Impact of the El Niño type and PDO on the Winter Sub-seasonal North American Zonal Temperature Dipole via the Variability of Positive PNA Events

In recent years, the winter (from December to February, DJF) North American surface air temperature (SAT) anomaly in midlatitudes shows a “warm west/cold east” (WWCE) dipole pattern. To some extent, the winter WWCE dipole can be considered as being a result of the winter mean of sub-seasonal WWCE events. In this paper, the Pacic SST condition linked to the WWCE SAT dipole is investigated. It is found that while the sub-seasonal WWCE dipole is related to the positive Pacic North American (PNA + ) pattern, the impact of the PNA + on the WWCE dipole depends on the El Niño SST type and the phase of Pacic decadal Oscillation (PDO). For a central-Pacic (CP) type El Niño, the positive (negative) height anomaly center of PNA + is located in the western (eastern) North America to result in an intensied WWCE dipole, though the positive PDO favors the WWCE dipole. In contrast, the WWCE dipole is suppressed under an Eastern-Pacic (EP) type El Niño because the PNA + anticyclonic anomaly dominates the whole North America. Moreover, the location changing Pacic midlatitude eastward +


Introduction
In the recent decades, North American east experienced frequent severe cold extreme weathers. For example, 2013/14 and 2014/15 winters are characterized by warm and drought in California with a strong cold anomaly in the east of North America, whose temperature anomalies show a North American "warm west/cold east" (WWCE) dipole in midlatitudes (Wang et al .2014;Lee et al. 2015;Hartmann 2015;Singh et al. 2016). Such a temperature dipole pattern was expected to frequently occur in the future (Wang et al. 2017;Chien et al. 2019). Thus, the physical cause of the North American WWCE dipole has attracted a great interest of scientists and has been an important research topic in recent years (Wang et al. 2014(Wang et al. , 2015Seager et al. 2016;Lee et al. 2015;Yu and Zhang 2015;Hartmann, 2015;Yu et al. 2016Yu et al. , 17, 2018Singh et al. 2016;Xie and Zhang 2017;Schulte and Lee 2017;B. Luo et al. 2020).
Many previous studies have mentioned the role of sea surface temperature (SST) in the winter WWCE dipole of 2013/14 and 2014/15. For example, Wang et al. (2014Wang et al. ( , 2015 indicated that the presence of the WWCE dipole in the North America in 2013/14 winter might be associated with the developing phase of El Niño. Seager and Henderson (2016) found that the SST anomaly pattern in the tropical Pacific and the south Indian Ocean might have contributed to California's drought during the 2013/14 winter. Hartmann (2015) suggested that the North American cold anomaly in the 2013/14 winter was related to the North Paci c mode of the SST anomaly. Peng et al (2019) further found that the abnormal atmospheric circulation related to the extreme climate on the North America for the 2013/14 winter was also in uenced by the tropical Madden-Julian Oscillation. However, Xie and Zhang (2017) pointed out that the cold winter in 2014/2015 is likely related to internal atmospheric variability, even though Peng et al. (2018) connected it to the effect of El Niño or North Paci c mode. Furthermore, some studies have suggested that the low sea ice concentration might modify the frequency of North American WWCE dipole in the future (Lee et al. 2015). Luo (2017, 2019) found that the westward shift of Greenland blocking due to the sea ice concentration decline in the west of Greenland could lead to such a WWCE dipole. More recently, B.  noted that the WWCE dipole not only depends on the North Paci c blocking, but also on the phase of the North Atlantic Oscillation. Of course, such a midlatitude temperature dipole has been shown to be related to North Paci c Oscillation/West Paci c pattern (Baxter and Nigam 2015).
The PNA pattern is an internal mode with 10-20 days timescale, characterized by a quadrupole structure from North paci c to North America (Feldstein. 2002, Franzke et al. 2011). The PNA has been shown to have an important impact on the weather and climate over North America (Yu et al .2016;Harnik et al. 2016). Some studies have indicated that the PNA as an atmospheric internal mode is modulated by the air-sea interaction (Straus andShukla 2002, Peng 2014).  found that the North Paci c winds could in uence the period, spatial shape, movement and amplitude of the PNA events with a timescale of 10-20 days, whereas the structure and strength of midlatitude westerly winds over North Paci c are related to the SST anomaly patterns on interannual/decadal timescales. Thus, it is inferred that the PNA events can be in uenced by the Paci c SST anomaly through the background wind condition change. While some studies have noted the effect of the PNA on the air temperature anomaly over North America (Harnik et al. 2016), what type of Paci c SST condition favors or suppresses PNA + leading to a strong North American WWCE SAT dipole is not clear so far. Thus, examining under what SST condition the PNA + can lead to a North American WWCE dipole has an important implication in understanding and predicting extreme weathers over the North America, which is the main purpose of our present paper. This paper is arranged as follows. The data and method are described in Section 2. In Section 3, atmospheric circulation patterns during the 2013/14 and 2014/15 winters are presented to establish the relationship of North American SAT anomaly with the phase of PNA. The different weather regimes of PNA + events and their connection to the type of El Niño are investigated in Section 4. In this section, we also explore why the type of El Niño can cause a large change in the PNA + . In Section 5, we examine a connection of the WWCE dipole to the combined effect of the CP-type El Niño and PDO. The conclusions and discussion are given in the nal section.

