1. Introduction

Among various monsoons in the world, the widely studied Asian summer monsoon (ASM) is considered the most distinct (e.g., Webster 1987; Webster et al. 1998), due to its complex interannual and subseasonal variability and multistage onset (Tanaka 1992; Wang and Xu 1997). The vast ASM discussed here includes several components, for example, the classical Indian summer monsoon (the eastern Arabian Sea coast), and the summer monsoon in the Bay of Bengal (BOB), Southeast Asia (the Indo-China peninsula), the South China Sea (SCS), and East Asia. It has been argued that the ASM can be divided into various parts according to the different characteristics and timing of the subseasonal evolution (e.g., Tao and Chen 1987; Wang and LinHo 2002; Ding 2004). On the other hand, Yanai et al. (1992) viewed the ASM as one system and identified two major transitions during the development of the ASM to a mature stage, namely, the first and second transitions occurring climatologically in mid-May and mid-June, respectively. Similar results can also be found in other studies (e.g., He et al. 1987; Tao and Chen 1987; Nakazawa 1992; Li and Yanai 1996; Lau and Yang 1997).

The first transition is characterized by the abrupt change in monsoon circulation and convection mainly in Southeast Asia, while the second transition corresponds to the abrupt change in a larger region covering South Asia, Southeast Asia, and southern East Asia. Hsu et al. (1999) studied the characteristics of the first transition and identified the abrupt change in the large-scale circulation, convection, and heating distribution throughout South and Southeast Asia. One of the significant regional features of the first transition is the sudden development of the convection and its southwesterly flow in the SCS, which has been identified as the major characteristics of the East Asian summer monsoon (EASM) onset. In view of the simultaneous change in a spatial scale that is much larger than the SCS, Hsu et al. (1999) suggested that the EASM onset (or the SCS summer monsoon onset by some studies) is only part of the first transition and is closely associated with other significant features occurring in South and Southeast Asia. This view is shared by several studies, revealing the importance of possible forcing in the Indo-China peninsula, the Bay of Bengal, and the Tibetan Plateau, etc. (e.g., Li and Yanai 1996; Li and Qu 1999; Lau et al. 2000; Liu et al. 2002; Zhang et al. 2002).

The first transition of the ASM and the onset of the EASM mark the onset of the rainy season in the regions surrounding the SCS, such as southern China, Hong Kong, and Taiwan. This can be understood by the fact that the southwesterly burst in the SCS meets the continental northeasterly flow in the SCS, and results in the quasi-stationary moisture convergent zone and persistent precipitation. This period, which is also called the “presummer rainy season” in southern China, Hong Kong, and Taiwan (Ding 2004), is a distinguishing point between the spring and summer seasons (Lau et al. 1988); it usually starts in mid-May and ends in mid-June during the first transition of the ASM.

This late-spring rainy season in southern East Asia is widely known as the “mei-yu season” in Taiwan and corresponds to the rainy season over southern China in the same period under the influence of the similar large-scale circulation. The mei-yu (meaning “plum rain” in Chinese; known as the baiu, in Japanese) is a unique feature in the East Asian summer monsoon, and it usually indicates the rainy season in the Yangtze River valley and Japan from mid-June to mid-July. It is characterized by the mei-yu front and the continuous rainfall from stratiform clouds at the beginning, and then there is a switch of the characteristics of precipitation from being continuous to convective in type in the later half of the mei-yu period. This characteristic is very different from the convective monsoon rains observed in the Indian summer monsoon. This feature does not exist only in the Yangtze River valley. It shifts northward along the East Asian coast from May to August (e.g., Tao and Chen 1987; Wang and LinHo 2002). The elongated zone of low-level convergence and precipitation first appears over southern China, Taiwan, and the northern SCS between mid-May and mid-June, marking the onset of the EASM. This period is also called mei-yu in Taiwan. The Yangtze River mei-yu and the Japanese baiu usually appear following the sudden northward shift of this convergent zone, which marks the end of the Taiwan mei-yu.

Although the timing of the first transition of the ASM and the onset of the Taiwan mei-yu are very close, few researchers have focused on the relationship between the large-scale monsoon system and the regional Taiwan mei-yu rainfall during this period. There are also reports on the existence of the intraseasonal oscillation (ISO), which propagates eastward from the western Indian Ocean to the SCS. The arrival of the ISO in the SCS often signals the EASM onset and the abrupt increase in rainfall around the SCS, for example, southern China, Taiwan, and the Indo-China peninsula (e.g., Nakazawa 1992; Ding and Liu 2001; Wu and Wang 2001; Chan et al. 2002). Many of these studies were based on one single case, and therefore cannot be generalized as a common feature for all years. Others constructed the composite based on calendar dates, ignoring the interannual variation of the onset date, under the assumption of strong phase locking between the ISO and seasonal cycle. Whether this assumption holds for all years is doubtful, in view of the different times of ISO occurrence in each year. The present study revisits this issue, based on a multidecadal dataset, and identifies the close relationship between the eastward-propagating ISO, the onset of the EASM (and the first transition), and the Taiwan mei-yu.

