Strengthened Regulation of the Onset of the South China Sea Summer Monsoon by the Northwest Indian Ocean Warming in the Past Decade

Yang AI, Ning JIANG, Weihong QIAN, Jeremy Cheuk-Hin LEUNG, and Yanying CHEN

1State Key Laboratory of Severe Weather (LASW) and Institute of Climate System,Chinese Academy of Meteorological Sciences, Beijing 100081, China

2Department of Atmospheric and Oceanic Sciences, Peking University, Beijing 100871, China

3Guangzhou Institute of Tropical and Marine Meteorology/Guangdong Provincial Key Laboratory of Regional Numerical Weather Prediction, China Meteorological Administration, Guangzhou 510641, China

4School of Marine Sciences, Nanjing University of Information Science &,Technology, Nanjing 210044, China

ABSTRACT Traditionally, a delayed (early) onset of the South China Sea summer monsoon (SCSSM) has been observed to follow a warm (cold) El Niño-Southern Oscillation (ENSO) event in winter, supporting high seasonal predictability of SCSSM onset. However, the empirical seasonal forecasting skill of the SCSSM onset, solely based on ENSO, has deteriorated since 2010. Meanwhile, unexpected delayed onsets of the SCSSM have also occurred in the past decade. We attribute these changes to the Northwest Indian Ocean (NWIO) warming of the sea surface. The NWIO warming has teleconnections related to (1) suppressing the seasonal convection over the South China Sea, which weakens the impacts of ENSO on SCSSM onset and delays the start of SCSSM, and (2) favoring more high-frequency, propagating moist convective activities, which enhances the uncertainty of the seasonal prediction of SCSSM onset date. Our results yield insight into the predictability of the SCSSM onset under the context of uneven ocean warming operating within the larger-scale background state of global climate change.

Key words: Indian Ocean, South China Sea summer monsoon, monsoon onset, ENSO

1. Introduction

The onset of the South China Sea Summer Monsoon(SCSSM) is characterized by an apparent seasonal circulation that causes near-surface winds to turn from easterlies to westerlies accompanied by bursts of deep moist convection(Xie et al., 1998; Wang et al., 2004; Shao et al., 2015). This seasonal transition generally occurs around mid-May, marking the beginning of the rainy season associated with the East Asian Summer Monsoon (Lau and Song, 1997). Its onset processes include an eastward retreat of the Western North Pacific Subtropical High (WNPSH), the rapid northeastward shift of the South Asian High (SAH), and the generation of the monsoon trough and low-level cross-equatorial flow (CEF) over the South China Sea (SCS) (Liu and Zhu,2016). The onset and development of the SCSSM exert significant socio-economic impacts in eastern China and affects global climate through its teleconnections (He and Zhu,2015). Therefore, the accurate prediction of the SCSSM onset is an important issue in the overall seasonal forecast(Zhu and Li, 2017).

As a critical stage of the seasonal cycle in Asian monsoon (Ding and Chan, 2005; Kueh and Lin, 2010), the onset process of the SCSSM is usually associated with a slow seasonal evolution superimposed with rapid, sub-seasonal fluctuations (Zhou and Chan, 2005; Wang et al., 2018; Liu and Zhu, 2021). Accordingly, the SCSSM onset can be regulated by factors that operate on different time scales. Contributing to the understanding and consequent prediction of SCSSM onset, various factors affecting the SCSSM onset have been documented, including the characteristics of the tropical oceans (Zhou and Chan, 2007; Yuan et al., 2008),the Tibetan Plateau (Wu and Zhang, 1998), subseasonal oscillations (Zhou and Chan, 2005), and high-latitude influences including the North Atlantic Oscillation (Liu and Zhu,2019), Arctic Oscillation (Gong and Ho, 2003) and Antarctic Oscillation (Nan and Li, 2003), in addition to synoptic activities, such as typhoons and midlatitude fronts (Mao and Wu, 2008; Tong et al., 2009). Under the influence of these factors, the onset timing of the SCSSM exhibits considerable interannual and interdecadal variability (Xie et al.,1998; Kajikawa and Wang, 2012; Feng and Hu, 2014; Luo et al., 2016).

