth, and is therefore used to denote the AM in the present work. As the momentum, mass, and energy are exchanged across the equator, there are interactions between the Southern Hemisphere (SH) and Northern Hemisphere (NH) (Guan and Yamagata, 2001;Graham et al., 2011;Wang et al., 2015). Generally, the global AM is approximately conserved, and the AM is able to oscillate between the NH and SH (Trenberth, 1984;Christy and Trenberth, 1985;Trenberth and Christy, 1985;Guan and Yamagata, 2001;Carrera and Gyakum, 2003;Lu et al., 2008;Guan et al., 2010).Guan and Yamagata (2001) identified the interhemispheric oscillation (IHO) and defined the IHO index (I_IHO) as the difference between the quantities of AM over the NH and SH, as follows:

$ \;\;\;\;\;\;\;\;\;\;\;\;\;{I\!_{\rm IHO}} = {M\!_{\rm NH}} - {M\!_{\rm SH}}, $\hspace{10000pt}

(1)

$ \;\;\;\;\;\;\;\;\;\;\;\;\;{M\!_{\rm NH}} = \frac{1}{{2\pi }}\iint_{\rm NH} {{p_{\rm sa}}\cos \varphi {\rm d}\varphi {\rm d}\lambda }, $\hspace{10000pt}

(2)

$ \;\;\;\;\;\;\;\;\;\;\;\;\;{M\!_{\rm SH}} = \frac{1}{{2\pi }}\iint_{\rm SH} {{p_{\rm sa}}\cos \varphi {\rm d}\varphi {\rm d}\lambda }, $\hspace{10000pt}

(3)

where ${\,p_{\rm sa}} = {p_{\rm s}} - {\bar p\,\!_{\rm s}}$ and ${\,\bar p_{\rm s}}$ is the mean climatology of p_s. To further focus on the AM variations (p'_s) in the NH, the influence of the IHO needs to be removed from the SAP anomaly p_sa (SAPA) to obtain p'_s, as follows:

$ {p'_{\rm s}} = {p_{\rm sa}} - \alpha {I_{\rm IHO}}, $

(4)

whereα is the regression coefficient for p_sa regressed onto I_IHO. It is known that there are long-term trends in the IHO (Guan and Yamagata, 2001), which is possibly induced by global warming. Thus, the trends in p'_s are removed when the IHO is regressed out from p'_s. Obviously, variations of AM are independent of the IHO and are able to redistribute throughout the NH. Following Guan et al. (2015), the area-averaged AM anomalies over liquid water oceans, lands, and ice-covered regions (M_O,M_L, and M_I, respectively) that are calculated based on p'_s can be expressed as below:

$ {M\!_{\rm{L}}} = \iint_{{\rm{Land}}} {{{p'}_{\rm s}}\cos \phi {\rm d}\phi {\rm d}\lambda } /\iint_{{\rm{Land}}} {\cos \phi {\rm d}\phi {\rm d}\lambda }, $\hspace{10000pt}

(5)

$ {M\!_{\rm{O}}} = \iint_{{\rm{Ocean}}} {{{p'}_{\rm s}}\cos \phi {\rm d}\phi {\rm d}\lambda } /\iint_{{\rm{Ocean}}} {\cos \phi {\rm d}\phi {\rm d}\lambda }, $\hspace{10000pt}

(6)

$ {M\!_{\rm{I}}} = \iint_{{\rm{Seaice}}} {{{p'}_{\rm s}}\cos \phi {\rm d}\phi {\rm d}\lambda } /\iint_{{\rm{Seaice}}} {\cos \phi {\rm d}\phi {\rm d}\lambda } . $\hspace{10000pt}

(7)

In this study, several statistical methods, including linear regression, correlation and composite analysis, are employed to investigate the variations of M_O and M_L in two different epochs, and the differences between them. The index M_I, defined in Eq. (7), is provided for reference. To examine the interdecadal changes, we obtain the interdecadal components of the PDO index and SAPA by calculating the 9-yr running mean. The interannual components of the time series are obtained by subtracting the interdecadal components from the original departure va-lues.

