Decadal Changes in Interannual Dependence of the Bay of Bengal Summer Monsoon Onset on ENSO Modulated by the Pacific Decadal Oscillation

The onset of the summer monsoon signifies the advent of the rainy season for a particular region in the Asian monsoon regime. Climatologically, the Asian summer monsoon onset firstly arises over the eastern Bay of Bengal (BOB) at the beginning of May, and then occurs over the South China Sea in mid-May, and finally over India in early June (Wu and Zhang, 1998; Mao et al., 2003, 2004; Mao and Wu, 2007). Because the onset timing usually determines the dates of ploughing and planting as well as the intensity of subsequent rainfall in agrarian societies in the monsoon regions (Webster et al., 1998), small variations in the timing and quantity of rainfall have the potential for significant consequences. As suggested by Mao and Wu (2007), the earliest onset over the BOB as a precursor in the entire Asian monsoon region may also affect the subsequent establishment of the summer monsoon over other sub-monsoon areas. Therefore, forecasting the timing of the BOB summer monsoon (BOBSM) onset and studying its year-to-year variability are more representative and important to understand the causes of the Asian summer monsoon regime.

The BOBSM onset date is defined as the reversal of the area-averaged meridional temperature gradient (MTG) in the middle−upper troposphere (200−500 hPa) over the eastern BOB by Mao and Wu (2007), who found that the interannual variability of the BOBSM onset is related to middle−upper tropospheric temperature anomalies around the Tibetan Plateau north of the westerly−easterly boundary surface, and suggested such temperature anomalies are partly caused by land-based conditions of local snow-cover anomalies over the Tibetan Plateau. More importantly, they found that the interannual variability of the BOBSM onset is also controlled by ocean-based conditions in terms of the Pacific sea surface temperature (SST) anomalies (SSTAs) in the preceding winter and spring; specifically, in relation to El Niño−Southern Oscillation (ENSO) events, with an early (late) BOBSM onset being preceded by a La Niña (El Niño) event. ENSO-related SSTAs are believed to affect the middle−upper tropospheric temperature anomalies over the Asian sector south of the westerly−easterly boundary surface through changes in the Walker circulation and local Hadley circulation. Subsequently, Jiang and Li (2011) noted that the BOBSM onset tends to follow the abrupt northward jump of the warmest SST axis and lags by about two pentads, and argued that ENSO may influence the BOBSM onset via changes in the subseasonal variability of SST in the northern Indian Ocean during spring.

However, the spatial distribution and strength of ENSO-related SSTAs could be modulated by the Pacific Decadal Oscillation (PDO), which is the dominant decadal variability in the Pacific (Mantua et al., 1997), leading to different decaying speeds and SSTA patterns of ENSO (Feng et al., 2014) and resultant decadal changes in the relationship between ENSO and climate anomalies over various regions such as North America (Gershunov and Barnett, 1998), South Asia (Kumar et al., 1999; Watanabe and Yamazaki, 2014), and East Asia (Wu and Wang, 2002; Zhou et al., 2007; Wang et al., 2008, 2012; Chen et al., 2013; Wu and Mao, 2016).

Recently, in investigating the spatial and interannual variations of spring rainfall anomalies over eastern China in relation to ENSO, Wu and Mao (2016, 2018) found four kinds of different SSTA distributions associated with in-phase and out-of-phase PDO−ENSO events, inducing different atmospheric circulation teleconnections. For example, if El Niño events occur in a warm PDO epoch, the resultant SSTA distribution is characterized by both enhanced positive SSTAs in the tropical central−eastern Pacific and enhanced negative SSTAs in the midlatitude North Pacific, in conjunction with weak negative SSTAs in the tropical western Pacific (TWP) and positive SSTAs in the tropical Indian Ocean. If El Niño events occur in a cold PDO epoch, not only positive SSTAs in the tropical central−eastern Pacific and tropical Indian Ocean, but also the negative SSTAs in the TWP, become very weak, because the ENSO-related SSTAs and PDO-related SSTAs tend to cancel each other out. Obviously, the PDO exerts a significant modulation on the distribution and intensity of ENSO-related SSTAs in the entire tropical Pacific. Of note is that the SSTAs and resultant convective anomalies over the TWP during ENSO events are a key factor responsible for the interannual variability of the BOBSM onset, as suggested by Feng et al. (2013), who found that the time series of the BOBSM onset dates is highly correlated with that of the TWP upper-ocean heat content during the preceding winter and spring for the period 1979−2012. This is because the anomalous convection over the western Pacific warm pool induced by the underlying ocean temperature anomalies can modify the strength of Walker circulation and the related lower-tropospheric westerlies over the tropical Indian Ocean flowing into the BOB in spring. On the other hand, ENSO-induced convection anomalies over the southern Philippines may also affect the position of the South Asian high in late April, modulating the upper divergence-pumping effect and thereby inducing the active convection for BOBSM onset (Liu et al., 2013, 2015). Since ENSO-related SSTAs in both the tropical western and eastern Pacific are modulated by the PDO, it is conceivable that the interannual relationship between BOBSM onset and ENSO may differ in different PDO epochs because the modified SSTA distributions of various PDO−ENSO events induce different atmospheric teleconnections, resulting in abnormal BOBSM onsets.

