Tropical–Extratropical Interactions Associated with East Asian Cold Air Outbreaks. Part I: Interannual Variability

Interannual variability of winter-mean East Asian cold air outbreaks (CAOs) and its relationship with the tropical climate system during 56 boreal winters (DJF) are investigated. The magnitude of CAO is quantiﬁed as winter-mean equatorward cold airmass (CAM) ﬂux below 280-K potential temperature across the 45 8 N latitude. EOF analysis shows that the interannual variation of East Asian CAOs is attributed mainly to the contributions from western and eastern CAOs. In particular, the western and eastern CAOs tend to be re-motely forced by La Ni ñ a and El Ni ñ o events, respectively. The western and eastern CAOs have distinct climate variability. The western CAO, which is enhanced under the climatic anomalies of high pressure over northern Eurasia and low pressure over the western North Paciﬁc, causes negative CAM anomalies over northern Eurasia and positive ones over midlatitude East Asia. In the tropical region, the western CAO negatively correlates with the eastern Paciﬁc and Indian Ocean SST, both of which enhance precipitation over the Maritime Continent. On the other hand, the eastern CAO is enhanced by the strong Aleutian low and results in positive CAM anomalies in the western North Paciﬁc and substantial negative anomalies in western North America. The eastern CAO positively correlates with the tropical SST anomalies and accordingly precipitation anomalies over the central Paciﬁc. ENSO inﬂuences western and eastern CAOs through upper and poleward Rossby wave trains excited by convective anomalies over the Maritime Continent and central Paciﬁc, respectively.


Introduction
Intermittent cold air outbreak (CAO) events during the winter season often cause severe damage to human activities (Chang et al. 1979;Chen et al. 2002;Tangang et al. 2008;Yang et al. 2009). CAO events are characterized by strong equatorward wind and a sudden drop of temperatures, which in turn significantly alters the weather in the affected region. In East Asia, a CAO event is known as the prominent feature of East Asian winter monsoon (Zhang et al. 1997;Wang and Chen 2014). Any variations within the CAO events contribute to the variation of the winter monsoon, which may be linked to climate anomalies in East Asia and even in other more remote areas. Therefore, it is of great importance to study the variability of East Asian CAO.
A quantitative definition of cold airmass (CAM) flow is necessary in order to accurately denote the variation of CAO events. A recent study has developed an analysis method for quantifying polar CAM using a threshold potential temperature u T , which represents global-mean polar CAO ). In the general circulation of mass-weighted isentropic zonal mean (MIM; Gallimore and Johnson 1981;Iwasaki 1989;Juckes et al. 1994), extratropical direct (ETD) circulation becomes visible and depicts simple meridional heat transport following wave-mean interactions. ETD circulation turns from downward in polar region to equatorward about at 458N and at 850 hPa. The equatorward branch of ETD is regarded as global-mean CAO where it is confined roughly below the 280-K isentrope (Iwasaki and Mochizuki 2012;Iwasaki et al. 2014). Using a mass integration from surface to u T 5 280 K, Iwasaki et al. (2014) identified two distinct equatorward streams in boreal winter: the East Asian (EA) stream and North American (NA) stream, as shown in Fig. 1b. The regions of these two streams are consistent with the location of major intermittent CAO events, which are largely affected by topography and land-sea distributions (e.g., Garreaud 2001). Meanwhile, no distinct equatorward streams are observed in Southern Hemisphere winter (Kanno et al. 2015b).
Hemispheric-scale CAOs are controlled by wave-mean flow interactions (Iwasaki and Mochizuki 2012), and their evolutions are revealed in Kanno et al. (2015a). In regional scale, Shoji et al. (2014) focused on the synoptic variability of East Asian CAOs. They created a CAO index defined as the anomalous equatorward CAM flux below 280 K at 458N integrated from 908E to 1808. The evolution of East Asian CAO is basically controlled by the development of both the Siberian high (SH) and stationary Aleutian low (AL), which exerts a large eastward pressure gradient force and then geostrophically induces the equatorward CAM flux. The CAO index was shown to be useful to explain CAM transport from the polar regions to the midlatitudes and subtropical East Asia and consistent with surface temperature anomalies. Furthermore, they created two more indices: western CAO and eastern CAO indices, simply defined as longitudinal integration at 908-1358 and 1358E-1808, respectively. They found that both CAOs have distinct characteristics of their synoptic evolutions (e.g., centers of SLP and northerly anomalies).