The interdecadal climatic shifts that occurred in the 20th century can be detected in many variables. The North Pacific index (Trenberth and Hurrel, 1994), regional climate records over the North Pacific and adjacent continents, and the leading principal component (PC) time series of monthly mean SST anomalies (SSTAs) for the winter–spring season during the period 1899–2002, all exhibit interdecadal fluctuations (as seen in Fig. 4 of Deser et al. (2004)); the global mean surface temperature increased rapidly during the late 1970s (IPCC, 2013), and the southern oscillation index also underwent an abrupt change in the late 1970s (Li et al., 2010). Based on this 1976/77 regime shift, the analysis period is divided into two epochs: 1959–1976 (epoch-I) and 1977–2002 (epoch-II). To eliminate the influence of interdecadal signals, the calculations and analyses are performed separately for these two epochs.

To examine the relationship between the interannual variability of a variable f' with the interdecadal variations of a physical quantity, the time series of the N_w -yr moving root-mean-squared (RMS) value of this variable f' is defined, which is written as

$ {\rm RMS}(t_i) = {[\frac{1}{{{N_w}}}\sum\limits_{j = - m}^m {{{f'}^2}} ({t_{i + j}})]^{{\textstyle{1 \over 2}}}},\,i = \overline {5,N - 4} , $

(8)

where t_i means the ith sampling in the annual time series. The length of the moving window N_w equals 9 in this study, while m = 4.N is the length of time series. Obviously, RMS(t_i ) is a time-series with N−8 data.

3 Eurasia–Pacific anti-phase relation pattern changes before and after 1976

To examine the interdecadal difference in the relations between M_O and M_L, we need to check both interdecadal changes in the correlations between variations of M_O and M_L, and in the spatial patterns related to M_O and M_L between the periods before and after 1976.

3.1 Relation of M_L with M_O

Before and after the 1976/77 regime shift, the AM fluctuates interannually over liquid water oceans, lands, and ice-covered regions, exhibiting two anti-phase relations: one between M_O and M_L and the other between M_O and M_I (Fig. 1). For the 1958–1976 period (epoch-I;Figs. 1a and 1c), significant negative correlations are found between M_O and M_L and between M_O and M_I, with values of –0.69 and –0.81, respectively (above the 99% confidence level). These relations among M_O,M_L, and M_I can still be observed in the 1977–2002 data (epoch-II;Figs. 1b and 1d), albeit the correlation coefficients are a little bigger or smaller than in epoch-I. These results suggest that, in both epoch-I and epoch-II, when anomalous AM accumulates over liquid water oceans, it is compensated by oppositely signed anomalies over lands and sea ice. Note that the anomalous AM variations over lands are positively correlated with those over the sea-ice covered Arctic region, with a correlation coefficient of 0.51 in the period before 1976, and 0.39 after 1976. These correlations verify the results of Guan et al. (2015), derived from the NCEP–NCAR reanalysis data for the period 1979–2012.

Amazingly, although the relations among the norma-lized time series of M_O,M_L, and M_I remain unchanged for epoch-I and epoch-II, there are considerable differences in the spatial patterns between these two different epochs, especially for M_L. Therefore, in the following, we focus on the M_L-related pattern change before and after 1976.

To examine the spatial patterns of AM variations during boreal winter for the two different epochs, composite analyses are presented for some strongly anomalous years during epoch-I and epoch-II based on M_L. From Fig. 1, years when the absolute values of M_L are larger than one standard deviation (σ) are selected for the composite analyses, as shown in Table 1.

3.2 Interdecadal change in the Eurasia–Pacific anti-phase relation pattern

Guan et al. (2015) discussed the M_L-related variations of AM in the NH in wintertime using the NCEP–NCAR reanalysis data from 1979 to 2012, demonstrating a teleconnection pattern in the anomalous SAP field. This pattern was referred to as the Eurasia–Pacific anti-phase relation (EPAR) pattern, and is characterized by a positive anomalous SAP center over Eurasia along with a nega-tive anomalous SAP center over the North Pacific when M_L > 0. This pattern looks zonal, which is quite different from the Arctic Oscillation (AO) pattern (Guan et al., 2015).Figure 2 shows this EPAR pattern by using ERA-40 reanalysis data.