Data And Method
The reanalysis data in winter (from December to February, DJF) used in this study is taken from the National Centers for Environmental Prediction-National Center for Atmospheric Research (Kalnay et al. 1996) from December 1950/February 1951to December 2018/February 2019(1950. This dataset includes the daily mean 500-hPa geopotential height (Z500) and surface air temperature (SAT) as well as the monthly horizonal (zonal and meridional components U and V) eld and vertical velocity with 2.5°×2.5° grids. For the winter SST anomaly, we used the monthly-mean sea surface temperature (SST) dataset with 1°×1° grids resolution taken from the Hadley Centre (Rayner et al. 2003(Rayner et al. ) from 1950(Rayner et al. -2018 All the anomalies at each grid were obtained by removing the seasonal cycle and linear trend. Furthermore, we used the daily PNA index provided by the NOAA Climate Prediction Center (https://www.cpc.ncep.noaa.gov/), which is obtained based on the Rotated Empirical Orthogonal Function (REOF or rotated EOF) of 500-mb height anomalies. Here, a PNA + (PNA -) event is de ned if the daily PNA index is above 1.25 (below -1.25) standard deviations (STDs) for at least three consecutive days. During the life cycle of PNA, lag 0 denotes the peak day of the PNA event. In the following discussions, we calculate the time-mean composite daily Z500 and SAT anomalies averaged from lag -5 to 5 days of individual PNA events to re ect the contribution of the mature PNA to the SAT anomaly over the North America. But the temporal variations of the composite daily SAT anomalies averaged over the east and west parts of the North America and the composite daily warm west/cold east SAT dipole of PNA events are presented for the entire PNA event lifetime (from lag -10 to 10 days) because the PNA event is of two weeks (10-20 days) (Feldstein 2002).
Here, we also used a combined regression-EOF procedure as in Kao and Yu (2009) to classify the El Niño-Southern Oscillation (ENSO) into CP and EP types. We subtracted the SST anomalies regressed with the Niño1+2 index in the eastern equatorial Paci c (10°S-0 o , 80°-90°W) from the original SST anomalies and then used the EOF analysis to obtain the CP-ENSO structure of the subtracted SST anomalies. Similarly, the EOF analysis of the SST anomalies regressed onto the Niño 4 index in the central Paci c region (5°S-5°N, 160°E°-150°W) and subtracted from the original SST anomalies is performed to obtain the EP-type ENSO structure. In this case, the principal component (PC1) time series of the rst empirical orthogonal function (EOF1) of the DJF-mean SST anomaly for a CP-type (EP-type) El Niño is de ned as the CP-type (EP-type) El Niño index. The SST anomaly averaged over the region (10°S-0° and 80°-90°W) or the region (5°S-5°N and 160°E°-150°W) is de ned as the Niño1+2 or Niño 4 index respectively. Moreover, the winter Paci c Decadal Oscillation (PDO) index was de ned as the PC1 time series of the EOF1 mode of DJFmean SST anomalies over the North Paci c (120 o E-120 o W, 20 o -70 o N) as in Ding et al. (2015).
In this paper, we used the k-means clustering method as used in Michelangeli et al., (1995), Ferranti et al., (2015) and Champagne et al., (2019) to classify the different regimes of SAT and Z500 anomalies over North America (25°N-70°N, 140°W-60°W) for PNA + events during the 1950-2018. Using such a method can help us to identify what types of PNA + patterns favor the WWCE dipole over North America. The kmeans clustering algorithm was constructed by minimizing the sum of the squares of distances between each sample and the corresponding cluster centroid based on an iterative process, in which the Euclidean distance was used. The detail of this clustering method can be found in Michelangeli et al. (1995).  (Wang et al. 2014, Lee et al. 2015Hartmann 2015;Yu and Zhang 2015;Peng et al. 2018;B. Luo et al. 2020). We rst analyzed the two winters to motivate our present study. Figures 1a-b show the DJF-mean Z500 and SAT anomalies during 2013/14 and 2014/15. It is noted that the winter-mean SAT anomaly shows an intense WWCE dipole along the zonal direction of North American mid-latitudes during 2013/14 (Fig. 1a) and 2014/15 (Fig. 1b) winters.
