Interdecadal change of the active‐phase summer monsoon in East Asia (Meiyu) since 1979

The timing of active‐phase East Asian summer monsoon (Meiyu) undergoes a marked shift since 1979. Diagnostic analysis indicates that active convection over Taiwan has occurred later in the season, from late May to early June, with a tendency of increasingly intense rainfall. This timing shift of convection results from a southward migration of Meiyu rainband, driven by an upper‐level cyclonic anomaly over eastern China and a lower‐level anticyclonic anomaly in the subtropical Western Pacific. Together, these two circulation patterns enhance both the moisture transport and baroclinic forcing. The role of Western Pacific warming and anthropogenic greenhouse gases in these changes is suggested.


Introduction
The East Asian summer monsoon (EASM) undergoes an active-break-revival sequence and the associated migration of the rainbands makes the timing of each phase geographically unique (Chen et al., 2004). This distinct lifecycle of EASM regulates rainfall and water supply in several Asian countries, including Taiwan. Located in the central region of EASM, Taiwan covers 36 000 km 2 of complex terrain with a population approaching 24 millions (location shown in Figure 1(a), inset). The active phase of EASM (Meiyu) produces the first influx of substantial water for agricultural, industrial and residential uses. Wang and Chen (2008) indicated that the active-phase EASM (interchangeable with Meiyu hereafter) contributes to ∼60% of Taiwan's early-summer rainfall. The phases of EASM relative to Taiwan are displayed in Figure 1(a) by the outgoing longwave radiation (OLR) averaged within 119 ∘ -122 ∘ E, 21 ∘ -25 ∘ N; here, OLR is shown as departure from 235 W m −2 to approximate convective rainfall regime, denoted as ΔOLR (=235 -OLR). This feature is critical because, as of April 2015, Taiwan underwent the most severe drought in its 67 years of recorded history and yet, the arrival of Meiyu mitigated the drought situation. However, predicting the timing and strength of active-phase EASM at longer range (>2 week) remains a challenge, making drought adaptation and planning difficult (M.-M. Lu, Central Weather Bureau, 2015, personal communication).
The Meiyu rainband is driven by the mid-tropospheric warm advection and transient eddies that are steered by the westerly jet, and these circulations induce instability and adiabatic ascent while the tropical warm pool supplies the moisture (Chen et al., 2004;Sampe and Xie, 2010). Previous studies have indicated that interannual variability of the EASM circulations is linked to the Tibetan Plateau thermal conditions and India Ocean sea surface temperature (SST) anomalies (Li and Yanai, 1996;Zhao et al., 2010;Liu and Wang, 2011;Hu and Duan, 2015). These processes are complicated by the varying mid-tropospheric temperature advection within the Meiyu rainband (Kosaka et al., 2011;Okada and Yamazaki, 2012). However, few studies have focused on the interdecadal variability of Meiyu. Among these, Li et al. (2010) found that EASM has shifted southward since 1958 probably due to the meridional asymmetric warming between the South China Sea (SCS) and East Asian continent. Luo and Zhang (2015) reported that peak Meiyu rainfall in southern China has tended to arrive later since 1993 due to weakened low-level southwesterly winds. Focusing on Taiwan, Huang and Chen (2014) observed a transition of Meiyu rainfall from the predominately frontal regime to an increase in the diurnal convection regime. Regardless, a mechanistic explanation of the Meiyu's interdecadal variation is lacking; this is analyzed herein.

