East Asian Winter Monsoon Impacts the ENSO-related Teleconnections and North American Seasonal Air Temperature Prediction

El Niño-Southern Oscillation (ENSO) is a key feature for seasonal weather and climate prediction in the extra-tropics since related sea surface temperature anomalies induce precipitation anomalies that generate poleward propagating Rossby waves and teleconnections. The East Asian winter monsoon (EAWM) is driven by processes originating over the Asian continent and, to a lesser degree, by ENSO-related tropical convection. EAWM also strongly affects convection and precipitation patterns over the western tropical Pacific by cold air outbreaks reaching equatorial latitudes. Hence, one can expect a modulating effect of EAWM on the generation of Rossby wave trains related to ENSO. By increasing the convective heating over the western Pacific, strong EAWM strengthens the Pacific Walker circulation, and weakens (strengthens) the El Niño (La Niña) related effects on the extra-tropics via a modulation of the Pacific North America teleconnection pattern. Our results indicate that, for seasonal prediction over North America, along with ENSO the variability of EAWM should also be taken into account. The climate anomalies over the North America for the same phase of ENSO are significantly different for strong and weak EAWM.

. The causal linkage between the ENSO events and interannual precipitation anomalies in the tropics is well reported 4,21 . However, the variability of tropical precipitation anomalies in the same phase of ENSO but under different extratropical circulation conditions is still not known. There is growing evidence that the extratropics can play an important role in ENSO evolution [22][23][24][25] . The primary objective of our present analysis is to investigate how and to what degree EAWM res modulates the ENSO-SST related rainfall anomalies in the tropics. This study is of potential global significance, because the associated local or regional convective heating differences may trigger varied tropical-extratropical teleconnections due to poleward propagating Rossby waves being excited at different locations and with different strength. Firstly, we present the composite rainfall anomalies over the Pacific Ocean under different background conditions as listed in Table 1 to show the modulating effect of the EAWM res on the ENSO-related rainfall anomaly patterns in the tropical Pacific Ocean. Secondly, we illustrate the background atmospheric circulation conditions responsible for these rainfall anomalies over the Pacific Ocean. Thirdly, since the variability of ENSO has far-reaching impact on weather and climate over North America, we investigate how the EAWM res is modulating the ENSO-related teleconnections and seasonal temperature anomaly patterns.

Results
SST related precipitation anomaly pattern modulated by EAWM res . The influence of EAWM res on the ENSO-induced tropical rainfall anomalies during boreal winter months is shown in Fig. 1. The left panel displays the mean rainfall anomalies during El Niño conditions. A composite from 25 El Niño events, Fig. 1a, is plotted for the purpose of comparison. We observe an increase in rainfall over the central and eastern Pacific, while drier conditions prevail over the western Pacific. These results are consistent with the previous studies 4 . Next, we sorted the El Niño events into strong and weak EAWM res conditions (Table 1). It should be noted that the subgroups of ENSO events have no bias against the ENSO intensity or spatial pattern changes. The intensity of SST anomalies in the Niño3.4 region (5°N-5°S, 120°-170°W) shows a roughly even distribution regardless of the EAWM res status (see Section Methods and supplementary Figure S1). Furthermore, the two types of ENSO events 26,27 have no obvious preference in any of the subgroups as shown in Table 1. It is also confirmed from the SST anomaly patterns for each of the four subgroups of ENSO that neither of them shows any consistent bias towards a central Pacific or an eastern Pacific type (supplementary Figure S2). SST anomalies in the weak EAWM res -El Niño are larger than those of the strong EAWM res -El Niño, but their differences are statistically insignificant. This could be attributed to the top 2 strongest El Niño events (1997/98, 2015/16) in the former subgroup.
