1. Introduction

A fundamental requirement for agricultural development and water resource management in the south and Southeast Asian region is a better predictive capability of internal and external variability of summer monsoon rainfall on time scales ranging from weeks to several years (e.g., Parthasarathy et al. 1988; Webster et al. 1998; Lal et al. 1999; Gadgil et al. 1999, 2002). As the Asian summer monsoon is a major component of global climate dynamics, it has the potential of imparting variability on the climates of other components (e.g., Webster and Yang 1992; Webster et al. 1998; Kirtman and Shukla 2000). Therefore, improvements in the predictability of the monsoon will lead to increased predictability elsewhere. The fundamental hypothesis of this study is that a greater understanding of the physical processes determining the mean seasonal monsoonal precipitation patterns will lead to better predictions and provide milestones for numerical climate modeling.

There is little disagreement on the fundamental mechanism that drives monsoon circulations. Essentially, this is a cross-equatorial pressure gradient formed by the differential heating between the landmasses of the summer hemisphere and the oceans of the winter hemisphere. However, the spatial and temporal variability of monsoon precipitation is far more complicated than that which can be expected from such a slowly varying cross-equatorial pressure gradient. In fact, the processes that produce the specific spatial distribution of even the long-term mean precipitation are not clearly understood. Different processes, phenomena, and geographic features have been suggested as modulators of monsoon precipitation, such as local and regional effects of orography (e.g., Grossman and Garcia 1990; Grossman and Durran 1984; Xie et al. 2006), propagation characteristics of the monsoon intraseasonal oscillation (MISO) that involve both the ocean and the atmosphere (e.g., Lawrence and Webster 2002; Jiang et al. 2004; Wang et al. 2005), propagation of biweekly signals from the South China Sea (Annamalai and Slingo 2001), and land–atmosphere feedback (e.g., Webster 1983). All of these features of the monsoon in isolation or in combination appear as possible candidates.

This paper is an attempt to determine the mechanisms responsible for the variability and structure of summer monsoon rainfall on different time scales and regions within Asia. Figure 1 provides a geographical reference for the regions discussed in the text. These regions include the east Indian Ocean (EIO), Bay of Bengal (BoB), central India (CI), Gangetic Plains (GP), Western Ghats (WG), and Cambodia (CMD). These areas were selected based on geography and seasonal average precipitation: BoB, WG, and CMD correspond to regional rainfall maxima associated with the summer monsoon, EIO is a key region in order to understand different features of the MISO, CI is an area over land that is strongly modulated by the MISO, and GP corresponds to a very important geomorphologic and hydrographic unit, vital for the development of India and Bangladesh, whose precipitation is determined by the Southeast Asian monsoon and its variability.

Figure 2a shows the spatial distribution of the mean cumulative rainfall from June to September (JJAS) over the period 1979–2004 using the Global Precipitation Climatology Project (GPCP) data discussed in the next section, as well as the mean annual cycle of rainfall over the six regions previously described (Fig. 2b). It is evident from Fig. 2a that there is a complex spatial pattern of rainfall associated with the Asian monsoon system, with localized precipitation maxima to the west of the WG foothills in India with a total of 850 mm during JJAS, to the east side of the BoB with 1430 mm, and over CMD with 1050 mm. While the overall driving mechanism of the monsoon is the slowly varying cross-equatorial pressure gradient force, the rainfall is strongly regional with maxima occurring over the ocean regions and neither over the Asian continent nor the Indian subcontinent as initially noted by Grossman and Garcia (1990). In fact, the BoB possesses the largest global maximum of boreal summer rainfall on the planet. It is interesting to note that total summer precipitation drops almost linearly between the Bay of Bengal (>1200 mm), across central India (800 mm), and into Pakistan (200–300 mm).

The rainfall maximum over the EIO (860 mm) does not have a monsoonlike annual cycle and is relatively constant throughout the year (Fig. 2b). However, it has been shown to be intimately related to monsoonal intraseasonal variability (e.g., Webster et al. 1998; Lawrence and Webster 2002). In contrast to the near-equatorial precipitation, summer rainfall farther north and east over the Bay of Bengal and Southeast Asian countries commences in May. Over most of continental India, however, rainfall starts later during June (Fig. 2b). In fact, a number of studies consider the EIO as the source region of monsoon intraseasonal variability that possesses the largest amplitude of all monsoon precipitation variations (e.g., Webster et al. 1998; Lawrence and Webster 2002; Waliser et al. 2003a, b) producing active and break periods of the south Asian monsoon on 25–80-day time scales.

