Impacts of ENSO on the seasonal transition from summer to winter in East Asia

The El Niño-Southern Oscillation (ENSO) has seasonally distinct impacts on the East Asian climate so that its seasonal transition depends on the phases of El Niño and La Niña. Here, we investigate the seasonal transition of surface temperature in East Asia from boreal summer to winter based on the warm/cold ENSO developing phases. During La Niña years, from summer to winter the continuous temperature drop in East Asia tends to be faster than that during El Niño, indicating a latter start and earlier termination of fall. This different seasonal transition in East Asia according to phases of ENSO is mostly explained by atmospheric responses to the seasonally-dependent tropical/subtropical precipitation forcings in ENSO developing phases. The anomalous positive precipitation in the subtropical North Pacific exists only in September and leads to the subtropical cyclonic flow during El Niño years. The resultant northerly anomalies on the left side of the subtropical cyclone are favorable for transporting cold advection towards East Asia. However, the positive subtropical precipitation disappears and teleconnection to East Asia is strongly controlled by the negative precipitation anomalies in the western North Pacific, modulating the anticyclonic anomalies in East Asia during the early winter (November). Therefore, these seasonally sharp precipitation changes in the tropics/extratropics associated with ENSO evolution induce distinctive teleconnection changes from northerly (summer) to southerly (winter) anomalies, which eventually affect seasonal transition in East Asia. Also, the Coupled Model Intercomparison Project Phase 5 models reasonably simulate the relatively rapid temperature transition in East Asia during La Niña years, supporting the observational argument.


Introduction
Considerable progress has been made towards understanding the extratropical atmospheric response to the tropical forcing associated with El Niño-Southern Oscillation (ENSO) sea surface temperature (SST) anomalies (Yeh et al. 2018). The Rossby wave dynamics lead to changes in the extratropical atmospheric circulation, as a direct response to the ENSOrelated tropical diabatic heating (Horel and Wallace 1981;Hoskins and Karoly 1981). Indeed, many previous studies suggested that the ENSO can influence the East Asian monsoon system in boreal summer (Nitta 1987;Huang and Wu 1989;Zhang et al. 1996;Kosaka and Nakamura 2006;Xie et al. 2009Xie et al. , 2016. The impacts of ENSO on the East Asian climate are also significant during the El Niño mature phase (i.e., winter), through the tropics-extratropics teleconnection (Rasmusson and Carpenter 1982;Zhang et al. 1999;Wang et al. 2000;Son et al. 2014;Gong et al. 2015;Kim et al. 2017. An observational study (Nitta 1987) hypothesized that during summers an anomalous anticyclone in the western North Pacific (WNP) can affect the summer climate in East Asia through atmospheric meridional dipoles, referred to as the Pacific-Japan (PJ) pattern. This lower-tropospheric teleconnection has been extensively considered as one of the common summer teleconnection patterns in East Asia * Jong-Seong Kug jskug1@gmail.com 1 1 3 (Nitta 1987;Huang 1992;Huang and Sun 1992). In its positive phase, there is a possibility that the PJ pattern brings a wetter and cooler summer in East Asia (Xie et al. 2009(Xie et al. , 2016Kosaka et al. 2013). Zhang et al. (1996) found that the suppressed convective activity over the equatorial western Pacific, as a result of warming in the eastern Pacific, intensifies the East Asian summer monsoon (EASM). Strong positive anomalies of outgoing longwave radiation over the Maritime Continent lead to the atmospheric cooling that accompanies the anomalous anticyclone. In a subsequent study, the intensified and westward shift of the anticyclonic anomalies in the WNP was suggested to be responsible for the negative precipitation anomalies over both southern and northern parts of China, and bounding positive anomalies over central parts (Zhang et al. 1999). Evidence presented in Wang et al. (2000) also shows that the recurrent anomalous anticyclone after an El Niño event is a result of an atmospheric Rossby wave response to the SST cooling over the WNP. The WNP anticyclonic anomalies persist until the following summer through positive thermodynamic feedback between the anticyclone and negative SST anomalies. In accordance with summer, the WNP anticyclonic anomalies act as a key system that bridges the ENSO and East Asian winter monsoon (EAWM), transporting warm and moist air towards East Asia (Wang et al. 2000). Previous studies on the ENSO impacts have mainly focused on the East Asian summer and winter monsoons, but the dynamical influences of ENSO are somewhat independent of the season (Zhang et al. 1999;Wang et al. 2000;Wu et al. 2003).
