Role of tropical intraseasonal oscillations in the South China Sea summer monsoon withdrawal in 2010

Based on the ensemble empirical mode decomposition of low‐level winds and outgoing longwave radiation, this paper analyzes the influence of tropical intraseasonal oscillations (ISOs) on one extremely late case of South China Sea (SCS) summer monsoon withdrawal (SMW). Compared to the climatological monsoon withdrawal of September 25, the South China Sea summer monsoon withdrawal in 2010 was delayed for approximately 1 month (October 26) and is the latest in the past 60 years. The quasi‐biweekly oscillation (QBWO) and Madden‐Julian oscillation (MJO) originating from the equatorial Indian Ocean propagate northeastward to the SCS. The accompanying circulation first induced a westerly burst event and prolonged the lifespan of the summer monsoon. After this transitory recovery, the circulation anomalies associated with the current propagation of ISOs into the SCS contributed to the large‐scale circulation adjustment, leaded to the shift of zonal wind (from westerlies to easterlies) and triggered the SCSSM withdrawal. This case study shows that, in addition to the well‐studied impacts of ISOs on the monsoon onset and active/break cycle, the ISOs can also trigger the monsoon withdrawal.

Based on the ensemble empirical mode decomposition of low-level winds and outgoing longwave radiation, this paper analyzes the influence of tropical intraseasonal oscillations (ISOs) on one extremely late case of South China Sea (SCS) summer monsoon withdrawal (SMW). Compared to the climatological monsoon withdrawal of September 25, the South China Sea summer monsoon withdrawal in 2010 was delayed for approximately 1 month (October 26) and is the latest in the past 60 years. The quasi-biweekly oscillation (QBWO) and Madden-Julian oscillation (MJO) originating from the equatorial Indian Ocean propagate northeastward to the SCS. The accompanying circulation first induced a westerly burst event and prolonged the lifespan of the summer monsoon. After this transitory recovery, the circulation anomalies associated with the current propagation of ISOs into the SCS contributed to the large-scale circulation adjustment, leaded to the shift of zonal wind (from westerlies to easterlies) and triggered the SCSSM withdrawal. This case study shows that, in addition to the well-studied impacts of ISOs on the monsoon onset and active/break cycle, the ISOs can also trigger the monsoon withdrawal.

K E Y W O R D S
Madden-Julian oscillation, quasi-biweekly oscillation, South China Sea, summer monsoon withdrawal, tropical intraseasonal oscillation

| INTRODUCTION
October 2010 witnessed the latest monsoon withdrawal over the South China Sea (SCS) and the heaviest autumn rainstorm on Hainan Island (located in the northern SCS) of the past 60 years (Figure 1a). According to the National Climate Center (NCC, 2016) of China Meteorological Administration, the South China Sea summer monsoon withdrawal (SCSSMW) in 2010 occurs in pentad 59 (fifth pentad of October), which is five pentads later than the climatology. Hainan Island experienced two sustained rainstorms (September 30 to October 9 and October 13 to 18; Figure 1b) that affected more than 2.5 million people and caused great economic losses (Cai et al., 2013;Qiao et al., 2015). Previous studies indicated that the tropical intraseasonal oscillations (ISOs), especially the quasi-biweekly oscillation (QBWO), have played important roles in these severe rainstorms (Jian and Zhang, 2013;Qiao et al., 2015;Wang et al., 2016;Li et al., 2017). Since Hainan Island is located in the northern SCS and these rainstorms occurred just before the SCSSMW, it is reasonable to question whether the ISOs affected the monsoon withdrawal in 2010 as well.
It is an established fact that the ISOs and the monsoon system are inherently linked. The seasonal march of monsoon circulation can strongly modulate the ISOs activities as manifested by, for example, the different behavior of ISOs in early and late summer (Yang et al., 2008;Hsu, 2012). The so called climatological ISOs are phase locked to the annual cycle and are very significant in the Asian monsoon region (Wang and Xu, 1997;Hsu, 2012). ISOs can not only trigger the monsoon onset (Wheeler and Hendon, 2004;Tong et al., 2009;Kajikawa and Wang, 2012;Lee et al., 2013) but can also modulate the active/break cycle of the rainy season (Ding and Chan, 2005;Mao and Chan, 2005;Hsu, 2012). However, it remains unclear if the ISOs affect the withdrawal of the monsoon as well. Previous studies concerning ISOs focus mainly on summer and winter seasons; however, the transition seasons (corresponding to monsoon onset/demise) are also very important and the ISOs often display strong seasonality and regionality (Yang et al., 2008;Hsu, 2012). Based on the case study of October 2010, we shall investigate the contribution of ISOs to this extremely late SCSSMW. The datasets and methods applied in this paper are described in Section 2. Section 3 investigates the influence of ISOs on the SCSSMW in 2010. The summary and discussions are presented in Section 4.