Among these factors, extensive studies have focused on understanding the role of the El Niño-Southern Oscillation(ENSO) in regulating the variation of SCSSM onset on both interannual and interdecadal time scales (Zhou and Chan,2007; Kajikawa and Wang, 2012; Liu et al., 2016a; Jiang and Zhu, 2021). This relevant seesaw sea surface temperature (SST) anomaly pattern in the tropical Pacific Ocean greatly affects the SCSSM onset by altering the Walker circulation. On the interannual time scale, an ENSO event from the previous winter acts as a good precursor for the seasonal forecast of the SCSSM onset, which tends to delay (or advance) the onset of SCSSM by strengthening (or weakening) the WNPSH, respectively. This relationship between ENSO and the SCSSM onset provides some seasonal predictability of SCSSM onset (Zhou and Chan, 2007; Martin et al., 2019). On interdecadal time scales, the SCSSM onset was, on average, advanced by about two weeks in 1994-2010 compared to 1979-93 (Kajikawa and Wang,2012). This observation was attributed to the combined effects among the interdecadal variations which include, a La Niña-like SST signature in the Pacific basin, an abrupt SST warming near the Philippine Sea (Xiang and Wang,2013), and the trend of western tropical Pacific upper ocean heat content (Feng and Hu, 2014).

However, despite the continuously warming SST and upper ocean heat content of the western tropical Pacific, the SCSSM onsets were generally delayed in the past decade(Jiang and Zhu, 2021). In addition, and somewhat unexpectedly, the SCSSM onset occurred late in 2018 following a La Niña event (Liu and Zhu, 2019) and extremely early in 2019 following an El Niño event (Hu et al., 2020; Liu and Zhu, 2020). The unstable relationship between ENSO and SCSSM onset resulted in the failure of the empirical model of SCSSM onset based on ENSO indices alone (Liu and Zhu, 2019; Hu et al., 2020; Liu and Zhu, 2020; Jiang and Zhu, 2021). Our recent work has partly attributed these changes to the frequent cold tongue La Niña events, emphasizing the effect of the ENSO diversity on the SCSSM onset on seasonal time scales (Jiang and Zhu, 2021). Meanwhile,we have also noticed that the SST in the Indian Ocean has exhibited an obvious warming trend that occurred simultaneously with the abovementioned significant variations in SCSSM onset in the past decade.

Previously, many studies have revealed a relationship between the SCSSM onset and the Indian Ocean (Liang et al., 2006; Yuan et al., 2008; Jing and Chong-Yin, 2011;Chen et al., 2012; Zhang et al., 2019). For instance, basinwide warming (cooling) in the tropical Indian Ocean can induce an anomalous reversed (intensified) Walker circulation over the Indo-Pacific region, causing a delayed (an early) SCSSM onset with an intensified (weakened) western Pacific anticyclone in April and May. The Indian Ocean also plays an important role in prolonging ENSO effects(Yuan et al., 2008). Furthermore, some studies have emphasized the impact of the Northwest Indian Ocean (NWIO) on the SCSSM onset (Chen et al., 2012). A larger (smaller) airsea heat flux over the NWIO can cause a later (an earlier)SCSSM onset, and this may be related to the Indian Ocean dipole mode, which can regulate the low-level westerlies across the equatorial Indian Ocean. In addition, an increasing number of studies have been devoted to the effect of the subseasonal oscillations and other high-frequency activities,and a large part of these high-frequency disturbances originate in the Indian Ocean (Shao et al., 2015; Wang et al.,2018; Liu and Zhu, 2021). Recent studies have revealed that typhoon Fani (2019) and 30-60 day oscillations originating from the Indian Ocean may have triggered an extremely early SCSSM onset in 2019 following an El Niño event (Hu et al., 2020; Liu and Zhu, 2020).