The rest of this paper is organized as follows. Section 2 describes the data and method used in this study. Section 3 presents the onset of the ASM and its link with the Taiwan mei-yu. Section 4 shows the relationship between the eastward-propagating ISO and the first transition of the ASM. Section 5 discusses how the moisture transport plays its vital role in the Taiwan mei-yu. Section 6 presents the onset types of the ASM and the moisture transport. Finally, section 7 highlights the conclusions and discussion.

2. Data and methodology

The daily rainfall data were provided from 21 weather stations (18 of them are evenly distributed around the main island of Taiwan, and the remaining stations are situated on the small islands surrounding Taiwan) operated by the Central Weather Bureau (CWB). The mean rainfall of these stations was computed to represent the all-Taiwan rainfall variation during the Taiwan mei-yu season (May and June). These weather stations were chosen because they provide the most comprehensive data starting from the 1950s. In addition, the daily rainfall from weather stations over China from 1958 to 2003 was used in this study. The Cressman objective analysis was applied to these station data to produce a 0.5° × 0.5° grid dataset for further use here. In addition to these station data, several variables from the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al. 2005) were analyzed as well. The ERA-40 data were available from 1958 to 2002 with a 2.5° × 2.5° spatial resolution. They are used in this study, because the other similar length reanalysis data from the National Centers for Environmental Precipitation–National Center for Atmospheric Re(NCEP–NCAR; Kalnay et al. 1996) search contains biases in the tropics (e.g., Trenberth and Guillemot 1988; Yanai and Tomita 1998).

To identify the timing of the first transition of the ASM, the onset is defined by the empirical orthogonal function (EOF) method described in Hsu et al. (1999). The main purpose of this method is to pinpoint the reversal of the wind directions (from northeasterly to southwesterly) in the large-scale flow over the southern edge of the Asian continent. The EOF analysis is applied to the 5-day running means of the 850-hPa streamfunction from 1 April to 30 June within the domain (30°S–50°N, 30°E–180°). Before applying the EOF analysis, the climatological mean, which is defined as the mean state from April to June, is removed from the ERA-40 data. The first EOF (EOF1), shown in Fig. 1, explains 52% of the total variance. This spatial pattern, which is the same as the pattern obtained by Hsu et al. (1999), is a zonally elongated dipole covering the Indian Ocean, South/Southeast Asia, and the western Pacific. The positive (negative) phase corresponds to the anticyclonic (cyclonic) circulation anomaly in southern Asia, the cyclonic (anticyclonic) circulation anomaly in the Indian Ocean, and the easterly (westerly) anomaly that encompass the regions between the Arabian Sea and the SCS. To test the sensitivity of this EOF method to the size of the domain, a much smaller domain covering only the SCS was chosen for the same EOF procedure to define the index. The leading EOF pattern and corresponding time series are almost identical to the results derived from the large domain. The index defined by Hsu et al. (1999) is not sensitive to the size of domain and is therefore adopted in this study.

Because the change from positive to negative phase signals a reversal of the large-scale circulations in the whole domain, Hsu et al. (1999) used the temporal variation of EOF1 amplitudes [herein “principal component 1” (“PC1”)] to define the onset of the first transition of the ASM. This approach was also incorporated in the present study. Composites of various variables based on the onset date are examined to reveal the evolution of the large-scale circulation before and after the onset, and are compared with the temporal rainfall pattern in Taiwan to infer the relationship between the large-scale circulation and the Taiwan mei-yu.

3. Onset of the Asian summer monsoon and Taiwan mei-yu

In Hsu et al. (1999), the timing of the onset is defined as 2 days after the PC1 falls below zero, but here we simply define the onset time as the first negative PC1 day when PC1 changes signs from positive to negative. This slight change does not result in notable differences. During the study period (1986–93) of Hsu et al. (1999), the PC1 value in most of the years drops sharply from positive to negative, and does not bounce back till the end of June, making it easier to define the transition time. However, when this method is applied to all the cases within the 45-yr time period (1958–2002), several exceptions are found. Although many of the cases still display the PC1 dropping sharply as described in Hsu et al. (1999), some of them show the PC1 fluctuating back and forth between positive and negative values, as seen in Fig. 2.