Results
The long-term change in the active-phase EASM is examined by analyzing the daily ΔOLR in Taiwan from mid-May to mid-July (x-axis) for each year from 1979 to 2014 (y-axis); this is plotted in Figure 1(b). The use of ΔOLR compensates for the lack of long, stable record of daily precipitation. Here, ΔOLR is subject to a 5-day and 5-year running mean to focus on the predominant intraseasonal variability that drives the EASM lifecycle (Chen et al., 2004). The peak of ΔOLR has undergone a timing shift from mostly late May before the 1990s to predominantly early June. There is also a tendency for ΔOLR to become stronger and more concentrated in mid-June (10th-15th) after 2003.
To illustrate this change, we compute the linear trend of ΔOLR for each day from 1979 to 2014 and superimpose it on Figure 1(b) as contours. Apparently, ΔOLR has decreased by 20 W m −2 in late May accompanied by an increase of 30 W m −2 in mid-June, estimated from the linear trend. Noteworthy is the change in the convective time span that has reduced from 3 weeks before 2003 to less than 2 weeks afterwards, suggesting more intense rainfall occurring within a shorter period of time. This feature echoes the finding of Huang and Chen (2014) that frontal rainfall regime in Taiwan has gradually been replaced by diurnal convection regime in May and June. The large-scale circulation and precipitation anomalies associated with the timing shift of Meiyu are examined by two epoch differences of the 250-and 850-hPa winds and ΔOLR: (1) between 1991-2002 and 1979-1990 to depict the timing shift and (2) between 2003-2014 and 1991-2002 to depict the precipitation intensification (these periods are indicated by arrows in Figure 1(b)), in June. The circulation and ΔOLR anomalies during 7th-20th June are plotted in Figure 2. In the earlier period, a robust upper-level cyclonic anomaly forms over eastern China (Figure 2(a), 'L') while a marked low-level anticyclonic anomaly extends from the SCS across the Philippine Sea (Figure 2(b), 'H'). Combined, these circulations induce strong southwesterly flows coupled with upper-level westerlies, promoting baroclinic instability in and around Taiwan. Correspondingly, a substantial increase in ΔOLR is observed in the northern SCS stretched across Taiwan, signifying an intensification of frontal rainfall regime. These circulation anomalies possibly reflect a stationary wave pattern superimposed on the westerly jet stream that was found to be influenced by the mechanical effect of the Tibetan Plateau (Wu and Chou, 2013).
For the latter period (after 2003), Figure 2(c) shows that the upper-level westerly winds enhance slightly, while a low-level cyclonic circulation appears in the vicinity of Taiwan (Figure 2(d), 'L'). Combined, these circulation changes delineate a meridional migration of Meiyu in the context of interdecadal variation. The change in ΔOLR is also substantial as it is shifted further south adjacent of the Philippines covering only the southern part of Taiwan. To clarify this implication, we plot in Figure 2(e) the latitude-time section of ΔOLR across Taiwan during 7th-20th June. Apparently, positive ΔOLR north of Taiwan has migrated southward from 26 ∘ to 20 ∘ N. Consequently, what used to be a relatively dry spell in Taiwan (i.e. between 18 ∘ and 24 ∘ N) has become increasingly convective in recent years. As is shown in Figure S1, Supporting Information, the earlier period of 24 May-6 June undergoes a decrease in convective activity as a result of this ΔOLR migration. These results provide a geographical reference for the timing change of Meiyu.
In order to connect the reported timing shift with the large-scale circulation change, we adopt a method designed to delineate the yearly evolution of a daily variable, following Wang et al. (2014). This method uses the empirical orthogonal function (EOF) of the covariance matrix of ΔOLR over Taiwan, by treating ΔOLR's daily interval as eigenvalue and its yearly interval as eigencoefficient. After applying a 5-day moving average (to capture the predominant intraseasonal variability of EASM), we obtain a set of EOFs representing the daily variation of ΔOLR and a set of principal components (PCs) representing the yearly variation. The first two EOFs are shown in Figure 3(a)-(d) representing the amplification of the temporally displaced ΔOLR, constituting collectively 32.7% of the total variance. The EOF 1 (Figure 3(a)) and EOF 2 (Figure 3(c)) show positive values in mid-June with an increasing trend of the PCs (Figure 3(b) and (d)), suggesting a tendency for enhanced convective activity and its timing shift in Taiwan. Next, we combine these two leading modes to reconstruct the ΔOLR changes in Taiwan while filtering out less relevant signals (Van den Dool, 2007). The combinations of EOFs/PCs 1 + 2 are shown in Figure 3(e) and (f). The distribution of EOFs 1 + 2 indicates maximum ΔOLR in mid-June and minimum ΔOLR in late May, and this feature has intensified as shown by the increasing trend in PCs 1 + 2 (significant at p < 0.05). Consequently, PCs 1 + 2 form an index enabling us to compare the change in subseasonal variability against interannual variations of any given variable.
By regressing PCs 1 + 2 upon the eddy streamfunction field (i.e. removing the zonal mean) for the month of June, the resultant regression coefficients depict the anomalous circulations accompanying the increased ΔOLR in Taiwan during mid-June. Figure 4 1958-1978 1979-1996 1997-2014  We next compare the interannual variations of the June stream function between the upper-level cyclone and the lower-level anticyclone, using values averaged from their center areas (domain outlined in Figure 4(a) and (b)). The variations of these two circulation features are not correlated (r < 0.16), as illustrated by the scatter plot of Figure 4(c). In other words, the circulation patterns in response to PCs 1 + 2 (i.e. increased ΔOLR) could only appear in the second quadrant of the scatter diagram, i.e. when negative 250 hPa values (trough) and positive 850 hPa values (ridge) coexisted. However, by adding the years onto the scatters, there is a discernible change in that the concurrence of the strengthened Western Pacific anticyclone with the deepened eastern China cyclone has increased after 1997 (indicated as red). This result is intriguing in that, although these two levels of circulation do not correlate, in the long run they have become increasingly cohesive in producing precipitation along the SCS-Taiwan corridor in the month of June.

