The NTUGCM's AMIP Simulation on the East Asian Summer Monsoon

The Asian summer monsoon accompanies two systems, the Indian monsoon and the East Asian monsoon. A major difference between the Asian and the Indian monsoon is that beginning in early May, the Asian monsoon takes about two months, to migrate over an extensive area from southern to northern China. At the same time, the associated heavy rainfall also moves from south to north. The monsoon rain-band associated with the Indian monsoon takes only one to two weeks to migrate from Southern to North­ ern India. The onset of the Southeast Asian southwest monsoon initiates the East Asian monsoon (EASM). Based on the NMC global spectral model, with significant modifica­ tions to virtually all of the physical parameters, we have developed a gen­ eral circulation atmospheric model i.e. NTUGCM. The NTUGCM was used to perform an AMIP integration from 1January1979 to 31December1988. For the JJA(June,July,August) climatology, the simulated stream function at 850 hPa captures most of the major features in the observed climatologi- , cal-mean fields. e.g., the subtropical high over the ocean, the cyclonic circulation over South Asia and the eddy straddling the equator in the Indian Ocean, as well as the Tibetan high, simulated in the 200 hPa stream func­ tion. In general, the NTUGCM gives a better simulation at 850 hPa than at 200 hPa and a better simulation around a large forcing region, e.g. South Asia, than around a weak forcing region, e.g. the Eastern Pacific and South American coast. The best simulation for precipitation is found in the Asian monsoon region where maximum precipitation occurs. The EASM rainfall distribution is the result of complex interactions between the atmosphere, the earth's surface and the tropical and extra tropi­ cal systems. The EASM rainfall distribution involves a wide range of spa­ tial and temporal scales from the mesoscale to the planetary scale. The NTUGCM is able to simulate the large scale features of the EASM and the sudden change of the monsoon rainfall which is associated with abrupt changes in large scale atmospheric circulation. This study also analyzes the onset characteristics of the EASM simulated by the model. The AMIP simu-1 lation of the NTUGCM gives the onset date of the EASM as 13 May. The present results are encouraging, however, there are still many ob­ vious issues that remain to be st udied, such a s the linkage between the monsoon's variation and surface boundaries forcing, the cause of the sea­ sonal northward jump of the subtropical high, the dynamic and thermody­ namic effects of the Tibetan Plateau and an improved model representing the physical parameters.

lation of th e NTUGCM gives the onset date of the EASM as 13 May.
The present results are encouraging, however, there are still many ob vious issues that remain to be studied, such as the linkage between the monsoon's variation and surface boundaries forcing, the cause of the sea sonal northward jump of the subtropical high, the dynamic and thermody namic effects of the Tibetan Plateau and an improved model representing the physical parameters.