The results of a detailed examination of the PC1 in Fig. 2 are summarized in Table 1, which lists the dates when the PC1 changes signs from positive to negative for each year. This summary demonstrates that 24 yr in the 45-yr period (1958–2002) can be classified as having a “sharp onset” (type 1 hereafter; e.g., see the 1969 case shown in Fig. 3a), when the PC1 falls sharply and changes sign from positive to negative only once in a 60-day period (30 days before and 30 days after the onset in April–June). The 5-day running mean of rainfall from 21 weather stations in Taiwan are also plotted in Fig. 2 to show the abrupt increasing Taiwan rainfall associated with such a sharp drop of PC1. However, within these 24 yr, five cases [1973, 1987, 1991, 1992, and 1996; denoted as having a late onset (“L”), in Table 1] are excluded in the subsequent analysis, because their onset dates are in June, which are too close to the end (30 June) of the April–June period chosen for the EOF method. Furthermore, they often exhibit different characteristics from those occurring in May.

In addition to this major type of onset (type 1), 15 cases (type 2 hereafter) in the remaining 21 yr are characterized by two sign changes in PC1 from positive to negative values in April–June. The time separation between the two PC1 drops is less than 30 days, and varies year by year. It is therefore difficult to conclude a common temporal pattern during the first transition for these particular cases. Further examination reveals that these 15 type-2 cases can be further divided into two groups—type 2a and 2b—when the temporal variation of Taiwan rainfall is also considered. Type 2a (10 cases, listed in Table 1) is characterized by the occurrence of heavy Taiwan rainfall during the second drop of the PC1 (e.g., see Fig. 3b), while type 2b (5 cases) is characterized by the occurrence of heavy Taiwan rainfall during the first drop of PC1 (e.g., see Fig. 3c). For the remaining unmentioned years, the PC1 swung back and forth between the positive and negative values more than twice, and did not exhibit a regular temporal pattern that appeared repeatedly in most of the other cases. As our study indicates, there are no obvious abrupt changes in the large-scale circulation and convection for these irregular years (not shown), and no further classification is done except for the given notation, “type X” in Table 1 (e.g., see Fig. 3d).

Interestingly, no matter what type, the PC1 exhibits an out-of-phase relationship with the Taiwan rainfall. As seen in both Figs. 2 and 3, larger Taiwan rainfall tends to occur with decreasing PC1, especially in the negative phase. There is also a tendency for the rainfall to become larger after the PC1 changes sign from positive to negative (i.e., onset), although there are a few exceptions (e.g., 1980) when the whole season was extremely dry. Note that the PC1 represents the temporal fluctuation of the large-scale monsoon circulation at 850 hPa and its negative (positive) phase implies the strengthening (weakening) of the southwesterly over the northern Indian Ocean, Indochina, and the SCS. The close temporal relationship described above suggests that Taiwan mei-yu rainfall is highly influenced by the fluctuation of the large-scale monsoon circulation represented by the EOF1, even though Taiwan is a relatively small island compared to the scale of circulation. This will be further illustrated later.

To contrast the temporal characteristics of the different types, the composites of PC1 (shown in a black line) and standardized all-Taiwan rainfall (5-day running mean; shown in a gray line) relative to the onset date for types 1, 2a, and 2b are shown in Figs. 4a,b,c, respectively. Day 0 denotes the onset day (boldfaced date in Table 1), and the negative (positive) value denotes the number of days before (after) the onset. The type-1 composite clearly shows that the Taiwan mei-yu rainfall increases abruptly when the PC1 changes sign and stays as positive anomaly relative to the seasonal mean for the remaining 30 days. A clear onset of the first transition of the ASM and a sharp increase in rainfall are clearly indicated.

In contrast, type-2a (Fig. 4b) and type-2b (Fig. 4c) composites show an up-and-down swinging pattern in both PC1 and Taiwan rainfall. Nevertheless, a clear out-of-phase relationship between PC1 and rainfall is again demonstrated. As defined above, PC1 drops twice in both type 2a and 2b. The clear, abrupt change of type 2a and 2b observed near day 0 (defined as the onset date, boldfaced date in Table 1) marks the major drop, while the secondary drop is relatively vague. For example, a weak secondary drop in type 2a is evident around days −25 to −15 and is associated with weaker rainfall fluctuation (but is still out of phase). The major rainfall increase does not occur until after day 0 when the PC1 drops significantly, indicating the well-established large-scale southwesterly flow. The PC1 in type 2b drops significantly around day 0 and bounces back slightly during days 10 to 25. The rainfall increases sharply around onset and fluctuates back and forth, along with the PC1 afterward.