In epoch-I (1958–1976), the spatial pattern related to M_L reveals an anti-phase relation between AM ano-malies over small continental regions, including Eurasia and North America, and those over regions in the Mediterranean, Northwest Pacific, and North Atlantic, as well as an in-phase relation between AM variations over lands and the Arctic (Fig. 2a). That is to say, when M_L > 0, ano-malous increases in AM are observed over lands at mid and high latitudes and in the Arctic, with small areas of significance over the southern part of the West Siberian Plain, the northern part of the Canadian Plain, and large circumpolar areas in the Arctic. Simultaneously, the AM decreases over oceans at midlatitudes, with areas of significance over the Mediterranean and relatively small areas over the Northwest Pacific and North Atlantic. This M_L-related pattern in Fig. 2a looks very similar to the well-known AO pattern (Thompson and Wallace, 1998).

However, in epoch-II, the M_L-related SAPA exhibits distinctly the anti-phase relation between AM anomalies over Eurasia and the North Pacific, with little contribution from the close in-phase relation between land areas and the Arctic ice cover (Fig. 2b). This pattern looks just as the EPAR pattern as explored in Guan et al. (2015). Compared with epoch-I, when M_L > 0, the areas domi-nated by significant accumulations of AM in Eurasia extend considerably, spanning a range from eastern Europe to eastern Siberia and moving southward over the Caspian Sea to arrive at the Iranian Plateau; the significant AM deficits over the Northwest Pacific also extend considerably to occupy nearly the entire sea surface of the North Pacific in the extratropics, with a center located to the south of the Aleutian Islands. Additionally, small areas of significant positive and negative AM anomalies are found over western North America and the Bermuda Triangle, respectively, which may be related to the EPAR. The Arctic ice cover is still controlled by anomalous AM accumulation, but to a less extent as compared to that before 1976 (Fig. 2a).

The spatial differences in the M_L-related SAPA between the two epochs can be clearly observed from Fig. 2c. Positive differences of AM anomalies are found over large areas of the Eurasian continent and the southwestern part of North America, with two centers located over the East European Plain and the western region of North America, respectively. Similar differences are also observed over the North Pacific at mid and low latitudes and over the North Atlantic south of 60°N. On the contrary, negative differences are distributed over the North Pacific north of 40°N, the Arctic sea-ice and adjacent continents. This means that the relation of AM variations between the Arctic and North Pacific is weakened in epoch-II. Notably, the interdecadal differences in the spatial patterns of the M_L-related SAPA are characterized by a north–south oscillation of the AM over the North Atlantic at mid and high latitudes, similar to the distribution of the North Atlantic Oscillation (NAO). However, because this NAO-like distribution of the SAPA is observed only prior to the shift, the differences of the SAPA as shown in Fig. 2c in the Arctic–North Atlantic sector indicate that the NAO/AO component in association with M_L variations weaken during epoch-II. That is to say, the influences of NAO on SAPAs over the Eurasian continent weaken after 1976.

The above aspects can be further explored through correlation maps. Most strikingly, the SAPA distributions related to M_L undergo a pattern shift from an AO-like pattern before 1976 to an EPAR pattern in the period after 1976 (Fig. 3). The negative correlations of the SAPA between Eurasia and the North Pacific are much stronger in epoch-II than in epoch-I. These results indicate that not only are the disturbances in SAP intensified in the Eurasia–North Pacific domain during epoch-II, but also the connections of the SAPA between Eurasia and North Pacific are strengthened in the period 1977–2002. The patterns as displayed in Fig. 3 are also cross-checked by using the NCEP-1 reanalysis data. It is found that the correlation patterns of M_L from NCEP-1 during the two epochs before/after 1976 look mostly similar to the patterns as derived from ERA-40 (Fig. 3), except for the Bering Sea region where some significant correlations appear during 1958–1976.

