1. Introduction

The climate of Asia is strongly controlled by the monsoon system. In particular, the water resources in East Asian countries depend largely on the precipitation during summer rainy season, which is strongly controlled by the East Asian monsoon system (Ramage 1971; Terjung et al. 1989; Ho and Kang 1988). The summer monsoon rainfall over East Asia exhibits diverse regional characteristics, which are known by different names in different countries: Mei Yu in China, Baiu in Japan, and Changma in Korea. The monsoon rainbands associated with these regional features undergo abrupt changes and appear at different phases of the monsoon cycle (Lau and Li 1984; Tanaka 1992; Ueda et al. 1995; Lau and Yang 1997). Therefore, the variations of regional rainbands are linked to large-scale monsoon changes. However the linkage is not well documented at present. One of the motivations of the present study is to examine the regional rainfall variations, as identified by principal modes of large-scale cloud variations over the entire Asian monsoon region.

Although the Asian monsoon is a complex phenomenon involving many timescales of variations, the seasonal variation is most distinctive. Thus an important task of monsoon research is to define the variation of the basic state—that is, the climatological seasonal cycle. An interesting aspect of climatological monsoon variations is the existence of intraseasonal variation superimposed on the smoother seasonal cycle. Several studies have suggested that a phase-locked intraseasonal variation controls the climatological onset date of regional rainy season over much of the Asian monsoon region (Nakazawa 1992; Tanaka 1992; Wang and Xu 1997). Kang et al. (1989) showed that about 40% of the amplitude of climatological variation during summer exists on the timescale of 30–60 days over the subtropical western Pacific and Indian region. Lau et al. (1988) also found a clear intraseasonal signal in the 10-yr mean rainfall over China.

The climatological intraseasonal oscillation (CISO) has regional differences (Wang and Xu 1997). The relationship between the intraseasonal variation in the Indian region and that in the western subtropical Pacific is relatively well documented (Yasunari 1980; Murakami 1984; Lau and Chan 1986; Kang et al. 1989). Lau and Chan (1986) showed that the intraseasonal variation in the western Pacific region has a quadrature-phase relationship with that occurring in the Indian region. Based on the 30–60-day filtered data, Kang et al. (1989) showed that the CISO originates in the Indian Ocean and propagates to the subtropical western Pacific before diminishing in the western Pacific near 20°N. It is noted that the dominant periodicity of CISO is about 30–40 days over both the Indian Ocean and the western Pacific south of 20°N (Lau and Chan 1986; Wang and Xu 1997). But the periodicity is about 60 days over the extratropics in East Asia and northern India (Wang and Xu 1997). Due to this regional difference of the dominant intraseasonal timescale, the linkage between tropical and extratropical intraseasonal variations would be difficult to identify, particularly using 30–60-day filtered data.

The intraseasonal variations of extratropical rainbands in East Asia can be inferred from the onset and revival of monsoon rainfall over the region (Lau and Li 1984; Kawamura et al. 1996). Climatologically, the development of the primary monsoon rainband associated with Mei Yu in China and Baiu in Japan is known to occur around mid June, whereas Changma in Korea appears in late June (Ho and Kang 1988). The different timing of the appearance of these regional rainbands appears to be related to the northward progression of the monsoon rain belt in East Asia. It is also well known that in East Asia a secondary rainy season appears during late August–early September, which is called “Shurin” in Japan (Matsumoto 1992) and “fall rainy season” in Korea. The autumn rainy seasons in China were analyzed by Kao and Kuo (1958). The time interval between the primary and secondary rainy seasons is about 2 months. From the observations, the regional rainfall over East Asia appears to fluctuate with an intraseasonal timescale. However, the large-scale structures and their evolution patterns associated with the extratropical CISO have not been well documented. In the present study, we identify the principal modes of CISO and show that the variations of East Asian rainbands are associated with a leading eigenmode of the CISO.

We also note that the onset and withdrawal of Mei Yu are associated with the northward migration and establishment of a subtropical high over the western Pacific (Lau and Li 1984). The Changma rainy season in Korea is known to be associated with the penetration of the subtropical Pacific high to the south of the Korean peninsula. Changma is terminated by the further northward movement of the high pressure system to the Korean peninsula. Thus the shift of rainbands in East Asia seems to strongly relate to the changes of the subtropical Pacific high over the western Pacific, but the detailed structural changes of lower- and upper-tropospheric circulations over the western Pacific and Asia region have not been fully documented. The circulation changes associated with the variations of the East Asian rainband are also documented in the present study.