Although both type 2a and 2b are characterized by two drops in PC1, the period separating the two drops varies from year to year. This variability obscures the appearance of the secondary drop in these two types. This inconsistency between the cases prevents the construction of clear characteristics of the large-scale circulation and convection associated with type 2 by the composite technique. Nevertheless, the composite of type 2a and 2b still indicates the tendency that the larger Taiwan mei-yu rainfall corresponds to a lower (or negative) PC1. This implies that the Taiwan rainfall in type 2a and 2b is also closely linked with the large-scale monsoon circulation change, even though this change is not as sharp as in type 1. In contrast, the 19 sharp-onset cases (out of the 45, slightly over 40%), which are identified as type 1, exhibit a greater similarity between the cases and are much more ready for a composite study to explore the major characteristics of large-scale monsoon circulation and convection and the relationship with Taiwan mei-yu rainfall. Understanding this relationship is valuable for further understanding the major factors affecting the Taiwan mei-yu rainfall. For this purpose, the following analysis will focus mainly on the common characteristics of the 19 sharp-onset cases.

Composite time series of PC1 and nonstandardized Taiwan rainfall for PC1 are shown in Fig. 5a. The ranges of one standard deviation for both PC1 and rainfall are also marked in the figure to indicate the variability between cases. The PC1 drops sharply from positive to negative during the period around day 0. At the same time, the Taiwan rainfall increases dramatically from about 5 to about 10–15 mm day⁻¹ in a few days. The rainfall remains above 10 mm day⁻¹ for almost 30 days after the initial onset. The contrast between the periods before and after the onset is clearly seen.

Because the mei-yu rainfall in Taiwan corresponds to the rainy season over southern China in the same period, the spatial distribution of the rainfall changes over China before and after the onset was examined. Using these observational records, the composites of the rainfall for day −30 to 0, day 0 to 30, and day 0 to 30 minus day −30 to 0 were made, and are shown in Figs. 6a–c, respectively. Although rainfall in the region south of the Yangtze River is seen throughout the whole period (Figs. 6a,b), the rainfall increase is observed in southwestern and southern China (Fig. 6c). On the contrary, the rainfall over central China decreases after the onset. The pattern shown in Fig. 6c indicates that the abrupt increase of rainfall occurs not only in Taiwan, but also in the coastal areas in southern China, the northeast Indo-China peninsula, and southwestern China. The rainfall increasing areas (20°–26°N, 95°–107°E; 20°–26°N, 107°–119°E) are marked by the boxes in Fig. 6c. Area-averaged rainfall was computed for these two regions and a similar composite of 5-day running mean rainfall for the 19 sharp-onset cases was obtained. The result shown in Fig. 5b indicates that, although the rainfall in the southern China region increases after day 0, as expected, the increase is not as significant as the rainfall change in Taiwan, and the rainfall increase is more abrupt in the left box (20°–26°N, 95°–107°E) than in the right box (20°–26°N, 107°–119°E). The sharp increase in the persistent rainfall in Taiwan during this period appears to be a more interesting feature than its counterpart in the southern China region.

The results shown above suggest that during the sharp-onset years, the changes in the large-scale monsoon circulation and convection during the first transition have a strong effect on the regional circulation and convection near Taiwan, which in turn induce the heavy and persistent rainfall in Taiwan. The following analysis will focus on this interesting feature.

4. Onset of the Asian summer monsoon and ISO

The result shown above demonstrates the strong relationship between the Taiwan mei-yu rainfall and the first transition. This section investigates how the changes in the large-scale monsoon circulation in the type-1 cases lead to the occurrence of the heavy and persistent rainfall in Taiwan after the onset.

Because the low-level zonal wind (u) is a major component of the monsoon circulation, which is associated with a reversal of the wind direction in South Asia and the Indian Ocean, the time–longitude section of the 5-day running means of the 850-hPa u averaged between the equator and 20°N for the sharp-onset cases is shown in Fig. 7a to gain an overview of the large-scale circulation’s evolution. A region of westerly flow appears near 90°E around day −20, and expands in both the westward and eastward directions. This is consistent with the previous finding (e.g., Hsu et al. 1999) that, in a climatological sense, the cross-equatorial jet becomes stronger to the south of Sri Lanka (80°–100°E) and off the eastern coast of Africa (40°–70°E), south of the equator, about 10 days before the onset. At day 0, when the onset occurs, the westerly region has extended eastward to almost 120°E. The zonal extension of the westerly region reaches a steady state several days later after the onset, and remains like that through day 20. At this time, the westerly flow extends from the western Indian Ocean all the way to the Maritime Continent, reflecting the full development of the southwesterly monsoon.

Several studies pointed out that the onset of the ASM is highly related to the movement of the ISO (e.g., Yasunari 1979, 1980; Krishnamurti and Ardanuy 1980; Tanaka 1992; Wang and Xu 1997; Zhang et al. 2002). The eastward- and northward-propagating ISO is an important factor modulating the timing of the monsoon onset in many cases, and such ISO movement can be traced back to the west of the onset region several days prior to the onset. Therefore, the movement of the ISO can be used as an indicator to monitor the timing of the monsoon onset.