Actually, Mao and Wu (2007), based on NCEP reanalysis data for a relatively short period (1958−2001), also noticed that the BOBSM onset−ENSO relationship seems to have experienced a decadal change. Prior to 1967 the BOBSM onset dates exhibited a weakly negative correlation with eastern Pacific SSTAs. Subsequently, the correlation has become stronger, with a significant correlation of 0.58 for the period 1968−2001. Similar decadal-scale variation was found to exist in South Asian summer monsoon onset over Kerala, India (Watanabe and Yamazaki, 2014), with strengthened correlation between the onset date over Kerala and the PDO for the period 1976−2002. Actually, many studies have suggested that the associations of ENSO with both East Asian winter monsoon and East Asian summer monsoon are modulated by the PDO (e.g., Wang et al., 2008; Mao et al., 2011; Chen et al., 2013; Feng et al., 2014). Since the BOBSM onset signifies the transition of the Asian winter monsoon to summer monsoon, its year-to-year variability is inevitably dependent on the PDO-modulated interannual fluctuation of the Asian winter circulation. Thus, the decadal changes in the interannual relationship of the BOBSM onset with ENSO have to be substantiated based on longer-period datasets.

Therefore, the objective of the present study is to examine whether the relationship between BOBSM onset and ENSO is modulated by the PDO, understanding the physical processes of how the ENSO-related SSTAs modified by the PDO induce different atmospheric circulation patterns to affect the BOBSM onset. Section 2 introduces the data and methods. Section 3 investigates the decadal changes in the dependence of BOBSM onset on ENSO under different PDO backgrounds. Section 4 examines the physical mechanisms for the PDO’s modulation of the BOBSM onset−ENSO relation. Finally, a summary and discussion are presented in section 5.

Daily atmospheric circulation datasets from NCEP1 (Kalnay et al., 1996) are used to calculate the BOBSM onset dates from 1948 to 2017, as done in Mao and Wu (2007). For comparison and validation, another atmospheric reanalysis product, JRA-55, is also introduced to determine the onset dates, as well as to diagnose the ENSO-related atmospheric circulation anomalies under different PDO backgrounds. Note that JRA-55 includes a more sophisticated data assimilation system and newly prepared dataset of past observations, being available since 1958, when regular radiosonde observations began on a global basis (Kobayashi et al., 2015). In addition, to confirm the decadal changes of the BOBSM onset−ENSO relation for as long a period as possible, the 20th Century Atmospheric Reanalysis from the European Centre For Medium-Range Weather Forecasts (ERA20C), which is available for the period 1900−2010 (Stickler et al., 2014), is also adopted to calculate the BOBSM onset dates and related atmospheric circulation anomalies under different PDO backgrounds.

Monthly SST data for the period 1900−2017 are extracted from HadISST 1.1, which consists of monthly mean fields of SST and sea-ice concentration on a 1° × 1° latitude−longitude grid from 1870 to the present day (Rayner et al., 2003). The monthly Niño3.4 and PDO indices used in this study are downloaded from http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/detrend.Nino34.ascii.txt and http://research.jisao.washington.edu/pdo/PDO.latest.txt, respectively. The monthly PDO index is defined as the leading principal component of monthly SSTAs in the North Pacific Ocean poleward of 20°N, with the monthly mean global average SSTAs removed to separate this pattern of variability from any “global warming” signal that may be present in the data (Mantua et al., 1997).

Following Mao and Wu (2007), the BOBSM onset date for a particular year is defined as the day when the following criteria are first satisfied: (1) the area-averaged upper-tropospheric (200−500 hPa) MTG over the BOB (5°−15°N, 90°−100°E) changes from negative to positive; (2) the MTG remains positive for more than 10 days.