In describing large-scale rainfall variability, a number of recent studies have used outgoing longwave radiation (OLR) data as a proxy for rainfall (Murakami 1984; Lau and Chan 1986), although there are a few exceptions (e.g., Lau et al. 1988). Lau et al. used observed station rainfall data, but their study was confined to China and surrounding land areas. The lack of observational data, particularly over the oceans, has hindered the description of large-scale rainfall variability over the entire Asian monsoon region. Also, there is some difficulty in using OLR to describe the rainfall variability over the extratropics and regions with inhomogenous surface conditions. On the other hand, the cloud data produced by the International Satellite Cloud Climatology Project (ISCCP) have been found to be very useful as a proxy for large-scale rainfall variations (Fu et al. 1996). For this reason, the present study used the ISCCP cloud data.

Section 2 describes the data utilized in the present study. Section 3 shows the climatological distribution of high clouds and their variability for the summer from May to August. Section 4 shows the principal modes of the climatological variation of high cloud over the Asian monsoon region. Also shown are the coupled modes of climatological variations of high cloud and surface pressure obtained by using singular value decomposition. In section 5, the climatological variations are separated into a smoothed seasonal component and an intraseasonal component. In section 6, the time evolution patterns of the main modes of climatological intraseasonal oscillation are obtained for the high cloud, and 850- and 200-mb geopotential height. A summary of the main results and concluding remarks are given in section 7.

2. Data

The 5-day mean data utilized in the present study are the ISCCP high cloud fraction and the sea level pressure, and 850- and 200-mb geopotential height data produced by the European Centre for Medium-Range Weather Forecasts. The data period of both datasets is five years from 1985 to 1989. The present study focuses on the summer monsoon for the 4-month period from May to August. May is included in this analysis because the first transition of monsoon rainfall in India and Southeast Asia occurs around late May (Murakami and Matsumoto 1994; Lau and Yang 1997). The spatial resolution of both datasets is 2.5° lat × 2.5° long. The period of 1985–89 is taken because the ISCCP data has no gaps for the period and the ECMWF employed an improved analysis scheme after 1985 (Trenberth and Olson 1988).

In 1983 the ISCCP began to produce a new global cloud climatology as a part of the World Climate Research Programme by using weather satellite datasets. The present study uses the ISCCP daily high cloud fraction. The high cloud is defined to be the cloud at levels higher than 440-mb (Rossow and Schiffer 1991). The ISCCP cloud detection methodology and the basic statistical properties of the dataset are described in detail by Rossow and Garder (1993a). The validations of the ISCCP cloud detection are given by Rossow and Garder (1993b) and Rossow et al. (1993). Rossow et al. made a detailed comparison of the ISCCP cloud climatology with three other climatologies, one based on surface observations and two based on satellite measurements, and suggested that the monthly average ISCCP data have a root-mean-square uncertainty of less than 10%.

A 5-day mean dataset is obtained by taking nonoverlapping 5-day averages of the daily data. The 73 pentad dataset for a climatological cycle is obtained by making a 5-yr mean for each calendar pentad. A 1–2–1 filter is then applied to remove the high-frequency noise less than about 7.5 days appearing in the climatological pentad data (Tanaka 1992).

3. Climatology and variance

Climatological mean distribution of high cloud fraction and its variations for the five summers of 1985–89 are examined in this section. The climatology is defined as a 5-yr average, and the summer is referred to as the period of May–August. The climatological distribution of high cloud fraction is shown in Fig. 1. Relatively large amounts of clouds appear over the west coast of the Indochina peninsula, the Bay of Bengal, and eastern India. Zonally elongated cloud bands are located along the intertropical convergence zone in the tropical western Pacific and the region from Japan to the central Pacific. The spatial pattern of high cloud distribution shown here is similar to the summer mean OLR pattern shown by Murakami (1980) and others, except for the zonally elongated pattern near Japan. However, Lau et al. (1988) clearly showed the rainband over the region from the east coast of China to Japan. Thus, the high cloud climatology shown in Fig. 1 matches the available observed rainfall climatology shown by Lau et al. (1988).