In addition, Wu et al. (2003) documented the spatialtemporal evolution of ENSO-related rainfall anomalies in East Asia. Because the rainband in East Asia exhibits large meridional migration over the seasons (Tao and Chen 1987), it is necessary to understand how ENSO affects anomalous rainfall changes differently with the season. According to Wu et al. (2003), the most robust influence of ENSO is seen as a positive correlation that migrates from southern China in the fall of the ENSO developing phase to eastern-central China and southern Japan through the following spring. Another is a negative correlation over northern China in summer and fall during the ENSO developing years. These two main rainfall anomalies in East Asia are induced by different anomalous atmospheric circulation systems (Wu et al. 2003). The evolution of positive rainfall in southern China is closely related to the lower-level anticyclone anomalies over the WNP. The anticyclonic anomalies are established over the South China Sea in the fall of ENSO developing years, move eastward to the Philippine Sea during the ENSO mature phase, and shift northward in spring and summer of ENSO decaying years. In comparison, the negative rainfall in northern China is associated with the northerly anomalies on the western side of the cyclonic anomalies that displace southwestward along the East Asian coast during the developing phase of ENSO. Lee and Julien (2016) reported that the climate impact of warm and cold phases of ENSO over South Korea. The temperature anomalies in El Niño years are below normal for summer through fall with opposite response in La Niña years. Additionally, the seasonal cycle of surface temperature over Japan can be amplified in La Niña years bringing hot summers and cold winters (Iwakiri and Watanabe 2020). Son et al. (2016) suggested that a significant negative correlation exists in September, the ENSO developing phase, between Niño3 SST and precipitation over the Korean Peninsula. Interestingly, the positive precipitation anomalies that exist over the subtropical North Pacific associated with ENSO can be considered a unique pattern only in this month. This subtropical diabatic heating in September plays a critical role in generating the cyclonic flow over the subtropical North Pacific as a Rossby wave response. The resultant northerly anomalies on the left side of cyclonic circulation act to decrease the precipitation over the Korean Peninsula in September. Therefore, Son et al. (2016) emphasized that the subtropical cyclone is a key component in linking ENSO signals to East Asia accompanying the northerly anomalies in September.
Several studies found that there is a relationship between the summer and winter monsoon in East Asia associated with ENSO. Chen et al. (2013) suggested that the EAWM and EASM are closely related only in ENSO years, comparing the link of the EAWM and EASM in ENSO and non-ENSO years. Additionally, the role of ENSO in the evolution of EAWM to the following EASM is investigated (Chen et al. 2013). Meehl and Arblaster (2002) emphasized that the tropospheric biennial oscillation (TBO) which encompasses most ENSO years is a fundamental feature of the coupled climate system over the entire Indian-Pacific region, and the ocean-atmospheric interaction over the region is the dominant contributor to Asian-Australian monsoon. The ENSO plays an important role in the in-phase EASM to EAWM transition, indicating stronger transition under El Niño than La Niña (Yu et al. 2021).
Despite the above studies on the seasonally-varying ENSO teleconnections in East Asia, detailed studies focusing on the ENSO-related seasonal transition have not been conducted sufficiently. In particular, the ENSO possibly contributes to the seasonal transition from summer to winter; owing to the East Asian perturbations associated with ENSO being considerably different in summer and winter, respectively. Although limited studies have investigated this season, there still remains a need to understand the ENSO impacts on the East Asian fall climate in terms of its spatial/temporal evolution, intensity, and dynamical processes. Therefore, in this study, we investigate the ENSO impacts on seasonal transition, in particular from summer to winter (i.e., fall) in East Asia.
The fall, namely autumn, can be defined diversely but simply refers to the season between summer and winter. Astronomically, fall is regarded in the Northern Hemisphere as from the fall equinox to the winter solstice. In meteorology, fall is generally considered to include the months of September, October, and November in the Northern Hemisphere; March, April, and May in the Southern Hemisphere. We compare the atmospheric seasonal transition and its regional impacts on East Asia, from summer to winter, related to the ENSO forcings using observational data. In addition, we assess how well the Coupled Model Intercomparison Project Phase 5 (CMIP5) models simulate the ENSO impacts on seasonal transition during fall in East Asia.
Section 2 describes the observed and simulated datasets. Section 3 examines the ENSO-related seasonal transition in East Asia with the observed results. The long-term climate simulations of CMIP5 models are investigated to support the observational findings, and the fidelity of seasonal transition in current models is also evaluated. A summary and discussion follow in Sect. 4.

Reanalysis and observation data
Observations used in this study are from a total of 400 weather stations in China, Korea, and Japan. The monthly surface temperature and precipitation data include 160 stations in China (China Meteorological Administration; CMA), 87 stations in Korea (Korea Meteorological Administration, KMA), and 153 stations in Japan (Japan Meteorological Agency, JMA). The daily mean temperature and precipitation data at 45 stations over South Korea from KMA are also used. The monthly SST data are derived from the Extended Reconstructed Sea Surface Temperature version 5 [ERSSTv5; Huang et al. (2017)]. The daily and monthly atmospheric variables are obtained using the NCEP-DOE AMIP-II Reanalysis (R-2) from the National Centers for Environmental Prediction/National Center for Atmospheric Research [NCEP/NCAR; Kanamitsu et al. (2002)]. Monthly precipitation data are also obtained from the Climate Prediction Center Merged Analysis of Precipitation ]CMAP; Xie and Arkin (1997)]. The hourly precipitation data of the European Centre for Medium-range Weather Forecasts [ECMWF; Hersbach et al. (2020)] fifth generation reanalysis (ERA5) are utilized on a daily scale. All reanalysis data cover the period from 1979 to 2018. Note the seasonal cycle and linear trend for the period 1979-2018 are removed from the data before analysis.

