Persistent Variations in the East Asian Trough from March to April and the Possible Mechanism

Over the mid-latitudes of the Northern Hemisphere, there is a quasi-stationary, long-wave trough in the mid-troposphere over East Asia, known as the East Asian Trough (EAT) (Zou et al., 1991; Chang et al., 2007; Huang et al., 2012). A strengthened EAT steers the southward invasion of a cold air mass from the Siberian-Mongolian region, causing cold outbreaks/surges (Lau and Lau, 1984; Chang et al., 2007; Cheung et al., 2015) and extreme precipitation/snowfall events (Gu et al., 2008; Sun et al., 2009) over East Asia. The EAT-related extreme events severely threaten social activity and human health. Therefore, it is important to research the EAT variations and related mechanisms.

The EAT is strong in winter; therefore, previous studies conducted extensive research on the influence and causes of winter EAT variations. In winter, the strengthened EAT induces cold anomalies along the coast of China, the Korean Peninsula, and Japan (Yang et al., 2005; Gao, 2007; Wang and Chen, 2010; Leung et al., 2017); in addition, precipitation will be less than normal over central-eastern China, South Korea, and southern Japan (Wang and Chen, 2010; Huang et al., 2013). The variability of the winter EAT is directly influenced by East Asian local factors, such as the Siberian High (Wu and Wang, 2002; Zhu et al., 2019), East Asian subtropical westerly jet stream (Yang et al., 2002; Jhun and Lee, 2004; Wu and Sun, 2017), and the sea surface temperature (SST) anomaly over the western North Pacific (Sun et al., 2016; Lei et al., 2020). In addition, some remote factors, like the El Niño-Southern Oscillation (ENSO) (Wang et al., 2000; Wang and He, 2012; Leung et al., 2017; Yu and Sun, 2020), Arctic Oscillation (AO) (Gong et al., 2001; Wang et al., 2005a; He et al., 2017), Eurasian teleconnection patterns (Liu et al., 2014; Oh et al., 2017), and Eurasian snow cover anomalies (Chen and Sun, 2003; Luo and Wang, 2019), have also been reported to affect the EAT.

Compared with winter, the EAT in spring becomes weaker; however, it still considerably influences the East Asian climate. For example, a strengthened EAT could cause a remarkable temperature decrease in Northeast China (Wang et al., 2005b; Dong et al., 2015). The spring EAT intensity is negatively related to the precipitation anomalies over the middle-to-lower reaches of the Yangtze River (Lu et al., 2014; Zhang and Sun, 2018), eastern Southwest China (Fan et al., 2016; Nan et al., 2022), and Northwest China (Lu, 2001; Sun and Guo, 2003). In terms of the impact factors, ENSO (Yuan and Yang, 2012) and the SST anomalies over the tropical Indian Ocean (Zhang and Sun, 2018) and the North Atlantic (Chen et al., 2018) are responsible for the spring EAT variability.

The above-mentioned studies have deepened our understanding of the spring EAT variation. However, the seasonal mean EAT does not reflect some important information, such as the month-to-month variation of the EAT. In retrospect, some features of the subseasonal variation in the East Asian winter climate have been revealed in many recent studies. For example, Xu et al. (2018) analyzed a noticeable subseasonal reversal of the winter surface air temperature over East Asia in 2014 and argued that this phenomenon was jointly caused by below-normal Arctic Sea ice in the previous September-October and a warm tropical central Pacific SST anomaly in winter. Lü et al. (2019) reported a phase-shift mode of the Siberian High from November to December-January, closely linked with the September Arctic Sea ice anomaly. Currently, the spring EAT month-to-month variation and the resultant climate anomalies have not yet been systematically studied. Spring is the plowing season in East Asia, and crop growth at this time is sensitive to air temperature and precipitation changes. Consequently, some questions arise: What is the relationship between the EATs in the three spring months? What is the intrinsic predominant mode of the spring EAT month-to-month variation and its possible mechanism? What are the resultant East Asian climate anomalies of the predominant mode? Motivated by these questions, this study aims to deepen our comprehension of the spring EAT month-to-month variation.

The remainder of this paper is organized as follows: the data and methods are introduced in section 2. Characteristics of the spring EAT month-to-month variation and the related climate anomalies are clarified in section 3. Section 4 investigates the possible factors responsible for the predominant mode of the spring EAT month-to-month variation. A discussion and conclusion are provided in section 5.

The monthly geopotential height, three-dimensional wind fields, and air temperature are from the Japanese 55-year reanalysis (JRA55, Kobayashi et al., 2015). The JRA55 reanalysis has a 1.25° × 1.25° horizontal resolution. The fifth generation ECMWF reanalysis product (ERA5) with a 1.0° × 1.0° horizontal resolution is also used to verify the results from the JRA55 reanalysis (Hersbach et al., 2020). The monthly SST data, also with a horizontal resolution of 1.0° × 1.0°, are from the U.K. Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) dataset (Rayner et al. 2003). Monthly land surface air temperature and precipitation data, with a 0.5° × 0.5° horizontal resolution, are from the Climate Research Unit (CRU) (Harris et al. 2020). The monthly snow cover extent data, with horizontal 89 × 89 grids, are from the Rutgers Global Snow Laboratory (Estilow et al., 2015). The monthly snow depth data, with a horizontal resolution of 0.5° × 0.5°, are from the land component of the ERA5 product (ERA5-land) (Muñoz-Sabater et al., 2021). The monthly AO index is obtained from https://www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/monthly.ao.index.b50.current.ascii.table.