The amplitude of high cloud variation is determined in terms of its standard deviation. Figure 2a shows the standard deviation of pentad data for the five summers, whereas the standard deviation of 5-yr mean pentads for the summer is shown in Fig. 2b. The difference between the two figures results from interannual and short timescale variations, which are not phase-locked to the climatological cycle. As seen in the figures, the regions of large variability generally coincide with the regions of large climatological-mean cloud fraction. In both Figs. 2a and 2b, relatively large variability appears over three regions: in India, the subtropical western Pacific, and the extratropical zonal bands extending from Japan. The three regions of large variability are more clearly separated in the climatological variation (Fig. 2b). Comparison of the two figures indicates that a large portion of summer variation is represented by the climatological variation over much of Asian monsoon regions. In particular, more than 60% of the mean amplitude of the variability for five summers can be explained by the climatological variation over the three regions of large variability. Here, the mean amplitude of variability is measured by standard deviation. It is also noted that the climatological variation is relatively weak over China, Indochina, the Indian Ocean, and the western Pacific along the 20°–25°N lat band.

4. Principal modes of the climatological variation of high cloud and the associated patterns of sea level pressure

Before examining the principal modes appearing in the pentad data, the monthly mean variations are examined. Figure 3 shows the month-to-month variations of climatological-mean high cloud fraction. The high cloud fraction in May (Fig. 3a) is characterized by three regional cloud bands over the eastern part of the Bay of Bengal, western China, and the zonally elongated region extending from the Yangtze River basin to the central Pacific. In June (Fig. 3b), the three cloud bands are connected among each other. From May to June, the maximum cloud band off the west coast of Indochina extends westward and covers the Bay of Bengal entirely and surrounding land regions. A noticeable increase of clouds appears in the South China Sea near the Philippines and the zonal cloud band near Japan. The relatively large cloud amount in southeastern China in May is shifted to central China in June. This cloud change appears to be related to the seasonal jump of the Mei Yu front (Lau and Li 1984). In July (Fig. 3c), the large cloud bands in India and northeast Asia move further to the north from their June positions. In August (Fig. 3d), relatively large clouds still prevail over India, the Bay of Bengal, and the subtropical western Pacific. However, the zonal cloud band in northeast Asia disappears.

The month-to-month variations of high cloud shown above are quite distinct over the Asian monsoon region. However, the monsoon variations are characterized by several major timescales associated with the seasonal march and intraseasonal oscillations (Wang and Xu 1997). To describe the climatological variations with a small number of principal modes, the empirical orthogonal function (EOF) analysis is applied to the climatological pentad data for the summer. For the EOF analysis, the latitude–longitude grid data is converted to the gridpoint data representing an equal area. For this purpose, the longitude data point of each latitude is reduced by a factor of the cosine of latitude. The first eigenvector shown in Fig. 4a explains 51.9% of the total variance over the domain. It is characterized by large positive components over India and the subtropical western Pacific. The associated time series (Fig. 4b) shows negative values for May and positive values for the rest of summer. The sign change occurs in early June. The results indicate that about half of the climatological variation during the summer is explained by the increase of large-scale cloud over India and the subtropical western Pacific from May to mid August.

The second eigenvector shown in Fig. 4c explains 19.9% of the total variance. Its spatial pattern is characterized by the two zonally elongated regions of large positive values in northeast Asia and the Indian region and two negative regions over the subtropical western Pacific and Tibetan region. The zonally elongated pattern extending from central China to the North Pacific is very similar to the East Asian rainband associated with Mei Yu and Baiu. The time series shown in Fig. 4d indicates that the second mode builds up during early summer, reaches a peak phase in mid June, and changes sign after mid July. In addition to the spatial pattern, the associated time series is consistent with the onset, mature, and retreat phases of the observed rainbands over East Asia (Lau et al. 1988). Over the region, dryness prevails for late July and August (e.g., Nakazawa 1992).