The tropical region is the primary source of global heating. Anomalous deep convection releases large quantities of latent heat, which is balanced by rising motion and then causes upper-level divergence anomalies. The divergence may lead to an important impact on the extratropical atmosphere by triggering poleward Rossby wave trains (e.g., Trenberth et al. 1998). The anomalous convections in the tropics are influenced greatly by persistent tropical sea surface temperature (SST) anomalies (e.g., Lindzen and Nigam 1987), particularly those related to El Niño-Southern Oscillation (ENSO). Previous studies have documented that ENSOinduced anomalous circulations have remote effects on the East Asian winter monsoon (Li 1990;Zhang et al. 1997;Wang et al. 2000;Wang and Chen 2014;Chen et al. 2014). The common agreement is that El Niño (La Niña) tends to bring a warmer (cooler) winter in East Asia. Considering the impact of ENSO to East Asia, it is interesting to investigate the influence of ENSO on East Asian CAO defined by a quantified equatorward CAM flux as such a study has not been carried out.
Identification of two potential independent CAO types (western and eastern CAOs) inspires us to further investigate the variability of East Asian CAOs. The current study aims to discuss their interactions with the tropical atmosphere at interannual time scale, which denotes a FIG. 1. Climatology of (a) cold airmass amount (color shaded with 25-hPa color contour interval) and its flux (vectors, hPa m s 21 ) and (b) cold airmass flux intensity (color shaded with a 150 hPa m s 21 contour interval) below u T 5 280 K during 56 winters. Brown contours represent topography with a 500-m interval. relationship between seasonal-mean CAOs and the tropical climate system. The interannual variability depicts variation of the total effects of many short-term CAO events. This paper is arranged as follows. Section 2 presents the data used in this study and the formulation of East Asian CAOs. Section 3 confirms the existence of western and eastern CAOs in the leading modes of East Asian equatorward flow. Section 4 reveals climate anomalies associated with the western and eastern CAOs. Section 5 discusses how CAOs relate with sea surface temperature (SST) and ENSO. In section 6, we perform simple numerical simulations to demonstrate dynamical linkage between East Asian CAO climate and anomalous tropical heating. Finally, a summary of the key findings and future study plans are presented in section 7.

Data and methods
We use atmospheric variables from the Japanese 55-year Reanalysis (JRA-55) dataset, which has 6-hourly time resolution, 1.258 3 1.258 horizontal resolution, and 37 vertical pressure levels (Kobayashi et al. 2015). JRA-55 covers more than half a century from January 1958 to the present. The monthly precipitation data are the Global Precipitation Climatology Project (GPCP) dataset provided by NOAA/OAR/ESRL PSD, which has 2.58 3 2.58 resolution and spans from 1979 to the present (Adler et al. 2003). We also use the Met Office Hadley Centre Sea Ice and Sea Surface Temperature (HadISST1; Rayner et al. 2003) as the oceanic dataset, which has 18 3 18 resolution and extends from 1870 to the present. We also use several climate indices, such as the SH (Gong et al. 2001) andAL (Overland et al. 1999), and several teleconnection patterns, such as the Eurasian (EU; Wallace and Gutzler 1981), western Pacific (WP; Wallace and Gutzler 1981;Linkin and Nigam 2008), and Pacific-North American (PNA; Wallace and Gutzler 1981) patterns. The indices are derived from reanalysis. The Arctic Oscillation (AO; Thompson and Wallace 1998) index is obtained from the NOAA/NCEP/ CPC website (http://www.cpc.ncep.noaa.gov/).
We average the data from December to February (DJF) to obtain the winter mean. The analysis period covers 56 boreal winters from 1958 to 2013, except that analysis involving precipitation only covers 35 winters from 1979. Despite the difference in data period, we confirm that the results are consistent for both periods. The winter of 1958 refers to the DJF 1958/59. The analysis uses EOF, linear regression, and correlation techniques. The confidence levels for correlation coefficient are calculated based on the two-sided Student's t test. In section 6, we perform numerical simulation using a simple linear atmospheric model. Model description and configuration are shown in the corresponding section.
The definition of East Asian CAO is similar to the index proposed by Shoji et al. (2014), which includes cold airmass equations formulated by Iwasaki et al. (2014). We begin with the definition of polar cold airmass amount DP below a threshold potential temperature u T : where p s and p(u T ) indicate the ground surface pressure and the pressure on a potential temperature surface, respectively. Then, horizontal flux of cold airmass F is given by where v is the horizontal wind field vector. Figure 1 shows the climatology of DP and F in Northern Hemisphere winter. Shoji et al. (2014) defined East Asian CAO index (CAOI) based on longitudinal integration of equatorward flux from 908E to 1808 at 458N: where F 2y 5 Ð ps p(u T ) 2y dp denotes the southward component of F with u T 5 280 K. The symbols f, l a, and g denote latitude, longitude, Earth radius, and gravitational acceleration, respectively. The western CAO index (W-CAOI) and eastern CAO index (E-CAOI) are simply defined as follows: and respectively. For correlation and regression analysis, each index is first normalized by its standard deviation. Although this study deals with winter-mean CAO index, its variation is consistent with interannual variation of frequency of independent CAO events (see appendix). This association is valid because the equatorward cold airmass flux is quantified in the identification of CAO events.

