Interdecadal Change in the Interannual Variation of the Western Edge of the Western North Pacific Subtropical High During Early Summer and the Influence of Tropical Sea Surface Temperature

The western North Pacific (WNP) subtropical high (WNPSH) has been identified as an important component of the East Asian summer monsoon system (Tao and Chen ^[1]; Wang and Chen ^[2]). The variations of the position, shape, and intensity of the WNPSH have important effects on the East Asian climate (Li and Lu ^[3]; Li et al. ^[4]; Wang et al. ^[5]). The western edge of the WNPSH is an important factor modulating the precipitation and temperature over China. On the one hand, southwesterlies prevail at the northwestern edge of the WNPSH, which could bring in abundant water vapor from the ocean and lead to above-normal precipitation (Chang et al. ^[6]; Yang and Sun ^[7]; Zhou and Yu ^[8]). On the other hand, subsidence dominates over the main body of the WNPSH and results in anomalous high temperature in situ (Luo and Lau ^[9]). For example, in the summer of 1998, the WNPSH was persistently located further westward and southward compared to the climatology, resulting in the persistent rain belt and consequent severe flood over the Yangtze River valley (Tao et al. ^[10]; Yang and Sun ^[11]). In the summer of 2017, the WNPSH abnormally intensified and was westwardextending, resulting in the anomalous high pressure and extreme high temperature over South China (Chen et al. ^[12]). Therefore, it is important to explore the variation of the western edge of the WNPSH, which is helpful for better understanding and prediction of the climate variation in the monsoonal region.

The western edge of the WNPSH exhibits clear interannual variation. An abnormally westwardextending (eastward-retreating) WNPSH is associated with an anomalous anticyclone (cyclone) over the WNP. Previous studies pointed out that the interannual variation of the western edge of the WNPSH is prominently affected by the tropical sea surface temperature anomalies (SSTAs) (Chen et al. ^[13]; Chung et al. ^[14]; Lu and Dong ^[15]). During El Niño events, the warm SSTA over the tropical central-eastern Pacific favors the warming over the Indian Ocean through the atmospheric bridge process (Alexander et al. ^[16]), which could persist until summer and favor an anticyclonic anomaly over the WNP via the Ekman divergence associated with the eastward-propagating Kelvin wave (Xie et al. ^[17]). The warm SSTA over the Maritime Continent could lead to local anomalous upward motion and further favor the anomalous subsidence over the WNP through a meridional-vertical circulation, resulting in the enhancement and westward extension of the WNPSH (Sui et al. ^[18]). Sui et al. further suggested that different interannual oscillation periods of the western edge of the WNPSH are related to different SSTA patterns, with the 2-3 years oscillation modulated by the SSTAs over the Maritime Continent and equatorial central-eastern Pacific while the 3-5 years oscillation by the SSTAs over the WNP, Indian Ocean, and equatorial eastern Pacific ^[18].

The interdecadal change and long-term trend of the western edge of the WNPSH have also received much attention. The geopotential height field shows that the WNPSH has obviously extended westward after the late 1970s, and the interdecadal warming of the Indian Ocean-western Pacific contribute to this interdecadal change (Ho et al. ^[19]; Hu ^[20]; Zhou et al. ^[21]). Recent studies argued that this increase of geopotential height is related to global warming, and the circulation of WNPSH actually retreats eastward after the late 1970s and tends to keep this trend under global warming (He et al. ^[22-23]; Huang et al. ^[24]; Wu and Wang ^[25]; Yang et al. ^[26]).

The above previous studies have revealed the characteristics and causes of the interannual variation and the interdecadal change in the mean state of the western edge of the WNPSH. On the other hand, many studies indicated that the interannual variation of the East Asian summer climate have witnessed obvious interdecadal change during the recent decades (Chen et al. ^[27]; He et al. ^[28]; Wang et al. ^[29]). Then, has the interannual variation of the western edge of the WNPSH also experienced interdecadal change in recent decades? In fact, Lu et al. have pointed out that the interannual variability of the western edge of the WNPSH obviously weakened in the late 1980s ^[30], but the relevant specific characteristics and plausible causes remain unclear. Therefore, this paper focuses on the recent interdecadal change in the interannual variation of the western edge of the WNPSH, aiming to reveal the specific characteristics of this interdecadal change and investigate the plausible causes from the perspective of the influence of tropical SSTAs. In addition, in consideration of the sub-seasonal variation of the WNPSH (Xue et al. ^[31]), this paper firstly compares the variations of the western edge of the WNPSH during different months, so as to better extract the relevant signals of interdecadal change.