In the BoB, precipitation occurs as a maximum parallel to the Burma mountains (Fig. 2a) as noted by Grossman and Garcia (1990) and Xie et al. (2006). But what are the main processes that determine the rainfall maximum? Is the maximum produced by the mean onshore flow from the basic southwesterlies of the monsoon producing rainfall through orographic uplift? Does the freshness and warmth of the bay itself play any role so that if there were no Burmese Mountains there would still be a regional precipitation maximum? Furthermore, for whatever reason there is a maximum in the BoB, what time scales of motion contribute to its existence? Zuidema (2003), Grossman and Garcia (1990), and Xie et al. (2006) examined this question but the first study considered only two summers while the latter two used monthly data. Thus it is difficult to determine the relative role of intraseasonal variability or synoptic events, for example, in producing the maximum. Indian monsoon rainfall precipitation has proven to be very difficult to model, which is evident from the results of the Atmospheric Model Intercomparison project as summarized by Gadgil and Sanjani (1998) and Sperber and Palmer (1996). In fact, simulation of mean monsoon precipitation in general has been elusive in models. It is important to determine whether or not these failures result because of the poor representation of orography, for example, or the general absence in many numerical models of specific phenomena such as intraseasonal variability.

The complexity of the summer monsoon rainfall is not restricted to spatial patterns but also to its variation in time. The monsoon possesses a strong temporal variability ranging from the diurnal, to the synoptic and intraseasonal, and then to interannual time scales. In fact, this complicated structure extends to the relationship between interannual variability of each of the monsoon regions we consider. These interregional correlations are summarized in Table 1 for the period 1979–2004. It is important to note that scatterplots (not shown here) do not show the presence of extreme outliers modifying the correlations artificially. The complexity may be seen by comparing the interannual correlations between the BoB, CI, and GP, which we may expect to be highly correlated. The JJAS seasonal rainfall correlation between the BoB and the GP is 0.37, which is higher and of the opposite sign to that between the BoB and CI (−0.32). The reason for this dissimilarity in magnitude and sign between the correlations is unclear. Furthermore, why is the correlation between the BoB and CMD larger than between the BoB and the GP? Why is the precipitation in EIO so poorly correlated with all of the monsoon regions? The All-India Rainfall Index (AIRI; Parthasarathy et al. 1987) correlates well with CI, moderately with GP and WG, but very weakly with BoB. Are the relative weaknesses of the correlations or even their absence between regions indicating that other higher-frequency variability links different parts of the monsoon?

The questions raised above will be addressed in the following diagnostic study. Section 2 describes the datasets used in the study. Section 3 provides an overview of precipitation in south Asia. Sections 4 and 5 describe further analysis and present results of the analysis. Section 6 offers some conclusions.

2. Data

A number of different datasets are used in the study. These are described briefly below.

3. Overview of summer monsoon precipitation

Even though the amplitudes of the rainfall maxima shown in Fig. 2a change from year to year, they remain persistent climatological features. However, as suggested by the correlations shown in Table 1, the respective changes in seasonal rainfall from year to year for the different regions do not appear to be interrelated, suggesting strong internal variability in the monsoon system, possibly associated with intraseasonal activity. Different authors have previously pointed out that a proportion of the interannual modulation of the monsoon rainfall is found to be essentially independent of influences external to the Indian Ocean (e.g., Torrence and Webster 1999; Kumar et al. 1999). Large amplitude and large-scale interannual variability is evident in the rainfall anomalies for each JJAS period between 1997 and 2004, relative to the long-term mean (figure not shown). Most often anomalies of different signs lie across different parts of south and Southeast Asia. Over India, the spatial complexity of the patterns calls into question, once again, the usefulness of using the AIRI as a measure of Indian monsoon rainfall variability. Only in 2002 is all of India dominated by an anomaly of the same sign. The identification of high-amplitude internal monsoon variability is very important, as it may obscure signals of the interaction between the monsoon system and other major phenomena of the global climate, such as the ENSO system.