Two leading modes of East Asian cold air outbreaks
First, we investigate whether the western CAO (W-CAO) and eastern CAO (E-CAO) appear dominantly in the prominent modes of East Asian equatorward flow. Figure 2b exhibits three leading EOF patterns of wintermean equatorward flux F 2y crossing 458N from 908E to 1808. The remarkable features are the quasi-monopole structures of EOF1 and EOF2 that are located in the eastern and western side of the climatological EA stream, respectively. The intersection point between these two modes is located around 1358E. The contribution of EOF1 and EOF2 is quite balanced. Their explained variances are 31.1% and 28.8%, while EOF3 has only 15.6%.
Low-level climate anomalies associated with EOF1 and EOF2 are shown in Fig. 3. In extratropical regions, EOF1 is characterized by negative pressure anomalies over the North Pacific (Fig. 3a). Northerly anomalies present over the western North Pacific to the coast of eastern Eurasia and sharply turn eastward north of 308N. This pattern clearly indicates that EOF1 is controlled largely by enhanced Aleutian low pressure system. On the other hand, EOF2 shows a positive pressure anomaly over northern Eurasia, which is associated with developing Siberian high. It couples with the negative pressure anomaly centered at 1508E (Fig. 3b), which is located rather to the west of the climatological mean of Aleutian low (not shown), such that the eastward pressure gradient forcing leads to anomalous strong equatorward flux east of Tibetan Plateau. In the tropical region, both EOF1 and EOF2 show significant signals, but they are quite different from each other. The structures of streamfunction and wind field anomalies resemble low-level circulation response to anomalous tropical convection as in Gill (1980). It indicates that there may be strong connection between CAO variation and tropical convection. Detailed explanations are presented in the following section.
Time series components of EOF1 and EOF2 are depicted in Figs. 4a and 4b, respectively. The predefined E-CAO index and W-CAO index are also shown. The variation of EOF1 is remarkably consistent with the E-CAO index, as well as EOF2 with W-CAO index. Correlation coefficients r between EOF1 and E-CAOI and EOF2 and W-CAOI are very high (0.95 and 0.97, respectively). These suggest that both E-CAO and W-CAO indices are appropriate to represent the variability of the first two modes of East Asian CAO. For simplicity and consistency with the previous study of Shoji et al. (2014), the following analysis uses the W-CAO and E-CAO indices defined in Eqs. (4) and (5) instead of the first two EOF modes.

Climate anomalies associated with CAO indices
a. W-CAO in northern Eurasia resembles a large anticyclonic circulation anomaly. The flux is dammed up by the Tibetan Plateau, yielding a significant northerly and northwesterly flux anomaly in East Asia. This equatorward flow results in a meridional contrast of CAM amount across the East Asian region. Decreasing CAM extends along the Ural Mountains and eastern Siberia to the north of the Eurasian continent, while increasing CAM exists from the south of Siberia to southern China and Japan. The negative CAM anomaly over northern Siberian seems weaker compared to positive CAM anomalies exerted in midlatitude East Asia, indicating a great diversity in the origin of cold air mass in the CAO events.
The signatures of W-CAO on sea level pressure (SLP), surface air temperature (SAT), 500-hPa geopotential height (Z500), and precipitation field are shown in Fig. 6. Correlation coefficients between W-CAOI and selected climate indices are shown in Table 1. The strengthened SH is evident and consistent with anomalous CAM flux (Figs. 6a and 5a). The correlation coefficient between the interannual variability of the W-CAOI and the SH index is 0.79, far exceeding the 99.9% confidence level. Strong W-CAO winter shows a significant drop of SAT from midlatitude to subtropical East Asia and the western North Pacific (Fig. 6b). The negative anomaly of the 500-hPa East Asian trough is evident, and some extratropical teleconnections are also clearly observed (Fig. 6c). The W-CAO exhibits a wave train-like height anomaly stretching from Europe to East Asia, resembling the positive phase of the EU teleconnection pattern (Wallace and Gutzler 1981). The meridional dipole-like anomaly between eastern Siberia and the North Pacific quite resembles the negative phase of the WP teleconnection pattern (Linkin and Nigam 2008). The correlation coefficient between W-CAOI and the EU (WP) index is 0.75 (20.45), exceeding the 99.9% confidence level. The AO (Thompson and Wallace 1998), as one of the strongest sources of variability in the North Pole region, also has a significant relationship with W-CAOI (r 5 20.47). The correspondence of W-CAO features with EU, WP, and AO signals indicates the contributions of extratropical variabilities to W-CAO.