The Japanese 55-year Reanalysis (JRA-55) monthly mean data provided by the Japan Meteorological Agency (JMA) is employed to analyze the atmospheric circulation, including the geopotential height, horizontal wind, and vertical velocity. The horizontal resolution is 1.25° × 1.25° and there are 37 vertical levels extending from 1000 hPa to 1 hPa (Harada et al. ^[32]; Kobayashi et al. ^[33]). The monthly mean SST is extracted from the Extended Reconstructed Sea Surface Temperature version 5 (ERSSTv5) provided by the National Oceanic and Atmospheric Administration (NOAA), which has a horizontal resolution of 2° × 2° (Huang et al. ^[34]). The analyzed period is 1958-2019.

A westward extension index is employed to depict the western edge of the WNPSH, which is defined as the most westward position of the 5880-gpm geopotential height contour over the region (10°-90°N, 90°E-180°). The same definition is adopted by the National Climate Center in China (NCC). In order to verify the westward extension index calculated based on the JRA-55 reanalysis data, the westward extension index time series released by the NCC is used for comparison (Liu et al. ^[35]). It is noted that the westward extension index time series from the NCC is calculated based on the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP / NCAR) reanalysis data. However, the NCEP/NCAR data has an unreal interdecadal change in the lower-tropospheric circulation around the late 1970s (Huang ^[36]). To guarantee the accuracy of the interdecadal change signals, the current study uses the JRA-55 reanalysis data to calculate the westward extension index of the WNPSH and analyze relevant atmospheric circulation. In the discussion section, in attempt to discuss the influence of global warming, this geopotential height index is detrended and together discussed with some other definitions that were previously proposed to eliminate the signal of global warming. The discussed definitions include: (1) 500-hPa relative vorticity index, defined as the 500-hPa relative vorticity averaged over (22.5°-30°N, 115°-140°E) (Yang and Sun ^[11]); (2) 850- hPa eddy geopotential height index, defined as the 850- hPa eddy geopotential height averaged over (15°-30°N, 120°-150° E), in which the eddy geopotential height is the difference between geopotential height and its zonal mean (Huang et al. ^[24]); (3) 500-hPa eddy geopotential height index, defined as the most westward position of the 0-gpm eddy geopotential height contour over 90°E- 180°. The eddy geopotential height is calculated by subtracting the average of the geopotential height over (0°-40°N, 180°W-180°E) from the original geopotential height (He et al. ^[22-23]); (4) 500-hPa eddy stream function index, defined as the most westward position of the 0-m² s^-1 eddy stream function contour over (10°-90°N, 90°E-180°). The eddy stream function is defined as the deviation of stream function from the regional average over (0°-40°N, 180°W-180°E) (He et al. ^[22]).

Composite analyses are performed on the differences between the abnormally westward-extending years and the abnormally eastward-retreating years of the WNPSH, so as to depict the anomalies associated with the interannual variation of the western edge of the WNPSH. The Student's t-test with a confidence level of 90% is used to estimate the significance of the composite analyses. The interannual variability of the westward extension index is measured by the standard deviation and the significance of its interdecadal change is examined by the F test. The decadal component of the westward extension index is extracted by the Lanczos filter with 11-year low pass. Spectrum analyses and Morlet wavelet power spectrum analyses with Markov red noise test are used to obtain the primary oscillation periods of ENSO.

Figure 1 shows the time series of the westward extension index of the WNPSH for different months in summer (Figs. 1a-h) and the corresponding 11-year sliding standard deviation (Figs. 1i-l). The index calculated based on the JRA-55 data exhibits similar interannual and interdecadal variations as the index released by the NCC, except for some differences in the specific values in each year. It is indicated that the index calculated based on the JRA-55 data is reliable and thus all the following analyses are based on it. As shown in Fig. 1, the westward extension index exhibits an interdecadal change in the interannual variability, which decreases after the early 1990s for June and July while increases after the mid 1970s for August (Figs. 1i-k). This difference between June-July and August, suggests that their influencing factors would likely to be distinct. Therefore, we divide the whole summer into June-July (named as early summer) and August (named as late summer). The current paper focuses on the recent interdecadal change in the interannual variation of the early summer WNPSH around the early 1990s, and the following analyses are based on the average of June and July unless mentioned otherwise. The interdecadal change in the interannual variation of the late summer WNPSH around the mid 1970s would be investigated separately in the future.

