Subseasonal-to-seasonal (S2S) forecast that bridges the gap between weather forecast and seasonal climate prediction has attracted more and more attention in recent years. Forecasts of this timescale are particularly important to reduce the economic losses and casualties caused by extreme or persistent weather events. At present, it is still a challenging issue to forecast with the timescale from two weeks to three months, that is, the weather-seasonal climate prediction gap (Hendon et al., 2000; Molteni et al., 2007; Neena et al., 2014; Li & Robertson, 2015). Improving the S2S forecast skill is an urgent need for the development of meteorological services, and also a current international trend (Vitart et al., 2012; Hoskins 2013; Mariotti et al., 2018).
Intraseasonal oscillation (ISO) is the major source of the S2S predictability. As one of the main components of tropical climate system variability, ISO exhibits significant seasonal variations (Wang & Rui, 1990; Madden & Julian, 1994; Salby & Hendon, 1994; Zhang & Dong, 2004). The tropical ISO is characterized by equatorially-trapped eastward-propagating convective variability, known as the Madden-Julian Oscillation (MJO) in boreal winter (Madden & Julian, 1994). In boreal summer, the tropical ISO has farther north variability centers and more complex propagation characteristics, which is called the boreal summer intraseasonal oscillation (BSISO) (Wang & Xie, 1997; Lawrence & Webster, 2002; Jiang et al., 2004; Li, 2014). The BSISO over the western North Pacific (WNP) plays a curial role in the evolution of East Asian summer monsoon (EASM) and associated rain belt by changing the large-scale circulation and moisture supply, and also has a high correlation with the extreme weather/climate events in East Asia (Lau & Chan, 1986; Chen & Chen, 1995; Wu & Wang, 2001; Lee et al., 2013; Zhu et al., 2003; Mao et al., 2010; Ren et al., 2013; Li et al., 2015a; Li et al., 2015b; Gao et al., 2016; Hsu et al., 2017; Chen & Zhai, 2017). It provides a physical foundation for the S2S prediction of EASM-related precipitation and atmospheric circulation (Waliser et al., 2003; Wang et al., 2009; Lee et al., 2010; Fu et al., 2013; Lin, 2013). An in-depth understanding of the evolution characteristics and propagation mechanism of the BSISO over WNP is of great significance for diagnosing and predicting the intraseasonal climate variability in East Asia.
It is necessary to extract the BSISO signals in practical research. The timescale-based band-pass filtering of relevant meteorological variables is an informative method (Hong & Ren, 2013; Truong & Tuân, 2019). Due to the need to understand the spatial pattern and temporal evolution, single or multivariate empirical orthogonal function (EOF) analyses on convection, precipitation, and wind field have become a conventional methodology (Zhu et al., 2003; Mao et al., 2010; Jiang et al., 2011; Kikuchi et al., 2012; Lee et al., 2013; Ren et al., 2020). Lee et al. (2013) defined two real-time multivariate indices (BSISO1 and BSISO2) for two BSISO modes through a multivariate EOF (MV-EOF) analysis on the daily outgoing longwave radiation (OLR) and zonal wind at 850 hPa over the Asian summer monsoon region. The first mode represented by BSISO1 indicates a northward and northeastward propagating mode with the convection anomalies originating from the equatorial Indian Ocean (IO) with a timescale of 30-60 days. This mode is also associated with the eastward propagating component along the equator like MJO in winter. The other mode represented by BSISO2 indicates a 10–30-day northwestward propagating mode with convection anomalies propagating northwestward from IO and the Philippine Sea, respectively. These two modes describe the major parts of spatio-temporal variations of BSISO and explain the ISO variabilities over the Asian summer monsoon region to a great extent. Therefore, they have been widely used in recent studies (Li, Mao & Wu, 2015; Hsu et al., 2016; Chen & Zhai, 2017; Hsu et al., 2017; Lee et al., 2017; Diao et al., 2018).
However, just as Lee et al. (2013) pointed out, the third and fourth principal components (PCs) of their MV-EOF decomposition which are defined as BSISO2 have different leading periods, respectively. A power spectra analysis shows that the leading period is around 30 days for PC3 and 10-20 days for PC4. Defining BSISO2 by these two PCs for the follow-up analysis may yield confusion. In addition, the northward propagating signals of convection anomalies over WNP which may remarkably affect the East Asian climate are split into two indices. On the other hand, using self-organizing map (SOM), Chu et al. (2017) found nonlinearity and asymmetry existing in the convection oscillation. The stationary dipole pattern over the eastern IO and the Philippine Sea at phase 1 and phase 5 occurs more frequently and lasts longer than in other phases. The propagating mode at phase 3 and phase 7 shows an obvious asymmetry in convective activity over the eastern IO which manifests as a slow-growing but a fast-decaying. However, the evolution of the convection anomalies over WNP obtained in the previous studies is generally considered to symmetric (Hsu & Weng, 2001; Tsou et al., 2005; Chen et al., 2015; Wang et al., 2018; Ren et al., 2018). In fact, by a pre-investigation, we found a considerable number of asymmetric intraseasonal oscillations of convection over WNP, with the convection weakening process slower than its strengthening process. It is worth thinking that why this asymmetric evolution of BSISO over WNP was not revealed in the previous studies. We speculate that the band-pass filtering of data with a narrower range of timescales used may impair the oscillation asymmetry. Also, just as Oettli et al. (2014) indicated, the results of EOF analysis are constrained by both orthogonality and linearity. Thus, the asymmetric BSISO over WNP needs to be clearly identified, and a new objective method rather than filtering or EOF decomposition is required to achieve this goal.
In this study, we propose to use a hierarchical cluster analysis to classify the BSISO events over WNP and then reveal their asymmetry features. The impacts of the classified BSISO events on the East Asian climate are further investigated. The rest of the paper is structured as follows. Section 2 introduces the reanalysis datasets and methods used in this study. Detailed procedures of the hierarchical clustering are also described in this section. Classification of the BSISO events and their spatio-temporal evolution characteristics especially the asymmetry features are presented in section 3. Different precipitation anomaly patterns associated with the classified BSISO events are examined in Section 4. Section 5 further explores the large-scale anomalous circulation and moisture supply to examine the causes for the BSISO-related precipitation anomalies. Final section is devoted to conclusions and discussion.