CMIP5 models
Historical simulations produced by the 40 Coupled Model Intercomparison Project Phase 5 (CMIP5) models were analyzed. A single ensemble member (r1i1p1, the first realization of the first initialization of the perturbed physics model) of each model is used in this study. Brief descriptions of the 40 models including their modeling centers are listed in Table 1. An overview of CMIP5 and the experiment design are described by Taylor et al. (2012). We utilized the monthly-mean outputs for a total of 100-year integrations from 1901 to 2000. Data of all CMIP5 models were also linearly detrended for the period 1901-2000.

Linear Baroclinic Model
A Linear Baroclinic Model [LBM; Watanabe and Kimoto (2000)] is used in this study to obtain a steady atmospheric response to prescribed heating in the tropics. The model is based on primitive equations linearized about the observed basic state. The basic state is used from the climatology during September and November for 1979-2018. Here, we use a dry version with a horizontal resolution of T21 and 20 sigma levels in vertical. The model is integrated for 30 days to obtain the quasi-steady state response to fixed diabatic forcing.
In LBM experiments, a given diabatic forcing denoted as H(x, y, z) is constructed by assuming a vertical structure of the diabatic heating as follows: H(x, y, z) = PRCP(x, y) × P(z),where PRCP(x, y) is a horizontal precipitation pattern over the tropical region (0°-360° E, 20° S-20° N), which is calculated from linear regression with respect to Niño3.4 (170°-120° W, 5° S-5° N) SST. P(z) is a vertical profile of heating with the vertical weighting at 400 hPa ( = 0.45). The model is forced by prescribed diabatic heating to mimic condensational heat release in deep convection associated with the ENSO-related precipitation. Figure 1 shows composites of the daily surface temperature and precipitation over South Korea from summer to winter (August-December) for the developing phases of El Niño and La Niña. The El Niño (La Niña) years are defined when the fall season (September-November; SON) Niño3.4 SST exceeds its 0.5 (− 0.5) standard deviation for the period 1979. Thus, 12 El Niño events (1982, 1986, 1987, 1991, 1994, 1997, 2004, 2009) and 1 3

Relationship between ENSO and seasonal transition in East Asia
13 La Niña events (1983,1984,1985,1988,1995,1998,1999,2000,2007,2010,2011,2016,2017) are selected. The temperature and precipitation data over South Korea are smoothed by the 11-day moving average to remove short-term fluctuations. The temperature over South Korea gradually decreases from summer to winter for both El Niño and La Niña years, showing the evident seasonal transition.
However, the decreasing trend appears to be stronger in La Niña years compared to that of El Niño years (Fig. 1a). Figure 1b shows the temperature difference between El Niño and La Niña events over South Korea. It shows that the South Korean temperature is higher in La Niña than in El Niño years from late summer to mid-fall (August and October), but the opposite is found from early November onwards. Such temperature differences between El Niño and La Niña years over South Korea are mostly significant at the 95% confidence level based on bootstrap analysis with 10,000 times random sampling. Over South Korea, heavy rainfall is predominantly experienced during summer, rapidly decreasing before winter, in both El Niño and La Niña years (Fig. 1c). The precipitation difference between El Niño and La Niña years shows that South Korean precipitation is higher in La Niña than in El Niño years until mid-October (Fig. 1d), almost consistent with temperature transition. According to Fig. 1, the surface temperature and precipitation over South Korea tend to be higher in La Niña than in El Niño years until October but reverses after October. Therefore, during the ENSO developing phase, the detailed temporal evolution of the ENSO warm and cold years is somewhat different over South Korea.
These temperature and precipitation differences in the El Niño and La Niña evolutions may affect the period of fall. Here, the fall period is defined as the mean daily temperature between the last sequence day above 5 ℃ and the first sequence day below 20 ℃ (Table 2) following Lee (1979). Averaged over 40 years, fall over South Korea totals 69 days, from 19th September to 26th November. During El Niño years, the number of days classified as fall over South Korea is 75. However, during La Niña, fall starts 5 days later and terminates 6 days earlier. Therefore, the period of South Korean fall for La Niña is 11 days shorter than that for El Niño years. This means that the seasonal transition in surface temperature from summer to winter over South Korea tends to be much faster during La Niña than that of El Niño years. Note that these results are robust to the temperature criteria used for the definition of the fall period.