The monthly simulations from the Community Earth System Model-Large Ensemble project (CESM-LENS; Kay et al. 2015) are also used to verify the reanalysis results. This project includes forty ensembles, run under historical forcing with different atmospheric initial conditions from 1920 to 2005. The horizontal resolution of the atmospheric component is approximately 1.0°. We give equal weight to each member and obtain the ensemble mean.

The analysis period in this study is from 1961 to 2020. However, the snow cover extent data is slightly shorter, from 1967 to 2020. The entire period of the CESM-LENS simulations is analyzed in this study. All monthly data are subtracted by their calendar monthly mean to obtain anomalies, and their long-term trends are also removed. A two-tailed Student’s t-test is used to estimate statistical significance. The spring season indicates the three months of March, April, and May. The area-averaged sign-reversed 500-hPa geopotential height anomalies over (25°–50°N, 110°–150°E) are defined as the EAT index, consistent with Yu et al. (2023).

The extended empirical orthogonal function (EEOF) method can efficiently separate the temporal-dependent modes of one variable in sequential time (e.g., Wang and An, 2005; Si and Ding, 2012; Zhang et al., 2022; Yu and Sun, 2023). In this study, anomalous geopotential heights in March and April are used to construct the matrix for EEOF analysis. The rule in North et al. (1982) evaluates the eigenvalue separation. The horizontal component of wave activity flux is calculated according to Takaya and Nakamura (2001). The anomalous Rossby wave source is calculated according to Sardeshmukh and Hoskins (1988) and is filtered by a spherical harmonic transformation (Sardeshmukh and Hoskins, 1984) to highlight the large-scale features.

The EAT indices in March, April, and May are shown in Fig. 1. These indices present noticeable interannual variability; the JRA55 and ERA5 reanalysis results are almost identical. We calculate the Pearson correlation coefficients between the EAT indices in different months to reflect the relationship among the EATs in spring months. The result shows that the March EAT variation is significantly correlated with the April one, with a coefficient of 0.38. In addition, the April and May EAT indices are also significantly correlated with each other, with a correlation coefficient of 0.46. Besides the relationship between the EATs in consecutive two months, we also analyze the relationship between the March and May EATs. The index correlation coefficient is weak and insignificant at 0.21.

Figure 1.  Standardized (a) March, (b) April, and (c) May EAT indices in 1961–2020 derived from JRA55 (black line) and ERA5 (red line).

Figure 2 shows the scatterplots of the March−April and April−May EAT indices. The boundaries of the dashed box in the figure are estimated by the percentile method (Wilks, 2019). For example, the upper and lower boundaries of the box in Fig. 2a are calculated as the $ Q3+1.5(Q3-Q1) $ and $ Q1-1.5(Q3-Q1 $), respectively. $ Q3 $ and $ Q1 $ denote the 75th and 25th percentiles of the standardized April EAT index. In Fig. 2, two red dots are outside each dashed box. The two dots indicate extreme values of the indices, which would threaten the robustness of the Pearson correlation coefficient because the correlation method is sensitive to extreme values (Wilks, 2019). Accordingly, we re-calculate the Pearson correlation coefficients after removing the red dots in Figs. 2a and 2b, respectively. After that, the correlation between the March and April EAT indices is almost unchangeable, with a value of 0.37, significant at the 95% confidence level. Aside from the two red dots, there are still two gray dots in the third quadrant of Fig. 2a. which also seem relatively far away from the other dots within the purple box. To strictly determine whether the two dots affect the relationship between the March and April EATs, we re-calculate the correlation coefficient between the March and April EAT indices after further removing the two gray dots. The correlation becomes 0.29, still significant at the 95% confidence level, further confirming the robustness of the relationship between March and April EATs. However, the re-calculated correlation coefficient between the April and May EAT indices substantially decreases to 0.25, which is insignificant. This result suggests that the robust relationship of the EAT variations is only realized between March and April.

Figure 2.  (a) Scatterplot of the standardized March and April EAT indices during 1961–2020. The solid and long-dashed (short-dashed) lines depict the linear regression of the April EAT index against the March EAT index during all years and during the remaining years after removing the years with the red (red and gray) dots. (b) Same as in (a) but for the standardized April and May EAT indices. Red dots denote the outliers. The boundaries of the purple dashed boxes are estimated by the method introduced in section 3, following Wilks (2019).

In addition, to prevent reanalysis-dependent results, we repeat the calculations using the ERA5 reanalysis. The results from the ERA5 reanalysis are identical to those from the JRA55 reanalysis, further confirming the robust relationship between the March and April EAT variations. Therefore, in the following analysis, we only focus on the EAT variation from March to April.