To see whether the ISO signals exist, the 61-day means were subtracted from Fig. 7a to reveal the evolution of the 850-hPa zonal wind anomaly more distinctly. The evolution shown in Fig. 7b is clearly marked by an eastward-propagating negative anomaly starting near 60°E and stopping near the date line. The negative anomaly is trailed by an eastward-propagating positive anomaly. One interesting feature is the occurrence of the maximum negative anomaly at 100°–120°E around day −10; another is the appearance of the positive anomaly around day 0. This feature signals the transition from the easterly to westerly wind and the occurrence of the low-level convergence in the SCS. Many studies have taken advantage of this unique phenomenon in identifying the onset of the SCS summer monsoon (e.g., Wang et al. 2004; Zhang et al. 2001). The index employed in this study appears to be appropriate in capturing the transition of the large-scale monsoon circulation. The eastward propagation is well organized and continuous without any high-frequency fluctuations between the positive and negative anomalies. This sharp contrast between the easterly and westerly regimes is consistent with the one-time rapid drop in the PC1 values from positive to negative.

The eastward-propagating feature seen in Fig. 7b shows a clear tropical intraseasonal oscillation [(TISO) or the Madden and Julian oscillation (MJO); the terminology is used according to the definition suggested by Lau and Waliser (2004)]. Similar eastward propagation can be seen in the 5-day running means of 200-hPa velocity potential averaged between 10°S and 20°N, which is often used to identify the TISO activity and is shown in Fig. 7c with the time mean between days −30 and 30 removed. The negative and positive velocity potential anomalies represent the upper-level divergence and convergence, where they are often accompanied by the low-level convergence and divergence in the deep convection region, respectively. The negative 200-hPa velocity potential anomaly propagates eastward from Greenwich, United Kingdom, at day −18, to 120°W, at day 10. At such a speed, it would take about 42 days to travel around the globe. This time scale corresponds to the typical periodicity of the TISO. The negative velocity potential anomaly is located between the 850-hPa westerly anomaly in the west, and the 850-hPa easterly anomaly in the east, as shown in Fig. 7b, indicating its close relationship with the convergence in the lower troposphere. It is interesting to note that the center of the negative velocity potential reaches 120°E around the onset date. This result suggests that the arrival of the eastward-propagating, large-scale convective system in Southeast Asia and the SCS may play an important role in inducing the onset of the rainy season in Taiwan and the surrounding region of the SCS.

Figure 8 presents the pentad means of the 200-hPa velocity potential anomaly from five pentads before to one pentad after the onset. A negative anomaly, which is seen north of Madagascar at days −20 to −16, propagates eastward along the equator, through the Indian Ocean and the Maritime Continent, and reaches the western Pacific at days 0–4. This evolution once again evidently exhibits the characteristics of the TISO. In addition to the east–west movement of the ISO, it also shows a northeastward movement from the equator to about 20°N, when the TISO reaches the eastern Indian Ocean. Interestingly, two negative centers appear at days 0–4. One is located near the date line, while the other, which is elongated in the northeast–southwest direction, is situated to the east of Taiwan and southeast of Japan. The former is obviously associated with the continuously eastward-propagating TISO. The latter that becomes stationary for the next 20–30 days and corresponds to the stationary front-like convergent zone that characterizes the Taiwan mei-yu. It is the development of this northeastward-propagating feature that initiates the Taiwan mei-yu. This northeastward excursion is not likely the TISO itself. However, the arrival of the TISO in the eastern Indian Ocean and the Maritime Continent seems to induce a new convection-favorable zone to the north. More discussion regarding this aspect will be illustrated later.

In Fig. 7a, the zonal wind exhibits a significant increase to the south of the Bay of Bengal (around 90°E). This increase can be referred to as the equatorial westerly wind burst (WWB). Several studies (e.g., Sui and Lau 1992; Yanai et al. 2000; Hung and Yanai 2004) have pointed out that the center of the active convection leads the WWB for several days. When the deep convection associated with the TISO moves to the Maritime Continent, it is an approximate indication of the WWB to occur near 80°–100°E. It is also the case here because this strong low-level westerly appears when the major convection center, as indicated by the negative 200-hPa velocity potential anomaly shown in Fig. 8, is located to the east in the western Pacific.

5. Role of moisture transport in the initiation and maintenance of Taiwan mei-yu

The heavy mei-yu rainfall in Taiwan occurs right after the sharp onset of the ASM, when the Somali jet forms and the TISO arrives at the region near 120°E. An abundant supply of moisture is needed to initiate and maintain the active convection associated with the Taiwan mei-yu. This process can be seen in Fig. 9, which presents the vector plots of the 11-day mean of the moisture transport (uqi + υqj; where q is the mixing ratio) at 850 hPa from days −30 to 30.