In the tropical region, a negative geopotential height anomaly spreads throughout the equator (Fig. 6c) and nearly overlaps with the negative SLP anomaly field (Fig. 6a). Dry anomalies in the precipitation field are visible over the central Pacific and Indian Ocean, while a large-scale wet anomaly is pronounced over the Maritime Continent (Fig. 6d). In the viewpoint of global mass-weighted meridional circulation (Iwasaki and Mochizuki 2012), the ETD circulation seems to enhance as a result of the contribution of equatorward flow in East Asia (Fig. 7a). Around the equatorial region, however, the W-CAO is remarkably associated with a weakening of northern global Hadley cell. Observational and diagnostic analyses have shown that ENSO largely affects the hemispherically symmetric variability of the Hadley cell, which tends to intensify in El Niño events and weaken in La Niña events (Fig. 7a;Seager et al. 2003Seager et al. , 2005. The zonal variation of precipitation anomalies in Fig. 6d indicates the enhancement of two Walker circulations in the Indo-Pacific region, which is also robust in the 200-hPa velocity potential anomaly field (Fig. 7b). The anomaly of divergent wind field at the upper level shows an enhanced local Hadley circulation over the western Pacific but a weakening one over the central Pacific. These signatures over the tropics quite resemble the major atmospheric response to a cool phase of ENSO event (i.e., La Niña). Over the Pacific basin, ENSO plays major roles in the variability of the local Hadley cell over the western and central Pacific (Fig. 7b;Wang 2004). The significant signals over the tropics suggest that ENSO may play a prominent role in W-CAO variation (Figs. 6d, 7a, and 7b).
b. E-CAO Figure 5b shows the anomaly of CAM amount and CAM flux associated with E-CAO winter. In contrast to W-CAO, the positive and negative CAM anomalies of E-CAO tend to distribute zonally rather than meridionally. The increasing CAM is observed mainly over the western North Pacific and coincided with the northerly anomalies. On the other hand, an enormous decrease of CAM is seen over the eastern North Pacific and western North America, overlapped with southerly flux anomalies.  Figure 8a shows an anomalous cyclonic circulation over the North Pacific corresponding to the strengthening of the AL. The correlation coefficient between E-CAOI and AL index is 20.75, far above the 99.9% confidence level (Table 1). Yet the correlation with the Siberian high index is very small. The distribution of temperature anomalies (Fig. 8b) is quite consistent with the CAM anomalies shown in Fig. 5b. Negative temperature anomalies over East Asia are less prominent compared to those observed in W-CAO. However, high positive temperature anomalies prevail particularly over western North America owing to a substantial loss of CAM. The CAM loss partly arises from the supply of CAM for outbreak events in the E-CAO region and an increase of diabatic heating due to warm airmass advection from the ocean (Fig. 5b). Figure 8c shows geopotential height anomalies in the midtroposphere. A northeastward wave train-like pattern extending from the subtropical central Pacific to North America resembles the positive phase of the PNA teleconnection pattern (Wallace and Gutzler 1981). The correlation coefficient of E-CAOI with PNA index indeed shows a significant relationship between E-CAO and PNA (r 5 0.77; Table 1).
Although W-CAO has a greater impact on East Asian temperature compared to E-CAO (Figs. 6d and 8d), the impact of E-CAO on precipitation should not be negligible, particularly over the Japan area (Fig. 9). The mountainous region in the center of the Japanese Archipelago causes distinct precipitation responses to the W-CAO between the Sea of Japan side (northern coast) and the Pacific side (eastern and southern coast) (Fig. 9a). In the Sea of Japan side, the northwesterly flow is blocked by the mountain, which consequently induces precipitation and causes a snowy winter around the region. Eastern and southern Japan experience less precipitation under this condition. On the other hand, E-CAO causes less precipitation over northeastern Japan but more precipitation over eastern and southern Japan (Fig. 9b). This finding highlights the importance of E-CAO on the increasing precipitation over the Pacific side of Japan. It is worth noting that, over a short time scale, cold air mass coming from the Pacific Ocean (resembling E-CAO-like events) can be a driving factor  for extreme snowfall events in eastern Japan (Yamazaki et al. 2015).