Spectral analysis of daily time series of rainfall reveals the importance of the intraseasonal variability present in monsoon rainfall. Figures 3a,b show the average wavelet spectra for the rainfall and OLR over the six regions obtained using techniques described by Torrence and Compo (1998). The solid line shows the average spectra and the shaded region encloses ±1 standard deviation. Since the wavelet analysis provides time and frequency localization of the distribution of the variance (power), it is possible to obtain a different average spectrum for every year (7 for GPCP rainfall, 26 for OLR). The mean average spectrum and the standard deviation of the spectrum are defined as the mean and the standard deviation of the power in all the yearly spectra for every period analyzed, respectively. The standard deviation is an indication of the variability of the periods in time as well as their change of amplitude. In general, rainfall in all regions shows high variability within a season on time scales ranging from about 7 to 60 days. Over EIO, the amplitude of the intraseasonal activity appears even stronger than the annual cycle, with a marked spectral peak at 30 days and a secondary peak around 15 days. This intraseasonal activity over the Indian Ocean is not an exclusive property of the summer season. In fact, variability in the intraseasonal band is usually stronger during winter (e.g., Salby and Hendon 1994). But, over most of the other regions considered, high intraseasonal activity is confined to the summer, with magnitude and periods that fluctuate notably from year to year as is evident in the average wavelet spectra and the magnitude of the deviation around the average. The intraseasonal activity during the wet phase of the Southeast Asian monsoon season is a well recognized feature since the work of Murakami (1976), Yasunari (1979), and Sikka and Gadgil (1980), who found intraseasonal variability associated with northward propagation of a zone of maximum cloudiness over the Indian Ocean. Since then, several studies (e.g., Cadet 1983; Krishnamurti and Subrahmanyam 1982; Knutson and Weickmann 1987; Hartmann and Michelsen 1989; Sengupta and Ravichandran 2001; Annamalai and Slingo 2001; Lawrence and Webster 2002) have reported the existence of intraseasonal spectral peaks in time series of precipitation, pressure, wind, OLR, and SST, as well as a tendency for northward propagation of the intraseasonal anomalies over the Indian Ocean.

From the summer average wavelet spectra, it is possible to identify some important aspects related to variability of rainfall within a wet summer season. For the BoB, in addition to the spectral peak corresponding to the annual cycle, there are two distinctive peaks at 10 and 30 days. A strong annual cycle peak is common to all regions except for the EIO. For CI, there are important spectral peaks at 6, 13, and 25 days, indicating significant variability differences with the BoB region.

While the wavelet spectrum of rainfall over the GP also shows intraseasonal 25–80-day variance during some years, it is not large enough to appear as a peak in the average spectrum. There is, however, a broad spectral peak between 4 and 9 days. The absence of a significant peak in the intraseasonal band poses a conundrum. Since the JJAS rainfall correlation is higher between the BoB and the GP than between the BoB and CI, why are the spectral characteristics of the variability closer between the BoB and CI?

For the WG, the average spectrum shows a broad peak between 5 and 50 days, indicating a more complicated temporal structure of the rainfall, perhaps due to the interaction between the combined effect of the large-scale monsoon intraseasonal variability and orography. Over CMD, the most distinctive peak after the annual cycle is located near 6 days (probably associated with propagating easterly waves from the Pacific Ocean) and a secondary peak around 45 days.

4. Intraseasonal variability during summer

The monsoon intraseasonal oscillation is the dominant large-scale feature of the monsoon (see review by Webster et al. 1998), which in general develops over the equatorial Indian Ocean and propagates eastward along the equator to the eastern reaches of the ocean (Yasunari 1979, 1981; Sikka and Gadgil 1980; Lawrence and Webster 2002; Webster and Hoyos 2004; Jiang et al. 2004). What local features modulate the intraseasonal monsoon rainfall signal so as to produce average spectra that are different in each region? Quite possibly the differences may depend on the specifics of the nature of the northward propagation, differences between individual oscillations, the different nature of the ocean–atmosphere in specific locations, and land–atmosphere feedbacks. In some manner, these processes aggregate to cause interannual variability of the intraseasonal mode.

Figure 4 shows the mean amplitude of the JJAS intraseasonal 25–80-day rainfall variance for each year from 1997 to 2004 as well as for all the years (multiannual). The mean amplitude in a spectral band is defined as the standard deviation of the filtered variable in the band, or alternatively, as the square root of the variance contained in the selected time span and spectral band. Areas of maximum intraseasonal activity vary strongly in magnitude and location from year to year, with the most recurrent feature being the high intraseasonal variability over the EIO and the BoB.