At days −30 to −20, there is little moisture transport in the Indian Ocean and South Asia. The major transport is from the easterly flow in the Philippine Sea. After 10 days, the cross-equatorial flow to the south of Sri Lanka starts transporting moisture northeastward in the central Indian Ocean, while the cross-equatorial flow off the east coast of Africa is also simultaneously intensifying. During the period from days −10 to 0, a strong eastward moisture transport develops suddenly between the equator and 10°N. At its western end, this Indian Ocean moisture supply channel is connected to the Somali jet, which transports the moisture from the Southern Hemisphere across the equator into the Northern Hemisphere. At the eastern end, it is connected to a northeastward transport route to the Indo-China peninsula. This moisture supply channel from the Southern Hemisphere in the western Indian Ocean to the Northern Hemisphere in the Indo-China peninsula is associated with the southwesterly flow in the southern flank of the cyclonic circulation in South Asia and the northern Indian Ocean, which quickly develops during these 10 days. At this stage, the westward-moving moisture transport in the Philippine Sea remains strong, while it is relatively quiet in the SCS.

It is worth noting that the TISO has reached the Maritime Continent at this time. During days 0–10, the enhanced convection in the Maritime Continent, resulting from the arrival of the TISO, is likely to enhance the cyclonic circulation mentioned above by inducing the equatorial Rossby wave as proposed by Gill (1980). The region of the eastward moisture transport therefore pushes further eastward into the SCS, while the easterly, which is associated with the ridge of the Pacific subtropical anticyclone and transports moisture westward, quickly retreats to the east. The major flow in the SCS has switched from easterly to westerly. This is again consistent with the design of many SCS monsoon onset indices (e.g., Wang et al. 2004; Zhang et al. 2001). In the next two 10-day periods, except for the region of northeastward moisture transport that extends continuously farther into Taiwan and the western North Pacific, the configuration of the moisture transport is basically the same.

The above evolution distinctly shows that after the sharp onset (day 0) of the ASM, the moist air can be continuously transported from the Indian Ocean to the SCS area via the moisture supply channel described above. This channel abruptly enhances after the onset, and quickly reaches the SCS in the following weeks. The development of this moisture supply channel from the western Indian Ocean to East Asia guarantees the continuous supply of the abundant moisture to the region near the SCS and Taiwan. This explains why the rainfall in Taiwan not only significantly increases after the onset of the first transition, but also becomes more persistent. Recently, a model study by Shi et al. (2005) found that cutting off the sensible and latent heat fluxes of the Indo-China peninsula results in the decreased rainfalls in the region spanning from the SCS and South China to southern Japan. This result agrees with the emphasis of the importance of the moisture supply channel shown here.

To fully see the contribution from the moisture transport to the Taiwan mei-yu in the first transition period, the divergence component of uq and υq at 850 hPa is calculated. The level of 850 hPa is selected here to represent the low-level moisture transport, because neighboring layers have similar features. The divergence component is shown in the form of streamlines and potentials in Fig. 10 in a manner similar to the composite shown in Fig. 9, but with the time means from days −30 to 30 removed. The convergence of the moist transport appears first at days −30 to −20 near eastern Africa, and gradually moves eastward along the equator at days −20 to −10 to the western Indian Ocean. Roughly 10 days before the onset, the moisture convergence center moves to the region near the Sumatra peninsula. During days 0–10, the moisture convergence center shifts northeastward to the western North Pacific and becomes a zonally elongated, stationary zone near 120°–160°E, along the latitudinal band between 20° and 30°N.

This feature is consistent with the significant increase in the moisture transport to the SCS, and the arrival of the TISO at the Maritime Continent. As seen in Fig. 10, the zonally elongated moisture convergence zone located between 120° and 160°E remains stationary for at least 30 days after the onset occurred. This long-standing system exists without much change, because the large-scale monsoon circulation does not switch back and forth between the westerly and easterly winds in these sharp-onset cases. Therefore, the moisture can be transported continuously to converge near Taiwan, leading to a persistent rainy mei-yu season in Taiwan.

The evolution discussed above can be divided into two quasi-stationary periods and one transitional period. The two quasi-stationary periods occur at the beginning and end of the evolution in the Africa/western Indian Ocean region and the western North Pacific, respectively. In between these two periods, the continuously eastward propagation characterizes the transitional period. This temporal evolution is also linked to the shift of the anomalous convergent zone from the Southern Hemisphere during the beginning of the evolution, to the Northern Hemisphere at the end of the evolution. All these characteristics denote the transition from the winter monsoon to the summer monsoon.