The precipitation anomalies during the two subgroups of El Niño show obvious differences in amplitude and spatial extents. In the strong EAWM res -El Niño group, the below-normal (<= −1.2 mm/day) rainfall is restricted to small regions near the Philippines, and west of Indonesia, whereas for the weak EAWM res -El Niño group (Fig. 1c), drier conditions prevail over the entire western Pacific, the Maritime Continent and northern Australia. The daily average rainfall anomalies reach as low as −2.8 mm/day over the region between 0-15°N and 110°E-155°E. During the strong EAWM res -El Niño years, the eastern edge of the "wetter than normal" region in the northern hemisphere is confined to a region south of Hawaii in the central-eastern Pacific, while during the weak EAWM res -El Niño years it can extend further eastward to 120°W. The differences in rainfall anomalies between the strong and weak EAWM res -El Niño years are shown in Fig. 1d. Clearly the modulating effect of EAWM res on the El Niño-related rainfall anomalies in the tropics can be observed. It is characterized by a positive center over the western and negative over the eastern Pacific Ocean. The anomaly ranges between ±2-3 mm/day, the same magnitude of mean El Niño-related anomalies. Hence, the EAWM res is playing an important role in modulating the El Niño-induced rainfall anomalies over the tropical Pacific Ocean. The stronger the EAWM res is, the weaker is the relationship between El Niño and tropical rainfall anomalies, and vice versa for weak EAWM res .
The right panel of Fig. 1 is similar, but for the La Niña conditions. Unlike for the El Niño conditions, the rainfall anomaly increases over the western Pacific Ocean, but decreases over the central equatorial Pacific Ocean (Fig. 1e). Strong EAWM res under La Niña conditions brings heavier precipitation over the Philippines, the Maritime Continent and northern Australia (Fig. 1f). In contrast, weak EAWM res under La Niña conditions favors abnormally low rainfall and drier conditions prevail in western Indonesia. The situation is very different over the central-eastern Pacific Ocean. Unlike during El Niño conditions, the EAWM res couples only weakly with the tropical rainfall anomalies during La Niña conditions (Fig. 1h). On the one hand, this may be explained by the smaller amplitude of SST anomalies associated with La Niña. On the other hand, this may be attributed to La Niña related low SST over central-eastern Pacific and the threshold value (27-28 °C) for the occurrence of convection. Therefore, the EAWM res has diverse influence on the ENSO-SST induced rainfall anomalies over the tropical Pacific Ocean.
The tropical rainfall anomaly patterns discussed above can be reconfirmed by several approaches. Firstly, as the subgroups have small numbers of samples for the composite analysis, we reconfirm our results using other datasets. The ECMWF's 20th century (ERA20C) atmospheric reanalysis and Climatic Research Unit (CRU) timeseries (TS) version 4.00 precipitation data are employed for larger samples. The common time period in the two datasets is 1900-2010, and the years for each of the categories are listed in supplementary Table S1. Although

Groups years CAO Frequency CAO Strength
Strong EAWM res -El Niño [19531970EP 1977EP 1978  limited by land precipitation only, results from these two longer datasets (supplementary Figure S3) are consistent with Fig. 1. Negative rainfall anomalies over the majority of the Maritime Continent during an El Niño is diminished by a strong EAWM res and enhanced by a weak EAWM res . Similarly, Borneo, east Sumatra, and Java experience more rainfall during strong EAWM res -La Niña condition compared to that in weak EAWM res -La Niña condition. Secondly, results from the NOAA 20 th Century reanalysis datasets are largely consistent with the former ones (supplementary Figure S4 and Table S2). Note that supplementary Figure S4 shows little significance over the central equatorial Pacific Ocean, and this could be attributed to the uncertainty in the reconstructed precipitation dataset which have large ensemble spread over the tropical central Pacific Ocean. Thirdly, we employed different EAWM indices 28,29 to get the ENSO-unrelated residual EAWM part. The variability of these "EAWM res " yields very similar rainfall anomaly patterns as presented in Fig. 1 (figures not shown).