The third eigenvector shown in Fig. 4e explains 9.9% of the total variance. The spatial pattern is characterized by positive values over northern India and a zonal band crossing the Korean peninsula and negative values over the south of the positive regions, particularly along the south and east coasts of the Asian continent. The associated time series has positive values from late June to late July and negative values from mid May to mid June and in August (Fig. 4f). The maximum positive values appear in early July. During this period Korea experiences the rainy season. Both the spatial pattern and the associated time series of the third eigenmode indicate that the third eigenmode appears to be related to the Changma rainy season in Korea. As will be discussed in section 6, this mode is related to the climatological intraseasonal oscillation.

The eigenvectors shown in Fig. 4 can be compared to the EOF modes of OLR based on three summers of 1975–77 obtained by Murakami (1980). His first mode is similar to the first mode shown in Fig. 4a. His second and third modes have some similarity to the present third and second modes, respectively, in the Indian and western Pacific regions. However, the rainbands in the northeast Asian and Tibetan regions appearing in Figs. 4c and 4e are not clear in the EOF modes in Murakami (1980). On the other hand, using rainfall data Lau et al. (1988) showed that the East Asian rainband in the south of Japan forms in May and reaches its peak phase in June. They also pointed out that OLR does not clearly show the East Asian rainband, prominently appearing in July.

The sea level pressure changes associated with the principal modes of high clouds shown above are obtained by using a singular value decomposition (SVD) method. SVD is based on the computation of a cross-covariance matrix that identifies pairs of spatial patterns that explain the largest variance of the mean-squared temporal covariance between two fields. The SVD method was first applied to meteorological data by Prohaska (1976) and recently by Wallace et al. (1992) and Kang and Lau (1994). A detailed explanation of SVD and a comparison with other methods such as canonical correlation analysis can be found in Bretherton et al. (1992).

Before applying SVD, the first EOF eigenmodes of high cloud and sea level pressure are removed from each dataset, since the modes simply explain the first harmonic annual cycle. We are more interested in the variations within the summer, particularly the second and third EOF modes shown in Figs. 4c and 4e. The first SVD mode with the first EOF modes retained indicates that the first EOF mode of high cloud shown in Fig. 4a is accompanied by a hemispheric seesaw of sea level pressure associated with the annual cycle (not shown). The annual cycle mode again appears as the first EOF mode of sea level pressure. The first SVD mode explains about 91% of total covariance between the two fields. As a result, other coupled modes explain a small fraction of the covariance. Without the first EOF modes, however, the next-leading modes are clearly identified by the SVD analysis.

Figure 5 shows the first SVD mode of high cloud fraction and sea level pressure after removing the first EOF eigenmode from each dataset. For the SVD analysis, the domain of high cloud is chosen to be the same as that of the EOF analysis shown above, but the domain of sea level pressure is slightly expanded to the area of 15°S–50°N and 40°E–160°W to include the Southern Hemisphere Tropics. It is noted that the SVD results are not sensitive to the expansion of the domain of sea level pressure. The first SVD mode explains 73.9% of the domain-integrated covariance between the two fields. The first singular vector of high clouds shown in Fig. 5a is nearly identical to the second EOF eigenvector shown in Fig. 4c. The associated singular vector of sea level pressure shown in Fig. 5b is characterized by negative values in the Asian continent and the North Pacific and positive values in the subtropical Pacific. The associated time series (Fig. 5c) indicates that this mode is associated with the development of subtropical Pacific high and low pressure systems over the continent during the early part of the summer. The continental low pressure results from the surface heating, and the oceanic high pressure system is due to relatively cooler temperature over the ocean (Ninomiya and Muraki 1986). Thus, the second EOF mode of high clouds shown in Fig. 4c appears to relate to the variations of thermal contrast between the Asian continent and the subtropical Pacific.