In the tropical region, positive anomalies associated with E-CAO are evident in SLP and geopotential height fields (Figs. 8a and 8c, respectively). Tropical precipitation is enhanced over the central Pacific and reduced over the Maritime Continent (Fig. 8d), which in turn weakens the Walker circulation (Fig. 7d). Large upper-level divergence over the central Pacific strengthens the overlaying local Hadley cell (Fig. 7d), which contributes to the intensification of the global Hadley circulation (Fig. 7c). These results confirm that the E-CAO is also closely associated with the tropical climate variability, particularly an El Niño-like event (Seager et al. 2003;Wang 2004).  Figure 10a shows the SST anomaly associated with W-CAO winter. Negative SST anomalies are present over East Asian coasts, indicating oceanic response to cold air outbreaks. The anomalies span mainly from the East China Sea around 308N to the equatorial region, which represents the route of W-CAO. In the tropical central and eastern Pacific, we notice a tongue of negative SST anomalies that resembles a La Niña-like pattern. This indicates the interannual variability of W-CAO is consistent with typical ENSO influence to the East Asian winter monsoon (e.g., Zhang et al. 1997), which suggests that the winter monsoon tends to be stronger (weaker) during La Niña (El Niño) events. A negative correlation is also observed between W-CAO and the tropical Indian Ocean (TIO) SST. To quantify the relationship between W-CAOI and tropical SSTs, we use SST indices as areal averages over the Niño-3 region (58S-58N, 1508-908W) and TIO region (108S-108N, 408-908E), respectively. The correlation coefficient between W-CAOI and Niño-3 (TIO index) is 20.28 (20.35) (Table 2), satisfying the 95% confidence level.

Relationship with SST and ENSO
The SST anomaly associated with E-CAO winter is shown in Fig. 10b. North Pacific SST exhibits a cooling structure, which is affected by equatorward cold air advection associated with the enhanced Aleutian low. In western North America, poleward advection of warm air from warm subtropical ocean yields positive SST anomalies along the coast of North America. In the tropical region, positive SST anomalies are particularly visible over the central and eastern Pacific, showing a resemblance to an El Niño-like pattern. The warm SST anomalies also appear in the Indian Ocean. The E-CAOI is positively correlated with Niño-3 (TIO index) with a correlation coefficient value of 0.35 (0.31) ( Table 1), significant at the 95% confidence level.
To clarify the evolution of ENSO-induced tropical SST anomalies, we calculate month-lagged correlation between CAO indices and SST indices. The monthly SST data are first smoothed by the 3-month running mean, and the trend is removed before calculation. Figure 11a shows that W-CAO winter is clearly preceded by significant eastern Pacific SST cooling that appears in autumn, peaks in early winter, and persists until spring. Generally, similar features are observed for E-CAO winter, but it is accompanied by eastern Pacific SST warming instead of cooling (Fig. 11b). Note that TIO SST evolutions follow the evolution of eastern Pacific SST with a lag time of several months. These features confirm the ENSO-like SST evolution. Indian Ocean SST responds to ENSO events possibly via an ''atmospheric bridge'' (Lau and Nath 1996;Klein et al. 1999). In the W-CAO pattern, the lag time between Niño-3 SST and TIO SST evolution is approximately 1-2 months. On the other hand, the lag time in the E-CAO pattern is longer (2-4 months), showing a close resemblance to typical ENSO-induced Indian Ocean SST signal as noted by Lau and Nath (1996) (3-6 months). In addition, the simultaneous correlation coefficient between the W-CAOI and TIO index is somewhat stronger than between the W-CAOI and Niño-3, unlike the pattern for E-CAO where the correlation coefficient between E-CAOI and TIO index is lower than that of between the E-CAOI and Niño-3 index. These may indicate that the Indian Ocean SST plays an important role in enhancing the ENSO impact on the W-CAO. Watanabe and Jin (2003) demonstrated an atmospheric response to the El Niño event using a moist linearized atmospheric model with and without the Indian Ocean SST anomaly. The El Niño-induced SST warming over the Indian Ocean was found to be important in amplifying the suppression of convection over the Maritime Continent through changes in the Walker circulation associated with the anomalous diabatic heating over the Indian Ocean (Watanabe and Jin 2003). In the lower troposphere, they showed that the suppression of convection over the Maritime Continent induces a subtropical Philippine Sea anticyclone (PSA), which is suggested by Wang et al. (2000) as a mechanism for why the East Asian winter monsoon tends to be warmer or weaker during El Niño. However, it seems that the PSA is not strong enough to suppress the equatorward CAM flow in the midlatitude. We suggest that the ENSO impact on the W-CAO is mainly delivered through uppertropospheric response, where the Rossby wave trains are excited by anomalous convection over the Maritime Continent. In