Possible mechanism. The Indo-Pacific Walker circulation is affected by both ENSO-and EAWM-related rainfall anomalies 30 . In Fig. 2, anomalies of the Pacific Walker circulation are shown for all ENSO events and the different combinations of ENSO and EAWM res ( Table 1) as anomalies of the 200hPa divergent wind and velocity potential. The variation of the Pacific Walker cell is closely tied to the zonal SST gradient anomalies between the central and western equatorial Pacific. However, it is noteworthy that the difference of zonal equatorial Pacific SST gradient in the two El Niño (La Niña) subgroups is statistically insignificant (see section Methods). Climatologically, during an El Niño event, the weaker zonal thermal and pressure gradients weaken the Walker cell ( Fig. 2a), as seen by anomalous ascending motion over the central-eastern Pacific Ocean and descending motion over the Maritime Continent (Fig. 2a). Stronger EAWM res tends to weaken the Walker circulation anomalies initiated by the El Niño ( Fig. 2b) i.e. convergence over the Maritime Continent has weakened to half the strength of its counterpart in the El Niño case. Figure 2c reflects the combined effect of weak EAWM res and the El Niño, resulting in a much stronger western-Pacific convergence and eastern-Pacific divergence. During the La Niña episodes (Fig. 2e), the anomalous Walker circulation behaves contrastingly, with anomalous ascending (descending) motions over the tropical western (central) Pacific Ocean. The divergent center over western Pacific extends southwestward during the strong EAWM res -La Niña case, whereas, it shrinks northward during the weak EAWM res -La Niña case. As a consequence, convergence of air masses over the tropical central-eastern Pacific is strengthened (weakened) by stronger (weaker) EAWM res . It should be noted that the difference of Walker circulation anomalies between the strong and weak EAWM res cases is weak during the La Niña events (Fig. 2h). This could be attributed to the suppressed convections in the central-eastern Pacific Ocean during a La Niña event.
In other words, La Niña-related negative SST anomalies superimposed on the climatological cold tongue would greatly decrease the occurrence of deep convection. That means the EAWM res would have limited impacts on tropical central-eastern Pacific during a La Niña event. What's more, the smaller amplitude of SST anomalies associated with La Niña may also explain the weak difference of the Walker cell. In general, the variability of the Walker Circulation shown in Fig. 2 is resulting in the aforementioned rainfall anomalies including the asymmetry between the El Niño and La Niña events (Fig. 1). Previous studies 13,14 reported that the cold air outbreaks (hereafter CAOs, also called cold surges) can trigger deep convective activity in the tropics. The CAOs are dominant feature of EAWM, which usually start with an intensification of Siberian-Mongolian High. Vigorous northerlies along the eastern flank of the Siberian-Mongolian high lead to a sharp rise in sea level pressure and sudden drop in surface temperature over East Asian region. When the CAO progresses rapidly southward, a frontal system of high pressure and near surface strong northerly winds affects the tropics. These induce low-level convergence and cyclonic vorticity over the tropical western Pacific sea surface. Hence, vertical lifting associated with the low-level convergence and cyclonic anomalies contribute to deep convection and heavy rainfall during the penetration of CAOs 11 . Diabatic heating associated with the precipitation further accelerates both the local East Asian Hadley cell and the Walker cells over the Indian and Pacific Ocean 14 . Therefore, the mid-high latitude originated CAOs could affect the convections over tropics.
We plotted the outgoing longwave radiation (OLR) in Fig. 3a as a proxy of tropical Pacific convective activity associated with the EAWM res . Due to the short time coverage of OLR dataset (1979-2016), we present the composite differences between strong and weak EAWM res during all ENSO years without separating further into El Niño or La Niña cases. As the weak EAWM res group contains more El Niño than La Niña events, weighted average is used to remove an El Niño bias. Similar averaging is performed for the strong EAWM res group prior to the analysis. Clearly, strong EAWM res favors enhanced convective activity over the tropical western Pacific and suppressed convective activity over the eastern Pacific Ocean (around 150°W). In Table 1 the frequency and strength of the cold air outbreaks in each of the subgroups is listed, indicating that strong EAWM res intensifies the CAO activity, and vice versa during weak EAWM res . Thus, stronger EAWM res with intensified CAO activity triggers enhanced convective activity over the Maritime Continent. The related diabatic heating anomalies associated with convection are the key to the Pacific Walker circulation variability (Fig. 3b). In general, strong EAWM res triggers above-normal convective heating anomalies over the western Pacific Ocean, enhancing the climatological Walker circulation. It counteracts the El Niño-induced weakened Walker circulation anomalies, but amplifies the La Niña-induced enhanced anomalies over the tropical Pacific Ocean. The situation is opposite but weaker during weak EAWM res , when the El Niño-induced weakened Walker circulation is amplified and the anomalies due to La Niña are suppressed.