6. Onset types of the ASM and the moisture transport

As shown above, when the sharp onset (type 1) of the ASM occurs, the moisture supply channel is well established once it connects all the way from the Indian Ocean to the South China Sea. Through this channel, the moisture can be transported continuously to the Taiwan area. Therefore, it is more likely to have a well-defined mei-yu season with more persistent rainfall in Taiwan. In contrast, if this moisture supply channel is connected and disconnected intermittently after the onset, a persistent mei-yu rainfall in Taiwan is less likely. Whether the moisture supply channel is connected or not can be quickly viewed by the lower-level flow in the coast of the ASM region. Although the moisture condition can play the role to either enhance or reduce the total moisture amount transported to the East Asian coast and the Taiwan region, the wind direction of the monsoon circulation plays a crucial role. Some early studies in the nineteenth century already explained mei-yu (baiu) with the monsoon wind reversals from being winter northeasterly to summer southwesterly (Suda and Asakura 1955). If the transition of the monsoon flow acts like an on-and-off switch, which connects and disconnects the moisture supply channel intermittently, the moisture cannot be continuously transported to the vicinity of Taiwan through the mei-yu period.

In contrast to type 1, this study found it difficult to make a composite for type 2a, 2b, and X because of the temporal inconsistency between the cases. The discrepancies between the cases of these types are much more significant than the type-1 cases. These cases do not exhibit a systematic eastward propagation as in the type-1 cases and the moisture supply channel tends to be connected and disconnected intermittently. Because type 2a, 2b, and X are not the focus of this study, the cases are only shown here to illustrate their differences from type-1 cases. For the case of type 2a, 1960 is a good example. A series of moisture divergence plots for the 1960 case are shown in Fig. 11, with the April–June seasonal mean removed. The first and second drop of the PC1 in the 1960 case occurred on 25 April and 18 May (Fig. 3b), respectively. In Fig. 11, no obvious moisture convergence is seen from the BOB to the SCS and Taiwan area after the first drop of the PC1. Instead, the moisture convergence occurred in the southwestern Indian Ocean. The moisture supply channel was not established during this period (not shown), and significant moisture convergence was not observed until the second drop of PC1 (fourth panel in Fig. 11). This occurrence of the moisture convergence was not preceded by an eastward-moving TISO starting from the western Indian Ocean, as in the type-1 cases shown in Fig. 10. Note that the center of the moisture convergence appeared near the Bay of Bengal, not near Taiwan as in the type-1 cases. This appeared to be a more transient feature, as reflected in the shorter rainy period seen in Fig. 3b and the weakening of the moisture convergence after the 15–25 May period, as shown in Fig. 11.

For the type-2b cases, 1995 is a good example and the corresponding moisture convergence plots are shown in Fig. 12. The PC1 of this case has the first and second drops occurring on 11 May and 3 June, respectively. The heavy Taiwan rainfall occurring after the first drop marks the beginning of the Taiwan mei-yu, but a clear break of the mei-yu rainy season occurred afterward (Fig. 3c). The rainfall did not resume until the second drop of the PC1, which occurred in early June. For this type, the on-and-off moisture supply channel is responsible for such discontinuity of Taiwan mei-yu rain. In the example of 1995, after the moisture convergence center moved from the Indian Ocean to the Indo-China Peninsula from late April to mid-May, similar to the eastward-moving TISO, the moisture convergence did not become stationary; instead it continued moving eastward and dissipating. A clear break of the moisture convergence therefore occurred between mid-May and early June, corresponding to the positive PC1 seen in Fig. 3c. When the PC1 drops again from positive to negative, the moisture supply channel is connected again in early June and the mei-yu rainfall in Taiwan restarts. The eastward-moving TISO in 1995 is not as smooth and clear as that of type 1, and the moisture convergence disappeared in mid-May to early June. For other type-2b cases, the periods between two drops of PC1 are different case by case. Despite the different lengths of the break periods, these cases are all characterized by the intermittently connected and disconnected moisture supply channel (not shown).

For the type-1 cases, the TISO plays an important role in the onset timing, and the signal of these TISOs is also very clear in the Indian Ocean. The mean amplitude of the TISO in May–June when it is in phase 2 or 3 [i.e., the index of the Real-Time Multivariate MJO series 1 and 2 (RMM1 and RMM2, respectively) as defined by Wheeler and Hendon (2004); data are from the Bureau of Meteorology Research Centre, Australian Government] is shown in Fig. 13. Although there are some exceptional years (indicated by star signs), almost all type-1 cases (13 out of 14 cases after 1975, with missing data in 1978) have clear TISO signals (e.g., the amplitude is larger than 1), while type 2 and X have much less consistent TISO signals (e.g., amplitudes smaller than 1). In the 17 strong TISO cases, 13 cases occurred in the sharp-onset (type 1) year, while there were only 2 cases in both type 2 and X. This result suggests that the sharp onset of the ASM tends to occur concurrently with the strong TISO signal. The strong relationship between the sharp onset and strong TISO is evident, although not exclusively.