Clearly, the warm anomaly associated with the anticyclonic anomaly in the west part of North America are located in higher latitudes than the cold anomaly in its east part. Thus, the domain-averaged SAT anomalies over the northwest (T W : 35°N-65°N, 125°W-100°W, red box) and east (T E : 25-55°N, 100°W-65°W, blue box) parts of North America, as denoted by T W and T E respectively, can be used to characterize the zonal SAT dipole. The difference T WE =T W -T E between T W and T E is de ned as the SAT dipole index.
Such an index de ntion is slightly different from that of Singh et al. (2016). To guarantee that the T WE index is continuous in winter, we do not require that T W is positive and T E is negative in two winters. But we require T W being positive and T E being negative to select individual WWCE events.
For 2013/14 and 2014/15 winters, the daily variations of the zonal SAT dipole index T WE and the PNA index from December to February are shown in Fig.1c-f. It is seen that the T WE index shows a notable subseasonal variability. When PNA is positive, the T WE index is mostly positive. But when PNA is negative, the T WE index is negative or shows a change from negative (positive) to positive (negative) . This indicates that the SAT WWCE dipole is mainly related to PNA + . It is also found from the wavelet power spectrum analysis that the timescale of the T WE variation is about 20 days ( Figure S1 in the supplementary le), crudely consistent with the lifetime of the PNA events (Feldstein 2002). Here, we used to calculate the contribution of the PNA to the winter WWCE SAT dipole, where T WE (PNA) represents the total sum of T WE indices for T W >0 and T E <0 from lag-10 to 10 for PNA events and T WE (DJF) denotes the total sum of T WE indices in winter for the same condition. It is found that the contribution of PNA to the winter WWCE dipole is about 64.7% during 2013/14 or 51.8% during 2014/15, in which the contribution of PNA + is 50.5% during 2013/14 or 51.8% during 2014/15. Thus, the PNA + can signi cantly in uence the two winter WWCE dipoles via the generation of sub-seasonal WWCE dipoles.
It is useful to show the time-mean elds of daily Z500 and SAT anomalies averaged from lag -5 to 5 days of PNAand PNA + events in Figs. 2a-d during the two winters, where lag 0 denotes the peak day of PNA. It is found that during the 2013/14 winter PNAcorresponds to an intense cold anomaly over the whole North American mid-high latitude region (Fig. 2a). But its combination with the PNA + (Fig. 2b) can produce an intense WWCE dipole over North America (Fig.2c). While the PNA + during the 2013/14 winter does not correspond to a strong cold anomaly in the east part of North America (Fig.2b), it can correspond to an intense WWCE dipole during the 2014/15 winter (Fig.2d). This suggests that the effect of the PNA + on the North American SAT anomaly exhibits a signi cant interannual variability.
We can see from the DJF-mean SST anomaly elds in the 2013/14 and 2014/15 winters as shown in Figs. 2e-f that there is a cold SST anomaly in the tropical region (Fig.2e). Thus, the weak role of the PNA + in the WWCE dipole during the 2013/14 winter is likely related to the tropical cold SST anomaly (Fig. 2e) and a negative PDO (PDO -) phase with a value of -0.5 STDs as noted below. However, there is a central Paci c (CP) El Niño during the 2014/15 winter. Thus, it is likely that a strong WWCE dipole in the presence of a PNA + (Fig. 2d) is associated with CP El Niño (Fig. 2f) and a positive PDO (PDO + ) positive with a value of 1.7 STDs as noted below. Thus, the role of the PNA + in the North American WWCE dipole might depend on the different SST anomaly patterns. Although the PNA + is is an internal atmospheric mode and driven by synoptic-scale eddy forcing , it does not imply that the PNA is not in uenced by the Paci c SST anomalies. Moreover, because the PNAevent does not correspond to a typical WWCE SAT dipole over North America, our emphasis in the following discussion is mainly placed on examining the effect of the varying PNA + associated with Paci c SST anomalies on the North American WWCE dipole. b) Composite result of PNA + events As noted above, an individual PNA + event can produce a sub-seasonal WWCE dipole pattern with the timescale of 10-20 days. To establish the linkage of the seasonal SAT pattern to sub-seasonal SAT anomalies, we should examine whether the frequency of sub-seasonal WWCE dipoles in uence the seasonal SAT pattern. First, the WWCE dipole for each day is de ned as an instantaneous daily WWCE dipole event if the value of T WE is above 1.0 STDs. Additionally, we require a limitation that Tw>0 in the west and T E <0 in the east domains of North America must be satis ed. Figures 3a-b show the time series of the winter frequency (total number of days) of daily WWCE events and the composite SAT and Z500 anomaly elds for all days of daily WWCE events during the winters from 1950 to 2018. It is seen that the Z500 anomaly mainly shows a zonally oriented wave train structure from the North Paci c to the Atlantic (Fig. 3b). As noted below, the zonal wave train looks like the PNA + during the CP-type El Niño winter. The composite DJF-mean SAT and Z500 anomaly during winters with high WWCE days shows a winter WWCE pattern (Fig. 3c). In contrast, the winter WWCE dipole cannot be seen during winters with low WWCE days (Fig. 3d). Thus, the winter WWCE pattern is mainly related to the frequency of the subseasonal WWCE dipole. In Figs. 3c-d, the winter with high (low) WWCE days is de ned as the normalized time series of the winter frequency of daily WWCE events being above 1.0 STDs (below -1.0 STDs).