INTRODUCTION
Climate is an extraordinarily complex environmental system. In the study of the many processes responsible for a given climate's structure and variation, the atmospheric scientist has found it is necessary to construct mathematical models of the climate in order to represent quantitatively the various interacting elements. In addition to the atmosphere, the scientific definition of climate also includes the average behavior of the oceans and the condition or state of the land surface and its associated vegetation. Each of these elements are linked to gether in a worldwide system, with changes in one part generally affecting the behavior of other parts, setting in motion a chain of events which may either reinforce or cancel the origi nal change. The number of possible mechanisms causing changes in climate are rather large.
Our knowledge of the physical basis of climate and climate variation is most usefully and comprehensively organized in terms of models, which are mathematical representations of the physical laws and govern the climate system's behavior. From such models the system's fu ture behavior or climatic state may then in principle be determined. In this sense, climate models represennhe most physically rational basis for studying climate. The general circula tion model (GCM) has been used extensively to study both global climate changes and ob served regional climate characteristics. A comparison of simulated and observed seasonal varia tions for the monsoon system can therefore contribute substantially to the evaluation and im provement of the inter-annual variability of the GCM model and lead to an understanding of short term climate variations. In this research we have focused on the GCM's simulation of the East Asian summer monsoon.
The monsoon is a part of the seasonal variations caused by the annual variation in solar fluxes to the earth's surface. Although convective heating plays the most dominant role as a driving force for monsoon circulation (e.g. Krishnamurti and Ramanthan, 1982), the land ocean heat contrast caused by the seasonal variations in solar radiation provides a basic frame work for the large-scale monsoon circulation. If inter-;annual variations in large scale monsoon circulation are determined entirely by lower boundary forces, they could, in principle, be simu lated perfectly by the GCM, integrated with specified observed lower boundary conditions.
On the other hand, it is possible that nonlinear internal atmospheric dynamics, being most powerful on intra-seasonal time scales, could limit monsoon predictability on inter-annual time scales. Therefore, using the GCM model to simulate monsoon circulation is a good ap proach to verify the climate model's capability. Yanai et al. (1 992) carefully analyzed the FGGE data and concluded that there are two transition phases to the Asian summer monsoon ( ASM ) in the early summer. The first transi tion is marked by the onset of the East Asian summer monsoon ( EASM ), while the second transition is marked by the onset of the Indian summer monsoon (ISM). Nakazawa( l 99 2) examined the climatological mean evolution of the ASM, and found the same characteristics. tuates in a more regional scale. In contrast, the circulation during the second transition fluctu ates on a much larger spatial scale (H su & Chen, 1997 ). The EASM rain band is typically a quasi-stationary large-scale feature that produces continuous precipitation. Embedded within the rain band are mesoscale vortices which produce intense local rainfall. In the EASM system there are two rain belts. One is linked to a monsoon trough in the SCS and the ITCZ in the Western Pacific and has a tropical nature. The other rain belt is formed through the joint contribution of a cold air mass from the polar continent and a warm and wet flow from the north side of the subtropical high( SH). This rain belt is known as the Mei-Yu and it has a subtropical nature.
The EASM rain band takes about two months, from early May, to spread over an exten sive area from Southern to Northern China. In the Indian Monsoon system, there is only one rain belt corresponding to the monsoon trough which is of a tropical nature, and it takes only about one to two weeks to migrate from Southern to Northern India. The establishment of the EASM over the SCS depends on the equatorial westerly wind near 70-90° E, or the southwest erly flow which may come from the south side of the SH and tum to a southwesterly direction on the west side of the SH and northwest side of the SH. However, the onset of the ISM relies on the cross-equatorial flow near 40° E. Therefore, the EASM cannot be simply viewed as northward and eastward extensions of the Indian summer monsoonal flow. According to a suggestion by Tao (1985) , the principal components of the EASM should include: the mon soon trough (or ITCZ) in the SCS and the Western Pacific, the cross-equatorial flow to the region east of 10 0° E, the SH over the Western Pacific, the upper-level easterly flow, the convection along the monsoon trough, the Mei-Yu frontal zones and the mid-latitude distur bances. The principal components of the ISM system suggested by Krishnamurti & Bahalme(l 97 6) include: the monsoon trough over Northern India, the southwest monsoon flow, the cross-equatorial flow, the Somali jet off East Africa, the Mascarene high, the anticy clone in the upper troposphere over the Tibetan Plateau, the tropical easterly jet stream and monsoon rainfall and the cloud cover over the Indian Peninsula. Therefore,the EASM is influ enced not only by the Indian monsoon, but also by the Western Pacific SH. Because of large inter-annual and inter-seasonal variations in the Western Pacific SH, the anomalies of summer monsoon rainfall and circulation are large in East Asia. Comparatively, the EASM is much more complicated and variable than the ISM.
However, the above studies emphasize the simulation of the ISM and pay little attention to the EASM.
In this paper, we show the recent AMIP simulation of the NTU (National Taiwan Univer sity) GCM on the EASM. Concerning the onset and progression of the EASM, some of the most important features from the climatological analysis and the NTUGCM simulation will be presented. In this report, section 2 provides a brief description of the dynamics and physics used in the NTUGCM. A short section follows the model description on the model climatol ogy of JJA (section 3). A range of important climate variables produced by the model over a 10-year AMIP run is presented. Section 4 is the simulated EASM where the observed fields are displayed for comparison. Section 5 provides a summary and conclusion.

MODEL STRUCTURE
The present NTU model, designed especially for simulation of the Asian monsoon, is based on the early NMC global spectral model described by Sela (1980). Subsequent changes include the substitution of a triangular truncation scheme for the model's original rhomboidal representation as well as significant modifications of virtually all of the physical parameterizations. Like most of the numerical weather prediction models, the climate model of the NTUGCM is based on primitive equations. The model simulates a comprehensive range of physical processes including radiation and precipitation, which act as forces on the dynamic equations.

Numerical and Computational Properties
Because of the complexity of the model equations, realistic solutions can be obtained only through numerical simulations. For convenience, we consider the numerical methods in three parts: horizontal, vertical and time.