Figure 6 shows the second SVD mode explaining 17.4% of the domain integrated covariance. The singular vector of high clouds is similar to the third EOF eigenvector shown in Fig. 4e. The SVD mode clearly shows the zonal cloud band from Korea to the North Pacific. The associated sea level pressure pattern shown in Fig. 6b is interesting to note. The pattern is characterized by the northward movement of a Pacific high in the central North Pacific and the penetration of the high pressure system to the western Pacific and all the way to the Bay of Bengal. Comparison of the two singular vectors indicates that fewer clouds appear in the regions of high pressure. On the other hand, more clouds appear in the northern boundary of a high pressure system where surface westerlies are intensified. Both time series associated with the second singular vectors have positive values in early May and July and negative values in the rest of the summer (Fig. 6c). It is noted that the period of large positive values in early and mid July is the rainy season over central and northern Korea. It is well known in Korea that Changma in Korea is accompanied by the penetration of a subtropical Pacific high to the south of the Korean peninsula (Lim 1997). The time evolution patterns of upper- and lower-layer circulations associated with this mode are further examined in section 6.

5. Separation of seasonal and intraseasonal variations

Recently a number of studies have investigated the climatological intraseasonal oscillation superimposed on a smoother seasonal cycle during the summer (Kang et al. 1989; Nakazawa 1992; Wang and Xu 1997). Kang et al. isolated the CISO component using a 30–60-day digital filter. But Wang and Xu (1997) showed that the dominant timescale of CISO is 30–40 days in the Tropics but is 60–70 days in the extratropics. Therefore, in order to investigate the CISO over the entire monsoon region, the CISO data has broad timescales. Wang and Xu (1997) obtained the CISO data from the climatological time series by eliminating their first four harmonics. Thus, the CISO data covers timescales of less than 90 days. Following Wang and Xu (1997), the present study obtains the smoothed seasonal cycle and the CISO components.

Figures 7a and 7b show, respectively, the standard deviations of the smoothed seasonal cycle and the CISO for the summer from May to August. Comparison of Fig. 7a to Fig. 2b indicates that the climatological variation is explained mainly by the smoothed seasonal cycle over the three regions of large variability: India, the subtropical western Pacific, and the extratropical zonal band near Japan. On the other hand, in regions of small seasonal cycle, the CISO amplitudes are comparable to or bigger than those of seasonal cycle. In particular, CISO has a relatively large amplitude in the western Pacific between 20° and 30°N, near Japan, the southern part of the South China Sea, and over the Indian Ocean. The ratio of the standard deviation of the intraseasonal variation to that of the seasonal cycle is shown in Fig. 7c. The figure indicates that the CISO amplitudes are bigger than those of the seasonal cycle over much of the monsoon regions, in particular two zonally elongated regions in the western Pacific and in the southern part of China. Wang and Xu (1997) also showed the standard deviations of the climatological seasonal cycle and intraseasonal variations using the OLR data from May to October. Their maps are similar to the present figures, except that the zonally elongated large variability near Japan is ill defined. The differences may result from two factors, namely different variables and different data periods used in the two studies.

The temporal phase propagation along the two important regions of large variability, the Indian and western Pacific regions, are examined by using the Hovmoeller diagrams. Figures 8a and 8b show the diagrams for the smoothed seasonal cycle along the longitude bands between 80° and 90°E and between 125° and 135°E, respectively. Along the Indian region (Fig. 8a), the seasonal component clearly shows a northward propagation from April to July and a retreat to the south after late August. At the time, a relatively large cloud band appears near the equator. Although the cloud band is more distinctive during the fall season, relatively large clouds straddle between 10°S and 10°N all year round in the Tropics. Along the western Pacific (Fig. 8b), the seasonal variations undergo a sudden shift of clouds from the Southern Hemisphere Tropics to the Northern Hemisphere Tropics during May. From May to August, a slow northward propagation of clouds can also be seen in the Northern Hemisphere subtropics. Also of note is the relatively large cloud region near 30°N appearing for the summer. The peak phase appears in mid June.

Comparison of Figs. 8a and 8b with the principal mode shown in Fig. 4a indicates that the large cloud bands along 15°–30°N in Fig. 8a and those along 10°–20°N in Fig. 8b, all appearing for the latter part of summer, are associated with the first eigenvector shown in Fig. 4a. As seen in Fig. 8, the cloud variations represented by the first eigenvector are clearly related to the first harmonic annual cycle. On the other hand, the second eigenvector (Fig. 4c) is related to the cloud band between 30° and 35°N appearing in Fig. 8b. It is also related to the cloud variation over the Indian region in the latitudes of 8°–20°N, where the clouds for late May–June are relatively large compared to those during the latter part of summer. The cloud variations related to the third eigenvector are not identified in Fig. 8 but are related to the intraseasonal variations shown below.