| DATA AND METHODS
The datasets applied in this study includes: (a) European centre for medium-range weather forecasts interim reanalysis (ERA-interim reanalysis) data (Dee et al., 2011)  The ensemble empirical mode decomposition (EEMD; Wu and Huang, 2009), which is an improvement of the empirical mode decomposition (Huang et al., 1998) designed for analyzing nonstationary and nonlinear time series, is used as an adaptive filter (without a priori determined basis) to obtain the QBWO and Madden-Julian oscillation (MJO) components. By using the EEMD, the problem of subjective selection of thresholds (for example: 10-20 days by Qiao et al. (2015) or 8-15 days by Li et al. (2017)) in other bandpass filters can be avoided. The Morlet wavelet (Torrence and Compo, 1998) is employed to analyze the spectrum characteristics and to define the statistically significant ISOs.
For the definition of SCSSMW, there are two criteria in NCC (2016): (a) steadily shifting zonal wind from westerlies to easterlies, and (b) the potential pseudo-equivalent temperature steadily less than 340 K, at 850 hPa in the key SCS region (110 -120 E, 10 -20 N; gray rectangle in Figure 3a). However, these criteria in NCC (2016) Zhang et al. (2014) quantitatively defined the steady state as: (a) The state is maintained at least for three pentads and can be interrupted for no more than two pentads thereafter, or (b) The state is maintained for two pentads, and is then interrupted for one pentad but immediately returns to the state before interruption. In fact, most of the SCSSMW dates defined by the subjective monitor (NCC, 2016) are the same as those from objective criteria (Zhang et al., 2014), and their correlation is 0.74 for 1951-2012 (significant at 99.9% confidence level). For the case of 2010, both the NCC (2016) Figure 1b shows the OLR and low-level winds averaged in 110 -120 E in 2010. The heavy convection in early-to-mid October corresponds to the heavy rainfall in Hainan. As documented in the introduction, the extreme autumn rainfall and the extremely late monsoon withdrawal in 2010 is worthy of investigation. Although there appear substantial easterlies after late-September (Figures 1b and 2a), the pseudoequivalent temperature is still greater than 340 K (see the SCSSM monitoring in the NCC website, that is, http://cmdp. ncc-cma.net/Monitoring/EastAsian/p8ciu2010126.gif), indicating that the SCSSM has not yet ended. The two criteria in NCC (2016) and Zhang et al. (2014) were not both satisfied until late October, when the SCSSM finally ended. Figure 2a shows the climatological mean and standard variation of 850 hPa zonal winds averaged in SCS. Climatologically, the transition of westerlies to easterlies occurs on September 25, which is very near to pentad 54 determined by the averaged withdrawal pentad of each year (Zhang et al., 2014;NCC, 2016). This climatological shift in zonal wind is mainly due to the westward intrusion of western North Pacific subtropical high. Other prominent changes during SCSSMW includes retreat of the monsoon trough/ rain belt over the SCS, anomalous low-level anticyclone over the northern SCS, and deceleration of the upper-level tropical easterly stream (see Hu et al., 2018 for detail). The red curve in Figure 2a shows the 850 hPa zonal wind for 2010. The persistent change from westerly winds to easterly winds occurs on October 26 (at this time, the pseudoequivalent temperature has shifted to less than 340 K). This result is also very close to pentad 59 and is approximately 1 month later than the mean monsoon withdrawal (Zhang et al., 2014;NCC, 2016).
To reveal the intraseasonal characteristics of zonal winds in Figure 2a, the wavelet spectrum is calculated and shown in Figure 2b. In October, the synoptic-scale (less than 1 week) oscillation and QBWO are very prominent and are significant at the 95% confidence level. Although the MJO with an approximate period of 30-60 days is also very strong in the wavelet spectrum, it does not pass the 95% significance test. The wavelet spectrum results of zonal winds over the large-scale SCS are in accordance with previous studies on several monsoonal parameters over Hainan Island, for example, Fourier spectrum for zonal winds (Jian and Zhang, 2013) and precipitation (Qiao et al., 2015), and wavelet spectrum for OLR (Wang et al., 2016) and rainfall (Li et al., 2017). respectively. Although MJO is not significant in the wavelet spectrum, it cannot be ignored for the following reasons: First, Figure 2c shows that the amplitudes of QBWO and    (Wheeler and Hendon, 2004) and boreal summer intraseasonal indices (Lee et al., 2013), show that the amplitude of MJO is very strong in October (not shown). Fourth, previous studies on the sustained rainstorms over Hainan Island in October 2010 (Qiao et al., 2015;Wang et al., 2016) have also mentioned the modulation of MJO. Lastly, both the QBWO and MJO are in the transition phase of westerlies to easterlies in October 26 (day of monsoon withdrawal), suggesting that they may contribute to the monsoon withdrawal. Thus, in the analysis below we shall investigate the propagation of both QBWO and MJO, and their possible influence on the SCSSMW in 2010. On October 18 and 20 (Figure 3a,b), the SCS is dominated by anomalous convection and a cyclone. Thus, the SCS is occupied by westerlies, which leads to the transitory resurgence of the summer monsoon. The positive OLR anomalies in the southern Bay of Bengal indicate suppressed convection. On October 22 (Figure 3c), the suppressed convection propagates northeastward, reaching the Indo-China Peninsula. Meanwhile, the center of active convection propagates from the northern SCS to the Taiwan Strait. On October 24 (Figure 3d), the suppressed convection over the eastern Indo-China Peninsula strongly developed, and the negative OLR anomalies around Taiwan had almost dissipated. On October 26 (Figure 3e, day of monsoon withdrawal), the SCS is dominated by positive OLR anomalies and an anomalous anticyclone. The averaged wind has shifted to easterly (see also Figure 2c). After that, on October 28 (Figure 3f ), the suppressed convection continues to move eastward, and the easterly winds over the SCS become very prominent, signifying a robust summer monsoon withdrawal. It should be noted that the QBWO originating from the Indian Ocean in Figure 3 is different from those originating from the tropical western Pacific, which are suggested to be the equatorial Rossby waves (Kikuchi and Wang, 2009;Chen and Sui, 2010;Lee et al., 2013). Previous studies on this eastward propagating QBWO in October 2010 suggest that it can be interpreted as the coupled Kelvin-Rossby wave (Qiao et al., 2015;Li et al., 2017). As stated in Section 3.1, the MJO also contributes to the SCSSMW in 2010. Thus, the EEMD filtered MJO component (the fourth intrinsic mode function) of OLR and winds at 850 hPa are shown in Figure 4. On October 10 (Figure 4a), the SCS is dominated by anomalous convection and exhibits cyclonic circulation. Similar to QBWO, the anomalous westerlies also contribute to the transitory resurgence of the summer monsoon. Meanwhile, there appears anomalously suppressed convection in the equatorial India Ocean. On October 14 and 18 (Figure 4b,c), the anomalous positive OLR in the equatorial Indian Ocean strongly develops and propagates northward. Later, on October 22 (Figure 4d), the positive OLR anomalies propagate northeastward and reach the northern SCS, although there appears to be some discontinuity over the Indo-China Peninsula. On October 26 (Figure 4e; day of monsoon withdrawal), the Indo-China Peninsula is dominated by an anomalous anticyclone and strong positive OLR anomalies in the northern SCS. Although the mean MJO component of zonal winds over the SCS is still westerly, it is rather weak and is on the transition phase to easterly (Figure 2c). On October 30 (Figure 4f ), the zonal winds over the SCS finally shift to easterly, which suggests that the eastward propagation of MJO also contributed to this SCSSMW.