short, ENSO-induced TIO SST anomalies contribute to the variability of Maritime Continent convection and its upper-level convergence/divergence, which favor the development of Rossby wave trains in East Asia and consequently influence W-CAO variability. The mechanism is further discussed in the following paragraphs. In the case of E-CAO, however, the appearance of Indian Ocean SST signal may be solely associated with ENSO and unnecessarily connected to E-CAO variation. ENSO is known to have global impact through atmospheric circulation anomalies, which are mainly triggered by SST-induced tropical convection anomalies (e.g., Trenberth et al. 1998). To see the direct impact of tropical convection, we build a winter precipitation index for two main locations averaged over the Maritime Continent (PRE-MC index; 108S-158N, 1008-1408E) and central Pacific (PRE-CP index; 58S-58N, 1708E-1408W). Both the PRE-MC and PRE-CP indices are strongly correlated with the Niño-3 SST index. The correlation coefficients are 20.821 and 0.904, respectively (Table 2). Therefore, the positive values of the PRE-MC and PRE-CP indices resemble La Niña and El Niño events, respectively. Correlation coefficients between PRE-MC and W-CAO and PRE-CP and E-CAO indices are quite high compared to between the Niño-3 and CAO indices (0.524 and 0.521, respectively), satisfying the 99.5% confidence level (Table 2). It implies that the ENSO impacts to western and eastern CAO are better explained by forcing of convections over the Maritime Continent and central Pacific. Figure 12 shows Northern Hemispheric circulation anomalies correlated and regressed with the PRE-MC index. A negative anomaly of upper-level geopotential height prevails over the tropics, and a negative phase of PNA-like pattern appears from the central Pacific to North America, indicating a La Niña-like event (Fig. 12a). In East Asia, subtropical positive height anomalies exist as an upper-level Rossby response to the convection over the Maritime Continent. Sardeshmukh and Hoskins (1988) confirmed that, in their vorticity equations, the upper-tropospheric divergence situated in the equatorial easterly wind could lead to a Rossby wave source in the subtropical westerlies. It excites subtropical anticyclonic anomaly and thus potentially delivers Rossby wave trains to the extratropics, causing change in the East Asian jet stream and development of negative geopotential height over midlatitude East Asia (i.e., East Asian trough). The anomalous cyclonic circulation triggers extratropical low-level tropospheric flow barotropically, as seen in the 280-K cold airmass flow field (Fig. 12b). It shows significant positive CAM anomalies and equatorward flow over inland East Asia. These conditions are favorable for developing more W-CAO events. Negative cold air mass and anomalous poleward flow are observed in the middle of the North Pacific, indicating fewer E-CAO events (Fig. 12b), although the amplitude is not as significant as W-CAO.  Over northern Eurasia, the EU teleconnection pattern is significantly correlated with PRE-MC index (Fig. 12a). We suggest that this signal appears not as a result of the impact of convection over the Maritime Continent but rather mainly as a result of the impact of synoptic-scale W-CAO events (associated with midlatitude waves across the EU pattern) to the tropical convection over some part of the Maritime Continent (i.e., South China Sea and Philippines). Previous studies have noted that CAO events can lead to more convection in the tropics (e.g., Compo et al. 1999). Further investigation using lagged analysis of daily data reveals that W-CAO events can both affect and be affected by tropical convection, indicating the impact and precursor of CAO. More detailed discussion about this topic will be presented in our next work. In short, the relationships in Fig. 12a may contain two-way interactions between the W-CAO and PRE-MC index. To elucidate the impact from PRE-MC, we perform simple model experiments using prescribed thermal forcing, which are presented in section 6.
On the other hand, Fig. 13 reveals circulation anomalies associated with the PRE-CP index. The PNA-like pattern is more robust compared to the pattern depicted by PRE-MC index and can be traced from the central Pacific to Atlantic (Fig. 13a). The associated wave train enhances negative geopotential height in the North Pacific and then strengthens the Aleutian low, which in turn induces positive cold airmass and equatorward flow anomalies around the E-CAO region (Fig. 13b). In East Asia, the weakening of W-CAO is observed but less significant compared to the strengthening of E-CAO. The above results imply the relative importance of the Maritime Continent and central Pacific convection to the W-CAO and E-CAO events.  (108S-158N, 1008-1408E). Black contours denote regression coefficients of (a) 200-hPa geopotential height with a 10-gpm interval and (b) 280-K CAM with a 5-hPa interval. Light and dark shading, exceed the 90% and 99% confidence levels of the variables' correlation coefficients, respectively. The vector field in (b) denotes regression coefficients with the 280-K CAM flux.