The positive correlation between the March and April EAT indices suggests the potential persistence of the EAT anomaly from March to April. To better describe this characteristic, we perform the EEOF analysis on the 500-hPa geopotential height anomalies over the EAT key region from March to April. The first EEOF (EEOF1) is distinctly separated based on the rule of North (North et al., 1982) and presents a persistent mode of the anomalous EAT from March to April (Figs 3a, b). The pattern in Figs. 3a and 3b is regarded as the positive EEOF1 in this study. The EEOF1 explains 38.63% of the total variance in the EAT month-to-month variation. The standardized PC1 is displayed in Fig. 3c, which exhibits a noticeable interannual variability. The correlation coefficient between PC1 and the March (April) EAT index is 0.82 (0.82), indicating the EEOF1’s capacity to describe the March and April EAT variations.

Figure 3.  Spatial pattern for EEOF1 of the monthly 500-hPa geopotential height anomalies over the EAT key region in (a) March and (b) April for the period 1961–2020. The corresponding standardized PC1 is shown in (c).

The positive EEOF1 is associated with a persistent and strengthened EAT. Corresponding to a persistent and strengthened EAT, the surface air temperature shows cold anomalies over East Asia in March and April (Figs. 4a, b). For precipitation, persistent and significantly reduced precipitation occurs over central-eastern China, South Korea, and southern Japan (Figs. 4c, d). The persistent temperature and precipitation anomalies in March−April could threaten the local social and agricultural activities more seriously than the anomalies in the individual month. Therefore, it is crucial to investigate the EAT month-to-month variations from March to April. Moreover, considering the potential persistence of the EAT anomalies from March to April, we examine the relationship between the March EAT and the April surface air temperature/precipitation (figure not shown). The strengthened March EAT could lead to significant cooling over northeast China and Japan in April, similar to the regressed results with respect to the positive PC1 (Fig. 4b). The pattern of the regressed April precipitation anomalies against the March EAT index is similar to that against PC1 (Fig. 4d), albeit with relatively weaker statistical significance. We can find an associated significant reduction in the precipitation over the middle reaches of the Yangtze River, North China, and southern Japan. These results imply the potential value of the March EAT on the April East Asian climate prediction, possibly achieved by its monthly persistence.

Figure 4.  Regressed (a) March and (b) April surface air temperature anomalies (units: °C) against PC1 during 1961–2020. (c−d) Same as in (a−b) but for the regressed precipitation anomalies (units: mm month^–1). Values significant at the 95% confidence level are dotted.

Many previous studies have shown the role of the boundary forcing of SST in causing persistent climatic anomalies (e.g., Wu et al., 2009; Li et al., 2012; Zhang et al., 2022; Yu and Sun, 2023) due to an SST known to have low variance. Therefore, the relationships between the EAT EEOF1 and the SST anomaly are analyzed.

The correlation coefficients between PC1 and the SST anomalies in March and April are shown in Fig. 5. From March to April, we see a persistent dipole SST pattern over the North Atlantic (Figs. 5a, b). Accordingly, an index measuring the dipole SST pattern over the North Atlantic is defined as the difference between the area-averaged SST anomalies over the regions bounded by (42.5°–57.5°N, 60.5°–20.5°W) and (20.5°–37.5°N, 80.5°–40.5°W) (boxes in Fig. 5a), marked as the NASSTI. The correlation coefficient between PC1 and the March (April) NASSTI is 0.40 (0.40), significant at the 95% (95%) confidence level. In addition, the correlation coefficient of the EAT index with the NASSTI in March (April) is 0.37 (0.32), significant at the 95% (95%) confidence level. The correlation between the March and April NASSTI is 0.83, indicating that the North Atlantic dipole SST pattern may impact the EAT EEOF1 through its robust monthly persistence from March to April.

Figure 5.  Correlation coefficients between PC1 and SST anomalies in (a) March and (b) April during 1961–2020. (c–d) Same as in (a–b) but between PC1_res and SST anomalies. Values significant at the 95% confidence level are dotted. The red rectangles are the domains used to define the specific SST indices.

To find more SST signals related to the EAT EEOF1, but independent of the dipole SST pattern over the North Atlantic, we further conduct the following analysis. The linear regression method is applied to remove the March NASSTI-related part from PC1 (marked as PC1_res), and the remaining signal is independent of the variability of the dipole SST pattern over the North Atlantic. The correlation coefficients between PC1_res and SST anomalies in March and April are then calculated. As shown in Figs. 5c and 5d, significant correlation coefficients exist over the tropical Indian Ocean in March and April. Accordingly, another index measuring the tropical Indian Ocean SST variation is defined as the area-averaged, sign-reversed SST anomalies over the region bounded by (15.5°S–5.5°N and 50.5°–100.5°E) (box in Fig. 5c), marked as the TIOSSTI. The correlation coefficient between PC1_res and the March (April) TIOSSTI is 0.31 (0.31), significant at the 95% (95%) confidence level. The correlation coefficient of the EAT index with the TIOSSTI in March (April) is 0.19 (0.33), which is insignificant (significant) at the 95% confidence level. The reason for the insignificant correlation coefficient in March will be explained in the following analysis. The correlation between the March and April TIOSSTI is 0.92, suggesting that the persistent tropical Indian Ocean SST anomalies likely contribute to the persistent EAT anomalies in March and April.