We checked the differences of fall days over South Korea for El Niño/La Niña and normal years. The sample size, mean, and standard deviation of fall period during 1979-2018 for El Niño, La Niña, and normal years are shown in Table S1. The difference of fall days between El Niño and normal years are statistically different at the 95% significance level (p < 0.0206). The difference between La Niña and normal years is also statistically significant (p < 0.0108). In addition, the El Niño and La Niña differences in their fall periods are significant at the 99% confidence level (p < 0.0001).
To assess whether the result over South Korea extends over a wider region, the same analysis as in Fig. 1 is also MRI, Japan NorESM1-M NCC, Norway NorESM1-ME NCC, Norway applied for surface temperature particularly over the Korea-Japan region (120°-145° E, 30°-40° N) using the reanalysis data (Fig. 2). Overall, temperature trends for Korea-Japan are similar to those of South Korea (Fig. 1a) with gradual decreases in temperature for both El Niño and La Niña years (Fig. 2a). The temperature difference between El Niño and La Niña years is also consistent with the station results over South Korea (Fig. 1b), indicating that temperature is higher in La Niña than in El Niño years until early-November and then the opposite occurs afterward (Fig. 2b). Similarly, Korea and Japan experience distinctly more rainfall in La Niña than El Niño until late October, but the opposite pattern occurs from November (Fig. 2d). Overall, the East Asian climate rapidly changes during La Niña years from summer to winter, showing a steeper slope than those during El Niño.
To examine the seasonal transition in East Asia during the ENSO developing phase, Fig. 3 shows the regressed temperature and precipitation in East Asia with respect to SON Niño3.4 SST using data from 400 weather stations throughout China, Korea, and Japan. In September, there are negative temperature anomalies overall at most stations (Fig. 3a). In particular, the negative temperature anomalies are significant over the lower reaches of the Yangtze River in China, Korea, and Japan. In November, the negative temperature anomalies exist over China but are insignificant at most stations, except for a few stations around Shandong (Fig. 3b). Interestingly, positive temperature anomalies are La Niña years. Temperature and precipitation data are filtered with an 11-day moving average. Significant differences, above the 95% confidence level, are displayed by gray shading using the bootstrap analysis with 10,000 times random sampling. The gray dotted lines indicate the zero line Table 2 The fall period and number of days at 45 weather stations over South Korea for El Niño, normal and La Niña years, based on the ± 0.5 standard deviation of SON Niño3.4 (170°-120° W, 5° S-5° N) SST The fall period is defined as the mean daily temperature between the last sequence day above 5 ℃ and the first sequence day below 20 ℃ as in Lee (1979) . 3c). It is noted that the overall correlations with the temperature difference over Korea-Japan are somewhat higher than those within each month (Fig. S1). The regressed precipitation anomalies in September are similar to the temperature pattern, but significant anomalies are mostly confined over Korea (Fig. 3d). In November, significant positive anomalies exist at most weather stations throughout southeastern China, Korea, and Japan (Fig. 3e). Consistent with the temperature difference ( Fig. 3c), the precipitation difference in East Asia between September and November shows the overall positive anomalies, again higher than those of each month's precipitation (Fig. 3f). The temperature difference between September and November with respect to SON Niño3.4 SST shows positive anomalies dominant over Korea and Japan (Fig. 3c). Simply, the temperature anomalies in East Asia have a negative tendency from September to November as shown in Figs. 1a and 2a. Therefore, the positive anomalies found in their difference indicate that the seasonal temperature tendency slows down as the El Niño increases, while the temperature decreases steeply from summer to winter when La Niña intensifies in amplitude. The seasonal precipitation tendency from September to November overall also shows positive anomalies in East Asia indicating the dramatic precipitation change during the La Niña evolution (Fig. 3f), as exhibited in Figs. 1c and 2c. Therefore, a key question remains of why the impacts of El Niño and La Niña forcing on the East Asian climate are seasonally different. During the ENSO developing phase, the temperature/precipitation tendencies in East Asia can be explained by the ENSO-related tropical forcing and resultant atmospheric circulation. The detailed underlying dynamical processes, that are associated with the different seasonal transitions in East Asia will be examined in the next section. between El Niño and La Niña years. Temperature and precipitation data are filtered with an 11-day moving average. Significant differences, above the 95% confidence level, are displayed by gray shading using the bootstrap analysis with 10,000 times random sampling. The gray dotted lines indicate the zero line Impacts of ENSO on the seasonal transition from summer to winter in East Asia 1 3

Seasonal ENSO teleconnections to East Asia
To investigate the seasonal transition responses to ENSO in East Asia, the regressed precipitation and lower-tropospheric circulation with respect to SON Niño3.4 SST are calculated (Fig. 4). In September, the typical ENSO-related precipitation pattern over the tropical Pacific is characterized by a dipole with negative precipitation anomalies in the western Pacific and zonally-extended positives from the equatorial central to eastern Pacific (Fig. 4a). The positive precipitation anomalies centered over the equatorial central Pacific (CP) induce a lower-level cyclonic flow as a Gill-type response, but its meridional extent is confined in the off-equatorial and some subtropical regions. Interestingly, the distinct positive precipitation anomalies appear over the subtropical North Pacific (150° E-160° W, 5°-20° N), which is a unique pattern only in this season (Son et al. 2016). The subtropical positive precipitation anomalies likely play a role in extending the subtropical cyclonic anomalies further northward by causing additional subtropical diabatic heating (Son et al. 2016). The resultant cyclonic anomalies are dominant over the subtropical North Pacific, accompanied by northerly wind anomalies on the western side of cyclonic circulation (Fig. 4d). The low-level northerly wind anomalies near the south of Korea are favorable for cold air and dry advection over the Korea-Japan region in September (Fig. 3a, d).