Linear model experiments
Remote response of East Asian equatorward flow to the tropical convection is examined using simple linear model experiments. A linearized steady-state baroclinic model enables us to calculate the steady direct atmospheric response from a prescribed thermal forcing in a given three-dimensional basic flow. The model is expected to provide clear dynamical evidence of how ENSO-like convection anomalies affect equatorward flow of W-CAO and E-CAO. Using this kind of model, Hoskins and Karoly (1981) first demonstrated the PNA pattern triggered by heating associated with El Niño events.
We use a dry linear baroclinic model (LBM) as in Watanabe and Kimoto (2000). It has a spectral T21 horizontal resolution with 20 sigma levels in the vertical. LBM includes three dissipation terms: biharmonic horizontal diffusion = 4 with 6-h e-folding decay time, weak vertical diffusion with damping time scale of 1000 days to suppress noise arising from finite difference, and linear drag resembling Rayleigh friction and Newtonian damping. The linear drag term is set to 1 day at the three lowest levels (model sigma levels .0.9) and the two uppermost levels (model sigma levels ,0.03) and 15 or 30 days at other levels. The basic state is zonally varying climatology from 1958 to 2013 winter. In this study, three types of experiments are carried out. The first two experiments demonstrate the atmospheric response to a monopole heating centered over the Maritime Continent and the central Pacific, respectively. The third experiment uses thermal forcing with a dipole structure that mimics the El Niño pattern: cooling over the Maritime Continent and heating over the central Pacific. An idealized Gaussian horizontal distribution is used for each anomalous heating center, while its vertical distribution follows a sinusoidal pattern with a maximum at model sigma level 0.36 (4 K day 21 ) and zero heating at the uppermost and lowest level. The spatial heating distribution in each experiment is denoted by gray shading in Figs. 14-16. Heating over the central Pacific is set to be wider in longitude compared to heating over the Maritime Continent. The time-integration approach is used to solve the linear response in LBM (e.g., Jin and Hoskins 1995). The model is run for 30 days, yet only results at day 15 are shown here because the extratropical responses tend to be steady in about two weeks after integration. Figure 14 exhibits the simulated atmospheric response to the heating over the Maritime Continent (first experiment). The negative geopotential height over the midlatitude East Asian coast is evidently enhanced by the poleward Rossby wave trains associated with heating over the tropics (Fig. 14a). Over the tropical region, the low-level cyclonic and anticyclonic patterns appear, respectively, in the northwest and northeast off to the center of heating (Fig. 14b), indicating vorticity response to the low-level convergence as shown in the simple model of Gill (1980). A clear barotropic response can be seen in the extratropical cold airmass field, as well as a robust development of equatorward flow over the W-CAO region (Fig. 14c). This equatorward flow transports more CAM to most of the East Asian countries. In the high-latitude region north of 458N, a significant decrease of cold air mass is also evident and consistent with the observation. No signature of an EU pattern is found over northern Eurasia. It clarifies that the significant EU pattern in the regression map (Fig. 12a) appears as a result of the impact of the EU pattern on the tropical convection.
The second experiment using heating over the central Pacific is quite common and has been investigated numerous times after Hoskins and Karoly (1981). Figure 15a shows the PNA-like pattern emanating from the central Pacific, which is generally consistent with the past studies. Development of low-level equatorward flow over the northwest Pacific is evident following the Aleutian low response to the barotropic Rossby wave trains, facilitating more E-CAO events (Fig. 15c).
Nevertheless, the center of the negative height anomaly over the North Pacific (Fig. 15a)  the dry LBM (Watanabe and Jin 2003). A suppression of convection (or cooling) over the Maritime Continent should be included to demonstrate a more realistic El Niño event. Figure 16 exhibits the atmospheric response by including both the central Pacific heating and the Maritime Continent cooling. Along the East Asian coast to North America, the upper height response (Fig. 16a) is quite consistent with the regression maps in Figs. 12a and 13a. A seesaw pattern between the equatorward flow of the W-CAO and E-CAO is evident (Fig. 16c), confirming that El Niño events tend to suppress W-CAO and enhance E-CAO. In contrast, a La Niña event (represented as central Pacific cooling and Maritime Continent heating) yields a reverse response in both of the East Asian CAOs compared with El Niño.

Concluding remarks
The winter-mean East Asian CAOs exhibit two major modes, which remarkably contribute to interannual variability of East Asian equatorward flow. The first and second modes clearly depict the eastern CAO (E-CAO) and western CAO (W-CAO) types, respectively. The W-CAO index, defined as equatorward flux at 458N from 908 to 1358E, and the E-CAO index, defined as equatorward flux at 458N from 1358E to 1808, have very robust correlations with the time series components of the EOF2 and EOF1 indices (r exceeding 0.95), respectively. This indicates that W-CAO and E-CAO closely represent the prominent modes of East Asian equatorward flow.