Aside from the SST anomalies over the North Atlantic and tropical Indian Oceans, there are also significant large-scale SST signals over the North Pacific in April (Figs. 5b and 5d). The SST anomalies over the region could be related to the EAT variations in April, but they cannot influence the preceding March EAT; therefore, we will not discuss the influence of the SST anomalies over the North Pacific because this study aims to explore the possible mechanism for the persistent variations of the EAT from March to April.

Further correlation analysis shows that the coefficients between the NASSTI and TIOSSTI in March and April are only –0.16 and –0.06, which indicates that the dipole SST pattern over the North Atlantic and anomalous SST over the tropical Indian Ocean are independent factors impacting the EAT EEOF1.

The physical processes responsible for the influence of the aforementioned two SST patterns on the EAT EEOF1 are then analyzed. Figure 6 displays the regressed anomalous horizontal winds, Rossby wave source, geopotential heights against the NASSTI, and the associated wave activity flux in March and April. Related to the positive NASSTI, is a negative NAO-like pattern with an equivalent barotropic structure over the North Atlantic (Fig. 6). Concurrently, a dipole pattern of the anomalous Rossby wave source is over the region slightly downstream of the negative NAO for these two months (Figs. 6b, d). As a result, a Rossby wave train is triggered, emanating from the mid-latitude North Atlantic, propagating eastward to East Asia, and finally strengthening the EAT. The persistent dipole SST pattern over the North Atlantic through this physical process could cause persistent EAT anomalies and contribute to the EEOF1 formation.

Figure 6.  (a) Regressed 500-hPa geopotential height anomalies (color; units: gpm) and 850-hPa horizontal wind anomalies (vector; units: m s^–1) against the standardized NASSTI in March during 1961–2020. (b) Same as in (a) but for the regressed 300-hPa geopotential height (contour; units: gpm) anomalies, anomalous Rossby wave source (color; units: 10^–11 s^–2), and the associated wave flux activity (vector; units: m² s^–2). (c–d) Same as in (a–b) but for the regression results in April. Values significant at the 95% confidence level are dotted. Only the anomalous 300-hPa Rossby wave source over the North Atlantic sector that is significant at the 95% confidence level is drawn. Black vectors in (a) and (c) are significant at the 95% confidence level. Vectors less than 0.05 m² s^–2 in (b) and (d) are not shown.

The atmospheric circulation anomalies associated with the TIOSSTI are shown in Fig. 7. Associated with the cold tropical Indian Ocean SST anomalies are regionally depressed vertical motions over the tropical Indian Ocean and intensified upward motion around the western tropical Pacific, which occur in March (Fig. 7a). Over the tropical regions, the vertical motion generally reflects the variation of the convective activity. Accordingly, the convective activity is depressed above the cold SST anomaly region in the tropical Indian Ocean and is strengthened over the western tropical Pacific. Such a change in the Walker circulation favors a low-level anomalous cyclone over the western tropical Pacific. In addition, the downward vertical velocity anomalies are associated with low-level divergence and further induce the low-level westerly wind anomalies over the tropical Indian Ocean. The resultant westerly wind prevails along the equator from the western Indian Ocean extending to the west of the Maritime Continent, contributing to the anomalous cyclone over the western tropical Pacific. The anomalous cyclone can cover the southern part of the EAT (Fig. 7b), leading to significant changes in southern EAT. Similar to in March, the cold tropical Indian Ocean SST anomalies in April can also strengthen the southern part of the EAT by changing the Walker Circulation over the tropical Indian Ocean to the tropical western Pacific (Figs. 7c, d). Accordingly, the persistent tropical Indian Ocean SST anomaly contributes to the EAT EEOF1.

Figure 7.  Regressed (a) 500-hPa omega (units: Pa s^–1) and (b) 500-hPa geopotential height anomalies (color; units: gpm) and the 850-hPa horizontal wind anomalies (vector; units: m s^–1) against the standardized TIOSSTI in March during 1961–2020. (c–d) Same as in (a–b) but for the regressed results in April. Values significant at the 95% confidence level are dotted. Vectors in (b) and (d) are significant at the 95% confidence level.

Compared to the regressed 500-hPa geopotential height anomalies over the EAT key region, we can find that the results with respect to the NASSTI are larger than (comparable to) the results with respect to the TIOSSTI in March (April). Such results are due to the different relationships of the EAT with the NASSTI and TIOSSTI in March and April. Overall, the impact of the dipole SST pattern over the North Atlantic on the EAT EEOF1 is larger than that of the tropical Indian Ocean SST anomaly.

The above subsection demonstrated the relationship between the EAT EEOF1 and the underlying SST anomalies. Aside from SST, other factors could potentially impact the EEOF1. As reviewed in the introduction section, the Eurasian snow cover variation can significantly influence the winter EAT. Along with the gradual warming in spring, the Eurasian snow cover substantially changes, especially over the Eurasian midlatitudes. Does the snow cover variability also influence the EAT EEOF1? To answer this question, we explore the key snow cover regions in this subsection.