The wintertime (December−February, DJF) averaged PDO index is used to identify the PDO phases. In this study, an 11-year Lanczos low-pass filter (Duchon, 1979) is applied to the wintertime PDO time series to obtain the decadal variability of the PDO, and then warm (cold) PDO years are classified as those when the filtered PDO index is above (below) zero, as done by Kim et al. (2014). Following Wu and Mao (2016), an El Niño (La Niña) year is identified when the detrended December−February Niño3.4 SSTA is greater than 0.55°C (less than −0.55°C). Note that this anomaly threshold is approximately equal to 0.6 times the standard deviation of the interannual variations of the Niño3.4 index. Based on this criterion, there are 32 El Niño and 32 La Niña events arising during the three warm PDO epochs (1900−1911, 1922−1945, 1978−2006) and three cold PDO epochs (1912−1921, 1946−1977, 2007−2014) for the period 1900−2017. As such, these ENSO events occurring in different PDO epochs can be further categorized as different PDO−ENSO combinations of PDO-related El Niño and La Niña events (Table 1), as done by Wu and Mao (2016, 2018). For the third cold PDO epoch, being cut off by the 11-year Lanczos low-pass filter, so far, it is difficult to infer unambiguously how long this epoch persists after 2014 and in which year the PDO changes into the positive epoch. Thus, we only consider the period 2007−14 as a cold PDO epoch. Note that the El Niño events in 2015 and 2016 are not taken into consideration due to uncertain PDO phases for the post-2014 period.

The influence of different PDO−ENSO events on the BOBSM onset during the subsequent spring (i.e., when ENSO decays) is then examined. Unless otherwise stated, ENSO-related circulation and rainfall anomalies in the following text refer to those in the spring following the peak of the PDO−ENSO events. The statistical significance of the composite differences is estimated by Student’s t-test.

Figure 1a displays the time series of BOBSM onset dates derived from JRA-55 and NCEP1 reanalysis data, with the onset dates exhibiting similarly strong year-to-year fluctuations varying from the first pentad of April to the last pentad of May. Focusing firstly on the period 1958−2017, when JRA-55 products are available, the long-term mean date of the BOBSM onset is the first pentad of May (about 1 May), and the interannual standard deviation (indicated by σ) is 11 days for this period. The correlation coefficients are calculated between the time series of the BOBSM onset dates and wintertime Niño3.4 index to reach 0.52 for the period 1958−2017, statistically significant at the 95% confidence level, indicating that the interannual variability of the onset dates is indeed correlated to ENSO events. However, such a significant correlation may not always remain unchanged during the whole period, instead having obvious decadal differences modulated possibly by the PDO. As shown in Fig. 1c, during the period 1978−2006 for the warm PDO epoch, the 11-year running correlations between the BOBSM onset dates and Niño3.4 index are stably greater than 0.602, with statistically significance above the 95% confidence level, while the correlations drop down rapidly and become less significant before 1978 for the cold PDO epoch. Especially, when the PDO transforms again into the cold epoch from 2007, the 11-year running correlations exhibit an evident declining trend, albeit the correlations have not become as insignificant as those during the pre-1978 period, due to the limited length of this epoch.

Figure 1.  (a) Time series of the BOBSM onset date (left-hand y-axis, Julian dates in the calendar year) and its anomaly expressed in multiples of σ (right-hand y-axis, σ refers to the interannual standard deviation of the BOBSM onset time series) derived from NCEP1, JRA-55 and ERA20C. The long-term mean onset date is 1 May, as indicated by a horizontal solid line. One standard deviation is shown by the dashed lines. (b) Time series of wintertime Niño3.4 index (bars; units: °C) and 11-year low-pass filtered PDO index (dark curve), with the red (blue) bars indicating El Niño (La Niña) years, and the parallel dark lines indicating the thresholds of ±0.55°C to identify the El Niño and La Niña years (as listed in Table 1). (c) Time series of 11-year running correlations between time series of Niño3.4 index and BOBSM onset date, with the dark dashed lines showing the critical values of the correlation at the 95% significant level. In (b, c), the vertical lines show the transition years from negative (positive) to positive (negative) PDO epochs.

To further demonstrate the PDO’s modulating effect, the El Niño and La Niña events are classified into warm and cold PDO epochs (Table 1), with 10 El Niño events occurring in the warm PDO epoch (abbreviated as EN_WPDO), 8 La Niña events in the warm PDO epoch (LN_WPDO), 8 El Niño events in cold PDO epochs (EN_CPDO), and 8 La Niña events in cold PDO epochs (LN_CPDO) from the period 1958−2017. During the warm PDO epoch, as shown by italic font in Table 1, all El Niño events are associated with the late BOBSM onsets, and the anomaly magnitude of the onset dates for almost all EN_WPDO events is at least 0.4σ (4.4 days, equivalent to around one pentad) later than the long-term mean onset date, except in 1988, while all La Niña events are related to the early BOBSM onset, with onset date anomalies for nearly all LN_WPDO events being at least 0.4σ earlier than the mean onset date. It turns out that during the warm PDO epoch, an El Niño (La Niña) event is indeed most likely to result in a late (early) BOBSM onset.