Simultaneous correlation and regression using W-CAOI and E-CAOI are used to reveal their characteristics on global climate patterns. In the extratropics, W-CAO supplies an abundant amount of CAM to continental East Asia and consequently causes cold winter over the region. On the other hand, E-CAO releases CAM mostly over the western North Pacific and gives less impact on East Asian land temperature. The strong E-CAO is also associated with robust warm winter over northwestern North America. Differences in the signature of W-CAO and E-CAO arise from different driving forces. The W-CAO and E-CAO are closely associated with Siberian high and Aleutian low variability, respectively. These imply that using W-CAOI and E-CAOI, we can observe the relative importance of zonal pressure gradient forcing in the East Asian winter monsoon system. Furthermore, W-CAO and E-CAO are significantly connected with several teleconnection patterns over the extratropical Northern Hemisphere. W-CAO winter is associated with positive Eurasian (EU), negative western Pacific (WP), and negative Arctic Oscillation (AO) patterns. On the other hand, the E-CAO is mostly associated with the Pacific-North American (PNA) pattern.
The W-CAO and E-CAO have a distinct relationship with the tropics. W-CAO is associated with greater precipitation over the Maritime Continent, weaker global Hadley circulation, and negative SST anomalies over the tropical eastern Pacific and Indian Ocean. Stronger Walker circulations are evident and consistent with the La Niña-like events. On the other hand, E-CAO exhibits greater precipitation over the central Pacific, stronger global Hadley cell, and positive SST anomalies over the eastern Pacific and Indian Ocean. The El Niño-like events tend to induce E-CAO. These signals indicate that there are notable interactions between extratropical CAO and tropical circulation systems, particularly those related to ENSO. The primary forcing of interactions between ENSO and East Asian CAO climate is tropical convection, which triggers global circulation anomalies through poleward Rossby wave trains. The main action centers of ENSO-induced FIG. 16. As in Fig. 14, but using idealized heating over the central Pacific and cooling over the Maritime Continent. Light and dark gray shading, in (a) and (b) denote positive heating and negative heating anomalies with a 1 K day 21 contour interval, respectively, omitting the zero contour. tropical convection anomalies are located over the Maritime Continent and central Pacific. Both are evidently shown to have a significant relationship with the W-CAO and E-CAO variability.
It should be noted that a complex dynamical system in the extratropics (i.e., internal variability) plays a dominant role in the variation of East Asian CAOs. Therefore, the relationships between tropical atmosphere and CAOs may be variable depending on the particular influence from internal modes in the extratropics, which is worth considering in a future study. Considering the long time span of the current period of study, it is also important to note that the W-CAO and E-CAO may experience some decadal or interdecadal variations (Fig. 4). In fact, many studies have documented an interdecadal feature of the East Asian winter monsoon (e.g., Wang et al. 2009Wang et al. , 2010, which may partially exist in both W-CAO and E-CAO variation. ENSO can explain the interannual variability of wintermean CAO. However, it cannot explain the variation of CAO events, which has a time scale about 1 week . Moreover, previous studies have documented that cold surges are able to impact tropical convection (e.g., Chang et al. 1979;Zhang et al. 1997). For our next work, we will focus on the interactions between CAOs and the tropical weather system in an intraseasonal time scale. A better understanding on this topic is expected to improve short-to medium-range forecast of both CAO events and tropical weather.
A CAO event is basically a synoptic-scale phenomenon ). Time averaging a CAO index in the whole winter may raise the question whether it can represent the annual frequency of the individual CAO events. This section assesses the reliability of the seasonal-mean CAO index in association with the CAO events.
The time scale of a CAO event is about 3-5 days according to autocorrelation of the daily CAO index . We identify individual CAO events using local maxima of daily CAO index data, which are greater than (or equal to) 1.5 standard deviations (period 1958-2013). We find 234 W-CAO-type events (annual rate 4.2 events yr 21 ) and 207 E-CAO type events (annual rate 3.7 events yr 21 ). Synoptic conditions associated with the events are shown by lag-composite analysis in Fig. A1. Circulation and temperature anomalies before and after the events are consistent and comparable with lag-regression analysis shown by Shoji et al. (2014). Therefore, identification of individual CAO events based on local maxima is sufficiently reliable. Figure A2 exhibits time series of annual frequency of CAO events and winter-mean CAO index. Both W-CAO and E-CAO show close relationship between their individual event frequency and winter-mean index, confirming the association between synoptic-scale event frequency and seasonal-mean index. Strong and weak winter CAO indices indicate more and less frequent CAO events, respectively. This association is valid because we utilize the quantified cold air mass to define the CAO event and index.