Figure 8 depicts the regressed Eurasian snow cover anomalies against PC1 in March and April. In March, the positive EEOF1 is significantly related to the increased snow cover over central Europe and Northeast Asia (Fig. 8a). In April, only Northeast Asia is a key region with a significant snow cover-EEOF1 connection (Fig. 8b). To focus on the persistence characteristic, we regard Northeast Asia as the key region of the snow cover. Accordingly, a snow cover index measuring the Northeast Asian snow cover variation is defined as the area-averaged snow cover concentration anomalies over the region bounded by (42.5°–55°N and 110°–135°E) (box in Fig. 8a), marked as the NEASCI. The correlation coefficient between PC1 and the March (April) NEASCI is 0.46 (0.39), significant at the 95% (95%) confidence level. The correlation between the March and April NEASCI is 0.61, indicating good persistence of the Northeast Asian snow cover variation from March to April. In addition, Han and Sun (2021) have also reported the predictive value of the March Northeast Asian snow cover on the April EAT.

Figure 8.  Regressed (a) March and (b) April snow cover concentration anomalies (units: %) against PC1 during 1961–2020. Values significant at the 95% confidence level are shaded by dots. The black rectangle is the domain used to define the snow cover indices.

We further analyze the relationship between the Northeast Asian snow cover and the two SST factors introduced in section 4.1. The absolute correlation coefficients between the simultaneous NEASCI and TIOSSTI are less than 0.1 in March and April, indicating their independent variations. The correlation between NEASCI and NASSTI in March is insignificant, with a value of 0.26; while the correlation coefficient between NEASCI and NASSTI in April is 0.28, which is barely significant at the 95% confidence level. This result indicates a significant linkage between the Northeast Asian snow cover and the North Atlantic SST anomalies in April. Therefore, in the following mechanism analysis, we will also discuss the influence of the Northeast Asian snow cover in April after removing the signal of the dipole SST pattern over the North Atlantic.

The longitude-pressure cross-section of the regressed air temperature anomalies averaged along 42.5°–55°N against the NEASCI in March is shown in Fig. 9a. Significant cold air temperature anomalies are observed from the surface up to 300 hPa, with the minimum anomalies existing in the lower troposphere. Northeast Asia becomes a cold source when the regional snow cover is above normal. As a consequence, the 1000–500-hPa thickness significantly decreases (Fig. 9b), causing the northern part of the EAT to deepen (Fig. 9c). When the increased snow cover anomalies persist into April, the air temperature below 300 hPa (Fig. 9d) and the 1000–500-hPa thickness also decrease over Northeast Asia (Fig. 9e), also strengthening the northern part of the EAT (Fig. 9f). As a result of this process, the Northeast Asian snow cover could contribute to the EAT EEOF1 in March and April.

Figure 9.  (a) Longitude-pressure cross-section of the regressed air temperature anomalies (units: °C) averaged along 42.5°–55°N against the standardized NEASCI in March during 1961–2020. Regressed (b) 1000–500-hPa thickness anomalies (units: gpm) and (c) 500-hPa geopotential height anomalies (units: gpm) against the standardized NEASCI in March. (d–f) Same as in (a–c) but for the regressed results in April. Values significant at the 95% confidence level are dotted.

The above correlation analysis shows a significant simultaneous correlation between the NASSTI and the NEASCI in April. To distinguish the individual influence of the Northeast Asian snow cover, we remove the signals associated with the April NASSTI from the NEASCI and re-calculate the regression results in Figs. 9d–f. The re-calculated regression results (figures not shown) are highly consistent with that in Figs. 9d–f, indicating that the pronounced influence process is mainly attributed to the snow cover anomalies over Northeast Asia.

Summarizing the above results, we can see that the EAT EEOF1 is substantially influenced by the SST anomalies over the North Atlantic and tropical Indian Oceans as well as the snow cover anomalies over Northeast Asia. To further investigate their combined effects, we construct an index (referred to as the Mult_JRA55 index) by regressing PC1 against the NASSTI, TIOSSTI, and NEASCI in March. Limited by the length of the snow cover data, the available period of the Mult_JRA55 index is from 1967 to 2020. The Mult_JRA55 index is significantly correlated with PC1, with a coefficient of 0.60, significant at the 95% confidence level. In addition, the correlation coefficients of the Mult_JRA55 index with the March and April EAT indices are 0.49 and 0.46, respectively, both significant at the 95% confidence level. Noting that only March information is used to construct the Mult_JRA55 index, the Mult_JRA55 index not only describes the status of the March EAT but also provides the prediction information for the April EAT.

The above analyses have shown that the predominant mode of the March−April EAT variation features a persistent anomaly from March to April. Such a mode is closely related to two SST factors and one snow cover factor. To further consolidate the reanalysis results, we analyze the CESM-LENS simulation, derived from the ensemble mean result of 40 ensembles during the entire simulation period of 1920–2005 in this subsection.