The interannual variations of the Indian Ocean SST are thought to be linked to ENSO events. However, along with the rapid warming trend in the NWIO and the changes in ENSO diversity, the above facts suggest that the regulation of the SCSSM onset by the NIWO seems strengthened. Evidence suggests that the NWIO warming may indeed be the cause for the unstable relationship of SCSSM onset with ENSO and the late-onset of SCSSM in the past decade.

The remainder of this paper is organized as follows. Section 2 describes the data and methods, section 3 outlines the subseasonal and seasonal processes, section 4 outlines the effects of NWIO warming, and section 5 provides a brief discussion and concludes.

2. Data and Methods

The daily and monthly atmospheric variables are taken from the National Centers for Environmental Prediction(NCEP)/Department of Energy (DOE) Reanalysis 2 (Kanamitsu et al., 2002) with a 2.5° × 2.5° resolution from 1979 to 2019. The National Oceanic and Atmospheric Administration (NOAA) interpolated outgoing longwave radiation(OLR) dataset (Liebmann and Smith, 1996) is used as a proxy for convection with the same resolution and temporal coverage as the NCEP/DOE Reanalysis 2. The monthly SST is from the Hadley Centre Sea Ice and SST data set version 1 (HadISST1) with a 1° × 1° resolution (Rayner et al.,2003).

The activities of the Madden-Julian Oscillation (MJO)are described by the Real-time Multivariate MJO (RMM)index provided by Japan Meteorological Agency (JMA) and OLR-based MJO index (OMI) provided by NOAA. An active MJO day is defined when the MJO amplitude exceeds one standard deviation.

The Ensemble Empirical Mode Decomposition(EEMD) analysis (Wu et al., 2009) and the Butterworth bandpass filter isolate the sub-seasonal and seasonal components. Meanwhile, the composition, correlation, and linear trend analyses are also used to describe the features of SCSSM onset. A two-tailed Student’s t-test is used for significance testing.

3. Changes in roles of seasonal and subseasonal processes on the SCSSM onset

According to previous studies, the SCSSM onset can be affected by seasonal and sub-seasonal processes (Zhou and Chan, 2005; Zhou and Chan, 2007; Tong et al., 2009; Zhu and Li, 2017; Wang et al., 2018; Liu and Zhu, 2021). Meanwhile, there may be changes in the leading influential position among these two types of processes that are based on different timescales. This section examines the differences in the nature of the roles of the seasonal and sub-seasonal processes on the SCSSM onset.

Though there are many definitions for the SCSSM onset from different aspects and approaches (Qian and Yang, 2000; Mao et al., 2004; Wang et al., 2004; Shao et al., 2015), most of them are generally consistent with one another. The SCSSM onset is defined by both circulation and convection factors following Shao et al. (2015) and is roughly classified into two types according to the relative contributions of the seasonal and sub-seasonal processes in the following analysis. A type I case reflects a year when the seasonal processes play a dominant role in the SCSSM onset with only a slight modification by sub-seasonal processes;in contrast, type II cases feature an SCSSM onset that is strongly modulated by the sub-seasonal processes. The classification is based on the differences among three SCSSM onset dates, which are defined by the original data (D0), a low-frequency component by EEMD (D1), and the sum of the low-frequency component and sub-seasonal component by EEMD (D2), respectively [Figs. S1, S2 in the electronic supplementary material (ESM)]. The low-frequency component is characterized by a smooth seasonal cycle (Figs. S3,S4 in the ESM), which also includes an annual cycle as well as interannual, interdecadal, and long-term trend signals(Fig. S1), and it regulates the seasonal processes of the SCSSM onset during the early summer (Fig. S4). Further details about the definitions and classification can be found in electronic supplementary material .