Figure 1.  (a, b, c, d) The westward extension index of the WNPSH calculated based on the JRA-55 (black lines) and its decadal component (gray lines). (e, f, g, h) The westward extension index of the WNPSH released by the NCC. (i, j, k, l) 11-year sliding standard deviation of the westward extension index calculated by the JRA-55 data. Columns 1 to 4 are for June, July, August, and June-July, respectively. The gray vertical lines in Fig. 1l denote the divisions of P1 (1966-1989) and P2 (1990-2014).

As shown by the 11-year sliding standard deviation of the westward extension index, the interdecadal change point in the interannual variability of the early summer index occurs at 1989 / 1990 (Fig. 1l). The westward extension index exhibits large interannual variability during 1966-1989 with the standard deviation reaching 17.1°, while small interannual variability during 1990-2014 with a standard deviation of only 9.4°. Therefore, we select 1966-1989 and 1990-2014 as period 1 and period 2 (abbreviated as P1 and P2 hereafter; depicted by the gray lines in Fig. 1l), so as to investigate this interdecadal change around the early 1990s. The interdecadal decrease of the standard deviation from P1 to P2 is significant at the 99% confidence level according to the F test. It is noted that the mean state of the early summer westward extension index also features an interdecadal westward extension around 1990 (Fig. 1d), which is related to global warming (Huang et al. ^[24]; Wu and Wang ^[25]). Hence, the results presented in the current paper manifest the differences in the interannual variation of the western edge of the WNPSH under different backgrounds, i. e., P1 (P2) with the background of eastward-retreating (westward-extending) WNPSH. It should be mentioned that in order to guarantee the interdecadal change in the interannual variability is not due to the change in the mean state, we also remove the linear trend of the western extension index and repeat the analysis of sliding standard deviation. The result is similar to that exhibited in Fig. 1l (later shown in Fig. 10a), ensuring the robustness of the interdecadal change in interannual variability.

In order to analyze the specific characteristics and plausible causes of the interdecadal change in the interannual variation, the time series of the westward extension index during P1 and P2 are standardized separately to select abnormal years. The years with standardized anomalies smaller (greater) than -0.5 (0.5) are selected as abnormally westward-extending (eastward-retreating) years. During P1, there are 10 abnormally westward-extending years (1966, 1969, 1970, 1971, 1977, 1979, 1980, 1983, 1987 and 1988) and 7 abnormally eastward-retreating years (1967, 1972, 1973, 1974, 1976, 1978 and 1984). During P2, there are 6 abnormally westward-extending years (1991, 1995, 1996, 1998, 2003 and 2010) and 8 abnormally eastwardretreating years (1994, 1997, 1999, 2000, 2002, 2004, 2011 and 2012). The specific characteristics associated with the anomalous WNPSH could be depicted by the composite differences between the abnormally westward-extending years and eastward-retreating years.

Figure 2.  Standardized anomalies of the westward extension index during (a) 1966-1989 and (b) 1990-2014. The gray dashed lines are the values of -0.5, 0 and 0.5.

To demonstrate the zonal movement of the WNPSH corresponding to the distinct interannual variability during P1 and P2, we use the 5880-gpm contour of the 500-hPa geopotential height to depict the WNPSH and compare the 5880-gpm contours in abnormal years with the climatology (Fig. 3). During P1, the western edge of the WNPSH is located at 148.8°E for climatology, while averagely extends to 131.4°E (17.4° westwards) in abnormally westward-extending years (Fig. 3a). For the abnormally eastward-retreating years, the 5880-gpm contour is not observed over the western Pacific in many years and the western edge averagely retreats 27.7° eastward to 176.5°E (the western edge position is regarded as 180° when the 5880-gpm contour is not observed over the western Pacific). During P2, the western edge of the WNPSH is located at 131.3° E for climatology, while averagely extends to 120.4°E (10.9° westwards) in abnormally westward-extending years and retreats to 141.7°E (10.4° eastward) in abnormally eastward-retreating years (Fig. 3b). The west-east movement range of the western edge is obviously larger during P1 than P2, corresponding to the larger oscillation amplitude and thus larger interannual variability of the westward extension index before 1990 compared to that after 1990 (Fig. 1d). The above analyses show that the amplitude of the interannual variability of the westward extension index actually manifests the zonal movement range of the western edge of the WNPSH. The larger zonal movement range during P1 indicates that the zonal range of the geopotential height anomaly over the WNP in the abnormal years is likely to be larger during P1 than P2.