It is found that there are 835 days of daily WWCE dipole events during 1950-2018. However, there are 549 days of daily WWCE dipole events associated with PNA + , if the daily WWCE dipole event is de ned to be associated with PNA + when the PNA + index is above 0.5STDs. Thus, nearly 66% of daily WWCE dipole events are related to PNA +. . We also note that nearly 77% of daily WWCE dipole events are linked to PNA + if the PNA + index is above 0.25 STDs.
To indicate whether the PNA + leads to a typical WWCE dipole over North America, we show the time series of the event number of individual PNA + events during 1950-2018 in Fig.4a. Statistical calculation shows that there are 79 PNA + events during 1950-2018. Figure 4b shows the corresponding time-mean elds of composite daily Z500 and SAT anomalies averaged from lag -5 to 5 days for the PNA + events. It is seen that there is a warm anomaly in the northwest side of North America for the PNA + (Fig. 4b) followed by an anticyclonic anomaly there. Correspondingly, a weak cold anomaly can be seen over the southeast side of North America. The variation of the composite daily T WE index from lag-10 to 10 days of PNA events is shown in Fig. 4c. It is found that the dailyT WE index is positive during the period from lag -4 to 6 days, indicating that the PNA + can lead to a relatively weak WWCE SAT dipole over North America. This also suggests that these PNA + events may include some cases related to a more or less typical subseasonal WWCE dipoles. As we will nd below, the PNA + event can lead to a more typical WWCE dipole under the CP-type El Niño SST anomaly condition than the EP-type El Niño SST condition.

North American Wwce Dipole And Its Link To The Different Regime Patterns Of The Positive Pna And The Paci c Sst Anomaly a) Preferred regime patterns of the positive PNA associated with the WWCE dipole
To examine under what SST anomaly condition the PNA + events lead to a strong North American WWCE dipole, it is useful to use a k-means clustering method as widely used in previous studies (e. g., Michelangeli et al., 1995;Champagne et al. 2019) to classify time-mean SAT anomalies averaged from lag-5 to 5 days over the North America (25°N-70°N 140°W-60°W) associated with the PNA + events. For 79 PNA + events during the 1950-2018 winters, the time-mean SAT anomalies associated with the PNA + events can be classi ed into six clustering regimes: C1, C2, C3, C4, C5 and C6. By compositing daily Z500 and SAT anomalies associated with the six clustering regimes, one can nd the preferred regime of the PNA + associated with the typical WWCE dipole. Figure 5 shows time-mean elds of composite daily Z500 and SAT anomalies averaged from lag -5 to 5 days of the PNA + events related to six clustering regimes (C1, C2, C3, C4, C5 and C6). It is found that C3 with a zonal wave train structure with an anticyclonic (cyclonic) anomaly in the west (east) part of the North America corresponds to a typical WWCE dipole (Fig. 5c) which resembles the large-scale circulation pattern associated with the WWCE dipole (Fig.3b), whereas C6 corresponds to a PNA + concurrent with a negative Arctic Oscillation (AO -) and a strong cold anomaly in the eastern North America (Fig. 5f). C4 corresponds to a weak WWCE dipole (Fig. 5d). Although C1 (Fig. 5a), C2 (Fig. 5b) and C5 (Fig. 5e) are the PNA + patterns, they do not produce typical WWCE dipoles over North American midlatitudes (Fig. 5e). However, C2 can have a cold anomaly over the whole North America midlatitudes (Fig. 5b). The above results lead us to infer that some of PNA + events can produce a typical WWCE dipole over North America, but some cannot. Through a comparison with Figs. 2a-d, we can nd that C3 (Fig.5c) looks like the PNA + event in the 2014/15 winter (Fig. 2d) in that they have a pattern correlation coe cient of 0.73 in the region (150°E-30°W, 20°-80°N). In the six clustering regimes, the C4 regime is most frequent (Fig. 5d) and resembles the composite pattern of all the PNA + events (Fig. 4b) because its pattern correlation coe cient with Fig. 4b is nearly 0.95. Figure 6 shows the time series of the event number of PNA + events in winter for each one of the six clustering regimes during 1950-2018. It is interesting to see that C3, C4 and C5 regimes show notable interannual and decadal variability (Figs. 6c, d, e), whereas the C6 (C1) regime takes place mainly during the 1950-1981 (1994-2018) winters (Figs. 6a, f). We can also see from Fig. 6 that the PNA + event in the 2014/15 (2013/14) winter belongs to the C3 (C4) type. Thus, the variability of the PNA + events can produce the different spatial pattern of the North American SAT anomaly.