| The propagation features of QBWO and MJO
Statistically, the onsets of SCSSM tend to occur when the active phases of MJO are located around the SCS (Lin et al., 2016;Hu et al., 2018). Previous studies also revealed that ISOs can be the trigger of the SCSSM onset. For example, Tong et al. (2009) reported that, by inducing the largescale easterly/westerly shift over the SCS, the MJO play an important role in trigger the monsoon onset. The case of October 2010 is very similar: the circulation anomalies associated with the current propagation of QBWO and MJO into the SCS contributed to the large-scale circulation adjustment (Figures 3 and 4), leaded to the shift of zonal wind (from westerlies to easterlies) and triggered the SCSSM withdrawal (Figure 2). Thus, it can be concluded that in 2010 the collaboration of two ISO modes, namely, the QBWO and MJO, contributed to the late withdrawal of SCSSM.

| SUMMARY AND DISCUSSIONS
Based on the EEMD filtered low-level winds and OLR, this paper analyzes the contribution of ISOs to the extremely late SCSSMW in 2010, and some of the major findings are as follows: 1. In 2010, the summer monsoon withdrawal in the SCS occurred on October 26, which is about 1 week after the severe autumn rainstorm in Hainan Island and is approximately 1 month later than the climatological monsoon withdrawal. 2. The QBWO and MJO originating from the equatorial Indian Ocean propagated northeastward to the SCS, leading to a westerly burst event and prolonging the lifespan of the summer monsoon. The extremely late SCSSMW in 2010 occurs in the transition phase (from westerlies to easterlies) of QBWO and MJO. 3. The case study of 2010 shows that, in addition to the well-studied impacts of ISOs on the monsoon onset and active/break cycle, the ISOs can also trigger the monsoon withdrawal by inducing the easterly/westerly shift.
The above EEMD filtered QBWO and MJO have been verified by first-order bandpass filtering (Murakami, 1979;Qiao et al., 2015). Compared to previous work focusing on the influence of the QBWO on the local rainfall in Hainan Island during the early half of October 2010, this paper analyzes the contribution of ISOs (including MJO) to the largescale circulation change (monsoon withdrawal) in the SCS during the late half of October 2010, which is a case study. The mechanism for the simultaneous arrival of QBWO and MJO, and the possible link to the La Niña background (Jian and Zhang, 2013;Qiao et al., 2015) remains unclear. The RMM index phase points for the SCSSMW dates suggest that statistically the MJO has a weak modulation effect on the SCSSMW (Hu et al., 2018), thus the impact of ISOs in other years needs further study. Syroka and Toumi (2004) mentioned the intraseasonal variations accompanying the South Asian monsoon withdrawal, and Murakami et al. (1986) highlighted the role of low-frequency oscillations in determining the monsoon withdrawal in Australia. Thus the impacts of ISOs on monsoon withdrawal in other monsoon regions are also worthy of investigation.