However, such an ENSO−BOBSM onset relationship becomes less significant in cold PDO epochs. As shown by italic font in Table 1, for EN_CPDO events, there are three events related to late BOBSM onset and five events related to early BOBSM onset, without showing any early or late preferences. Similarly, for LN_CPDO events, although almost all events are associated with early monsoon onset, only half of those events lead to onset anomalies 0.4σ earlier than the mean onset date. That is, the influence of ENSO on BOBSM onset is not significant during the cold PDO epochs, which is demonstrated by both the running correlations (Fig. 1c) and category statistics (Table 1).

Furthermore, a longer time series of BOBSM onset derived from ERA20C is used to validate such decadal changes of the BOBSM onset−ENSO relation. As shown in Fig. 1a, the interannual fluctuations of the BOBSM onset date calculated using ERA20C data are in good agreement with those derived from the JRA-55 and NCEP1 products during their period of overlap, 1958−2010, although some differences exist among these three datasets. For example, the BOBSM onset date in 1989 is 16 May based on ERA20C, which is over one month later than the onset date of 7 April derived from both JRA-55 and NECP1 (Fig. 1a), indicating that the onset dates derived from ERA20C may be biased in a few years. Again, the 11-year running correlation is used to explore the decadal changes of BOBSM onset dates relying on Niño3.4 index over a longer time period. On the whole, the running correlations between the ERA20C-based onset date and Niño3.4 index exhibit high correlation coefficients during warm PDO epochs (1900−1911, 1922−1945, 1978−2006) but low correlation coefficients during cold PDO epochs (1912−1921, 1946−1977, 2007−2014), except the abrupt decrease around 1989 resulting from the aforementioned data problem of ERA20C. A similar slight decline of the running correlations appears around 1933, which may also result from the data quality of ERA20C, because the data prior to 1948 are not as good as JRA-55 and NCEP1. It is worth noting that the running correlation during the short cold PDO epoch of 1911−1921 even shows a declining trend, though such a decline is not as large as the cold PDO period of 1946−1977, which may result from the cold PDO period of 1911−1921 being to too short for 11-year running correlation to fall down. Similarly, the BOBSM onset dates derived from ERA20C during 1900−1957 are also classified into four types of PDO−ENSO events, as shown by italic font in Table 1. For warm PDO epochs, the BOBSM onset dates and ENSO show a close relationship, with 7 out of 9 El Niño (9 out of 10 La Niña) years associated with later (earlier) BOBSM onset than the mean onset date. However, in cold PDO epochs, the BOBSM onsets present no early or later preferences, with two early onset years and three late onset years in all five El Niño years, and three early onset years and three late onset years in all six La Niña years.

The above facts show that decadal changes in the dependence of BOBSM onset on ENSO events do indeed exist, and such decadal changes vary with the PDO phases. A closer correlation between BOBSM onset and ENSO arises during warm PDO epochs, while less correlation occurs during cold PDO epochs.