To further quantify which of the six clustering regimes is responsible for the North American WWCE dipole, the temporal variation of T WE during the life cycle of the PNA + (from lag-10 to 10 days) is shown in Figure 7 for the six clustering regimes of the PNA + . It is found that in the six clustering regimes, the WWCE dipole is strongest during the PNA + life cycle for C3 (Fig. 7c). Thus, C3 is an optimal PNA + pattern that promotes a typical WWCE dipole. Although C4 and C6 favor North American WWCE dipoles (Figs. 7d, f), C6 mainly appears before 1981 (Fig. 6f). Furthermore, we show the correlation coe cients between the WWCE dipole index (Fig. 3b) and the event number of each regime in the six clustering regimes (Fig. 6) in Table 1. It is noted that C3 (C4) shows a signi cant positive correlation of 0.34 (0.39) (p<0.05) with the WWCE dipole index. Thus, the North American WWCE dipole is associated with the C3-and C4-type PNA + events. Below, we will examine the SST anomalies associated with the six clustering regimes to understand what types of SST anomalies cause the PNA + events to have different North Atlantic SAT anomaly patterns. b) Paci c SST anomaly patterns linked to different clustering regimes While the PNA + is an internal mode, its variability depends on the type of ENSO as an external forcing (Straus and Shukla 2002;Yu et al. 2012a, b). In other words, the PNA + event in strength and location may become different as the ENSO-type SST anomalies are different. To establish the linkage of the six clustering regimes with different Paci c SST anomaly patterns, we show the composite DJF-mean SST anomalies of the six clustering regimes in Figure 8. It is seen that C3 (C5) corresponds to a typical CPtype (EP-type) El Niño in addition to C3 having a PDO + -like SST signal (Figs. 8c, e), whereas C4 corresponds to a rather weak CP-type El Niño (Fig. 8d). This indicates that the C4-type PNA + is in uenced by a weak CP-type El Niño. Although C6 corresponds to a CP-type El Niño (Fig. 8f), the El Niño signal is relatively weak compared to that of C3 (Fig. 8c). Our calculation of the correlation coe cients (Table 2) with the CP and EP El Niño indices (Fig. 9 below) shows that C3 (C4) has modest signi cant positive correlations of 0.22 and 0.29 (0.22 and 0.22) (p<0.1 for 0.22 and P<0.05 for 0.29) with the CP El Niño and PDO indices. But C1 (C5) not associated with the WWCE dipole has a signi cant positive correlation of 0.29 (0.37) (p<0.05) with the EP El Niño index. Thus, the WWCE dipole associated with C3 is related to a CP-type El Niño (Fig. 8c), whereas C5 without a WWCE dipole is linked to a strong EP-type El Niño (Fig.  8e). This result leads us to infer that the different types of El Niño might be an important factor in uencing the WWCE dipole through the change of the PNA + .
To examine the possible role of CP-and EP-type El Niños in the WWCE dipole, we show the winter SST EOF1 anomaly for the CP-type (EP-type) El Niño based on the combined EOF-regression method in Fig.9a  (Fig. 9b) and the corresponding PC1 time series in Figs. 9c-d. Here, a CP-(EP-) type El Niño winter is de ned if the CP (EP) El Niño index is above 0.5 STDs. It is easy to see that there are 12 CP-type El Niño winters (0.3/year) during 1950-1989, but 11 winters (0.38/year) during 1990-2018. Thus, the CP-type El Niño has slightly increased since 1990, consistent with the previous results (Kao and Yu 2009;Kug et al. 2009;Lee and McPhaden 2010;Yu et al. 2012a). Of course, the different spatial pattern of the North American SAT anomaly in the CP-or EP-type El Niño winter might be associated with the wave trains in response to tropical convection anomalies (Peng et al. 2018), the presence of PNA + does not necessarily require the appearance of tropical convection (Franzke et al. 2011).