The rainfall variability over the tropical western Pacific Ocean in response to ENSO was suggested to be directly related to the low-level circulation anomalies around the Philippines 15,16 . However, the cyclonic/anticyclonic flow over this region also has a close connection with the EAWM res . Feng and Chen 18 reported that during an El Niño event, the Philippine anticyclone tends to be weak (strong) in presence of a strong (weak) EAWM res , and vice versa during the La Niña years (see ref. 18 Figs 4,5). Furthermore, according to the numerical experiments by Feng et al. 31  Continent plays a dominant role for the existence of the Philippine anticyclone. Positive convective diabatic heating anomalies during strong EAWM res balance the El Niño-related Maritime Continent negative heating anomalies, which weakens the Philippine anticyclone. On the other hand, convective cooling during weak EAWM res strengthens the Philippine anticyclone, and vice versa for the La Niña. Thus, in the western Pacific the EAWM res modulates the ENSO-related rainfall anomalies, mediated by the variation of low-level Philippine anticyclone.
The EAWM res has diverse impacts on the ENSO-related tropical rainfall anomalies in the warm and cold phases: First, the EAWM res has a weaker modulation effect during the La Niña than the El Niño events. Such asymmetric behavior is consistent with the fact that rainfall anomalies have a saturation effect in La Niña but more intense in the El Niño conditions 32 . Second, a stronger EAWM res amplifies the rainfall anomalies over the region 5°S−15°N, 120°E-180°E under El Niño (Fig. 1d) and 15°S-15°N, 110°E-150°E under La Niña (Fig. 1h) conditions, respectively. The southwestward tilt in the precipitation anomaly pattern during La Niña conditions is in line with the trajectories of the cold air outbreaking events 16 . The Siberian High and the Aleutian Low experience a systematic eastward and westward shift in El Niño and in La Niña winters, respectively 16 . Hence, during El Niño cold air outbreaking events are more frequent across the Philippines. In contrast, during La Niña conditions cold air outbreaking events are shifted westward to the Sunda Strait 16 .
Impacts on the US air temperature forecast. ENSO is reported to have significant impact on climate in the mid and high latitudes, particularly over the North America through the Pacific-North American (PNA) teleconnection pattern 33 . The longitudinal shift in the PNA pattern, due to the movement of the driving diabatic heating centers, is resulting in different climatic anomalies over North America 34 . Considering these complex interactions, it is interesting to investigate whether EAWM res is modulating the ENSO-related climate anomalies (e.g. air temperature) over the North America. During the El Niño winter months, often the southern part of the United States (US) experiences colder than normal conditions, whereas the northern part of North American are warmer. The situation is just opposite during the La Niña winters (Fig. 4a and e). Taking also the information of EAWM res into account, the North American air temperature anomalies display quite different patterns: the strong EAWM res -El Niño case is favoring a southeast (cold)-northwest (warm) contrast (Fig. 4b); weak EAWM res -El Niño initiates warmer conditions in northeastern North American. These variabilities in air temperature anomalies are consistent with the shift in the PNA-like pattern (Fig. 4, purple lines), which is generated by the latent heating centers in the tropics 27,34 . Stronger (Weaker) EAWM res , with central (eastern) Pacific latent heating (induced by above-normal rainfall anomalies, Fig. 1b and c), powers the westward (eastward) shift of the PNA-like pattern and the air temperature anomalies over the North America. Besides, the strength of Aleutian Low over North Pacific during the strong and weak EAWM res -El Niño is related to the combined effect of El Niño and EAWM res. El Niño and strong EAWM res contribute to an intensified Aleutian Low but weak EAWM res acts opposite. Contrastingly, in the La Niña winters, the temperature difference between the two La Niña subgroups shows a quadrupole pattern, which is also consistent with the variation of 500 hPa geopotential height (Z500) (Fig. 4h). A wave train along Japan, west of Bering Sea, south to the Gulf of Alaska and eastern North Pacific is observed in Z500. The positive Z500 anomaly over Bering Sea and negative anomaly over Alaska favors stronger northerly winds between the two, which leads to a band of negative SAT over Alaska. Positive SAT anomalies over the western United States could be attributed to the southwesterly anomalies between the negative (north) and positive (southeast) Z500 anomalous centers over the eastern North Pacific. Results from a linear baroclinic model illustrated that the atmospheric linear response to an idealized western tropical Pacific thermal forcing is characterized by a wave train transporting from east of Japan to North America 35 . The wave train pattern in the North Pacific sector in Fig. 4h is similiar with the numerical results 35 , implying that the precipitation anomalies over the tropical western Pacific accounts for the different distributions of Z500 and SAT anomalies during two subgroups of La Niña. Surface air temperature differences between strong and weak EAWM res for similar ENSO conditions can vary by 1-2°C, which is comparable to the anomalies due to ENSO alone. This indicates that EAWM res is an additional important factor which may interfere with the seasonal North American air temperature prediction.