Horizontal
The horizontal representation of the model is spectral with transformation to a Gaussian grid for calculation of nonlinear quantities and physical processes. The present model has been coded for variable spectral resolution. However, in this study we used the spectral tri angular 42 (T42), roughly equivalent in to a grid spacing of 2.8 X 2.8 degrees.

Vertical
The vertical domain for the present model is from surface to lhPa. For a surface pressure of lOOOhPa, the lowest atmospheric level is 962hPa. Finite-difference sigma vertical coordi nates with specification of layer locations after Brown (1974) and Phillips (1975) are adopted.
The Quadratically conserving vertical finite-difference approximations of Arakawa and Mintz (1974) are utilized. There are 13 irregularly spaced sigma levels. For a surface pressure of lOOOhPa, 3 levels are below 800hPa and 4 levels are above 200hPa (Figure 1).

Time integration
The time integration was accomplished using the so-called semi-implicit method by Rob ert (1969) with a moderate time filter. The time step is 20 minutes for dynamics and physics, while the radiation /cloud calculations were done once every 3 hours.The operating system for the present model was run on a DEC-3000/600 computer using a single processor in a UNIX environment. It took about 8 minutes of CPU time per simulated day for the AMIP simulation.

Dynamic and Physical Properties
The primitive-equation dynamics are expressed in terms of vorticity, divergence, tem perature, specific humidity and the logarithm of surface pressure. Horizontal diffusion was applied to the temperature, vorticity, divergence and moisture fields. The diffusion is repre sented in a simplified spectral form based on Laursen and Eliasen (1989), where the diffusion coefficient varies for different variables and levels. The gravity wave drag parameterization by Chouinard et al. (1986) was included in the model, where the momentum transports due to the sub-grid scale gravity waves excited by stabile stratified flow over irregular terrain were parameterized. For the atmospheric chemistry, carbon dioxide concentrations were at the AMIP prescribed value of 345ppm and the monthly mean zonal averaged climatological ozone mix ing ratios were specified as a function of latitude and height using data from Rosenfield et al. (1987).
The determination for the land or sea category of surface type was made from NCAR' s 10 minute by 10 minute Navy topography dataset, which included the percentage of water-cover at any point. The data was averaged to the model's grid points. Any grid-box whose average water percentage was 60% or more was defined as a water point. The spatially variable surface albedo was prescribed by the climate data obtained from the Central Weather Bureau. Over oceans the albedo is specified as 0.09 and over ice 0.60. Monthly mean values were linearly interpolated to the current time. Surface orographic heights were determined by averaging the Navy 10 by 10 minute data over each model grid box. The averaged orography was then passed through a Lanczos (1966) filter. The model physics were formulated in a simplified parametric form. There are:

Radiation
The radiation transfer parameterization scheme is based on the broadband methods devel oped by Ou and Liou ( 1988), which calculated the transfer of infrared and solar fluxes in both clear and cloudy regions. In a clear atmosphere, the entire infrared spectrum is divided into five bands; three for water vapor, one for carbon dioxide and one for ozone absorption. The parameterization of these broadband infrared emissivities follow Liou and Ou (1981) and Ou and Liou (1988). The solar spectrum consists of 25 bands-six for water vapor, one for carbon dioxide (which overlaps the 2.7 micron band for water vapor) and 18 for ozone. The param eterization of the corresponding broadband solar absorptivities was documented by .
In a cloudy atmosphere, low and middle clouds are treated as black bodies in the infrared radiation transfer calculation. The broadband infrared emissivity, reflectivity and transitivity for high clouds as well as broadband solar absorption, reflection and transmission for various cloud types are computed based on a prescribed vertical liquid water content. Cloud radiative properties were calculated based on the parameterizations developed by Liou and Wittman(l979). For purposes of the radiation calculations, clouds were assumed to be ran domly overlapped in the vertical (Kau et al.1995).

Convection
The parameterization of convection follows the approach of Kuo (1965 ) as adapted by Phillips ( 1979 ) . Penetrative convection occurs in the presence of large moisture convergence accompanied by a moist unstable lapse rate under conditions of moderately high relative hu midity. The convective cloud is assumed to dissolve instantaneously through a lateral bound ary, thereby imparting heat and moisture into the environment. In a vertical column, the total moisture available from convergence is divided between a fraction b that moistens the envi ronment and the remainder (1-b) that contributes to the latent heating (precipitation) rate. The moistening parameter b is prescribed as a function of the SST, i.e., the b value linearly decreases to zero by a factor when the SST is between 296 K to 299K and is set to zero when the SST is equal to or greater than 299 K (Kau et al., 1995) .