Relative roles of these tropical/subtropical positive precipitation forcings and their teleconnections in September might be divided in a linear analysis. To linearly separate the tropical and subtropical precipitation signal, partial regression is applied (Fig. S2). While the cyclonic responses to the tropical forcing cannot play a role near East Asia, the linearly separated positive precipitation anomalies in the subtropical North Pacific are responsible for extending the subtropical cyclonic anomalies to East Asia in September. Therefore, we can conclude that the subtropical precipitation forcing plays an important role in modulating the East Asian climate during September.
In November, the teleconnection pattern in East Asia is quite different from that in September. In addition, the distinct positive precipitation anomalies which occur over the subtropical North Pacific in September clearly disappear in November, and the overall negative anomalies in the WNP are dominant (Fig. 4b). The negative WNP precipitation anomalies are responsible for the strong anticyclonic anomalies near East Asia conveying the southerly anomalies on the western side of the circulation (Son et al. 2014;Kim et al. 2017. The lower-tropospheric anticyclonic anomalies in the WNP are vertically baroclinic, therefore are accompanied by cyclonic anomalies in the upper troposphere. The upper-tropospheric divergence related to the cyclonic circulation drives the anticyclonic anomalies near East Asia (Fig. 4e) through Rossby wave energy propagation (Horel and Wallace 1981;Hoskins and  Karoly 1981). Therefore, in November the WNP precipitation anomalies can modulate temperature and precipitation variability in East Asia.
The precipitation difference between September and November shows negative in the WNP and positive in the equatorial CP (Fig. 4c). The WNP (110°-150° E, 0°-15° N) and equatorial CP (160E°-120° W, 5° S-5° N) precipitation anomalies in November are 1.46 and 1.42 times stronger than those in September, respectively. For the ENSO-related seasonal difference in atmospheric circulation, anticyclonic anomalies in East Asia accompany the southerly anomalies (Fig. 4f). The anticyclonic anomalies near East Asia in the circulation differences between September and November are directly linked to the temperature and precipitation results evident at weather stations in Fig. 3c, f. Based on the circulation patterns in September and November, the meridional wind directions show a tendency, from northerly to southerly anomalies, over East Asia (Fig. 4d, e). From September to November, the southerly tendency in East Asia occurs slower during El Niño, and faster during La Niña. These results clearly demonstrate that the seasonal transition of atmospheric circulation over East Asia, from summer to winter, is relatively rapid during the La Niña years. Therefore, in East Asia, the atmospheric tendency of ENSO developing warm and cold phases is mostly explained by the dynamical atmospheric responses to the different tropical/ subtropical precipitation forcings in September and November, respectively.
The same analysis as in Fig. 4 using CMIP5 simulations is repeated to further understand the ENSO-related seasonal transition in East Asia (Fig. 5). The CMIP5 models simulate the observed precipitation pattern reasonably well, exhibiting the negative anomalies in the western Pacific and zonally-extended positive anomalies spanning the equatorial central to eastern Pacific in September and November (Fig. 5a, b). Although the CMIP5 models represent well the dominant observed features, the ENSO-induced precipitation centers in the tropics tend to be shifted westward (Ham and Kug 2015). In addition, the CMIP5 models also underestimate the magnitude of regressed precipitation anomalies related to the El Niño forcing (Kim et al. 2017).
In September, the negative precipitation anomalies are confined south of the equator, while the equatorial CP and subtropical positive precipitation anomalies exist (Fig. 5a) though they are weaker than the observed (Fig. 4a). The weaker subtropical precipitation anomalies lead to weaker cyclonic circulation anomalies over the subtropical North Pacific, and their center locates further northeastward (Fig. 5d) compared to the observations (Fig. 4d). The resultant northerly wind anomalies on the western side of cyclonic circulation are also weaker but remain significant near East Asia. In November the CMIP5 models tend to simulate well the negative precipitation anomalies in the WNP and positives in the equatorial CP, although they underestimate the magnitude (Fig. 5b) compared to the observed pattern (Fig. 4b). As a result, the anticyclonic anomalies are similarly weaker and confined in the subtropics south of 40°N (Fig. 5e) than the observed (Fig. 4e). It is also shown here that the subtropical positive precipitation anomalies, the dominant feature of September, diminish during November in the case of the CMIP5 model pattern.