To further check the pattern changes in the SAPA, we performed the EOF analysis for the SAPA over the Eurasia–Pacific domain. The leading EOF mode (EOF1) explains 26.44% of the total variance for the period 1958–1976 and 39.82% for 1977–2002. This EOF1 for epoch-II is well separated from the second leading mode (EOF2), but not for epoch-I, according to the criteria proposed by North et al. (1982). The correlation between the time series of coefficients of EOF1 and M_L is 0.28 for the period 1958–1976, whereas it is 0.79 for 1977–2002, indicating that variations of M_L are closely related to the principal mode of EOF1 in the later epoch. The spatial distribution of EOF1, as seen in Fig. 4b, looks similar to the EPAR pattern, whereas it does not in Fig. 4a.

4 Possible role of the PDO in modulating the EPAR pattern

The interdecadal change in the EPAR pattern from epoch-I to epoch-II is remarkable, as mentioned above. However, what causes this change is still not clear. To examine this, we consider the PDO as the modulator of the EPAR pattern before/after 1976.

4.1 General features of the PDO-related SAPA

Temporal variations of the PDO can be described by the PDO index. From Fig. 5a, the PDO index is composed of both interdecadal and interannual components. Overall, the PDO is in its negative phase (cold phase) during the period from 1950 to the late 1970s, whereas it is in its positive phase (warm phase) after the 1970s. The 9-yr running mean RMS of the interannual component of the PDO index varies roughly in opposite phase against the interdecadal component of the PDO index in period, especially before the 1980s. This suggests that interannual perturbations in SSTs north of 20°N are stronger than normal during the cold phase of the PDO, whereas they are weaker than normal during the PDO's warm phase.

The spatial pattern of SSTAs of the PDO in its warm phase is characterized by a stronger and colder SSTA center over the North Pacific and a stronger and warmer SSTA in the tropical region (Mantua et al., 1997;Mantua and Hare, 2002). Both the interannual and interdecadal components of the PDO as seen in the PDO index share the same spatial pattern (Yang et al., 2004). Correspondingly, the SAPA is characterized by a nega-tive center in the North Pacific and a positive but weak center over the Arctic during 1958–2002 (Fig. 5b).

It has been reported that the current of the Kuroshio extension underwent an interdecadal shift around 1976/77 (Taguchi et al., 2007). On interdecadal timescales, the SAPA change may be attributable to the atmosphere being heated anomalously due to the warmer SSTA near the Aleutian Islands and colder SSTA near the Kuroshio extension in the North Pacific (Fig. 5b). The disturbance of the atmosphere in association with the SSTA, as displayed in Fig. 5b, can be explained roughly by the geostrophic relations. Since the air and sea interact with each other via heat, vapor, and momentum exchanges in the atmospheric boundary layer, the air temperature anomalies (T'_as) near the sea surface are strongly influenced by the SSTA (T'_os). That is,T'_as∝T'_os holds near the sea surface. According to the thermal wind equations, the anomalous zonal wind u'_as near the underlying sea surface is negatively proportional to the meri-dional gradient of the SSTA $ - \partial {T'_{\rm os}}/\partial y$ , i.e., ${u'_{\rm as}} \propto - \partial {T'_{\rm os}}/\partial y$ , suggesting that anomalous easterly winds dominate in the region north of 40°N while ano-malous westerly winds dominate in the region south of 40°N (Appendix A).

Note that there are some discrepancies between the distribution of regression coefficients of SAPAs from the 20th century and ERA-40 reanalysis datasets onto the interdecadal component of the PDO index during the period 1958–2002. These discrepancies, found mainly over midlatitudes of Asia, are possibly due to some errors introduced into processes of data assimilation.

4.2 Possible modulations by the PDO on the interannual variability of the EPAR

The PDO possibly modulates the interannual variations of AM, but differently in different periods.

Firstly, when the PDO is in its negative phase during 1958–1976, the sea level pressure (SLP) and geopotential height at 500 hPa (H500) are anomalously higher compared with those for the period 1977–2002 (Figs. 6a–d), which indicates both the Aleutian low and East Asian trough (Huang et al., 2013) are anomalously weaker north of 40°N before 1977. Meanwhile, the zonal component of westerly flow is stronger during epoch-II than during epoch-I (Figs. 6e and 6f), suggesting the role of the westerly flow acting as the waveguide being stronger during epoch-II than during epoch-I.