Cloud formation
The cloud cover and liquid water content required for the radiation calculations were determined diagnostically following Gelyn (1 981), Liou and Zheng (1984) , and Slingo and , Ritter (1985). This scheme is based on whether the ambient relative humidity exceeds pre scribed threshold values (Gelyn, 1981). The amount of cloud is a quadratic function of the excess relative humidity. The model allows partial or total cloud cover in seven vertical sigma layers (layers 3 to 9) , while the top two layers and the bottom three layers are specified as being cloudless( i.e. no stratiform cloud below 850 hpa). The model-generated multi-layered cloudiness is then placed into high, middle and low cloud decks (K au et al., 1995).

Precipitation
Precipitation can result either from convection or from large-scale condensation, when the local specific humidity exceeds the saturated humidity at the environmental temperature/ pressure. Evaporation of convective precipitation is parameterized as a function of convective rain intensity and saturation deficit. Before falling to the surface, large-scale precipitation must saturate all layers below the condensation level by evaporation. Precipitation that reaches the surface is assumed to be ·uniformly distributed over a grid box.

Planetary boundary layer(PBL)
The PBL top is defined as the lifting condensation level and is assumed at the lowest model layer. The surface wind velocity was determined using the parameterizations of the Drag Law. The surface temperature over land is calculated from the surface heat budget equa tion (Liou and Zheng, 1984 ) using a force-restore method . The surface fluxes are computed, using the bulk aerodynamic method, to a diffusive flux from the surface layer into the layer above. Three mechanisms are considered in terms of fluxes into the atmosphere: surface fric tion, sensible heat fluxes from the surface, and the evaporation from the surface (moisture flux) are expressed as a bulk formula. The values of wind, temperature and humidity required for the bulk formula were taken to be those at the lowest atmospheric level, which was as sumed to be within a constant-flux surface layer. The drag coefficients for surface stress vary geographically from a minimum of 0.0013 over the open oceans to a maximum of 0.009 over the Himalayas. Thermal and moisture drag coefficients are the same and allow the transfer process to be highly wind speed selective (Sela, 1980).

MODEL CLIMATOLOGY OF JJA
The NTUGCM was used to perform an AMIP integration from 1 January 1979 to 31 December 1988. SST was prescribed to force the model. Two kinds of SST temporal means were used-a sequence of 36 monthly SSTs taken between 1979 and 1981, and 366 sequences of weekly SSTs between 1982 and 1988. The initial condition was the circulation of 31 De cember 1978, which was taken from the ECMWF FGGE data set. In this section, we describe the model climatology of JJA. The following two sets of observed data were used for model verification: (a) the general circulation statistics derived from the climatology of the ECMWF analysis for 1985-1995;( b) Legates/MSU precipitation climatology produced by Wallace et al. ( 1994, personal communication ).

Zonal Means
Cross-sections of the zonal-averaged zonal winds are shown in Figure 2. On the whole, the location and strength of the simulated jets in the extra-tropics of the Northern and Southern Hemispheres are quite similar to the observed. The polar night jet in the Southern Hemisphere was also simulated. However, the tropical easterly flow in the model is too weak in the middle troposphere. Figure 3 shows cross sections of the zonal�averaged temperature. In general the simulated zonal-averaged temperature agrees with the observation. The model bias indicates that in the tropics, the lower troposphere is colder(these results are very similar to ECHAMO(Dumenil and Schlese , 1987)).This error is caused primarily by a lack of latent heat exchange between the ocean surface and the atmosphere , predominately in the tropical Pa cific. Another contributing factor might be the excessive radiative cooling caused by tropical cirrus cloud . The zonal averaged meridional wind and vertical veloc ity are shown in Figures 4 and 5, respectively. From these figures we see that the basic struc ture of the zonal mean Hadley and Ferrel cells are simulated . However , the upper tropo spheric wind peak is too weak , suggesting a deficiency in the convection scheme . This fea ture is also consistent with the vertical structure of the temperature errors as shown in Figure3.