Results show that all type I events occurred before 2009, and in contrast, the SCSSM onset was strongly controlled by sub-seasonal processes during the past decade,shown as type II events (Fig. 1a). Most extremely late or early onset events were concerning, such as in 2018 (Liu and Zhu, 2019) and 2019 (Hu et al., 2020; Liu and Zhu,2020), each of which was classified as a type II event. Meanwhile, type I events rarely deviate from their normal onset date (e.g., 2000-05).

Along with the changes of the SCSSM onset type, the SCSSM onset date also shows a decadal variability during past decades (Fig. 1a), late-onset (P1: 1979-93), early-onset(P2: 1994-2009), late-onset (P3: 2010-19). Notably, the early-onset of P2 has been attributed mainly to the west Pacific warming (Kajikawa and Wang, 2012). Though the west Pacific continued to warm (Figs. 2a, b), the SCSSM onset was generally delayed in P3. There is about a 9.5-day delay of D0 in P3 compared to P2.

Fig. 1. (a) The D0 (bars), D1 (solid green line with a square), and D2 (red dashed line with a triangle) of SCSSM onset from 1979 to 2019 and the averaged D0 (black lines)/D1 (green lines) in P1 (1979-93), P2 (1994-2009), and P3 (2010-19) periods are shown. The green squares and red triangles along the lower axis indicate type I and type II events, respectively. The red/blue bars indicate El Niño/La Niña years. Panel (b) shows the correlations between D0/D1/D2 and DJF Niño-3.4 index (black/red/blue lines) for the three periods (P1, P2, and P3).

Meanwhile, the correlation between D0 and the averaged December, January, and February (DJF) Niño-3.4 indices during the previous winter has also dropped significantly in P3, partly attributed to the more frequent cold tongue La Niña events (Jiang and Zhu, 2021). However, the correlation between D1 and the Niño-3.4 index (green bars in Fig. 1b) in the previous winter is statistically significant over the whole period, including P3 (Fig. 1b). This finding suggests that ENSO affects the SCSSM onset primarily through seasonal processes, and such a relationship was still valid in P3. Weakened correlations between D0/D2 and ENSO in P3 imply that sub-seasonal processes disturb the seasonal impacts from ENSO on the SCSSM onset and reduce the SCSSM onset seasonal predictability. The above results demonstrate that the SCSSM onset date has been overall delayed since 2010, which is not consistent with the results of previous studies. Furthermore, the recently low SCSSM seasonal predictability based on ENSO is mainly attributed t o sub-seasonal processes.

4. Impacts of NWIO warming on the SCSSM onset

It is shown that neither the interdecadal relationship of SCSSM onset with the SST over the west Pacific Ocean nor the interannual relationship of SCSSM onset with ENSO activity can explain the SCSSM onset time in the past decade, suggestive of another controlling factor that cannot be neglected. Considering that the SST change over the Indo-Pacific Ocean can modulate the SCSSM onset via both seasonal processes and the activities of subseasonal oscillations (Zhou and Chan, 2005; Tong et al., 2009), the effects of the NWIO warming are a candidate mechanism for the changes in SCSSM onset in the past decade.

In the context of climate change, the Indo-Pacific Ocean is an area that has experienced significant sea-surface surface warming. This phenomenon can be analyzed through a least-squares linear regression utilized to estimate the linear trend (Fig. 2a). In this formulation, the coefficient of determination (R2) is computed as the ratio of [the explained sum of squares (SSE) and the total sum of the squares (SST)], expressed mathematically by R2=SSE/SST, which can be interpreted as the proportional variation of the linear trend (Fig. 2b). Evaluation of R2shows that the NWIO exhibits its most significant warming trend that is largely independent of interannual variation (Fig. 2c). After removing the ENSO signal by linear regression, the rapid NWIO warming in May (Fig. S5 in the ESM), along with enhanced convection, induces the descending air associated with the anticyclone over the SCS (Fig. 2d), which may account for the recent changes in the SCSSM onset.