Figure 3.  The 5880-gpm contour of the 500-hPa geopotential height for abnormally westward-extending years (red contours), eastward-retreating years (blue contours), and the climatology (black contours) during (a) 1966-1989 and (b) 1990-2014. The red, blue, and black vertical lines denote the average western edge of the WNPSH for abnormally westwardextending years, eastward-retreating years, and the climatology, respectively.

Figure 4 displays the composite differences of the geopotential height and horizontal winds between the abnormally westward-extending years and eastwardretreating years. For both periods, there is a highpressure anomaly over the WNP (WNPHA) throughout the whole troposphere, but presents obvious differences in the shape and intensity. During P1, the WNPHA extends widely in the zonal direction. Taking the 12-gpm contour of the 500-hPa geopotential height anomaly for example, it spans 97.5° in the zonal direction, with the western edge at 102.5°E and the eastern edge at 162.5°E (Fig. 4c). Besides, there are two positive anomaly centers at 500 hPa with the amplitudes exceeding 18 gpm, which are located near 135° E and 180° respectively. The WNPHA at 850 hPa also extends widely in the zonal direction, with the 12-gpm contour spanning 65° in the zonal direction (Fig. 4e). In contrast, the WNPHA during P2 features apparently narrower zonal extension and smaller amplitude than that during P1. For P2, the 12-gpm anomaly contour spans 40° from ~100°E to ~140°E in the zonal direction at both 500 hPa and 850 hPa (Figs. 4d and 4f). Moreover, there is only one positive anomaly center with amplitude exceeding 18 gpm at 500 hPa, which is located near 130° E (Fig. 4d). The amplitude of the 850-hPa WNPHA center is 18 gpm for P2 while 21 gpm for P1 (Figs. 4f and 4e). During P1 (P2), the wider (narrower) zonal extension of the WNPHA corresponds to the larger (smaller) westeast movement range of the western edge of the WNPSH and thus stronger (weaker) interannual variability of the westward extension index. Therefore, the following sections will compare the formation processes of the WNPHA during the two periods, so as to investigate the causes of the interdecadal change in the interannual variation of the western edge of the WNPSH.

Figure 4.  Composite differences of the geopotential height (shadings; units: gpm; significant areas are dotted) and horizontal winds (vectors; units: m s^-1; black vectors are significant) at (a, b) 200 hPa, (c, d) 500 hPa, and (e, f) 850 hPa between the abnormally westward-extending years and eastward-retreating years during (a, c, e) 1966-1989 and (b, d, f) 1990-2014. The black contours are the 12-gpm geopotential height anomaly.

This section explores the relationship between the local WNPHA and the atmospheric circulation anomalies over other remote regions, based on the composite differences between the abnormally westward-extending years and eastward-retreating years. Fig. 5 shows the composite anomalies of the vertical velocity at 500 hPa and the velocity potential, horizontal divergence, and divergent winds at 200 hPa and 850 hPa. During P1, the subtropical WNP around 20° N is dominated by anomalous descent, which extends from 105° E to 160° W and has two anomalous centers with one over the northern South China Sea and the other to the east of Taiwan (Fig. 5c). To the south of the anomalous descent, there is anomalous ascent over the tropical WNP around 10°N, with two anomalous centers located around the southern South China Sea and to the west of tropical central Pacific. Accompanying with the anomalous descent (ascent), there are anomalous upper tropospheric convergence (divergence) and lowertropospheric divergence (convergence) (Figs. 5a and 5e). The anomalous upper-tropospheric divergent winds over the tropical WNP blow toward the subtropical WNP, causing the anomalous upper-tropospheric convergence and anomalous descent over the subtropical WNP. In turn, the anomalous lower-tropospheric divergent winds over the subtropical WNP blow southward, resulting in the anomalous lower-tropospheric convergence and anomalous ascent over the tropical WNP. It is indicated that the anomalies over the tropics and subtropics may be linked by meridional-vertical circulation.