Global Projections
Simulated stream function and zonal winds at 850hPa are shown in Figure 6. The model captures most of the major features in the observed climatological mean fields, e.g. the SH over the oceans, the cyclonic circulation over South Asia and the eddy straddling the equator in the Indian Ocean. However, the circulation appears to be more zonal than the observed, suggesting weaker amplitudes for the stationary eddies. The SHs in the Northern Hemisphere  are weaker than the observed, but their counterparts in the Southern Hemisphere are stronger. The meridional gradient in the northern Indian Ocean is too strong, indicating that the simu lated Somali jet is too strong (Figure 6). The simulated cyclonic circulation in South Asia covers the region from northern Africa to 100° E which is about the same size as observed. Figure? shows the stream function and zonal wind at 200hPa. The Tibetan high at 200hPa is well simulated. The Mexican high is missing and is replaced by a cyclonic circulation in the model climatology, which is accompanied by equatorial westerly flow from the dateline east ward to the Greenwich line as shown in Figure 7. In reality, the upper troposphere in the equator should be occupied by easterly flow as one can see in Figure 7a. This deficiency can also be seen in the zonal average( Figure 2). On the whole, the NTUGCM gives a better simu lation at 850hPa than at 200hPa, and a better simulation around a large forcing region, e.g. South Asia, than around a weak forcing region, e.g. Eastern Pacific and South America. The velocity of potential fields at 850hPa and 200hPa are shown in Figure 8. The tropical divergence field is as a whole weaker than the observed . In Asia and the Pacific, the centers of the observed velocity potential at both 850 hPa and 200hPa are located to the east of the Philippines . The model simulates both 850 hpa and 200 hpa velocity potential centers as being very close to the observation. The curvatures and the gradients of the observed velocity poten tial indicate two other convective regions, one in Northern Africa and the other in the Eastern Pacific and Southern America. However, the simulated results exhibit weaker convective ac tivities in those two regions. The above deficiencies can be seen clearly in the precipitation field ( Figure 9). There is generally good agreement between the observed and simulated pre cipitation patterns. The best simulation is found in the Asian monsoon region where the maxi mum precipitation occurs. However, the model precipitation is less than the observed precipi tation both in the tropics and in the extra-tropics (outside the Asian monsoon area). This defi ciency is consistent with the zonal means' biases discussed in the previous section.

EAST ASIAN SUMMER MONSOON
The ASM constitutes a major circulation feature in the Northern Hemisphere. Tao and Chen(19 87) have suggested that the EASM should be studied as a separate system because it behaves quite differently from the ISM. In fact, the EASM possesses a wide range of spatial and temporal scales of variability including the seasonal cycle, intra-seasonal oscillations, sub-seasonal scale, inter-monsoon interactions, sub-synoptic scale variability and super-clus ter organization in the Western Pacific (Lau, 1992). Therefore, the EASM is much more complex than the ISM. A detailed discussion of the EASM should involve a diverse range of spatial and temporal scales . However, in this section we have only presented some of the important features of the simulated EASM.

Monthly Mean Precipitation
Climatically, the summer monsoon rainfalls in East Asia include the pre-summer heavy rainfall in Southern China, the Mei-Yu season in Taiwan, the Mei-Yu over the Yangtze and Huaihe River Valleys and the mi d-summer heavy rainfall in Northern China. FigurelO shows the simulated monthly variations for the precipitation from May to August. It can be seen that up into two with one over Northeastern China/Southern Japan and one over the southwestern part of China. In August, the basic rainfall pattern remains unchanged, although the amounts of rainfall appear to diminish considerably. The above features agree very well with the monthly mean rainfall climatology presented by Lau(1992). Figure 11 shows the latitude-time cross sections of the 10-year mean annual cycle of Asian monsoon rainfall. The rainfall averages occurred between 50° E and 120° E for MSU and AMIP, respectively. This figure indicates a similarity in the phases of monsoon transitions between the observed and simulated rainfall. In both panels, monsoon rainfall attains a maximum magnitude in June and reaches its northern most position during July-August. Similar to the observed, the model monsoon exhibits a sudden northward jump during a period from late boreal spring to early summer. The model produces a good precipitation rate during the summer monsoon period (slightly displaced to the south, however). While for the rest of the year the precipitation rate tends to be less than that observed.  1EJ in May is located over the southern tip of the Indochina Peninsula. At this time the wind speed over India is weak. In June, the TEJ rapidly intensifies. The strongest easterly wind may be observed over the southern tip of the Indian Peninsula. During the same period the easterly jet markedly extends northward and a branch of easterlies occurs from 20 to 25° N over the western Pacific emerge with one branch which is located at the lower latitude. The above two branches of the easterly jet merge over lndo-China. In July, the easterly jet over South Asia significantly intensifies, and it also extends southward down to about 10 °S. August is a month when the easterly jet reaches its maximum intensity. The evolution of the simulated monthly mean upper-level easterly jet in general is in good agreement with the above-mentioned fea-    tures. For example, in July, the TEJ has two branches. One is located around 20 to 25° N, extending from the northeastern part of the SCS westward to the Bay of Bengal and Southern India. The other is located around 5 to 10° N (barely seen), however, over the southern part of SCS. These two branches merge over the Arabian Sea.