Discussions and conclusion
A tendency has been observed in June for the low-level anticyclonic anomaly in subtropical Western Pacific and upper-level cyclonic anomaly in eastern China to occur together more frequently. This feature promotes frontal instability and subsequent convection in early June over Taiwan, delaying its Meiyu season. For the upper level, previous studies analyzing the change in mid-latitude stationary waves have noted an amplified short-wave regime and associated increases in weather extremes (Screen and Simmonds, 2013;Teng et al., 2013;Wang et al., 2013b;Screen and Simmonds, 2014). Other research (Wang et al., 2013b;Cho et al., 2015) has indicated an intensification of the Eurasia-South Asia short-wave train in the month of June (Yasunari et al., 1991;Ding and Wang, 2005). This reported wave pattern consists of (from west to east) a deepened trough in western Nepal, a strengthened ridge over Bhutan and an enhanced trough over eastern China -these are shown in Figure 5. The cause of this changing wave-train pattern is under debate, and our testing of SST regression with PCs 1 + 2 (not shown) does not reveal any robust linkage with any known climate mode. However, the SST in subtropical Western Pacific has tended to warm by 40% associated with the 30-W m −2 increase of ΔOLR in June, based on linear regression. This increase in local SST coincides with the ongoing warming trend in the Western Pacific. The variation of lower-level circulations in the Western Pacific has been widely documented. Yet, most studies only focused on the typical summer season of June-August, rather than the seasonal transition of May or June. Nevertheless, those studies have uniformly found a link between the strengthened North Pacific subtropical anticyclone and the increased SST under the anthropogenic global warming. The strengthened subtropical anticyclone adds thermal contrast between land (b) Reg. PC1+2 with June E 250 mb (contour) and trend of June 250 mb (shaded) CI:4*10 5 m 2 s -1 10 5 m 2 s -1 (a) Reg. PC1+2 with June E 250 mb (contour) and trend of June E 250 mb (shaded) CI:4*10 5 m 2 s -1 10 5 m 2 s -1 Figure 5. (a) The regression pattern of June E 250 hPa with PCs 1 + 2 (contours) overlaid with the linear trend slopes of the June E (shadings). Green-dotted areas indicate significance at p < 0.05 for the regression. (b) Same as (a) but for the short-wave regime (i.e. zonal wavenumbers 5 and beyond) applied in each field, following the trending pattern as depicted by Wang et al. (2013b). and ocean (Li et al., 2012) and further warms the northern Indian Ocean (He and Zhou, 2015) while enhancing thermal contrast in the subtropical Western Pacific (Wang et al., 2013a). These reported changes in oceanic thermal property and land-sea contrast have a detectable anthropogenic footprint and could be linked to the finding of this study.
To reconcile with the previous findings, we did conduct a preliminary analysis using the historical single-forcing experiment of the Community Climate System Model Version 4 (CCSM4) derived from the CMIP5 archive (Taylor et al., 2009). By reproducing Figure 1 using daily precipitation output of CCSM4, which is shown in Figure S3, it is observed that only the anthropogenic greenhouse gases (GHG) forcing simulates the timing shift of the active-phase EASM in a way similar to the observation. Neither the natural forcing nor the aerosol forcing generated any persistent change in the occurrence of peak rainfall. The preliminary result of Figure S3 suggests a possibility that anthropogenic GHG can influence the timing change of Meiyu rainfall in Taiwan. Subsequent analysis using the full archive of CMIP5 outputs will be the focus of future study.

Supporting information
The following supporting information is available:    Figure 1, but for CCSM4 historical simulation precipitation in recent 36 years with different forcing: (a,b) anthropogenic greenhouse gases (GHG), (c,d) natural including solar and volcanic forcing (Nat), and (e,f) anthropogenic aerosol (Aero). The yearly distribution of daily precipitation is the departure from seasonal means. Notice the rather weak Meiyu phase of rainfall than the observation, as well as the peak rainfall shift in (b) that is coincident with Figure 1(b).