Figure 5.  Composite differences of (a, b) 200-hPa and (e, f) 850-hPa velocity potential (gray contours; units: 10⁶m² s^-1; contour intervals: 0.5 × 10⁶m² s^-1; solid and dashed contours denote positive and negative values, respectively), horizontal divergence (shadings; units: 10^-6 s^-1; significant areas are dotted) and divergent component of the horizontal winds (vectors; units: m s^-1; black vectors are significant), and (c, d) 500-hPa vertical velocity omega (shadings; units: Pa s^-1; significant areas are dotted) between the abnormally westward-extending years and eastward-retreating years during (a, c, e) 1966-1989 and (b, d, f) 1990-2014.

Based on the zonal ranges of the two anomalous ascent centers, we employ the vertical sections averaged between 100° - 150° E and 160° E-180° to analyze the meridional-vertical circulation during P1 (Figs. 6a-b). For the western vertical cell, anomalous ascent occurs around 10°N and turns into northerly winds at 200 hPa, then descends near 20°N and turns into southerly winds at 700 hPa, forming a meridional-vertical circulation (Fig. 6a). For the eastern vertical cell, anomalous ascent appears around 5° N and turns into northerly winds at 250-100 hPa, then descends around 20°N and turns into southerly winds at the lower troposphere, forming another meridional-vertical circulation (Fig. 6b). These results confirm that the anomalous vertical motions over the tropical and subtropical WNP are linked by meridional-vertical circulation. It is noted that the anomalous ascent over the tropical Pacific extends widely in the zonal direction and reaches far eastward to the tropical central Pacific, favoring the wide zonal extension of the descending anomaly over the subtropical WNP. Correspondingly, the WNPHA presents wide zonal extension for P1 (Figs. 4c and e).

Figure 6.  Composite differences of the vertical circulation (vectors; units: m s^-1 for the horizontal velocity and -10^-2Pa s^-1 for the vertical velocity omega; black vectors are significant) and vertical velocity omega (shadings; units: 10^-2Pa s^-1; significant areas are dotted) between the abnormally westward-extending years and eastward-retreating years during (a, b) 1966-1989 and (c, d) 1990- 2014: averaged along (a) 100°-150°E, (b) 160°E-180°, (c) 105°-120°E, and (d) 0°-25°N.

In contrast, for P2, the anomalous descent over the subtropical WNP mainly concentrates at 105° - 155° E, which presents narrower zonal range but wider meridional range compared to that for P1 (Fig. 5d). There is also anomalous descent east of 155° E but presents obviously weaker amplitude and narrower meridional range, corresponding to the weaker highpressure anomaly compared to the western part (Fig. 4d). Thus, we focus on the western-part strong descending anomaly. To the southwest and west of the anomalous descent, there are anomalous ascent centers located over the Maritime Continent and northern Indian Ocean respectively (Fig. 5d). Judged from the divergent winds, the anomalous descent and ascent centers are closely connected (Figs. 5b and 5f), which indicates that they may be linked by vertical circulation. Based on the locations of the two anomalous ascending centers, we select the vertical sections averaged along 105° - 120°E and 0° - 25° N to verify the vertical circulation. For the meridional-vertical section along the Maritime continent, anomalous ascent occurs around 10° S and turns into northerly winds at 250-150 hPa, then descends around 15°N and turns into southerly winds at the lower troposphere, forming a meridional-vertical circulation (Fig. 6c). For the zonal-vertical section along the northern Indian Ocean, anomalous ascent appears between 60°-90°E and turns into westerly winds at 250- 100 hPa, then descends around 135° E and turns into easterly winds at 1000-400 hPa, forming a zonal vertical circulation (Fig. 6d). In general, the significant anomalous ascent is located more westward and southward during P2 than P1, thus the main body of the anomalous descent over the WNP is also located more westward and southward and has narrower zonal extension during P2. Correspondingly, the WNPHA presents relatively narrower zonal extension during P2 (Figs. 4d and f).