Precipitation
Based on the latitude-time cross-section of the 10-day rainfall level in Eastern China  Figure 19 shows the time-latitude cross-section of zonal wind at 200hPa along 125° E between 10 to 50° N. Bef ore mid-May, the upper level westerly jet is located at 37.5° N and the center of the easterly jet is at 15° N. The northward shift of the upper level jet occurs rapidly in 5 to 10 days at the beginning of June after which the westerly core is located over a stationary position near 40° N. At the same time, the center of the easterly jet moves northward to 20° N. At the end of July, a second northward transition occurs with the westerly jet shifting further in a similar manner to 45° N giving rise to the monsoon onset over North and North eastern China (Lau et al., 1988). The seasonal variations of the upper-level westerly and east erly jets is believed to be related to the movement of the South Asian upper-level anticyclone.

Zonal wind
The abruptness of the transition and the large extent of the area affected by the monsoon rainfall are unique features of the EASM rainfall that do not have a counterpart in the Indian region (Lau, 1992).

223
SH is one important component of the EASM system. The SH over the Western Pacific assumes a marked seasonal variation in respect to its position. As the summer monsoon devel ops from spring to summer, the SH continuously moves northward. The seasonal shift in the position of the SH shows up as an alternation of gradual shifts and abrupt jumps, which are closely related to the advance and retreat of maj or rain bands in Eastern China. In general, three abrupt northward jumps may be observed (Ding, 1994 ). The first northward jump occurs in the middle or last part of June, with the ridge line of the high moving over to 20-25° N. This indicates a northward jump of the rain-belt over the Yangtze and Huaihe River Valleys, her alding the end of both the pre-summer season over south China and the Mei-Yu season over Taiwan. The second northward jump of the SH occurs in the early and middle part of July. The ridge line of the high reaches to 25-30° N, starting the rainy season in the Yellow River and

South Asian anticyclone
The seasonal migration of the high-level anticyclone is strongly related to the establish ment of the process of the summer monsoons over Asia. He et al. (1987)  integration. From the movement of the 200hPa anticyclone center we can find the center moves rapidly northeastward during the first three sub-periods. After sub-period 4 (16-20 May) the center shifts westward to 90° E near 26° N, where it then stays for the next four sub-periods (2 1 May to 10 June) . The anticyclone center then moves northwestward aft er sub-period 8 (6-10 June). The appearance of the western center over the Saudi Arabia-Iran region occurs in sub-period 11(20-24 June) .

Cross-equatorial· flow
The summer monsoon in South Asia is believed to be closely related to the cross-equato rial air flows. Findlater (1969) discovered the existence of the Somali low-level jet anchored off Eastern Africa. Tao  the flow at 105-110 ° E being stronger than the other two. Wang and Li( 1982) indicated that the beginning and maintenance of the Mei-Yu is also related to the cross-equatorial flow at about 110° E. Figure22 shows the climatological mean of the time-longitude cross-section of the simulated cross-equatorial flow at 850hPa in the equatorial region from 30 to 180° E. From this Figure, we can see two strips of stable and steady southerly flows. One is located around 40° E, which is the Somali Jet, and the other is between 140-150° E. These two stable south erly flows are in good agreement with the observation. There is another stable southerly flow that wanders between 110 and 120° E, where its intensity varies greatly with time. During the period of May, we can also find a southerly flow that extends eastward from 70° E to 100° E. At the same time there is another steady cross-equatorial flow located at 85-90° E.