From the ENSO-related precipitation difference between September and November, the CMIP5 models reproduce well the strengthening of positive precipitation anomalies in the equatorial CP (Fig. 5c). In the WNP region, the negative precipitation anomalies also intensify between September and November, consistent with the observations (Fig. 4c). Therefore, the atmospheric differences between September and November show the anticyclonic anomalies near East Asia accompanying the southerlies on the western side of circulation (Fig. 5f). These results suggest that the CMIP5 models have an ability in reproducing the observed rapid transition in East Asia during the La Niña years, although the simulated magnitudes are relatively weaker than those of the observed.
The resultant temperature and precipitation anomalies in East Asia using the CMIP5 models (Fig. S3) are also weaker compared to those of the observed in Fig. 3. These systematic errors in ENSO teleconnection towards East Asia can be explained by the weak tropical precipitation response to a given SST forcing in the CMIP5 models (Kim et al. 2017). There are significant differences between September and November over China in the CMIP5 models, but these signals are not observed. However, the CMIP5 models simulate well the regressed temperature and precipitation patterns particularly over the Korea-Japan region, and partly support the observational arguments.

Role of seasonal transition of tropical precipitation on teleconnections in East Asia
To further investigate the relationship between the tropical/subtropical precipitation transition and teleconnection response, the correlation between the precipitation transition (i.e., difference between subtropical precipitation index in September and WNP precipitation index in November) and temperature transition in East Asia is shown in Fig. 6. Overall, the significant negative anomalies are dominant in East Asia especially centered over the Korea-Japan region. This indicates when the precipitation change is strong in the tropics/subtropics, the temperature transition over Korea-Japan becomes faster from September to November. Therefore, the distinct transition from the disappearance of the positive subtropical precipitation in September to the establishment of the negative WNP precipitation anomalies in November can be considered to play an important role in inducing the atmospheric teleconnection and its impacts on East Asia. The same analysis in Fig. 6 is also repeated in the CMIP5 simulations to investigate the relationship between the tropical/subtropical precipitation transition and temperature responses in East Asia (Fig. S4). The CMIP5 models represent well the observational pattern with weak but significant negative correlations over Korea-Japan implying that the tropical precipitation transition can contribute to the atmospheric responses in East Asia from September to November. The anomalous precipitation transition in the tropics/ subtropics is considered to be responsible for the transition of the atmospheric circulation between September Fig. 6 Correlation coefficient (shading) between difference of subtropical precipitation index in September and WNP precipitation index in November, and air temperature difference of September and November. Values over the 95% confidence level based on the student's t test are stippled and November. However, the climatological mean state change between September and November also can play an important role in inducing the anomalous atmospheric circulation. We check what factor plays a more dominant role between the basic state and tropical precipitation forcing in teleconnection changes using the LBM (Watanabe and Kimoto 2000). The LBM experiments are carried out using the prescribed forcings (0°-360° E, 20° S-20° N) obtained from the regressed tropical precipitation with respect to the SON Niño3.4 SST in September (Fig. 4a) and November (Fig. 4b), with basic states in September and November, respectively. When we use the September forcing with the September basic states, the LBM simulates the subtropical cyclonic circulation (Fig. 7a), as suggested in the observed (Fig. 4d). With the November tropical forcing and basic states, the model simulates the anticyclonic circulation in East Asia (Fig. 7b), similar to the observed pattern (Fig. 4e). For the simulation difference between September and November, the anticyclonic tendency is evident in East Asia (Fig. 7c), representing well the observed faster transition during the La Niña years (Fig. 4f). Therefore, the LBM has an ability in simulating the atmospheric teleconnection patterns during ENSO developing phase.
If we replace the basic state by the difference between September and November in the September forcing simulation, the anticyclonic circulation seems to be weak and confined south of 40° N (Fig. 7d). The delta basic state indicates the effect of the basic state changes, derived from the difference between the September forcing/November basic state experiment and the September forcing/September basic state experiment. However, the tropical forcing is replaced by the difference between September and November for the given September basic state, the anticyclonic circulation clearly locates near East Asia (Fig. 7e). These simulations which replace the forcing and basic state between September and November have a problem in representing the atmospheric teleconnections toward East Asia, compared to that of the original simulation difference (Fig. 7c). Although the magnitude is weak in the replaced simulations, the LBM results imply that the tropical forcing can be responsible for the teleconnection evolution between September and November. Of course, our linear analyses here have a limitation to sufficiently consider the effect of the basic states due to strong nonlinearity. Therefore, further investigation is needed to quantify how much two factors contribute to the seasonal atmospheric transition related to the ENSO in the coupled model experiments.