Before analyzing the EAT EEOF1 in the CESM-LENS simulation, we first calculate the correlation coefficient between the EAT indices in March and April, which is 0.51 and significant at the 95% confidence level. Such a result confirms the possible persistence of the EAT anomaly from March to April. Figure 10 displays the EEOF1 of the EAT variation from March to April in the CESM-LENS simulation. There are visible differences in the EEOF1 details between simulation (Fig. 10) and reanalysis (Figs. 3a, b). For example, the centers of the anomalous EAT in EEOF1 are located further east in the model simulation than in the reanalysis, especially in April. However, the CESM-LENS simulation can generally reproduce a persistent mode of the EAT variation from March to April. We further analyze the relationship of PC1 with the East Asian temperature and precipitation in the CESM-LENS simulation. The corresponding regression results are shown in Fig. S1 in the Electronic Supplementary Material (ESM). In the CESM-LENS simulation, the persistent strengthened EAT in March−April is associated with the persistent cold surface air temperature anomalies over East Asia (Figs. S1a and S1b in the ESM), consistent with the observational results (Figs 4a, b). For precipitation, the positive EEOF1 could lead to significantly less precipitation over central-eastern China, South Korea, and southern Japan in March (Fig. S1c in the ESM), consistent with the observational results (Fig. 4c). In April, the EEOF1-related precipitation signals over southern Japan are maintained; however, significant signals are largely weakened over central-eastern China (Fig. S1d in the ESM), which differs from the observational result (Fig. 4d). This is possibly due to the simulated eastward displacement of the EEOF1 center in April (Fig. 10b).

Figure 10.  EEOF1 patterns of 500-hPa geopotential height anomalies over the EAT key region in (a) March and (b) April in the CESM-LENS simulation.

To confirm the influences of the SST and snow cover factors on the EAT EEOF1, we use the same index definitions as those in the reanalysis. The correlation coefficients between PC1 and the NASSTI, TIOSSTI, and NEASCI in March and April in the model simulation are shown in Table 1, all of which are significant at the 95% confidence level. Regarding the relationships among the impact factors, the SST anomalies over the tropical Indian Ocean are independent of the other two factors in March and April, as evidenced by their maximum absolute correlation coefficient being less than 0.13. The dipole SST pattern over the North Atlantic is closely related to the downstream Northeast Asian snow cover, with a correlation coefficient of 0.35 and 0.29 in March and April, respectively, both significant at the 95% confidence level. Therefore, in the following analysis, when conducting regression analysis against the NEASCI, the part related to the NASSTI will be linearly removed from the NEASCI (denoted as NEASCI_res) to concentrate on the individual influence of the Northeast Asian snow cover.

In the CESM-LENS simulation, a negative NAO-like pattern appears in March and April consistent with the reanalysis; this is associated with the positive dipole SST pattern over the North Atlantic (Fig. 11). As a consequence, a Rossby wave train is triggered, which propagates eastward to strengthen the EAT in both March and April. Such a process is also consistent with the reanalysis.

Figure 11.  (a) Regressed 500-hPa geopotential height anomalies (color; units: gpm) against the standardized NASSTI and the associated wave flux activity (vector; units: m² s^–2) in (a) March and (b) April in the CESM-LENS simulation. Values significant at the 95% confidence level are dotted. Vectors less than 2.5 × 10^–3 m² s^–2 are not shown.

The cold SST anomaly over the tropical Indian Ocean could cause anomalies of sinking motion over the tropical Indian Ocean and anomalies of rising motion over the western tropical Pacific (Figs. 12a, c). An anomalous cyclone is therefore established over the western tropical Pacific, which can strengthen the southern part of the EAT (Figs. 12b, d).

Figure 12.  Regressed (a) 500-hPa omega (units: Pa s^–1) and (b) 500-hPa geopotential height anomalies (color; units: gpm) and the 850-hPa horizontal wind anomalies (vector; units: m s^–1) against the standardized TIOSSTI in March in the CESM-LENS simulation. (c−d) Same as in (a−b) but for the regressed results in April. Values significant at the 95% confidence level are dotted. Vectors in (b) and (d) are significant at the 95% confidence level.

The increase in Northeast Asian snow cover leads to a regional cold source (Figs. 13a, d), significantly cooling down the air column above, as indicated by a decrease in the 850–500-hPa thickness (Figs. 13b, e) and causing negative 500-hPa geopotential height anomalies (Figs. 13c, f). The 850 hPa level is selected as the lowest layer for analysis to avoid the missing values beneath it due to the terrain in the model simulation. Accordingly, the persistent Northeast Asian snow cover could cause the anomalous EAT in March−April.

Figure 13.  (a) Longitude-pressure cross-section of the regressed air temperature anomalies (units: °C) averaged along 42.5°–55°N against the standardized NEASCI_res in March in the CESM-LENS simulation. Regressed (b) 850–500-hPa thickness anomalies (units: gpm) and (c) 500-hPa geopotential height anomalies (units: gpm) against the standardized NEASCI_res in March in the CESM-LENS simulation. (d–f) Same as in (a–c) but for the regressed results in April. Values significant at the 95% confidence level are dotted.

From the above analysis, we can see that the CESM-LENS simulation well reproduces the EAT EEOF1 from March to April. The impacts of the dipole SST pattern over the North Atlantic, the SST anomaly over the tropical Indian Ocean, and the snow cover anomaly over Northeast Asia on the EAT EEOF1 are also effective in the simulation. Such results further confirm the important roles of the three factors in the persistent variations of the EAT from March to April.