Onset of the EASM
The onset of the EASM (or the first transition of the ASM) has received much less atten tion than that of the ISM. Hsu and Chen( 1997) This process coincides with the seasonal variation of heating over the Tibetan Plateau and the resultant seasonal evolution of temperature and height fields. The maximum temperature anoma lies, located near 90° E, persist from June to August. The thermal condition of the Tibetan Plateau may cause not only the variations in the circulation system over the plateau and its surrounding areas, but also the planetary scale circulation anomalies (Huang, 1985). Figure   23b indicates that the convection activities from 90 to· 110° E start in early April. The convec tion between 75 and 80° E also starts in early April but is weaker than its counterpart between 110 and 120° E. The convection is restricted in these two regions and remains localized until mid-May when it occurs simultaneously over a mu ch broader region . This later onset of the convection coincides with the first transition shown in Figure 23a. In early June, the convec tion simultaneously occurs over an even larger region, from the Arabian Sea to the Philippines and coincides with the second transition. As.depicted in Figure 23b   to 150° E occurs on 9 June. Also, we can see the maximum temperature anomalies located near 90° E and these persist from June to August. From Figure 24b we can see that the large increase in convection around 110° Eis on 12 May, and in early June, the convection simulta neously occurs over the three bay areas in South Asia. These two events agree very well with   Figure 24c indicates that the simulated precipitation between 5 and 20° N sud denly increases from 2.5 mm/d to 5 mm/d on 10 May and then persists until September. The rain band propagates northward in late May and early June. From Figure 24d we can see that an abrupt switch of the 850hPa zonal wind from easterly to westerly, between the equator and 15° N, occurs in early May. The westerly then persists throughout August. Note that the west erly wind greatly intensifies in mid-May and gradually moves northward to mid-latitude. Com paring the model sim4lations of Figures 24c and 24d with observations, we can see the general agreement between them. However, the simulated onset day of the EASM seems earlier than the climatology, and the rainband is a few degrees north of that observed. In order to see the onset date of the EASM more clearly , Figure 25 plots the evolution of the SH ridge line at 850 hPa and five-day means from 28 April to 27 May are plotted. From this Figure we can see the SH retreats in an eastward direction over the SCS and the onset date of the EASM could be set around 13 to 17 May(i.e. sub-period 4). Figure 26 shows the time-latitude cross-section of southwesterly wind at 850hPa, averaged between 110 to 130° E. From this figure we can notice that the southwesterly wind greatly increases between 10° N and 25° N around 14 May. Based on the above discussion, we therefore conclude that the AMIP simulation of the NTUGCM for the onset date of the EASM is set at 13 May.