Since the onset process of the BOBSM occurs in early May, the spring (March−May) mean SSTAs are thus analyzed. Figure 2 displays the composite spring SSTAs in the decaying episode of various PDO−ENSO events including EN_WPDO, LN_WPDO, EN_CPDO and LN_CPDO events. Obviously, the amplitudes of SSTAs associated with in-phase PDO−ENSO events [i.e. EN_WPDO (Fig. 2a), LN_CPDO (Fig. 2d)] are larger than those of out-of-phase PDO−ENSO events [i.e. LN_WPDO (Fig. 2b), EN_CPDO (Fig. 2c)] over the tropical central−eastern Pacific as well as the midlatitude North Pacific, as demonstrated by Wu and Mao (2016), who suggested that the in-phase superimposition of PDO-related and ENSO-related SSTAs can enhance the ENSO-related SSTAs, but the out-of-phase superimposition tends to weaken them. Notably, the zonal distribution of SSTAs in the warm PDO epoch (Figs. 2a and b) exhibits evident differences from that in the cold PDO epoch (Figs. 2c and d), especially in the tropical Indian and Pacific oceans. For the warm PDO epoch, i.e., during spring when the EN_WPDO event decays, significant negative SSTAs are present over the TWP, with positive SSTAs over the tropical central−eastern Pacific and most of the tropical Indian Ocean, manifesting a zonally tripole pattern of positive−negative−positive SSTAs over tropical oceans, thus resulting in a positive SSTA gradient from the TWP to Indian Ocean (Fig. 2a). Meanwhile, a similar but opposite SSTA pattern occurs during the decaying episode of the LN_WPDO event, with a negative−positive−negative distribution characterized by positive SSTAs over the TWP (Fig. 2b), resulting in a negative SSTA gradient from the TWP to Indian Ocean. However, during the cold PDO epochs, the SSTA distributions over the tropical Indian and Pacific oceans for either EN_CPDO (Fig. 2c) or LN_CPDO (Fig. 2d) events do not show a distinct tripole pattern, without significant SSTAs over the TWP. Actually, relatively weak positive SSTAs are present only over small areas in the northern tropical Indian Ocean and tropical central Pacific for EN_CPDO events (Fig. 2c), while strong negative SSTAs occur over almost the entire tropical Indian Ocean and extend over the South China Sea and Maritime Continent, as well as over the central−eastern Pacific, for LN_CPDO events (Fig. 2d). As suggested by Feng et al. (2014), the decaying speed of ENSO under different PDO backgrounds is the major reason for the different SSTA patterns of PDO−ENSO events. For the EN_WPDO events, the strong El Niño-type SST forcing persists from the preceding winter to spring, resulting in strong positive SSTAs over the tropical central and eastern Pacific, as well as the tropical Indian Ocean, in association with negative anomalies over the TWP; however, the El Niño-type SST forcing is much weaker for the EN_CPDO events, which is quickly weakened in spring, resulting in weak positive anomalies over the equatorial central Pacific and Indian Ocean in spring. Similarly, for the LN_CPDO events, the strong La Niña-type SST forcing also persists into spring, resulting in strong negative SSTAs over the tropical central and eastern Pacific as well as tropical Indian Ocean, but without significant positive anomalies over the TWP. Also, the La Niña-type SST forcing for LN_WPDO events is much weaker than that for LN_CPDO events, resulting in relatively weak negative anomalies over the tropical central Pacific and Indian oceans; plus, significant positive anomalies appears over the TWP, which may be coupled with the southwest−northeast-oriented anomalous cyclone over the western North Pacific through local air−sea interactions.

Figure 2.  Composite distributions of spring (March−May) SSTAs (shading; units: °C) for (a) El Niño events in warm PDO epochs (EN_WPDO), (b) La Niña events in warm PDO epochs (LN_WPDO), (c) El Niño events in cold PDO epochs (EN_CPDO) and (d) La Niña events in cold PDO epochs (LN_CPDO), based on HadISST1.1 for the period 1958−2017. Stippling denotes anomalies statistically significant at the 95% confidence level.

To explore how the PDO−ENSO events induce anomalous BOBSM onsets through atmospheric teleconnections, anomalous atmospheric circulations forced by the ENSO-related SSTAs under different PDO epochs are examined. Figure 3 illustrates composite fields of SSTA-induced wind anomalies in the lower and upper troposphere in spring. The atmospheric response to SSTAs of the EN_WPDO event is characterized by significantly anomalous 850-hPa easterlies prevailing over the tropical Indian Ocean (Fig. 3a) and anomalous 200-hPa westerlies extending northward to 30°N (Fig. 3b), forming a westerly wind shear in the vertical direction, which indicates that the MTG remains negative according to the thermal wind relation, and thus the BOBSM onset will certainly be delayed. In fact, the anomalous 850-hPa easterlies (Fig. 3a), along with upper-tropospheric westerlies (Fig. 3b), are induced directly by the zonal interoceanic SSTA gradient between the negative SSTAs over the TWP and the positive SSTAs over the tropical Indian Ocean (Fig. 2a), which in turn are coupled with the anomalous Walker circulation forced by positive SSTAs in the equatorial central−eastern Pacific (Fig. 2a), as suggested by Mao and Wu (2007) and Feng et al. (2013). The strong anomalous anticyclone over the western North Pacific forced by the SSTAs also plays an important role in inducing the wind anomalies over the BOB. As shown in Fig. 4a, the strong negative velocity potential anomalies in association with the anomalous anticyclone over the western North Pacific induce strong anomalous divergent southeasterlies (northeasterlies) over the BOB (tropical Indian Ocean) in the lower troposphere, coupled with strong anomalous northwesterlies (southwesterlies) over the BOB (tropical Indian Ocean) converging towards the negative velocity potential anomalies over the western North Pacific in the upper troposphere (Fig. 4b), leading to strong easterly wind anomalies in the lower troposphere and westerly wind anomalies in the upper troposphere over the BOB (Figs. 3a and 3b). Note that, besides the anomalous anticyclone over the western North Pacific, another anomalous anticyclone is present over the BOB (Fig. 3a), which is unfavorable for local convective activity developing over the eastern BOB, thus leading to a late BOBSM onset. Similar but opposite circulation patterns are observed for LN_WPOD events (Figs. 3c and d), with significant anomalous westerlies in the lower troposphere over the tropical Indian Ocean (Fig. 3c) and anomalous easterlies in the upper troposphere over the Maritime Continent, southern Indochina Peninsula and northern BOB (Fig. 3d). Also, the anomalous cyclone over the western North Pacific in association with positive velocity potential anomalies induces strong anomalous convergent westerlies over the BOB in the lower troposphere (Fig. 4c), coupled with strong divergent easterlies over the BOB diverging from the negative velocity potential anomalies over the western North Pacific in the upper troposphere (Fig. 4d). This easterly wind shear in the vertical direction represents a positive MTG, with the low-level anomalous westerlies favoring active convection to arise over the eastern BOB, as suggested by Mao and Wu (2007), indicating that the BOBSM onset is earlier than normal. Obviously, the anomalous low-level westerlies and upper-level easterlies are forced by the zonal interoceanic SSTA gradient between the positive SSTAs over the TWP and the negative SSTAs over the tropical Indian Ocean (Fig. 2b). In contrast, during the cold PDO epochs, no significant anomalous signals are observed in both the lower and upper tropospheric circulations for EN_CPDO events (Figs. 3e and f), as well as the anomalous velocity potential and divergent winds (Figs. 4e and f), and thus the preference of an early or late BOBSM onset becomes ambiguous (Table 1). This is because the SSTA-related external forcing is very weak. Note from Fig. 2c that no strong SSTA signals are present over the tropical Indian Ocean, especially over the TWP, to produce strong external forcing.