Intermodel diversity of seasonal transition in East Asia
To examine the individual CMIP5 model performance in simulating the seasonal transition in East Asia, Fig. 8 shows the regressed air temperature tendency over the Korea-Japan reproduces the positive temperature tendency over the Korea-Japan region but underestimates the observational magnitude accounting for only 21%. The seasonal transitions over the Korea-Japan region from September to November among the 40 individual models are diverse. Interestingly, most models simulate the positive magnitude apart from only 6 models, but no individual models exceed the observed strength.
The 40 CMIP5 models simulate diverse ENSO-related seasonal transitions in East Asia (Fig. 8) owing to a large diversity in the simulated tropical precipitation responses to the ENSO forcing among the CMIP5 models (Kim et al. 2017). To examine what determines the speed of seasonal transition in East Asia associated with ENSO, we classified the models into two groups, 10 good models ( Fig. 8. Note that the poor models include the 6 models which reproduce the negative transition sign opposite to the observed result. Figures 9 and 10 display the multi-model composites of the regressed atmospheric teleconnection and precipitation pattern, respectively, with Niño3.4 SST for both the good and poor model cases. In the good models, the dominant cyclonic anomalies over the North Pacific induce the northerly wind anomalies near East Asia during September (Fig. 9a), although the center of anomalous cyclonic circulation locates more northeastward than observed (Fig. 4b).
The regressed precipitation anomalies for the good models show that the positive precipitation anomalies excessively extend to the western Pacific (Fig. 10a). Thus the observed negative precipitation anomalies in the western Pacific are no longer evident in the good models. The distinct positive precipitation anomalies specifically located in the subtropical North Pacific are weak but not significant among the good models. In November, anticyclonic anomalies in East Asia accompany the southerly anomalies on the western side of the circulation feature (Fig. 9b), suggesting the possible role of the negative WNP precipitation anomalies. The negative precipitation anomalies are shown to be dominant in the WNP for the good models in November (Fig. 10b).
The atmospheric circulation tendency between September and November during the El Niño developing phase demonstrates that the anomalous southerly wind is associated with the strong anticyclonic anomalies in East Asia (Fig. 9c), consistent with the observed pattern (Fig. 4f). In the precipitation difference between September and November, the negative difference in the WNP is clearly evident (Fig. 10c). It is conceived that for the good models the seasonal transition pattern in the WNP is most realistic (Fig. 4c). Therefore, the strong anticyclonic tendency with the southerly anomalies in East Asia between September and November is presumably a response to the strengthening of negative precipitation anomalies in the WNP. These results suggest that the good models simulated well the observed rapid seasonal transition of atmospheric circulation in East Asia during the La Niña developing phases due to the realistic evolution of tropical precipitation anomalies.
Conversely, the poor models simulate the cyclonic anomalies over the North Pacific north of 40°N in September (Fig. 9d). The cyclonic anomalies induce significant westerly wind anomalies on the southwestern side of circulation in East   Fig. 8 The area-averaged regression differences (November-September) in air temperature over the Korea-Japan region (120°-145° E, 30°-40° N) regressed with respect to SON Niño3.4 (170°-120° W, 5° S-5° N) SST from observations (blue), CMIP5 MME (red), and individual models (pink). Black intervals indicate the 95% confidence level based on the student's t test Asia, quite different from the observed teleconnection pattern (Fig. 4d). The poor models simulate the WNP precipitation anomalies close to zero, and the observed positive precipitation anomalies in the subtropical North Pacific are not clearly captured (Fig. 10d). In November, the anticyclone anomalies are confined south of 30° N (Fig. 9e), therefore difficult to have an influence on the East Asian climate. In the atmospheric circulation tendency between September and November, the poor models simulate insignificant atmospheric circulation in East Asia (Fig. 9f). The ENSO-related negative precipitation anomalies in the WNP become stronger from September to November (Fig. 10f), but their magnitude in their difference is weaker than in the good models (Fig. 10c). As a result, the poor models do not simulate sufficient seasonal transition of atmospheric circulation in East Asia during September and November. Therefore, the good models have better performance in the seasonal transition of atmospheric teleconnection in East Asia associated with ENSO compared to the poor models.