Secondly, the variances of both the SLP anomaly (SLPA) and H500 are relatively smaller before 1977 than after 1977 over the North Pacific (Figs. 6a–d). These re-sults suggest that the Rossby wave-related teleconnections between Eurasia and the North Pacific may be weaker before 1977. This can be further confirmed by the correlation of the 9-yr moving RMS value of the SAPA, as derived from the 20th century reanalysis, with the interdecadal component of the PDO index (Fig. 7). The indication is that the interdecadal component of the PDO, as obtained from the PDO index by using a 9-yr running mean, may also affect the perturbation strength of the SAPA. The large and significant positive correlations are in the mid and high latitudes of both the North Pacific and Atlantic, whereas large and significant nega-tive correlations are situated over regions around the Tibetan Plateau, the eastern part of North America, and the lower latitudes of oceans. No significant correlations are found over high latitudes of continental Eurasia. These results suggest that, during the 20th century, the interdecadal component of the Pacific SSTA related to the PDO phenomenon may modulate the strength of the SAPA; the interannual variability of the SAPA is intensified in the North Pacific, while it is weakened in land regions in midlatitudes when the PDO is in its positive phase.

Thirdly, the periodicities of the SAPA over regions between Eurasia and the North Pacific are more different during 1958–1976 than during 1977–2002 (Fig. 8); the correlations of spectral variance between the regions EC-I (45°–65°N, 60°–90°E) and NP-I (25°–45°N, 165°–135°W), and between EC-II (42.5°–62.5°N, 50°–90°E) and NP-II (27.5°–47.5°N, 155°E–165°W), are found to be –0.33 and 0.36, respectively. This results in the corre-lations of p'_s between EC-I and NP-I, and between EC-II and NP-II, being –0.19 and –0.74, respectively. Note that why the periodicities of p'_s over the regions between Eurasia and North Pacific are so different during epoch-I to those during epoch-II needs to be investigated in the future.

Based on the above, it is reasonable to expect that a significant out-of-phase relation in AM variations between Eurasia and North Pacific should be observed after 1976, but not before 1977.

4.3 Influences of interannual variations of the SSTA in the North Pacific on the EPAR

The interdecadal variations of the PDO may modulate the SAPA's intensification in the North Pacific. This suggests that interannual variations of the SAPA over Eura- sia may be significantly correlated with those over the North Pacific. To clarify this possibility, we begin by exa-mining the correlations between both M_L and M_O and the interannual component of the PDO index, as listed in Table 2. From the results in Table 2, it can be seen that there are significant correlations between both M_L and M_O and the interannual variations of the SSTA north of 20°N in epoch-II. The critical values of absolute correlation coefficients at the 90%/95% confidence level, based on the t-test, are 0.37/0.43 for the period 1958–1976, and 0.32/0.37 for the period 1977–2002, respectively. Clearly, the correlations are slightly bigger in epoch-II than in epoch-I, for both M_L and M_O, which looks simi-lar to the interdecadal change in correlations of M_L with M_O over these two periods (Figs. 1a and 1b).

Interestingly, the correlation patterns of p'_s with the interannual component of the PDO index in epoch-II look considerably different to that in epoch-I in the Eurasia–North Pacific domain (Fig. 9). Note that the EPAR pattern (Guan et al., 2015) can be clearly seen in Figs. 2b and 3b. From Fig. 9b, for epoch-II, the correlations are also distributed like the EPAR pattern over the NH, with a positive correlation center over high-latitude Eurasia and a negative center over the North Pacific. However, this pattern is not clearly seen in Fig. 9a for epoch-I. This means that the interannual variations of the SSTA in the North Pacific north of 20°N, which shares the same spatial pattern of the interdecadal component of the PDO (Yang et al., 2004), have significant influences on the SAPA, leading to the EPAR pattern being stronger and clearer in epoch-II than in epoch-I.

5 Climate anomalies related to EPAR pattern change

Because of the differences in the spatial distribution of AM migrations after the climatic shift of 1976, accompanied by anomalous circulations, these migrations must exert different impacts on climate variations between the two different epochs.