Figure 3.  Composite anomalies of spring (March−May) (a) 850-hPa winds and (b) 200-hPa winds (vectors; units: m s⁻¹) for El Niño events in warm PDO epochs (EN_WPDO) based on JRA-55 for the period 1958−2017. Panels (c, d), (e, f) and (g, h) are the same as in (a, b) but for La Niña events in warm PDO epochs (LN_WPDO), El Niño events in cold PDO epochs (EN_CPDO) and (d) La Niña events in cold PDO epochs (LN_CPDO). Shading denotes anomalies statistically significant at the 95% confidence level.

Figure 4.  Composite anomalies of spring (March−May) divergent winds (vectors; units: m s⁻¹) and velocity potential (shading; units: 10⁶ m² s⁻¹) at (a) 850 hPa and (b) 200 hPa for El Niño events in warm PDO epochs (EN_WPDO) based on JRA-55 for the period 1958−2017. Panels (c, d), (e, f) and (g, h) are the same as in (a, b) but for La Niña events in warm PDO epochs (LN_WPDO), El Niño events in cold PDO epochs (EN_CPDO) and (d) La Niña events in cold PDO epochs (LN_CPDO), respectively. Only shown are the velocity potential anomalies that are statistically significant at the 95% confidence level and the divergent wind anomalies where at least one of the zonal and meridional components is statistically significant at the 95% confidence level.

For the LN_CPDO events, a significant anomalous cyclone in the lower troposphere is present over the western North Pacific (Fig. 3g) in response to the strong negative SSTAs over the tropical central−eastern Pacific and eastern Indian Ocean (Fig. 2d) along with positive SSTAs over the midlatitude North Pacific. In turn, such an anomalous cyclone inevitably generates active convection locally, as a result of the thermally-forced response (Gill, 1980) to anomalous convective heating over the western North Pacific, and an anomalous upper-tropospheric anticyclone is induced north of the Indochina Peninsula with significant anomalous easterlies over the northern BOB (Fig. 3h), favoring the development of active convection over the eastern BOB. However, there are no significant positive SSTAs over the TPW (Fig. 2d) to cause a significant zonal SSTA gradient with the negative SSTAs over the upstream Indian Ocean. As shown in Fig. 4g, the positive velocity potential anomalies in the lower troposphere over the western North Pacific are much weaker than those during LN_WPDO events (Fig. 4c), with the anomaly center located more eastward. As a result, the anomalous convergent westerlies are also weaker than those in the LN_WPDO events. Furthermore, the anomalous circulation anomalies over the BOB are dominated by local anomalous divergent northwesterlies in the upper troposphere, thus without anomalous flows from the tropical Indian Ocean (Fig. 4h). Consequently, the anomalous low-level westerlies are very weak and insignificant over the BOB (Fig. 3g); therefore, not enough moisture is transported into the BOB to generate strong convection, although the favorable divergence condition associated with anomalous easterlies exists in the upper-troposphere, merely resulting in slightly early BOBSM onset (Table 1). In brief, under the cold PDO epochs, the SSTAs over the TWP become insignificant for both EN_CPDO and LN_CPDO events, resulting in weak interoceanic SSTA gradients between the tropical Indian Ocean and TWP, which can hardly induce significant circulation anomalies, especially with respect to the lower-tropospheric anomalous winds over the BOB, and thereby the BOBSM exhibits no preference for early or late onset.