The results using CMIP5 models have important implications for the model fidelity of ENSO teleconnection in East Asia. In particular, the good models simulate patterns that resemble more the observed tropical/subtropical precipitation anomalies which play a dominant role in linking ENSO to the East Asian climate system. A reasonable inference is that the good models reproduce well the observed seasonal transition from boreal summer to winter in East Asia, further supporting the importance of diabatic forcing in the tropical and subtropical North Pacific. Fig. 9 Composites of regressed geopotential height (shading; m), and wind (vector; m/s) anomalies at 850 hPa for the good (left panels) and poor (right panels) models with respect to the SON Niño3.4 (170°-120° W, 5° S-5° N) SST in a, d September, b, e November, and for c, f difference between November and September. Values over the 95% confidence level based on the student's t test are represented as vectors. The good and poor models are classified as the 10 highest and 10 lowest models, respectively, based on the regression values in Fig. 8

Summary and discussion
The seasonal transitions of East Asian climate during the ENSO developing warm and cold phases are investigated using observations and CMIP5 models. It is clearly shown that during the La Niña years, East Asia experiences a rapid seasonal transition from summer to winter, with a later start and earlier termination of fall, than those during El Niño years. Further, the evident difference in fall duration between El Niño and La Niña years can be dynamically explained by the seasonally different ENSO teleconnection. In September, East Asia experiences northerly wind anomalies that transport cold and dry advection as a direct response to the distinct El Niño-related positive precipitation anomalies in the subtropical North Pacific only in this month. Conversely, in November the WNP precipitation forcing is critical in developing the anticyclonic anomalies in East Asia with anomalous southerly winds associated with El Niño. In the El Niño-related seasonal transition for atmospheric circulation from September to November, the southerly anomalies in East Asia indicate that the northerly tendency (i.e., the seasonal transition of the background state) slows down during El Niño years; that is, the tendency increases during La Niña years.
The CMIP5 models simulate well the overall observed tropical/subtropical precipitation and atmospheric teleconnection patterns associated with ENSO in September and November, respectively. However, it is evident that the CMIP5 models tend to underestimate the atmospheric circulation in East Asia because the simulated tropical/subtropical precipitation anomalies are on average somewhat weaker and shifted westward than observed. In addition, the seasonal transition in East Asia is rather diverse from model to model, indicating that the transition magnitudes of the CMIP5 models are mostly weaker than observed. The performance of CMIP5 models in simulating the seasonal transition in East Asia is still limited, but some models (i.e., here the good model cases) tend to simulate more realistic Fig. 10 Composites of regressed precipitation (mm/day) anomalies for the good (left panels) and poor (right panels) models with respect to the SON Niño3.4 (170°-120° W, 5° S-5° N) SST in a, d September, b, e November, and for c, f difference between November and September Values over the 95% confidence level based on the student's t test are stippled. The good and poor models are classified as the 10 highest and 10 lowest models, respectively, based on the regression values in Fig. 8 tropical precipitation and atmospheric responses to ENSO forcing, supporting the observational findings.
However, most CMIP5 models have problems in simulating the observed precipitation anomalies located in the subtropical North Pacific associated with ENSO that are distinct only in September. In a comparison of the model simulations with the observed, there are several systematic errors in simulating the ENSO-related tropical/subtropical precipitation. The CMIP5 models tend to have a stronger Intertropical Convergence Zone (ITCZ) in the central and eastern part of off-equatorial regions and weaker along the equatorial regions, than the observed in September. This pattern results from the well-known biases in the coupled models, referred to as the double ITCZ and cold tongue biases (Wittenberg et al. 2006). Although the CMIP5 models overestimate the climatological precipitation in the subtropics during September, the simulated precipitation responses to the El Niño forcing in the subtropical North Pacific are not evident. One may suggest that the precipitation variability in the subtropical North Pacific is not closely related to the El Niño itself, given the correlation coefficient with Niño3.4 SST in September is only 0.20. However, the correlation with the observed indices for the 40-year period examined is much higher (i.e., 0.54) than that of the CMIP5 MME. As the El Niño-induced precipitation anomalies in September are mostly confined in the equatorial Pacific, it is meaningful to have a better understanding of the relationships between tropical ENSO SST and subtropical precipitation biases in model simulations.
In this study, we conclude the linkage of tropical/subtropical precipitation and atmospheric teleconnection associated with ENSO is important for understanding the seasonal transition in East Asia. However, a question still arises of why the teleconnection responses in East Asia to El Niño and La Niña years are slightly different? Recently, Luo and Lau (2020) suggested that northeast Asia experiences more (less) frequent summer extreme heat weather during the La Niña (El Niño) developing phases as a result of the circumglobal teleconnection (CGT) pattern associated with ENSO. According to Hardiman et al. (2018), the response of southeastern China summer rainfall to El Niño/La Niña is found to be asymmetric, with no significant response following La Niña years. Therefore, it is likely that the East Asian climate does not respond to the El Niño and La Niña forcing linearly. Although the different transition speeds of fall in East Asia associated with El Niño and La Niña developing phases are the focus of the current study, the detailed spatial pattern and intensity of tropical/subtropical diabatic forcing between El Niño and La Niña, and its impact on East Asia, will be pursued in further study.