4.1. On the seasonal timescale

The correlations between SST anomalies in February,March, and April (FMA SSTA) and the onset date of the SCSSM are examined (Figs. 3a-d). Consistent with previous studies (Yuan et al., 2008; Zhang et al., 2019), the correlation pattern of type I years in Fig. 3a reveals the effect of an ENSO-related signal on the seasonal processes of the SCSSM onset, while type II events seem only to be correlated with the SSTAs local to the SCS, particular in the NWP (Fig. 3b). Compared to the period of 1979-2009, in the past decade, the correlations increase in the NWIO(5°S-20°N, 50°-80°E) and decrease in Northwest Pacific(NWP, 0°-20°N, 120°-150°E), implying a more robust relationship between the NWIO SST and the SCSSM onset date over the past decade (Figs. 3c vs. 3d).

The SST around the warm pool region is near the convection threshold (Jiang and Zhu, 2020). Along with the significant warming in the NWIO, the difference in atmospheric conditions between P3 and P2 exhibits an anomalous zonal circulation over the north Indo-Pacific Ocean averaged over 5°-15°N (Fig. 3e). The descending branch around the SCS has the net effect of strengthening the western Pacific anticyclone in May and weakening the cross-equatorial flows over the SCS and NWP (Fig. 3f). Meanwhile, strong low-level convergence and convection are also found in NWIO (Fig. 3f).In this way, the suppressed convection and reduced crossequatorial flows have collectively served to delay the SCSSM onset.

The persistence and memory of sea-surface temperature anomalies (SSTAs) and circulations from the previous winter to spring can maintain the linear relationship of the ENSO-SCSSM, thus assisting the seasonal forecasting of SCSSM onset (Zhou and Chan, 2007; Zhu and Li, 2017; Martin et al., 2019). Meanwhile, the Indian Ocean SST warming (cooling) often follows and may be indirectly associated with a fading El Niño (La Niña) event, such as is described by the Indian Ocean Capacitor effect (Xie et al.,2009), which plays an important role in prolonging the ENSO effects (Yuan et al., 2008). However, any NWIO SSTA cooling (post-La Niña) is partially offset by the background, rapid warming which has taken place over the past decade (Fig. S5). It means that when a La Niña event occurs, the NWIO can hardly drop to a cooling state due to the rapid warming trend in NWIO. Considering the frequent cold tongue La Niña events in recent decades, the Indian Ocean Capacitor effect from ENSO is speculated to have weakened (Fig. 2c). The consistency between Fig. 3f and Fig. 2d further confirms the dominance of the role of the NWIO warming trend as opposed to ENSO concerning the regulation of the SCSSM. The suppressed convection over the SCS induced by the rapid NWIO warming trend not only delays the onset of the SCSSM but also weakens the ENSO influence.

Fig. 2. (a) The trend of monthly SSTs during 1979 to 2019, in which the dotted area indicates trends that are significant at the 95% confidence level. (b) The coefficient of determination (R2) that corresponds to (a). (c)The monthly SSTA averaged within the NWIO region (shading), which is relative to the climate state of 1981-2010. The solid black line indicates the linear trend for the SST at NWIO, which is significant at the 95% confidence level. (d) The regressed 850 hPa wind (vectors) and OLR (shading) in May onto the NWIO SST after the removal of the ENSO signal by linear regression (bars in Fig. S3) from 1979 to 2019. The red/blue vectors denote the differential significance of southerlies/northerlies at the 95% confidence level.The two black rectangles from left to right denote the concerned area of NWIO and NWP, respectively.