The different tropical atmospheric circulation anomalies between the two periods may be modulated by different tropical SSTAs. Fig. 7 displays the composite anomalies of the Indo-Pacific SST and lowertropospheric horizontal winds evolving from the preceding winter to the simultaneous early summer. For both periods, warm SSTA appears over the tropical central Pacific in the preceding winter and then gradually decays in spring (Figs. 7a-d), presenting the El Niño decaying phase. The warm SSTA over the tropical central Pacific could favor the warm SSTAs over the Indian Ocean and Maritime Continent, via inducing anomalous descent over these regions accompanied with an anomalous Walker circulation (Alexander et al. ^[16]). However, the early summer SSTAs over the tropical central Pacific are distinct between the two periods, which presents persistent warm SSTA for P1 while transforms into cold SSTA for P2 (Figs. 7e-f). It is indicated that the decaying speed of the El Niño type SSTA is different between the two periods, featuring slow decaying El Niño for P1 while rapid transformation from El Niño to La Niña for P2.

Figure 7.  Composite differences of the SST (shadings; units: ℃; significant areas are dotted) and 850-hPa horizontal winds (vectors; units: m s^-1) in the (a, b) preceding winter, (c, d) preceding spring and (e, f) simultaneous early summer between the abnormally westward-extending years and eastward-retreating years during (a, c, e) 1966-1989 and (b, d, f) 1990-2014. Green boxes denote the areas used for calculating the key SSTA indices: northern Indian Ocean (0°-20°N, 50°-100°E), tropical central Pacific (10°S-10°N, 170°E-180°-170°W), and Maritime Continent (10°S-10°N, 100°-150°E).

The different SSTA patterns in early summer for the two periods have different effects on the atmospheric circulation over the WNP. For P1, the significant warm SSTAs in early summer are located at the local WNP, northern Indian Ocean, southern South China Sea, and tropical central Pacific (Fig. 7e). The local warm SSTA over the WNP is accompanied by an anomalous anticyclone at 850 hPa, which indicates that the SSTA is a response to the atmospheric circulation since anticyclonic anomaly could lead to less cloud cover and more solar radiation at surface to heat the SST. As a response to the atmospheric bridge effect of the tropical central Pacific warming, the warm SSTA over the northern Indian Ocean could persist to early summer and cause the WNPHA via the Ekman divergence induced by Kelvin wave (Fig. 7e; Klein et al. ^[37]; Xie et al. ^[17]). The warm SSTA around the southern South China Sea could favor local ascending anomaly and further induce anomalous descent over the WNP via the meridionalvertical circulation as indicated by Fig. 6a, which could further strengthen the WNPHA. Over the tropical central Pacific, significant warm SSTA occurs around 180° and it is accompanied by anomalous ascent centered around (5°N, 165°E) to its northwest (Figs. 5c and 7e), indicating the anomalous ascent is triggered by the warm SSTA through Gill respond (Gill ^[38]). This anomalous ascent further strengthens the anomalous descent and WNPHA through the meridional-vertical circulation (Fig. 6b), which favors the eastward extension of the WNPHA to central Pacific (Figs. 4c and e).

For P2, the significant warm SSTAs in early summer are located at the local WNP, Maritime Continent, and northern Indian Ocean (Fig. 7f). The same as P1, the local warm SSTA over the WNP is a response to the atmospheric circulation featuring an anticyclonic anomaly. The warm SSTA around the Maritime Continent is favored by the phase transition from El Niño to La Niña (Chung et al. ^[14]). This anomalous warming could induce local anomalous ascent and subsequent anomalous descent over the WNP through the meridional-vertical circulation (Fig. 6c), favoring the WNPHA. Different from P1, the warm SSTA over the northern Indian Ocean during P2 induces the anomalous descent and WNPHA via the zonalvertical circulation (Fig. 6d). This zonal-vertical circulation associated with the northern Indian Ocean warming is also emphasized by Wu et al. ^[39]. Overall, the significantly anomalous warm SSTs and ascent and the resultant anomalous descent over the WNP are located more westward and southward during P2 than P1. As a result, the eastward extension of the WNPHA is obviously farther during P1 than P2.