Compared to warm PDO epochs, BOBSM onset is less dependent on ENSO (EN_CPDO and LN_CPDO) events during cold PDO epochs, which can be attributed to the absence of significant SSTAs over the TWP (Fig. 2). Without the strong TWP SSTAs, larger zonal SSTA gradients between the TWP and central−eastern Pacific, as well as between the TWP and tropical Indian Ocean, cannot be generated to induce anomalous zonal circulations (Fig. 3). This validates the important role of the TWP SSTAs in affecting the BOBSM onset, as emphasized by Feng et al. (2013).

In addition, Fig. 5 shows the composite circulation anomalies of different PDO−ENSO events based on ERA20C for the period 1900−2010. For the EN_WPDO (LN_WPDO) events, anomalously strong 850-hPa easterlies (westerlies) prevail over the tropical Indian Ocean (Figs. 5a and c), which are coupled with 200-hPa westerlies (easterlies) extending to 30°N (Figs. 5b and 5d), forming a westerly (easterly) wind shear in the vertical direction, and thus favoring the MTG to remain negative (change as positive) according to the thermal wind relation. As such, the BOBSM onset is certainly delayed (advanced), consistent with the mechanism revealed by the JRA-55 datasets (Figs. 3a-d). In contrast, following the EN_CPDO events, both the 850-hPa and 200-hPa wind anomalies are not induced over the tropical Indian Ocean and BOB in spring (Figs. 5e and f); thus, the preference of an early or late BOBSM onset becomes ambiguous. As for the LN_CPDO years, although the strong 200-hPa easterly wind anomalies tend to favor the development of active convection over the BOB (Fig. 5h), there are no significant circulation anomalies in the lower troposphere bringing moisture to support the development of convection (Fig. 5g), similar to the situations derived by JRA-55 (Figs. 3h and 3g), and thus the anomalous preference of BOBSM onset is also uncertain. Therefore, the circulation anomalies derived from the longer-term ERA20C data provide similar results as from JRA-55, indicating that the mechanism revealed by JRA-55 is reliable and the modulating effect of the PDO on the BOBSM onset−ENSO relation is robust.

Figure 5.  As in Fig. 3 but based on ERA20C for the period 1900−2010.

Considering that the summer monsoon onset is the most important sub-seasonal phenomenon of monsoon variability (Webster et al., 1998), which is a result of the phase-lock of seasonal evolution with sub-seasonal fluctuations of atmospheric circulation. Thus, monthly circulation anomalies prior to the BOBSM onset are again examined, expecting to demonstrate the anomalous sub-seasonal fluctuations. Given that the long-term mean BOBSM onset date is around 1 May, the monthly evolutionary features of circulation anomalies in March and April not only represent the anomalous precursory signals but also reflect the sub-seasonal variations before or during the BOBSM onset. Figure 6 shows the monthly lower-tropospheric circulation anomalies in March and April forced by SSTAs of different PDO−ENSO events. For EN_WPDO events, the significant anomalous 850-hPa easterlies are induced over the tropical Indian Ocean in March (Fig. 6a), and then migrate northward into the southern BOB in April (Fig. 6b). Although the easterlies over the tropical Indian Ocean in April become somewhat weaker as compared with March, an anomalous anticyclone is formed over the BOB with significant anomalous northerlies over the eastern part of the BOB, which tend to suppress the development of convective activities over the BOB, resulting in late BOBSM onset. In contrast, following LN_WPDO events, significant anomalous westerlies are produced over the tropical Indian Ocean and southern BOB in March (Fig. 6c), and subsequently become stronger in April (Fig. 6d), thus favoring early BOBSM onset. However, for the EN_CPDO years, no significant 850-hPa wind anomalies are present over the tropical Indian Ocean and BOB in March (Fig. 6e). Although anomalous easterlies occur over the southern BOB in April (Fig. 6f), such weak anomalies are less likely to trigger a much later BOBSM onset. Opposite situations are observed for the LN_CPDO events, with significant anomalous westerlies over the southern tropical Indian Ocean and Maritime Continent along with anomalous northeasterlies over the western North Pacific in March (Fig. 6g). However, such anomalies almost disappear in April (Fig. 6h), and there are no significant wind anomalies over the BOB in both March and April (Figs. 6g and 6h), thus without any preference of a late or early BOBSM onset. Therefore, the monthly evolution of 850-hPa wind anomalies again demonstrates that an anomalous late (early) BOBSM onset is likely to follow an El Niño (La Niña) event during warm PDO epochs, while anomalous BOBSM onset is less related to an ENSO event during cold PDO epochs.