Summary and Discussion
This study is mainly dedicated to a better understanding of the modulating effects of the EAWM res on the ENSO-related rainfall and circulation variability in the tropics. Composite results suggest that a strong EAWM res weakens the El Niño-induced Pacific rainfall and Walker circulation anomalies, while a weak EAWM res amplifies them. A strong EAWM res brings more cold air outbreaks to the low-latitudes. It intensifies the tropical deep convection and diabatic heating over the western Pacific. On the one hand, the anomalous heating weakens the low-level Philippine anticyclone and diminishes the consequent negative rainfall anomalies during the El Niño. On the other hand, the strong EAWM res induced western Pacific heating, pumps the Pacific Walker circulation in a positive feedback loop. Over the eastern Pacific the stronger Pacific Walker circulation favors stronger surface divergence and less rainfall. Weak EAWM res , with below-normal heating anomalies over the western Pacific, will amplify the anomalies related to the El Niño. A similar EAWM res intervention can be seen during the La Niña events. However, the EAWM res appears to have a small contribution to the eastern Pacific rainfall due to the cold SST anomalies over there inhibiting convective rainfall anyway. Through its impact on West Pacific rainfall and the Pacific Walker cell, the EAWM res also affects the ENSO-related temperature variability over the North America. Strong EAWM res warms the western part and cools the eastern part of North America during an El Niño, whereas it cools the northwestern edge of North America and warms the western part of US during a La Niña. ENSO, thus, is not the only potential predictor for North American seasonal forecasts; the EAWM is similarly important and should be accounted for in the future.
The stratification of ENSO events into four subgroups lead to small sample sizes, and one may concern about the robustness of the composite. Hence, we add some supplementary analysis. As mentioned in results section (a), more samples (listed in supplementary Table S1) are obtained by using datasets which cover a longer period, the ERA20C atmospheric reanalysis and CRU data. Besides precipitation (supplementary Figure S3), composite anomalies of Walker cell and Z500, SAT (supplementary Figures S5, 6) are consistent with the former results in Figs 2 and 4, respectively.

Datasets. The National Centers for Environmental Prediction -National Center for Atmospheric Research
(NCEP-NCAR) monthly mean atmospheric reanalysis data 36 are used in this study. This datasets have a horizontal resolution of 2.5° latitudes by 2.5° longitudes, 17 levels in the vertical direction, and a time period of 1948-present. For sea surface temperature (SST) we used the 1.0° × 1.0° horizontal resolution monthly mean Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) dataset 37 . The HadISST data covers the time period from 1870 to the present. In addition, for rainfall data we employed the globally gridded (2.5° × 2.5°) monthly-mean NOAA Precipitation Reconstruction (PREC) data from 1948 to the present 38 . The NOAA Outgoing Longwave Radiation (OLR) 39 data, available from 1979 to 2016, is used to signal the activities of convection.
We use the data averaged from December to February to represent boreal winter mean. The analysis period spans from 1948 to 2016, except for that related with OLR (from 1979 to 2016). Conventionally, the winter of Indices and methods. The EAWM res index 17 , which represents the interannual variability of EAWM res , is calculated by removing the ENSO-related part by linear regression. The original EAWM res index is multiplied with −1, so that a positive index corresponds to northerly winds and a strong EAWM res state. Strong (Weak) EAWM res winters are selected based on a threshold of + (−) 0.5 standard deviations of EAWMI res. We then obtain 20 strong and 20 weak EAWM res winters over the period of 1948-2016, respectively.