5.1 Surface air temperature anomalies

The anomalous surface air temperatures regressed onto M_L for epoch-I and epoch-II are shown in Figs. 10a and 10c. When M_L > 0 during epoch-I, the temperatures are lower than normal in East Asia, high-latitude conti-nental Eurasia, and midlatitude North America, but higher than normal over high-latitude North America and east of the Mediterranean. During epoch-II, the area of lower temperatures in Asia enlarges considerably, extending eastward from the Caspian Sea to the coast of East Asia. However, the lower temperature region in Europe is no longer significant. Moreover, significantly higher than normal temperatures are found over eastern Siberia and the southwestern part of North America, whereas the temperatures in Greenland are not strongly affected. These features are further confirmed in more detail over China by analyzing the observed data from 160 China stations (Figs. 10b and 10d). It is found that anomalously lower temperatures prevail in most areas in the east of the Tibetan Plateau during epoch-I (Fig. 10b), whereas during epoch-II the anomalously lower temperature mainly appears over North China (Fig. 10d).

To examine how the temperature is influenced by the EPAR, we plot in Fig. 11 the associated circulation ano-malies by regressing the relative quantities onto M_L. From Figs. 11a and 11b, the SLPA patterns change remarkably from epoch-I to epoch-II, which is consistent with the patterns of the SAPA. Anomalous northerly winds at 850 hPa dominate in East Asia in the mid and high latitudes during both epoch-I and epoch-II (Figs. 11c and 11d), but more significantly in the latter. These anomalous northerlies facilitate the occurrence of lower than normal winter temperatures. Simultaneously, these northerly winds are favorable for the transport of dry air from high latitudes northward to East Asia.

5.2 Rainfall anomalies

The influences of AM variations over lands on rainfall anomalies are widespread in the NH. For positive M_L during epoch-I, less than normal rainfall is observed over Asia westward of 90°E and in most regions of China (Fig. 12a). Because the water vapor converges from the north of Europe to the region near the Mediterranean, more rainfall is concentrated over southern Europe, whereas less rainfall is found over the north. In addition, less rainfall occurs over the southern Atlantic coasts because of the divergence of water vapor in these regions. In the northeastern part of North America, more rainfall is found. During epoch-II (Fig. 12c), the region of reduced rainfall over Eurasia at high latitudes shifts eastward compared with that during epoch-I, and the anoma-lous rainfall over northern Europe is no longer obvious. Moreover, less rainfall is also observed over the northeastern part of China, Korean Peninsula, and Japan, but not over other regions of China. Because of the distinct water vapor divergence over the region to the north of the Gulf of Mexico, less rainfall is received there. The features over China can be further detected from Figs. 12b and 12d. Significant rainfall anomalies are found over some regions of China, showing a southwest–northeast oriented belt from the Yunnan–Guizhou Plateau to North China. However, during epoch-II, significantly less rainfall can only be observed over scattered areas in Northeast China.

6 Summary

Based on the climatic shift that occurred around 1976, two epochs are defined in this study: epoch-I (1958–1976) and epoch-II (1977–2002). On this basis, we investigate the EPAR pattern changes in association with interannual variations in AM over lands (M_L) during epoch-I and epoch-II. The possible causes of the EPAR pattern changes and the related climate anomalies are also explored for epoch-I and epoch-II. The results can be summarized as follows.

The AM fluctuates interannually over lands and liquid water oceans in both epoch-I and epoch-II, exhibiting an anti-phase relation between M_L and M_O. This relation suggests that when anomalous AM accumulations or losses occur over liquid water oceans, they are compensated by oppositely signed anomalies over lands. Although the correlation between M_L and M_O does not change much from epoch-I to epoch-II, the M_L-related SAPA pattern changes considerably before/after 1976.

A teleconnection pattern, the EPAR, proposed by Guan et al. (2015), is observed during epoch-II, whereas it is not so significant during epoch-I. This EPAR pattern reveals an intimate connection between the AM ano-malies over Eurasia at the mid and high latitudes and those over the North Pacific at the midlatitudes during the period 1977–2002.

The formation of the EPAR teleconnection pattern is closely related to at least three factors, including anoma-lous AM flows, Rossby wave activities, and forcings due to diabatic heating, as discussed in Guan et al. (2015). However, the interdecadal change in this EPAR pattern of AM during boreal winter before and after 1976 is found to possibly be attributable to the PDO.