Figure 6.  Composite anomalies of (a) March and (b) April 850-hPa winds (vectors; units: m s⁻¹) for El Niño events in warm PDO epochs (EN_WPDO) based on JRA-55 for the period 1958−2017. Panels (c, d), (e, f) and (g, h) are the same as in (a, b) but for La Niña events in warm PDO epochs (LN_WPDO), El Niño events in cold PDO epochs (EN_CPDO) and La Niña events in cold PDO epochs (LN_CPDO). Shading denotes anomalies significant at the 95% confidence level.

In this study, the NCEP1, JRA-55 and ERA20C atmospheric reanalysis datasets, as well as Hadley SST reanalysis products, over the period 1900−2017 are used to reinvestigate the interannual variability of the BOBSM onset in association with ENSO events. Significant decadal changes are found to exist in the dependence of the BOBSM onsets on ENSO events, and such decadal changes vary with the PDO phases. During the three warm PDO epochs (1900−1911, 1922−1945, 1978−2006), the BOBSM onsets are highly correlated with ENSO events, in which an El Niño (La Niña) event is most likely to result in a late (early) BOBSM onset, with the anomaly magnitude of the onset dates being greater than at least one pentad for most ENSO years. However, during the three cold PDO epochs (1912−1921, 1946−1977, 2007−2014), the BOBSM onsets are less dependent on ENSO events, in which neither El Niño nor La Niña events produce any early or late preference of BOBSM onset. That is, the dependence of the BOBSM onset on ENSO is not always unchanged; rather, it undergoes evident decadal changes, with a much closer relationship of the BOBSM onset with ENSO during warm PDO epochs but an uncertain relationship during cold PDO epochs.

The physical process of the PDO modulating the ENSO−BOBSM onset relationship is through the variations in SSTA patterns in the tropical and North Pacific. During the warm PDO epochs, the SSTAs of an El Niño (La Niña) event are superimposed on the PDO-related SSTAs, forming an obvious SSTA distribution characterized by significant positive (negative) SSTAs over the tropical central−eastern Pacific and Indian Ocean along with negative (positive) SSTAs, especially over the TWP, for the EN_WPDO (LN_WPDO) event. Thus, the significant anomalous lower tropospheric easterlies (westerlies) together with upper-tropospheric westerlies (easterlies) over the BOB are induced directly by the zonal interoceanic SSTA gradient between the TWP and tropical Indian Ocean, which is not conducive (conducive) to the development of local active convection over the eastern BOB, thus leading to an abnormally late (early) BOBSM onset. In contrast, during the cold PDO epochs, the superimpositions of the PDO-related SSTAs with El Niño-related (La Niña-related) SSTAs lead to insignificant SSTAs over the TWP and thus a weak zonal SSTA gradient. Consequently, no significant circulation anomalies, especially lower-tropospheric anomalous winds, are induced over the BOB, and thus there is no preference for early or late BOBSM onset for EN_CPDO and LN_CPDO events.

Note that this study just focuses on the SSTA pattern and the related anomalous circulations of the different PDO−ENSO events, but the differences of ENSO-related SSTAs in warm and cold PDO epochs, especially why significant SSTAs over the TWP only appear in warm PDO epochs, should be further discussed. Subsequently, a series of modeling simulations will be performed to reveal and verify the PDO modulating mechanisms. In this study, the PDO is deemed as a decadal background to modify the BOBSM onset−ENSO relationship, in which the classification for the warm (cold) PDO epochs is simply based on the filtered PDO index being greater (less) than zero, without prescribing other thresholds according to the PDO intensity. Recently, some studies have examined the influence of the PDO on climate variations at relatively short time scales. For example, Xue et al. (2018) defined the PDO index above (below) zero as warm (cold) PDO phases but based on 8-yr low-pass filtered data, resulting in much shorter PDO epochs. Watanabe and Yamazaki (2014) even used one-third highest and one-third lowest PDO indices based on 6-yr low-pass filtered data to categorize warm and cold PDO events. In such a manner, the modulating effect of the ENSO−BOBSM onset relationship by the PDO may be recognized more clearly in the future.

Acknowledgments. This research was jointly supported by the National Key Research and Development Program of China (Grant No. 2018YFC1506004), the Priority Research Program of the Chinese Academy of Sciences (Grant Nos. XDA19070404 and QYZDY-SSW-DQC018), the Natural Science Foundation of China (Grant Nos. 41705065, 41876020 and 41730963), the SOA Program on Global Change and Air−Sea Interactions (Grant No. GASI-IPOVAI-03), the Foundation of Sichuan Education Department (Grant No. 18ZB0122), and the Open Foundation of the Plateau Atmosphere and Environment Key Laboratory of Sichuan Province (Grant No. PAEKL-2017-Y6).