Moreover, it is useful to show the time-mean composite daily Z500 and SAT anomalies averaged from lag -5 to 5 days associated with PNA + events in the CP-and EP-type El Niño winters in Figs.9e-f. It is noted that in the CP-type El Niño winter the PNA + events become a zonal midlatitude wave train like C3, which can produce an intense WWCE dipole over North America (Fig. 9e). But such a WWCE dipole is hardly seen for the EP-type El Niño winter (Fig.9f) because the anticyclonic anomaly of the PNA + almost occupies the whole North America and resembles the C5 regime. Thus, it is suggested that the PNA + events in the CP-type and EP-type El Niño winters played different roles in the North American WWCE dipole because of their spatial patterns being different. c). Physical mechanism of the CP-and EP-type El Niño SSTs in uencing the PNA + Here, we further examine why the CP-and EP-type El Niño SSTs have different in uences on the PNA + events. Because the variation of PNA events depends on the change of the North Paci c midlatitude westerly wind eld , it is not di cult to infer that the type of El Niño can in uence the PNA + events probably through changing midlatitude westerly wind due to the change of the winter meridional circulation such as Hadley cell. In fact, because the PNA + is sub-seasonal timescales (10-20 days), the interannual Hadley cell and midlatitude westerly winds associated with the type of El Niño can be considered as the background conditions in uencing the sub-seasonal PNA + . In this paper, the climatological plus the interannual circulation associated with the type of El Niño is considered as the background condition of sub-seasonal PNA + even though CP-and EP-type El Niño SSTs may require the same climatological condition. Such a consideration is reasonable because CP-and EP-type El Niño SSTs can correspond to different interannual Hadley cells. Below, we further calculate the winter Hadley cell and midlatitude westerly winds related to CP-and EP-type El Niño SSTs. According to Wang (2004) and Feng and Li (2013), the winter Hadley cell can be well characterized by the divergent component of the winter wind and vertical velocity.
We show the height-latitude pro les of the DJF-mean anomalous Hadley cell averaged over 150°E -120°W and the horizontal elds of DJF-mean U500 anomalies in Fig. 10 for the CP-and EP-type El Niño winters. It is found that the sinking movement of the Hadley cell is relative weak in the subtropical Paci c (10 o -30 o N) for the CP-type El Niño (Fig. 10a), but strong for the EP-type El Niño (Fig.10b). Their difference is signi cant (Fig. 10c). Clearly, on interannual timescales the Hadley cell is signi cantly modulated by the type of El Niño, though it is in uenced by enhanced convection in the central-east tropical Paci c associated with enhanced upward motions and enhanced poleward divergent wind anomalies in the subtropical Paci c (Fig.S2), as noted in Feng et al. (2017). Moreover, while the PNA + is also in uenced by high-frequency convectively tropical disturbances during the CP-and EP-type El Niño winters ), some studies have indicated that the PNA + events are driven by synoptic-scale eddies in the Paci c storm track (Feldstein 2002;Franzke et al. 2011). As noted above, because the type of El Niño can in uence the interannual Hadley cell and midlatitude westerly winds, the PNA + is inevitably changed by the background wind condition ) associated with the type of El Niño, no matter how the PNA + is generated.
To examine the linkage between the type of El Niño and the spatial distribution of winter midlatitude westerly winds and why the EP-type El Niño is favorable for the North American WWCE dipole, we show the composite winter 500-hPa zonal wind anomalies in Figs. 10d-f for the CP-and EP-type El Niño winters. It is seen that the winter midlatitude westerly winds are stronger and extend more east for an EPtype El Niño (Fig.10e) than for a CP-type El Niño (Fig.10d). Because the descending branch of the strong Hadley cell mainly appears in the latitudes 10 o -30 o N for the EP-type El Niño and favors an intensi ed subtropical high (not shown), the midlatitude westerly winds are enhanced in the north of 30 o N according to the geostrophic adjutsment. As seen from Fig. 10f, the winter midlatitude westerly winds are stronger in the latitudes 30 o -50 o N for the EP-type El Niño than for the CP-type El Niño. For this reason, the anticyclonic anomaly of the PNA + shifts more east so that it occupies the larger region of the North America for an EP-type El Niño (Fig. 10f) than for a CP-type El Niño (Fig. 10e). Thus, in the EP-type El Niño winter the eastward shift of the PNA + is related to enhanced Hadley Cell, which is also consistent with the modeling result of Yu et al. (2012b). These results explain why the PNA + events associated with the EP-type El Niño cannot produce the North American WWCE dipole. The physical mechanism of the type of El Niño in uencing the North American WWCE dipole can be simply described by the pathway: the type of El Niñodifferent interannual Hadley cellsdifferent interannual Paci c midlatitude zonal windschanges in sub-seasonal PNA + in spatial shape and locationchanges in the North American WWCE dipoles.