The PDO may modulate the interannual variations of the SAPA and the EPAR pattern in the Eurasia–North Pacific domain. On the one hand, it is found that the intensities of the Aleutian low, East Asian trough, and westerly flow acting as a waveguide are significantly stronger during the warm phase of the PDO, as compared to those in the cold phase, indicating the intensifi-cation of Rossby wave–related teleconnections between Eurasia and the North Pacific. On the other hand, the interannual variations of the SSTA in the North Pacific north of 20°N, which shares the same spatial pattern of the interdecadal component of the PDO, are significantly correlated with SAPA variations in the Eurasia–North Pacific domain; a positive correlation center is found over mid- and high-latitude continental Eurasia, and a negative correlation center over the North Pacific, in epoch-II. This pattern looks much like the EPAR pattern that appears in the Eurasia–North Pacific domain in epoch-II. In epoch-I, the EPAR-like correlation pattern in association with the interannual component of the PDO index is not observed. These results suggest that the interannual variations of the SSTA related to the PDO also possibly affect the teleconnection pattern on the interannual timescale in the Eurasia–North Pacific domain during boreal winter.

The climate anomalies associated with the AM variations during epoch-II are distinct from those in epoch-I in some places. When M_L > 0 before 1976, lower than normal temperatures are observed over the northeastern part of Asia, including Northeast China, the Korean Pe-ninsula, and Japan, as well as the southwest coast of North America; higher temperatures are found over the regions around the Mediterranean and in the northeastern part of North America. Rainfall is also affected in Eurasia, including southern Europe, regions close to the Arctic Ocean, west of 90°E in Asia, and in the midlatitudes of East Asia, along with the northern part and southeast coast of North America and the southeast of Greenland. After 1976, the areas of lower temperature in Eurasia and the southwestern part of North America expand, and the air temperatures over Alaska and Greenland are also affected. The anomalous rainfall area over the Eurasian continent at the mid and high latitudes also expands, but the area of affected rainfall in the southeast of Greenland vanishes.

Note that the North Pacific Gyre Oscillation (NPGO) is another dominant mode in the North Pacific, which represents the second EOF/PC of both SSTAs and sea surface height anomalies over the region (25°–62°N, 180°–110°W). The PDO pattern emerges as the first EOF/PC. The observed strengthening of the NPGO mode since 1993 reported in many studies (Bond et al., 2003;Douglass et al., 2006;Di Lorenzo et al., 2008) may represent a response to anthropogenic forcing and global warming. It has been reported, based on analyses of climate projections by using the coupled climate model GFDL2.0 (experiments GFDL CM2.0, 20C3M and GFDL CM2.0, SRES A2) (Di Lorenzo et al., 2008), that the NPGO variance is amplified by 38%, whereas the PDO variance is reduced by 58%, in the period of 2000–2100 compared with the period 1900–2000. However, the PDO index utilized in this study is defined as the EOF1 of the SST in the Pacific basin north of 20°N, and calculations are performed after subtracting the global mean SST from the Pacific Ocean SST to eliminate the impact of “global warming” (Zhang et al., 1997). The PDO index is derived from a specific pattern, suggesting the time variations of the spatial mode. Moreover, the IHO of AM is removed. The IHO has a long-term trend and may respond to “global warming.” However, studying the variations in the hemispheric average of AM along with global warming is beyond the scope of the present study. Of course, whether there are linkages between the NPGO and hemispheric mean AM variations still needs further investigation.

Note again that interdecadal changes in the spatial patterns of the correlations of p'_s with M_O, and those with M_I, from epoch-I and epoch-II, are also observed. Correspondingly, the winter climate is also affected. However, why these interdecadal changes occur needs to be clarified in the future.

Finally, it should be noted that the results in the present study are only from reanalysis data. Numerical studies using general circulation models are needed in the future for examining the mechanisms through which the EPAR in boreal winter is affected by the SSTA variations in the North Pacific north of 20°N, on both interannual and interdecadal timescales, in association with the PDO.