SUMMARY AND CONCLUSIONS
Based on the NMC spectral model with significant modifications of virtually all of the physical paramters, we have developed the NTUGCM. In this study the NTUGCM was used to perform an AMIP integration from 1 January 1979 to 31 December 1988. In general the model appears to simulate the global climate with a degree of skill similar to that of most comparable models (e.g. Boer f!.f al. 1991). For the latitude-pressure distribution of zonal means of atmospheric variables for the seasons of JJA, the location and strength of the simulated jets in the extra-tropics of the Northern and Southern Hemispheres are quite similar to those ob served. However, the model bias indicates that the troposphere is colder,that the tropical east erly flow is too small in the middle troposphere and that the Hadley and Ferrel cells are weaker than those observed. These biases are caused primarily by a lack of latent heat exchange be tween the ocean surface and the atmosphere, mostly in the equatorial Central and Eastern Pacific. Another contributing factor might be the excessive radiative cooling caused mainly by tropical cirrus cloud. For the global projection of the JJA climatology, the simulated stream fu nction at 850 hPa captures most of the major fe atures in the observed climatological-mean fields, i. e. the SH over the ocean, the cyclonic circulation over South Asia and the eddy straddling the equator in the Indian Ocean, as well as the Tibetan high, simulated in the 200 hPa stream function. In general, the NTUGCM gives a better simulation at 850 hPa than at 200 hPa and a better simulation around a large forcing region, e.g. South Asia, than around a weak forcing region, e.g. the Eastern Pacific and South American coast. The best simulation for precipitation is found in the Asian monsoon region where maximum precipitation occurs .
It is well known that the land-ocean heat contrast caused by the seasonal variations in solar radiation provide a basic framework for the large-scale monsoon circulation. If inter annual variation in large scale monsoon circulation is determined presumably by lower boundary  (1) A northward progression of the EASM rainfall(Figurel8), which is dominated by convec tive activities.
(2) A rapid intensification of the lower westerly and upper easterly monsoon circulation as well as the development of the monsoon trough during the period from May to June (Fig   ures 13&18 ).
(3) Three northward jumps of the SH, which over the Western Pacific, assumes a marked sea sonal variation in respect to its position. The three northward jumps of the SH during the This study also highlights the simulation of the onset characteristics of the EASM. HC pointed out that the onset of the EASM is characterized by a sudden change in large-scale atmospheric circulation. These features are: (1) The development of the low-level cyclonic circulation and the upper-level anticyclone in South Asia, (2) sudden eastward retreat of the subtropical Pacific anticyclone at 850 hPa which occupies the SCS in April and early May, (3) the 850 hPa wind in the SCS abruptly changes from southeasterly to southwesterly.
Based on the above characteristics, the AMIP simulated mean onset date for the EASM by the NTUGCM is therefore set as 13 May, which is close to the climatological overall mean. From the above discussions, concerning the onset and progression of the EASM, we may confirm that the present model has the ability to simulate the seasonal evolution of the EAS M. Thus, the NTUGCM may be a useful tool for understanding the physical mechanisms responsible for the variability of the broad scale EASM.
The present results are encouraging. They indicate that further predictability research on the ASM and long range forecasting of the EASM rainfall may be possible in the near future.
However, the present study is just a prelude and there is much work that must be done on the simulation of the EASM. There are four issues which merit future attention: (1) The variability of the ASM associated with the Earth 's surface fo rcing involves many complex processes on a variety of time scales. In this study, the model is forced by globally observed SST. We have not discussed the relationship between the monsoon and SST anomalies of the Pacific and Indian Oceans. However, Yang and Lau( 1997) pointed out that the impact of SST anomalies on the monsoon is nonlinear with warm and cold events . The change in tropical Western Pacific SST is sometimes important (Soman and Slingo, 1997).
There are even more uncertainties in the role of the Indian Ocean in the variability of the Asian monsoon (Palmer et al., 1992;Nigam, 1994;and Soman and Slingo, 1997). Whether the ocean plays an active or passive role in the monsoon and to what extent the ocean influences the monsoon system are unclear. HC pointed out that the orography, land-sea distribution and SST as well as different surface conditions in the Indochina Peninsula and India cannot be neglected during the first transition of the ASM and the onset of the EASM.
Therefore, the linkage between the monsoon's variation and surface boundary forcing is very important. In addition ,the application of the PBL parameterization scheme is crucial to the model's simulation.
(2) Although arguments have been put forward to explain this unique event ,the cause of the seasonal northward jump of the subtropical high has not yet been clarified. It has been revealed that the seasonal northward jump of the SH concurrently occurs with the abrupt meridional shift, from the south to the north, of the monsoon vertical circulation cell in East Asia. Lin & Miao ( 1981 ) further pointed out that there is a tendency for the change in the mean meridional cell to lead to the northward jump of the SH. The internal dynamics of the SH need more theoretical and numerical studies.
(3) The dynamic and thermodynamic effect of the Tibetan Plateau is important to the develop ment and maintenance of the ASM. However, in the early stage of the seasonal transition this effect does not function well, because the cold air advection and the adiabatic cooling eliminates the warming caused by the sensible heat flux ( Feng , 1992 ). Theoretical and numerical studies have shown that the upper-level anticyclones can be forced either by a realistic heating distribution (e.g. Hoskins and Rodwell, 1995) or by an idealized, deep heating profile (e.g. Gill, 1980). The Tibetan Plateau can obviously help establish the up per-level anticyclone and, possibly the first transition of the ASM to some extent, but is unlikely to be the sole and most important factor. HC also pointed out that the heating and lifting effects of the high terrain in the Indochina Peninsula can be more promint than heat ing effect of the Tibetan Plateau during the first transition. The exact role of the Tibetan Plateau in maintaining the summer circulation has no unified results yet. Sensitivity studies of the dynamic and thermodynamic effect of the Tibetan Plateau using numerical modeling are needed.
( 4) In view of the fact that the intraseasonal variations are linked to a variety of fu ndamental physical processes, such as latent heat release (including large-scale and convective heat ing) and boundary fluxes , a thorough understanding of the phenomenon and better repre sentation of the model physics(such as PBL, cumulus, cloud/radiation, and topography) have a potential impact on the prediction of long-range variations of the tropical and East Asian monsoonal circulations.