Observations and the CESM-LENS simulation have shown the significant impacts of the two SST factors and the one snow cover factor on the persistent EAT variation in March−April. In this subsection, we design more experiments to confirm the impacts of these factors on the EAT.

Three experiments are conducted to verify the influence of SSTs by the Community Atmospheric Model version 5 (CAM5). In the model configuration, we select the “F_2000_CAM5” as the component set and finite volume grid which has a horizontal resolution of 1.9°(lat) × 2.5°(lon) with 30 hybrid sigma pressure levels. The first CAM5 experiment is a control run (referred to as CAM5_CTL) and has a 45-year integration forced by the global climatological SST and sea ice concentration. The first five years are used for model spin-up, and the remaining 40-year simulation is then analyzed. The second CAM5 experiment is a North Atlantic SST-sensitivity experiment (referred to as CAM5_NASST). CAM5_NASST has an ensemble of 40 members; each starting from different initial conditions and integrated for three months from 1 February to 30 April. The integration with a one-month lead is designed to make a more coordinated connection between the atmosphere and boundary conditions in March−April. The 40 initial conditions are taken from the restart files of CAM5_CTL (every 1 February of each year the 40-year EXP_CTL). The SST boundary condition in CAM5_NASST is constructed by superimposing the monthly SST anomalies over the North Atlantic (20.5°–57.5°N, 80.5°–19.5°W) on the climatological monthly SST from February to April (Fig. S2 in the ESM). The superimposed monthly SST anomalies are the regressed SST anomalies against the standardized March NASSTI, which are then multiplied by 2 to highlight the SST forcing. In addition, to avoid unrealistic discontinuities, the superimposed anomalies are linearly decreased to 0°C across five grid points over the four borders. The third CAM5 experiment is a tropical Indian Ocean SST-sensitivity experiment (referred to as CAM5_TIOSST). This experiment is the same as CAM5_NASST, but the superimposed monthly SST anomalies are the regressed SST anomalies over the tropical Indian Ocean (15.5°S–5.5°N, 50.5°–100.5°E) against the standardized March TIOSSTI and then multiplied by 2 (Fig. S3 in the ESM).

The difference between the SST-sensitivity experiment and CAM5_CTL reflects the impacts of the specific SST forcing. As shown in Figs. 14a and 14b, we can see that the persistent positive dipole SST pattern over the North Atlantic could cause the persistent strengthened EAT in March and April through the triggered Rossby wave train, despite the relatively weak significance of the EAT response in March. The atmospheric circulation responses to the anomalous cooling tropical Indian Ocean are shown in Figs. 14c and 14d. The cold tropical Indian Ocean SST anomaly facilitates an anomalous cyclone over the western North Pacific and strengthens the EAT in March and April, albeit the EAT response is only weakly significant in April.

Figure 14.  Composite simulated (a) March and (b) April geopotential height (color; units: gpm) at 500 hPa between CAM5_NASST and CAM5_CTL and the associated 300-hPa wave activity fluxes (vector; units: m² s^–2). Composite simulated (c) March and (d) April 500-hPa geopotential height (color; units: gpm) and 850-hPa horizontal wind (vector; units: m s^–1) between CAM5_TIOSST and CAM5_CTL. Responses of the (e) March and (f) April 500-hPa geopotential height to the diabatic forcing in LBM_NEASC. Values in (a–d) significant at the 95% confidence level are dotted. Vectors less than 0.05 m² s^–2 are not shown in (a–b). Black vectors in (c–d) are significant at the 95% confidence level.

Because snow cover is a diagnostic variable in the CAM5, it is not easy to simulate the solo impact of the snow cover using this model. To explore the impact of the snow cover on the March−April EAT, we perform one experiment with the Linear Barotropic Model (LBM) (Watanabe and Jin, 2002). The dry LBM is often used to explore the impact of the snow via the diabatic heat forcing experiment (e.g., Jia et al., 2018; Ma et al., 2021). A horizontal resolution of T42 (equivalent to approximately 2.8°) with 20 vertical sigma levels is used in the LBM simulation. The LBM simulation is integrated for 20 days with a steady forcing. Since previous studies have reported that the atmospheric response to the prescribed mid-latitude forcing will be more stable and representative after 15 days (e.g., Jia et al., 2018; Ma et al., 2021; Yu et al., 2023), we analyze the averaged simulations over days 16–20.

To mimic the impact of the resultant regional cold source due to the more Northeast Asian snow cover, we design an LBM experiment with a diabatic heat forcing over Northeast Asia (42.5°–55°N, 110°–135°E), referred to as LBM_NEASC. The horizontal distributions and the vertical profiles of the heat forcing in LBM_NEASC are shown in Fig. S4 in the ESM. According to Figs. 9a and 9d, the heat forcing has a minimum of –1.0 K d^–1 at a height of σ = 0.830 (in σ-coordinate). In response to the persistent cooling over Northeast Asia, the northern part of the EAT strengthens in March and April (Figs. 14e, f).

Summarizing the above sensitivity experiment results, we can conclude that the two SST factors and one snow cover factor could contribute to the persistent EAT anomaly in March and April. Results in this subsection offer further support to the observation analysis and CESM-LENS simulation.