Figures 9a and 9b show the Hovmoeller diagrams of the CISO components along the longitudes of the Indian region (80°–90°E) and the western Pacific (125°–135°E), respectively. Along the Indian region (Fig. 9a), CISO appears year round in the Tropics between 10°S and 10°N. On the other hand, in the Northern Hemisphere extratropics, the intraseasonal activity has a seasonality. The extratropical intraseasonal variations are characterized by northward propagation, particularly for the period from mid March to August. The northward propagation of intraseasonal oscillation over the Indian region has been documented by several studies (Yasunari 1981; Lau and Chan 1986). It is also noted that the starting latitude of northward propagation moves southward with time from 20°N in mid March to the equator in early July. The northward propagation of CISO in the extratropics is also distinctive along the longitude 125°–135°E, particularly from May to August (Fig. 9b). The northward propagation starts from the latitudes about 20°N and ends near 40°N. In September and October, on the other hand, CISO propagates southward. This southward propagation may be related to the retreat of the summer monsoon system. It is also noted that in the Tropics between 10°S and 20°N, CISO is relatively weak during summer, indicating that intraseasonal variation is only slightly phase locked to the climatological cycle in the Tropics.

The spatial pattern of the northward propagating CISO shown in Figs. 9a and 9b are examined by making the 15-day averages of the CISO component with a 1-month interval. The time mean for 1–15 May (Fig. 10a) shows positive values in a zonally elongated region between 30° and 40°N and negative values in the subtropics, particularly in the oceans along the Asian continent. This spatial pattern is very similar to that of the third EOF eigenvector shown in Fig. 4e. It is noted that the time series associated with the eigenvector shown in Fig. 4f has positive values for 1–15 May. On the other hand, the time mean of 1–15 June shown in Fig. 10b has negative values in most of the region between 30° and 40°N and positive values to the south, thus its spatial pattern is almost a negative phase of Fig. 10a. One month later during 1–15 July (Fig. 10c), the CISO component has a similar spatial pattern to that of 1–15 May. During 1–15 August (Fig. 10d), its spatial pattern has some similarity to that of Fig. 10b, although in the region west of 110°E the positive and negative regions are shifted to the south compared to those of Fig. 10b. The four figures indicate that the CISO has a large spatial structure over the Asian and western Pacific regions and a timescale of about 2 months.

6. Time evolution of a leading eigenmode of climatological intraseasonal variation

In the present section, we examine how the large-scale cloud patterns shown in Fig. 10 vary for one cycle, and whether the evolution pattern can be identified as a principal mode of CISO. For these objectives, an extended EOF (EEOF) analysis is applied to the CISO data for the period of 10 April–20 September. The time period for the analysis is extended to see more clearly the positive phases of the CISO mode (Fig. 4e) appearing in early May and early September. The EEOF mode is obtained based on the lag covariance matrix with lags from −10 days (−2 pentads) to +10 days (+2 pentads).

The first and second EEOF modes of high cloud thus obtained are shown in Fig. 11. The eigenvector maps are obtained with a 5-day interval, but the maps are shown with a 10-day interval. The first and second eigenvectors explain 17.2% and 15.3% of the total variance, respectively. Both eigenvectors are orthogonal to each other but have a quadrature phase relationship, as indicated by the time series associated with the eigenvectors (Fig. 11g). Therefore, both eigenvectors explain different phases of the same CISO mode. The first eigenvector map of zero lag shown in Fig. 11b is similar to the third eigenvector of climatological variation shown in Fig. 4e. The time series associated with the two eigenvectors are also similar to each other. Therefore, the first and second EEOF modes show the time evolution pattern of large-scale clouds associated with the CISO mode shown in Fig. 4e.