Combined Effect Of The Enso And Pdo Types On The Pna And North American Air Temperatures
While the CP-type El Niño favors the North American WWCE dipole via changing the zonal location of the PNA + and its spatial shape, the PDO as a decadal signal of the North Paci c SST likely modulates the effect of the CP-type El Niño. Here, we further explore whether the phase of PDO modulates the North American WWCE dipole associated with the CP-type El Niño. A CP-type El Niño and PDO + (PDO -) combination is de ned if the CP index has a value ≥0.5 STDs and the PDO index (Figure11 a) is positive above (negative below) zero. To examine the impact of the CP-type El Niño combined with the different phase of PDO on the North American WWCE dipole related to the PNA + , we rst show the DJF-mean SST anomalies in Figs. 11b-d for the CP-type El Niño and PDO + combination and the CP-type El Niño and PDOcombination as well as their difference. It is found that the phase of PDO can modulate the intensity and region of the positive Paci c SST anomaly associated with the CP-type El Niño (Figs. 11b-c). Thus, it likely in uences the PNA + and associated North American SAT anomaly.
Figures 11e-g show the time-mean composite daily Z500 and SAT anomalies averaged from lag -5 to 5 days of PNA + events for the two combinations and their difference. It is seen that the PNA + events can produce a typical North American WWCE dipole (Fig. 11e) if the Paci c SST anomaly corresponds to the combination of a CP-type El Niño and a PDO + (Fig. 11b). But the North American WWCE dipole is relatively weak (Fig. 11f) under the Paci c SST anomaly with the combination of a CP-type El Niño and a PDO - (Fig. 11c). Their difference can be seen from the daily variations of T W , T E and T WE =T W -T E (Fig.   11h-j), even though the strong warm anomaly are located in the Northwest North America relative to the cold anomaly over the east part of North America. Thus, the PDO + (PDO -) tends to enhance (suppress) the role of the CP-type El Niño in the North American interannual WWCE dipole by strengthening (suppressing) cold and warm anomalies over the east and west parts of North America.

Conclusion And Discussions
A notable warm west/cold east (WWCE) zonal temperature dipole occurred over North America during 2013/14 and 2014/15 winters, which is characterized by an anticyclonic anomaly over the northwest part of North America and a negative height anomaly over the east part of North America. It is found that the PNA events with a timescale of 10-20 days contribute to the sub-seasonal WWCE SAT dipole in the two winters. By inspecting the composite results of PNA events in 2013/14 and 2014/15 winters, it is found that although the positive PNA (PNA + ) events occurred in the two winters, the effect of the PNA + is different between the two winters. Such a difference is likely related to the different SST anomalies conditions. The most notable difference between the 2013/14 and 2014/15 winters is that the two winters correspond to different SST anomalies over Paci c. Thus, it is speculated that the PNA + could be signi cantly in uenced by the winter SST anomalies in Paci c, even though it is an internal sub-seasonal mode. This motivates us to examine under what oceanic condition the PNA + events can lead to a typical WWCE dipole.
In order to establish the link of the PNA + events with the oceanic condition, it is useful to classify the North American SAT anomaly associated with the PNA + events using the K-means clustering method. By compositing daily Z500 anomalies associated with the six clustering regimes, one can know what types of PNA + correspond to what types of North American SAT patterns and Paci c SST anomalies. It is noted that C3 corresponds to a typical North American WWCE dipole with a strong cold (warm) anomaly over the east (northwest) part of North America. While C4 corresponds to a WWCE dipole, the cold anomaly over the eastern North America is relatively weak. It is also found that while C3 and C4 have a signi cant correlation with the CP-type El Niño and PDO + . The PNA + events associated with C5, which does not produce a WWCE dipole, is strongly related to the EP-type El Niño. This suggests that the type of the El Niño can in uence the PNA + events and associated sub-seasonal North American SAT anomaly to result in a notable interannual variability of the winter WWCE dipole as a winter-averaged result of sub-seasonal WWCE dipoles.
In addition, we further examine why the CP-type or EP-type El Niño can in uence the PNA + events by calculating the interannual winter-mean zonal winds and Hadley cell associated with the type of El Niño.