The Oceanic Niño index (ONI), which is defined by the Climate Prediction Center (CPC) as the 3-month running mean of ERAAT.v4 SST anomalies in the Niño-3.4 region (5°N-5°S, 120-170°W), is used as an ENSO index here. According to the CPC, when the ± 0.5°C threshold of ONI is met for a minimum of 5 consecutive overlapping months, it can be recognized as an El Niño/La Niña event. Based on these criteria, we obtain 25  All these selected years are listed in Table 1. It should be noted that the oceanic Niño index (ONI) is linearly independent of the EAWM res index. And the intensity of ENSO events shows a roughly even distribution regardless of the EAWM res status (supplementary Figure S1).
Eastern-Pacific (EP) and Central-Pacific (CP) ENSO events are identified by normalized winter mean Niño 3 index and El Niño Modoki index (EMI) 26 . When the normalized EMI exceeds 0.7 (−0.7), meanwhile greater (less) than the normalized Niño 3 index, a CP El Niño (La Niña) is defined. In the same way, an EP El Niño (La Niña) is identified when the normalized Niño 3 index greater (less) than 0.7 and also greater (less) than the EMI. Years that both of the two Niño indices don't exceed a 0.7 standard deviation would not be classified to any type. Thus, two types of ENSO are separated as listed in Table 1.
The zonal SST gradient along the equatorial Pacific Ocean between the central and western Pacific is defined as the difference in area averaged SST between the Nino 4 (160°E-150°W, 5°S-5°N) region and the western Pacific Ocean (90°-150°E, 5°S-5°N). Result of 1000 bootstrapping 45 combinations shows that the zonal SST gradient difference between the strong EAWM res -El Niño group and the weak EAWM res -El Niño group (the strong EAWM res -La Niña group and the weak EAWM res -La Niña group) is insignificant at the 90% confidence level based on a two-tailed test. The result is same if considering the zonal SST gradient between the eastern (Nino3 region, 150°W-90°W, 5°S-5°N) and the western equatorial Pacific.
Cold Air Outbreak. The winter frequency of CAO events is obtained from Abdillah 46 ( Figure A2; W-CAO over East Asian; period 1958-2013). Table 1 listed the average winter frequency of East Asian CAO events in each group. As the samples sizes are small, we used the bootstrap method 45 , resampling 1000 times, and one-tailed test to estimate the mean differences from two samples. The winter frequency of CAO events in the strong EAWM res -El Niño group is significantly higher than that in the weak EAWM res -El Niño group at a confidence level of 95%. Between the two groups during the La Niña, the difference reaches the 90% confidence level.
We adopt an alternative approach to estimate the interannual variation of CAOs. A profound characteristic related with CAOs in the tropics is the disturbance of sea level pressure (SLP) 13,14 . Usually, the time scale of a CAO event is about 5 days 47 , therefore a 3-6 day band pass filter is applied on the SLP anomaly over the western tropical Pacific region to define the East Asian cold air outbreaking events. Thus, an annual CAO strength index is defined as an area-averaged variance of the filtered anomalies during winter in the region between 0-15°N and 105°E-125°E. We checked the robustness of the index by considering smaller and bigger areas, however, the result remains the same. The average strength of the CAO strength index (normalized) is −0.1 for El Niño (25 events) and 0.1 for La Niña (18 events) conditions, reflecting the influence of ENSO on cold surges 46 . But when we include the EAWM res variability, the CAO strength index for the strong EAWM res groups is much higher than for the weak EAWM res groups. The values are 0.6&-0.2 for El Niño and 0.7&-0.1 for La Niña, respectively. The difference of CAO strength indices between the strong EAWM res -El Niño (La Niña) group and the weak EAWM res -El Niño (La Niña) group exceed the 95% (95%) confidence level according to a one-tail bootstrap statistical estimate. This indicates that strong EAWM res intensify the CAO activities, and vice versa during weak EAWM res , regardless of El Niño or La Niña background. Divergent wind calculation. The horizontal wind can be divided into two components: non-divergent (or rotational) and divergent wind 48 .