4.2. Sub-seasonal timescale processes

Recently, the sub-seasonal processes played a more important role in triggering the SCSSM onset (Fig. 1a) (Hu et al., 2020; Liu and Zhu, 2020). The sub-seasonal disturbances are strongly dependent on the seasonal state of the SST (Wu, 2010; Liu et al., 2016b; Wu, 2018), and the tropical Indian Ocean and NWP are the central areas for the eastward and westward propagating subseasonal oscillations affecting the onset of SCSSM, respectively. (Zhou and Chan, 2005; Wang et al., 2009; Li et al., 2017; Wang et al.,2018) (Fig. 4a). The warming NWP and NWIO SSTs can activate more sub-seasonal convection, which reduces the seasonal predictability of SCSSM onset. This has evidentiary support considering that even when following an El Niño event, the SCSSM onset can still arrive early due to the subseasonal disturbances from the Indian Ocean, similar to what occurred in 1995, 2003, and 2019 (Fig. 1a, Fig. S7 in the ESM): As a consequence of the Indian Ocean Capacitor effect, the Indian Ocean tends to become anomalously warm in spring following the El Niño event, which can serve to trigger additional convective disturbances which can go on further to affect the SCSSM onset (Fig. 4b).

High SSTs over the tropics are usually accompanied by enhanced convective activity (Roxy, 2014; Jiang and Zhu,2020). This premise is supported by a negative shift in the probability distribution of the averaged OLR over the NWIO observed over the past decade, which likely occurred in response to the rapid warming in NWIO (Fig. 4c). It follows that more sub-seasonal and synoptic convective complexes were activated. Liu and Zhu (2020) have proven that the tremendous effect exerted by typhoon Fani (2019) on the SCSSM onset in 2019 originated in the Indian Ocean. In addition, the activities related to subseasonal oscillations in the Indian Ocean, such as the MJO (Madden and Julian,1971), also tend to increase (Figs. 4d and S6 in the ESM).As an ocean-atmosphere-coupled convective phenomenon,MJO activity is highly dependent on tropical SSTs, with higher MJO activity typically occurring when SSTs are higher (Arnold et al., 2013). The eastward propagation of the MJO plays an important role in the establishment of the large scale near-surface westerly over the SCS (Tong et al.,2009), which can affect the SCSSM directly (Lawrence and Webster, 2002; Li et al., 2017). Hence, the more active subseasonal oscillations originating from the NWIO may lead to the anomalous interannual variability of the SCSSM onset in recent years, which also serves to inhibit the seasonal impact from ENSO and further reduces the seasonal predictability of the SCSSM onset. However, the relationship between Indo-Pacific warming and MJO is quite complicated and needs further study (Roxy et al., 2019; Huang et al., 2021).

Fig. 3. Correlation coefficient (shading, 0.1 intervals) between D0 and FMA SSTA during (a) type I event years and(b) type II event years. Panels (c) and (d) are the same as (a) and (b) except for the D1 during 1979-2009 and 2010-19. The black dots denote correlations significant at the 95% confidence level. Panel (e) is the zonal circulation difference between 2010-19 and 1994-2009 in May averaged over 5°-15°N, in which the streamlines are plotted based on zonal velocity (m s-1) and the vertical velocity (Pa s-1) multiplied by 100. The vertical velocity difference(Pa s-1, 0.01 interval) is shown as shading, in which dotted areas denote its significance over 90% confidence level.(f) The difference of wind at 850 hPa (vectors, m s-1) and OLR (shading, 1 W m-2 interval) fields are shown consistent with the periods in (e). The red/blue vectors and dotted areas denote the differential significance of southerlies/northerlies and OLR at the 90% confidence level.

Fig. 4. The evolution of composite 10-90 day bandpass filtered OLR (shading, 5 W m-2 intervals) averaged over 5°-15°N between 20 days before and after the SCSSM onset date based on D0 (a) during in 1979-2019 and (b) during El Niño years.The dotted areas tested at the 95% confidence level. (c) The probability distribution of OLR in May averaged over the NWIO during 2010-19 (solid line), 1979-2009 (dashed line), type I years (red line), and type II years (blue line). Panel (d)shows the activity days of MJO in phases 2-3 in May based on the RMM index from the JMA. The black line is the 5-year running averaged series for the days, and the blue lines are the linear trend resulting from a linear least-squares regression in which both trends are significant at the 95% confidence level.