The time evolution of the CISO mode is characterized by the northward movement from the western Pacific to northeast Asia and from the Bay of Bengal to northern India. At −10 day lag (Fig. 11a), negative components appear in the western Pacific around 25°N and the Indian ocean and positive components along the zonal band from northern India to the south of the Korean peninsula. For the next 10 days (Fig. 11b), the positive and negative components are intensified and moved northward. In particular, the zonal cloud band crossing the Korean peninsula is well organized at this time. The positive components move farther to the north. At the +10-day lag (Fig. 11c), they are found in the northern parts of Korea and Japan. The negative components in the south of Japan and the Bay of Bengal are also moved northward. It is noted that the phase of Fig. 11c corresponds to the phase between the lag −10 day (Fig. 11d) and 0 day (Fig. 11e) of the second EEOF mode. Thus, one can find from Figs. 11e and 11f that the northward propagating components die out in northeast Asia and other positive components initiate in the subtropical western Pacific near 20°N and in the Indian Ocean. It is also noted that the spatial pattern of the map between Figs. 11e and 11f is a mirror image of Fig. 11a, indicating that the northward propagating CISO mode oscillates with a timescale of about 60 days.

The principal modes of lower- and upper-tropospheric heights are also obtained by using the EEOF analysis. The objective of this analysis is to identify the modes of height variations associated with the principal mode of high clouds shown above. Figure 12 shows the first and second EEOF modes of 850-mb geopotential height in a domain of 10°–70°N and 60°E–160°W. The first and second modes explain 21% and 17.9% of total variance, respectively. As seen in the eigenvectors (Figs. 12a–f) and associated time series (Fig. 12g), the second mode has a quadrature relationship with the first mode. The time series shown in Fig. 12g are similar to those of high clouds shown in Fig. 11g, indicating that the modes of 850-mb height are closely associated with the high cloud variations shown in Fig. 11.

The time variations of 850-mb height inferred from the two EEOF modes are characterized by the northward movement of pressure systems, particularly over East Asia. The northward moving system originates in the subtropical western Pacific south of Japan and the Indian Ocean (Fig. 12a). At 0-day lag (Fig. 12b), the 850-mb height pattern is related to that of high clouds shown in Fig. 11b. Comparison of both figures indicates that the rainband in northeast Asia and dryness in the western Pacific are related to low and high pressure systems at 850 mb, respectively. At +10 day lag (Fig. 12c), the high pressure system in the south of Japan is expanded to the north and broadened to the Asian continent by combining with the high pressure system that moved from India. At the same time, the low pressure system in northern Japan and Sakhalin is much weakened. Ten days later (the phase between Figs. 12e and 12f), the spatial pattern becomes a mirror image of Fig. 12a. At that time, the high pressure system and dryness prevails over northeast Asia.

Figure 13 shows the first and second EEOF modes of 200-mb geopotential height. The first and second modes together explain 43% of the total variance. As in the cloud and 850-mb height, the second EEOF has a quadrature relationship with the first mode. Although there is little phase difference between the two time series particularly for late April–May, the associated time series (Fig. 13g) generally coincides with the first EEOF time series of high cloud and 850-mb height. The variations of 200-mb height inferred from the EEOF modes are also characterized by northward movement particularly over the western Pacific and northeast Asia. The northward propagating signal is originated in the northwestern Pacific around the East China Sea (Figs. 13a and 13b). The East Asian rainband (Fig. 11b) is maintained with the 200-mb height pattern of Fig. 13b, which shows positive heights over the south of the cloud band and negative heights over the north of the cloud band. Thus the cloud band is associated with the intensification of westerly flow (Lau and Li 1984). This pressure pattern is similar to that of 850 mb, although the 200-mb centers are located to the northwest of the corresponding centers at 850 mb, indicating that the pressure system associated with the cloud band in East Asia has a vertical structure with some northwestward tilt with height. It should be noted that the barotropic nature of the extratropical pressure system associated with intensified convection is shown by Kurihara and Tsuyuki (1987). It is also noted that the northward moving system from Korea and the southward moving pressure system from Siberia merge over Manchuria, making a zonally elongated pressure pattern in northeast Asia (Figs. 13b and 13f). An inspection of Figs. 13a–f indicates that the CISO at 200-mb level seems to be related to the alternative variations between a zonal pattern (Figs. 13b, 13d, and 13f) and wave patterns (Figs. 13a, 13c, and 13e).