7. Conclusions and discussion

In summary, this study clarifies the relationship between the large-scale ASM and the mei-yu in Taiwan. The major findings are highlighted as follows:

1. For about half of the years in 1958–2002, the first transition of the ASM can be classified as sharp onset. This type of onset is associated with an abrupt reversal of the monsoon flow from northeasterly to southwesterly. For the remaining years that are not sharp-onset cases, the wind directions of the monsoon flow swings back and forth more than once in April–June, and no abrupt change in the large-scale circulation is observed.

2. The evolution of the large-scale circulation and convection in the sharp-onset years is characterized by a clear eastward-propagating TISO from eastern Africa and the western Indian Ocean south of the equator to the Maritime Continent and the western North Pacific. Upon the arrival of the TISO in the Maritime Continent, the sharp onset occurs. After the onset, a moisture supply channel in the lower troposphere is well established across the Indian Ocean. This channel consists of the Somali jet, transporting the moisture from the Southern Hemisphere to the Northern Hemisphere, and the southwesterly monsoon, transporting the moisture across the Indian Ocean to the SCS and the western North Pacific.

3. A new moisture convergent zone develops in the SCS and the western North Pacific upon the arrival of the TISO at the Maritime Continent. It is presumably due to the Rossby wave response to the convection in the Maritime Continent. This response would enhance the cyclonic circulation and result in the southwesterly surge and moisture convergent zone in the SCS and the western North Pacific. This zonally elongated convergent zone becomes stationary in the region for the next 30-day period after onset, which corresponds to the Taiwan mei-yu.

4. The heavy mei-yu rainfalls in Taiwan often occur after the sharp onset of the ASM. When the moisture supply channel is established, the conduit of the moist air is abruptly enhanced from the Indian Ocean all the way to the South China Sea. The moisture can then be efficiently and persistently transported to the SCS and surrounding areas to provide a favorable condition for the maintenance of the quasi-stationary mei-yu front and the embedded convective systems. This marks the onset of the Taiwan mei-yu season.

5. Because the sharp onset of the ASM implies a clear sudden change in the large-scale monsoon circulation along with a continuous monsoon flow after the onset, the moisture in the sharp-onset cases can be transported more efficiently and continuously via the monsoon flow to the SCS and Taiwan areas in contrast to the non-sharp-onset cases.

6. The sharp onset of the ASM tends to occur concurrently with the strong TISO signal in most of the years. The strong connection between the sharp onset and strong TISO is evident in this study. The following process is proposed to explain the covariability between the TISO and large-scale monsoon onset. Before the onset, when the 850-hPa streamfunction covering the southern Asian continent is still in the anticyclonic phase, there is a low-level moisture divergence in the tropical western Pacific. This circulation configuration is consistent with the Gill solution (Gill 1980), which demonstrates the effect of tropical convection (heating) on the maintenance of the tropical large-scale flow. The anticyclonic flow starts weakening, when the low-level convergence in a TISO moves eastward from the eastern Indian Ocean to the tropical western Pacific. The arrival of the low-level convergence at the tropical western Pacific induces a cyclonic circulation (again, according to the Gill solution) to replace the original anticyclonic circulation, and signals the occurrence of the first transition.

The above results have some practical implications. One example is the close relationship between the occurrence of the major rainfall in Taiwan after the sharp onset and the clear eastward-moving TISO. In addition, almost all sharp-onset cases of the ASM exhibit clear TISO signals over the Indian Ocean in the May–June period. Because the TISO is a signal that can be traced back to the western Indian Ocean 20–30 days prior to its arrival at the Maritime Continent, a close monitoring of the TISO would be informative for weather forecasters in Taiwan to project when the Taiwan mei-yu begins. In contrast to the sharp-onset cases, most non-sharp-onset cases exhibit unclear TISO signals, and the moisture supply channel is usually not well established during the Taiwan mei-yu period. Therefore, a persistent mei-yu rainfall in Taiwan is less likely to occur in these cases.

While the TISO is often observed before the onset of the East Asian summer monsoon, the TISO does not seem to occur every year; instead, it exhibits significant interannual variability. Interestingly, whenever there is a strong TISO signal, there seems to be a greater chance to have a sharp onset. What mechanism leading to this interannual variability and the coupling between the strong TISO and sharp onset is an intriguing but unsolved question.