5. Conclusions and Discussion

Much effort has been devoted to the seasonal forecast of the SCSSM onset. Aside from the interannual variations,the changes of the SCSSM onset also exhibit decadal shifts.In the past decade, the SCSSM onset seems to be delayed,and the robustness of the ENSO-SCSSM relationship has also become weaker. Furthermore, sub-seasonal processes have recently played a more dominant role in triggering the SCSSM onset than seasonal processes. These changes have reduced the seasonal predictability for the SCSSM onset.

Our results attribute these changes to the rapid warming in NWIO, which is illustrated in Fig. 5. Compared to the past decades (1979-2009), the recent NWIO warming has enhanced seasonal moist convection, inducing an anomalous zonal circulation, which is characterized by a descending branch over the SCS and a weakened cross-equatorial flow into the SCS. These dynamics jointly suppress the convection over the SCS and delay the SCSSM onset date. In addition, the persistent warming trend in NWIO has reduced the Indian Ocean Capacitor effect, which acts to reduce the ENSO influence. Moreover, the warming SSTs in the NWIO have activated more sub-seasonal disturbances,which has dominated the interannual variations of the SCSSM onset over the past decade.

Fig. 5. Schematic diagram of the modulation of the NWIO warming on the SCSSM onset. Under the background context of global warming, the NWIO displays a significant increasing trend over the past 41 years (Figs. 2a, b).From a seasonal timescale perspective, warm SSTs trigger more convection which leads to the change of zonal circulation (Fig. 3a), within which the anomalous descending branch suppresses the seasonal convection activity in SCS, strengthens the anomalous anticyclone in the NWP, and reduces the intensity of cross-equatorial flows from the Indian Ocean and northern Australia (Fig. 3b), resulting in the delay of the SCSSM onset in the last decade (Fig. 1a).Meanwhile, the NWIO warming inhibits the influences from ENSO (Figs. 3b, d) and reduces the seasonal predictability of the SCSSM onset. From a sub-seasonal timescale perspective, the convergence of an enhanced Somali jet and an anomalous easterly flow from the SCS produces more convective activities (Figs. 4c, d); eastward propagation of these convective complexes can also reduce the seasonal predictability of the SCSSM onset.

Though the NWIO has gradually warmed since the 1980s, the delay in the SCSSM onset was not observed unitl recently. The following reasons could explain this. First, the NWIO affects the SCSSM onset mainly through the air-sea heat flux, particularly through deep moist convection and the associated latent heat flux (Chen et al., 2012). Hence,the impacts of the NWIO may have emerged in the past decade due to the absolute temperature of the NWIO SST beginning to exceed the moist convective threshold (Roxy, 2014;Jiang and Zhu, 2020). On the other hand, the interannual variations of the Indian Ocean SST are usually tied to the ENSO events, and the ENSO events have previously exerted strong effects on the SCSSM onset (Zhou and Chan,2007; Yuan et al., 2008). Compared to the periods before 2009, the frequent cold tongue La Niña events favor a delay in the SCSSM onset (Jiang and Zhu, 2021). It is worthy to note that we mainly have focused on the effect of the tropical ocean in this study, and the influences from the mid-latitude wave activities (Liu and Zhu, 2019) cannot be ruled out.

Our results provide further insight into the impacts of global warming on seasonal forecasting. In the context of global warming, the NWIO SST warming with enhanced convection could not only inhibit the influences from other potential precursors (e.g., ENSO) on the SCSSM, but also activate more sub-seasonal disturbances which propagate into the SCS, both of which reduce the seasonal predictability.

Acknowledgements.This study was jointly supported by the National Natural Science Foundation of China (Grant Nos.42005011, 41830969), the Scientific Development Foundation of the Chinese Academy of Meteorological Sciences (CAMS) (Grant No. 2020KJ012 and 2020KJ009), the Basic Scientific Research and Operation Foundation of CAMS (Grant Nos. 2021Z004). This study was also supported by the Jiangsu Collaborative Innovation Center for Climate Change.

Electronic supplementary material:Supplementary material is available in the online version of this article at https://doi.org/10.1007/s00376-021-1364-8.