Overall, the principal mode of CISO during summer is characterized by northward movement, particularly over the western Pacific and northeast Asia. Associated with the mode, the zonal cloud band from eastern China to Japan appears in early May, early July, and early September. The timescale of the mode is about 60 days. The cloud originates in the subtropical western Pacific centered near 20°N and propagates to the north before dying out in northeast Asia near 40°N. The northward propagation is also evident in India and the surrounding oceans. The principal modes of the lower- and upper-tropospheric heights are also characterized by northward movement. Also noticed in the height variations is the southward propagation of a large-scale wave from northern Siberia to East Asia. The southward and northward propagating components merge in northeast Asia, making the zonally elongated low pressure system near the cloud band at 850 mb and to the north of cloud band at 200 mb. These cloud and height variations are illustrated schematically in Fig. 14.

7. Summary and concluding remarks

The present study identified the principal modes of climatological variation of high clouds over the entire Asia monsoon region during the summer from May to August and the associated circulation changes. The first mode is characterized by an increase of large-scale clouds over India and the subtropical western Pacific from May to mid August. This mode is associated with the first harmonic annual cycle of the hemispheric monsoon characterized by a low pressure system in the summer hemisphere. The second mode shows the large-scale cloud changes accompanied by the East Asian rainband referred to as Mei Yu and Baiu. This mode is associated with the development of the subtropical monsoon generated by low pressure over the Asian continent and the subtropical Pacific high. The third eigenmode of high clouds is characterized by the zonal cloud band from northern India to the Korean peninsula and dryness over the oceans along the south and east coasts of the Asian continent. This mode is related to the mature phase of the Changma rainy season in Korea and is accompanied by the northward movement of the cloud band originated in the subtropical western Pacific. The associated pressure patterns are also characterized by a northward propagation. The mode repeats with a timescale of two months from early May to early September and thus produces a climatological intraseasonal oscillation.

It is mentioned that the data period of 5 yr used may not be sufficiently long to describe the climatological variation of the seasonal cycle. However, the usable high cloud data are limited at present. Nevertheless, major descriptions of the present study may not be sensitive to the use of longer data, since the results are robust for different analysis methods such as the EOF and SVD and composite methods. It is also noted that the present results are not changed by the analysis done without the 1987 El Niño summer. Many of our results are consistent with those well known by regional forecasters such as the timing of regional rainbands in East Asia and northward propagation of the rainband. Mei Yu and Baiu appear in mid June and Changma in late June–early July, and the regional rainbands are known to be associated with an abrupt jump and/or northward propagation of the rainband by the evolution of the summer season (Lau and Li 1984). However, what we found here is a distinction between the zonal cloud band near Japan, whose structure is similar to that of Mei Yu and Baiu, and the cloud band associated with Changma in Korea. Baiu is related to the smoothed seasonal variation of Asian summer monsoon. On the other hand, the Changma cloud band is associated with the northward propagating CISO with a timescale of about 2 months. It is also pointed out that although the two modes are clearly separated in the eigenmode analysis, the combination of the seasonal and intraseasonal modes should control the variations of regional rainbands in East Asia during summer.

The present study focuses mainly on the description of the climatological variations of high cloud and associated circulation patterns, though the associated dynamics can be discussed based on the results. In particular, the northward moving CISO is interesting to note. The westward vertical tilt of geopotential height indicates that the pressure system is related to the northward heat transport. The excessive surface heating from warm sea surface temperature in the western Pacific during summer forms local convection and a low pressure system in the lower troposphere. The convection grows until the circulation system has a large-scale wave to transport the access energy poleward. By that time, the large-scale convection system moves northward up to the northern boundary of the ocean, where the convection becomes weak due to reduced moisture supply. When the maximum convection appears in East Asia, the north–south temperature gradient between northern Asia and the ocean to the south intensifies upper-level westerlies over East Asia. It is noted that the variations of 200-mb height associated with CISO are characterized by the alternation of zonal and wave circulation patterns. This height variation is reminiscent of a kind of index cycle over the region. The zonal and wave patterns may be related to the variation of north–south temperature gradient over the region. A possible role of an index cycle in the intraseasonal oscillation over the Indian region was also suggested by Yasunari (1981). At present, the CISO in East Asia is poorly understood. More work is needed in order to understand the dynamics governing the extratropical CISO.