The above analyses indicate that the key SSTA areas affecting the WNPSH are obviously different in the two periods. In order to quantitatively evaluate the changes in the relationship between each key SSTA area and the western edge of the WNPSH, we define three SSTA indices based on the significant SSTA areas (green boxes in Figs. 7e and f). The first is the northern Indian Ocean index (denoted as NIO-SSTA), which is the averaged SSTA over (0°-20°N, 50°-100°E). The second is the tropical central Pacific index (denoted as TCPSSTA), which is the averaged SSTA over (10° S-10°N, 170° E-180° - 170° W). The third is the Maritime Continent index (denoted as MC-SSTA), which is the averaged SSTA over (10°S-10°N, 100°-150°E). Fig. 8 shows the correlation coefficients between the westward extension index and the three SSTA indices during the two periods. During P1, the westward extension index is significantly correlated with the NIO-SSTA and TCPSSTA with a correlation coefficient of -0.48 and -0.35 respectively, but is insignificantly correlated with the MC-SSTA. These significant negative correlations indicate that the warm SSTAs over the northern Indian Ocean and tropical central Pacific favor the westward extension of the WNPSH during P1. During P2, the correlation between the westward extension index and the TCP-SSTA dramatically reduces to 0.07, indicating the zonal movement of the WNPSH is not affected by the simultaneous tropical central Pacific SSTA. Meanwhile, the correlation coefficients between the westward extension index and the NIO-SSTA and MCSSTA have improved to - 0.63 and - 0.70 respectively; both are significant at the 99% confidence level, indicating that the warm SSTA over the northern Indian Ocean and Maritime Continent favors the westward extension of the WNPSH during P2. It is confirmed that the key SSTA areas affecting the western edge of the WNPSH experience interdecadal change, passing from northern Indian Ocean and tropical central Pacific to northern Indian Ocean and Maritime Continent.

Figure 8.  Correlation coefficients between the westward extension index and the three SSTA indices for 1966-1989 (black bars) and 1990-2014 (gray bars). The black (gray) dashed line denotes the 90% confidence level for the correlation coefficient in 1966-1989 (1990-2014).

The above analyses reveal that the simultaneous tropical SSTAs are important for the anomalous atmospheric circulation associated with the interannual variation of the western edge of the WNPSH, and the early summer tropical SSTAs are closely related to the evolution of ENSO. Accordingly, we hypothesize that the evolution rate of ENSO might also experience an interdecadal change around the early 1990s. To verify this hypothesis, we perform power spectrum analysis and wavelet analysis on the winter Niño 3.4 SSTA index, which is computed as the winter (from preceding December to February) SSTA averaged over (5°S-5°N, 170°W-120°W) (Fig. 9). It is shown that the Niño 3.4 SSTA index exhibits a significant oscillation period at 3- 4 years during both P1 and P2 (Figs. 9a and b). Moreover, P1 is characterized by another significant period around 5 years while P2 has no significant period exceeding 4 years, indicating the evolution rate of ENSO is slower during P1 than P2. The wavelet analysis further reveals that the interdecadal change in the dominant oscillation period of ENSO exactly appears around 1990 (Fig. 9c). The dominant oscillation period is 3-5 years with the maximum spectrum value at about 4 years before 1990, while it becomes 2-4 years with the maximum spectrum value at about 3 years after 1990. Therefore, it is confirmed that the evolution rate of ENSO does experience an interdecadal change from slow to fast in the early 1990s. As the leading mode of the interannual variation of tropical SSTAs, different evolution rates of ENSO would favor different tropical SSTA patterns in early summer, which is important for the distinct simultaneous tropical SSTAs influencing the western edge of the WNPSH between the two periods.

Figure 9.  Power spectrum of the winter Niño 3.4 SSTA index during (a) 1966-1989 and (b) 1990-2014. Dashed lines denote the 90% confidence level according to the Markov red noise spectrum. (c) Wavelets power spectrum of the winter Niño 3.4 SSTA index during 1966-2014 (shadings; significant areas are denoted by crosses).

In this study, we investigate the interdecadal change in the interannual variation of the western edge of the WNPSH from the perspective of the influence of tropical SSTAs. The results show that the interannual variability of the western edge of the WNPSH in early summer experienced an interdecadal decrease in the early 1990s. The change in interannual variability actually manifests the changes in the zonal movement range of the western edge of the WNPSH and the zonal range of the WNPHA in abnormal years. During P1 from 1966-1989 (P2 from 1990-2014), the wider (narrower) zonal extension of the WNPHA corresponds to the stronger (weaker) interannual variability of the western edge of the WNPSH.