Boundary forcings are generally slow-varying features and are instrumental in climate prediction (Kim et al., 2012; Tian and Fan, 2015; Ma and Sun, 2021). Our analysis shows that the anomaly signals of the three factors for the EAT EEOF1 appear in late winter. Therefore, we evaluated the statistical prediction skill of the EAT EEOF1 using the information of the three factors in the preceding February. The correlation analysis indicates that the correlation coefficient between PC1 (PC1_res) and the February NASSTI (TIOSSTI) is 0.36 (0.27), significant at the 95% (95%) confidence level, suggesting the potential usage of February signals of the NASSTI and TIOSSTI in the prediction of the persistent variations in the EAT from March to April. However, the correlation between PC1 and the February NEASCI is insignificant, with a value of 0.20. One possible reason for such a weak correlation could be the weak variations in snow cover concentration over Northeast Asia in February. This hypothesis can be validated by the fact that the standard deviation of the February NEASCI is almost half of the March index. According to previous studies (e.g., Han and Sun, 2018), the snow depth is usually used as proxy data to reflect the snow cover changes when the snow cover concentration has little variability. We then defined an index as the area-averaged snow depth anomalies over the same region of the NEASCI and referred to it as the NEASDI. The correlation coefficient between PC1 and February NEASDI is 0.29, significant at the 95% confidence level. In addition, the correlation between the February NEASDI and March NEASCI is significant, with a value of 0.55, confirming the rationality using the February NEASDI as a proxy. We also checked the relationship between the EAT EEOF1 and the SST/snow anomaly in January, which has a 2-month lead. The correlation coefficient between PC1 (PC1_res) and the January NASSTI (TIOSSTI) is 0.25 (0.26), and 0.26 between PC1 and the January NEASDI. All of these correlation coefficients are significant, but less so than that of PC1 with the SST/snow anomaly in February. Therefore, it is more skillful to use the SST and February snow information in predicting the EAT EEOF1 in March−April.

In retrospect, He et al. (2017) pointed out that the winter AO is valuable for predicting the East Asian spring climate, such as the surface air temperature over Northeast China. Accordingly, we examine the relationship between the winter AO and the EAT EEOF1. The correlation coefficient between the January (February) AO index and PC1 is low at 0.04 (–0.14). Despite the weak linkage between the winter AO and the EAT EEOF1, we analyzed the relationship between the winter AO and the dipole SST pattern over the North Atlantic, the latter is one of the factors impacting the EAT EEOF1. The correlation coefficient between the January (February) AO index and the March/April NASSTI is –0.14/–0.05 (–0.31/–0.18). Only the correlation coefficient between the February AO index and the March NASSTI is significant at the 95% confidence level. However, compared to the high correlation coefficient of 0.89 between the NASSTI in February and March, the contribution of the February AO to the March North Atlantic SST anomalies is minor. The value for the spring North Atlantic SST anomaly prediction by the winter AO is lower. Therefore, the winter AO cannot be regarded as an effective predictor for the EAT EEOF1 in March−April.

Then, we used the leave-one-out cross-validation method to predict PC1. The predicted PC1 is shown as the red line in Fig. S5 in the ESM. The variability of the predicted PC1 is significantly correlated to the PC1, with a value of 0.35. Furthermore, we analyzed the prediction skill of PC1 by the leave-p-out cross-validation (p = three and five are used). The validation periods are consecutive and do not overlap. Like the leave-one-out cross-validation, the variability of the predicted PC1 constructed by the leave-three (five)-out cross-validation is also significantly correlated to PC1, with a value of 0.40 (0.42), implying that the EAT EEOF1 could be statistically and stably predictable based on the signal of the aforementioned three factors in the preceding February.

The EAT significantly influences the East Asian climate in spring. Therefore, it is important to systematically research the spring EAT variability. This study investigates the characteristics of the spring EAT month-to-month variation and the possible mechanisms.

A close and robust positive relationship exists between the March and April EAT. However, the relationship between the April and May EAT is non-robust. Consistent with the correlation analysis, the EEOF method shows that the predominant mode of the EAT variations from March to April features a persistent EAT anomaly in these two months. The EAT EEOF1 is significantly related to a dipole SST pattern over the North Atlantic SST. The positive dipole SST pattern, with warm SST anomalies east of Newfoundland Island and cold SST anomalies east of Bermuda, could trigger a Rossby wave train to further strengthen the EAT in March and April. Moreover, the SST anomaly over the tropical Indian Ocean could change the Walker circulation over the tropical Indian Ocean-tropical western Pacific and cause an anomalous circulation over the western tropical Pacific, leading to anomalies in the southern part of the EAT in March and April. Aside from the SST factors, the Northeast Asian snow cover could change the regional thermal conditions and lead to persistent EAT anomalies from March to April. Further analyses based on the model simulations confirm the role of the SST and snow cover factors in forming the EAT EEOF1.

Moreover, the signals of the above-mentioned three factors can be traced back to February. The cross-validation results suggest that these three factors provide a valuable source for predicting the EAT EEOF1 in March−April, indicating the potential use of these factors in the operational prediction of the persistence of the EAT anomalies from March to April.