It is revealed that the type of El Niño can in uence the position of the anticyclonic anomaly of the PNA + through changing interannual Hadley cell and associated midlatitude westerly winds over North Paci c.
For the EP-type El Niño, the interannual zonal winds over North Paci c midlatitudes are intensi ed due to intensi ed Hadley cell, which cause the eastward shift of the sub-seasonal PNA + event and make its anticyclonic anomaly appear in the whole North America. Such a PNA + event does not produce a North American WWCE dipole. The reversed is seen for a CP-type El Niño. Under a CP-type El Niño condition the interannual Hadley cell is weakened, which leads to reduced interannual midlatitude zonal winds over North Paci c favoring the westward shift of the sub-seasonal PNA + and the appearance of the anticyclonic (cyclonic) anomaly of the sub-seasonal PNA + in the west (east) part of North America. Such a sub-seasonal PNA + shift can generate an intense North American sub-seasonal WWCE dipole. Because the presence of a strong winter WWCE dipole is mainly related to intense or frequent sub-seasonal WWCE dipoles, the type of El Niño can modulate the winter WWCE dipole over North America through changes in the sub-seasonal WWCE dipoles. As a result, the winter North American WWCE dipole can show a notable interannual variability. Furthermore, it is shown that while the CP-(EP-) type El Niño favors (suppresses) the North American sub-seasonal WWCE dipole associated with the PNA + events, the phase of PDO as a decadal signal can modulate the role of CP-type El Niño in the variability of the North American WWCE dipole. In particular, the negative phase of PDO tends to suppress the WWCE dipole of the PNA + event under the CP-type El Niño.
However, it must be pointed out that in this paper we do not examine the role of other circulation in the WWCE SAT dipole, such as the North Atlantic Oscillation and North Paci c Oscillation/West Paci c pattern. Also, the role of tropical convection activity associated El Niño and its combined effect with other processes were not discussed. These problems deserve a further study. Tables   Table 1. Correlation coe cients between the WWCE index (Fig 3b) and the six clustering regimes: C1, C2, C3, C4, C5, C6 (Fig. 6). The two asteriks represents the coe cient with a 95% con dence level.    In panels a-d, the region of the SAT anomaly (color shading) deviated from the DJF-mean eld during 1950-2018 winters with the 95% con dence level based on a two-sided Student's t-test is only plotted. composite daily Z500 (contours, contour interval is 20gpm) and SAT (color shading and unit: K) anomaly elds for all days of North American WWCE dipole events during 1950-2018 winters. The composite DJF mean Z500 (contours, contour interval is 10gpm) and SAT anomaly for the winters of with (c) high WWCE days and (d) low WWCE days. The winter with high (low) WWCE days is de ned that the normalized time series of the winter frequency of daily WWCE events is above (low) 1.0STDs. The region of the SAT anomaly (color shading) with the 95% con dence level based on a two-sided Student's t-test is plotted. composite daily Z500 (contours, contour interval is 20gpm) and SAT (color shading in the unit of K) anomalies averaged from lag-5 to 5 days of PNA+ events, where the region of the SAT anomaly (color shading) with the 95% con dence level based on a two-sided Student's t-test is plotted. (c) Temporal variations of composite daily TWE index during the PNA life cycle (from lag-10 to 10 days).

Figure 5
Time-mean elds of composite daily Z500 (contours, contour interval is 20gpm) and SAT (color shading in the unit of K) anomalies averaged from lag-5 to 5 days of PNA+ events related to six clustering regimes: (a) C1 (13 events), (b) C2 (15 events), (c) C3 (9 events), (d) C4 (19 events), (e) C5 (16 events) and (f) C6 (7 events), where the region of the SAT anomaly (color shading) with the 95% con dence level based on a two-sided Student's t-test is plotted.   Fig. 5. The dot represents the SST anomaly region above the 95% con dence level based on a two-sided Student's t-test. winters whose indices exceed 0.5 STDs, and the region of the SAT anomaly (color shading) with the 95% con dence level based on a two-sided Student's t-test is plotted. difference. The dot represents the region being the 95% con dence level based on a two-sided Student's ttest. Niño combination and (d, g, j) their difference. The CP-type El Niño and PDO+ combination is de ned as being a winter with the CP index above 0.5 STDs (Fig. 9c) and PDO index above 0. Similarly, the CP-type El Niño and PDO-winter is de ned as being a winter with the CP index above 0.5 STDs (Fig. 9c) and PDO index below 0. The gray shading denotes the difference between the two curves being signi cant above the 90% con dence level based on 5000 Monte Carlo simulations.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.