The interdecadal difference in the WNPHA is related to the influences of different tropical SSTA patterns. Albeit the WNPHA for both periods corresponds to the El Niño type SSTA from the preceding winter to spring, the decaying speed of the El Niño type SSTA is distinct between the two periods. The tropical SSTA evolution features slow El Niño decaying pattern for P1 while rapid transformation from El Niño to La Niña for P2, causing obviously different SSTAs and resultant atmospheric circulation anomalies in early summer. For P1, the key SSTAs areas in early summer influencing the WNPSH are located at the northern Indian Ocean, southern South China Sea, and tropical central Pacific. The warm SSTA over the northern Indian Ocean could cause the WNPHA via the Ekman divergence induced by Kelvin wave. The warm SSTA around the southern South China Sea could strengthen the WNPHA via the meridional-vertical circulation. Over the tropical central Pacific, significant warm SSTA triggers anomalous ascent located at its northwest through the Gill response. This anomalous ascent further strengthens the anomalous descent and WNPHA through the meridional-vertical circulation, favoring the WNPHA to extend eastward to central Pacific. For P2, the key SSTAs areas in early summer are located over the Maritime Continent and northern Indian Ocean, which could favor local anomalous ascent and further cause the WNPHA through meridional-vertical circulation and zonal-vertical circulation, respectively. Overall, the significantly anomalous warm SSTs and ascent and the resultant anomalous descent over the WNP are located more westward and southward during P2 than P1. As a result, the WNPHA presents narrower zonal range and extends less eastward during P2 than P1, corresponding to the smaller interannual variability of the western edge of the WNPSH during P2.

The different early summer tropical SSTAs influencing the western edge of the WNPSH between the two periods are related to the different evolution rates of ENSO. The dominant oscillation period of ENSO experienced an interdecadal reduction around the early 1990s, favoring the El Niño SSTA to change from slow decaying type during P1 to rapid transformation type during P2. Correspondingly, the early summer tropical SSTAs associated with the variation of the western edge of the WNPSH are different between the two periods, causing the interdecadal change in the interannual variation of the western edge of the WNPSH.

The geopotential height index adopted in the current study shows an obvious trend in the mean state, which is indicated to be related to global warming (Huang et al. ^[24]; Wu and Wang ^[25]). Is the interdecadal change in the interannual variability of the western edge of the WNPSH related to global warming? We discuss this problem from two aspects. Firstly, we remove the long-term trend of the geopotential height index adopted in the current study (Fig. 10a). Secondly, we compare the geopotential height index with other indices that help eliminate the signal of global warming (Figs. 10b-e). The 11-year sliding standard deviation of different indices turn out to exhibit discrepancies. The detrended 500-hPa geopotential height index, 500-hPa relative vorticity index, and 850-hPa eddy geopotential height index consistently present interdecadal decreases in interannual variability around the late 1980s to early 1990s (Figs. 10a-c). In contrast, the interannual variability of the 500-hPa eddy geopotential height index increases around the early 1990s (Fig. 10d), while that of the 500-hPa eddy stream function index decreases around the late 1970s and increases again around the mid 1990s (Figs. 10e). These discrepancies suggest that the WNPSH experiences different changes in different aspects, and it needs further investigation to decide which index is the best to describe the WNPSH. To some extent, the detrended 500-hPa geopotential height index, 500-hPa relative vorticity index, and 850- hPa eddy geopotential height index indicate that the interdecadal change revealed in this study is not totally due to global warming. This viewpoint is also supported by previous studies documenting that the changes in the relationship between WNPSH and ENSO are modulated by the atmospheric internal variability (He and Zhou ^[40]; Song and Zhou ^[41]; Wang et al. ^[42]; Wu et al. ^[43]). More studies combing observations and climate model simulations are required to further reveal the relative role of atmospheric internal variability and global warming in the historical and future changes of WNPSH and its relationship with ENSO.

Figure 10.  11 year-sliding standard deviation of different WNPSH indices: (a) detrended geopotential height index adopted in this study (units: °), (b) 500-hPa relative vorticity index (units: 10^-6m² s^-1), (c) 850-hPa eddy geopotential height index (units: gpm), (d) 500-hPa eddy geopotential height index (units: °), and (e) 500-hPa eddy stream function index (units: °).

Data Availability Statement: The circulation data is available at https://rda.ucar.edu/datasets/ds628.1, and the SST data is available at https://psl.noaa.gov/data/gridded/data.noaa.ersst.v5.html.