Figure 5 shows the cumulative intraseasonal rainfall for JJAS, obtained by simply adding the bandpassed rainfall time series for each day from 1 June to 30 September at every individual grid cell highlighting the locations where prolonged active and break phases of the monsoon occur. In essence, the distribution of positive and negative anomalies of rainfall associated with cumulative MISO activity can be regarded as the most important internal characteristic of interannual monsoon variability given its basinwide influence as well as the relative magnitude of the intraseasonal peak compared to variability at other time scales. Here we define internal variability as variability that does not arise through remote teleconnections with other phenomena, but rather from instabilities within the system. Among other sources of internal variability that could potentially introduce interannual monsoon anomalies are the variability at shorter time scales (e.g., 4–20-day variability) and the Indian Ocean Dipole (IOD; Saji et al. 1999; Webster et al. 1999). It is possible that the overall MISO activity during a season is partly modulated by long-term changes in the background state of the Indian Ocean induced by internal dynamics associated with the IOD, for example, or by external forcing such as ENSO. In the case that the former situation is true, the cumulative MISO would still be representing internally generated interannual variability. In the case of the latter, the cumulative MISO would be a mixture of internally and externally induced variability.

In the manner described above and similarly to Palmer (1994) and Webster et al. (1998), we hypothesize that summer intraseasonal activity, and especially its northward component, may be considered an instability of the monsoon system, which in general is not strongly affected by the changes in the long-term background state of the ocean–atmosphere system outside the Indian Ocean Basin and different teleconnections. The MISO occurs within the envelope of the summer monsoon, without preferred timing, and considerably altering the entire seasonal monsoon circulation. In addition, there exists outside the monsoon region no forcing of a time scale equal to that of the MISO. Little significant statistical relationship has been found between seasonal MJO activity and the ENSO cycle (Hendon et al. 2000; Slingo et al. 1999), except during extremely large warm events when MJO activity is suppressed. Hendon et al. (2000) suggested that most of the year-to-year variability of the MJO is internally generated independent of any slowly varying boundary forcing. In addition, Lawrence and Webster (2001) found that the summertime Intraseasonal Oscillation (ISO) is relatively uncorrelated with ENSO except for weakly positive correlation at the beginning of the Southeast Asian monsoon season. As a consequence it appears that it is necessary to consider the distribution of the interannual internal variability in order to explain some of the spatial features of JJAS rainfall anomalies. Perhaps the most prominent variation from the average is the 2002 drought over India, which coincides with large amplitude negative anomalies of MISO rainfall over the Indian subcontinent, mainly occasioned by a very strong and prolonged midseason break. It is also important to note that Gadgil et al. (2004) show some evidence that in extreme cases, and in particular during 2002, a composite state of ENSO and the atmospheric component of the IOD suggest unfavorable conditions for the overall summer monsoon. From this point of view, 2002 is a good example of how external and internal forcing at different time scales played a role in determining the seasonal rainfall.

Figure 6 shows the variability of the 25–80-day bandpassed daily rainfall plotted against 4–20-day rainfall variance, for all maxima in a 4–20-day variance band and for all regions of the analysis (Fig. 1). Spectral analysis showed high variability in the 25–80- and 4–20-day bands in the regions of analysis. There are two major points that suggest that the variability in the 4–20-day band is mainly contained in the 25–80-day band, especially in the EIO, BoB, and CI. One is that the 4–20-day variance time series (not shown) presents intraseasonal variability, and the other is that, as observed in Fig. 6, most of the major peaks in the 4–20-day variance correspond to a maximum or a very high value (above 1 standard deviation) in the 25–80-day bandpassed rainfall. This indicates that rainfall variability in the 4–20-day band is preferentially present during active periods of the monsoon intraseasonal oscillation.

Similarly, most of the monsoon subsynoptic variability appears to be contained within the intraseasonal envelope of convection. Figure 7a shows the average wavelet spectrum of the 3-hourly TRMM rainfall over the BoB, illustrating in addition to the variability in the 4–20- and 25–80-day bands that there is a very strong diurnal cycle of precipitation. The amplitude of the diurnal cycle is modulated by the envelope of convection in the 4–20- and 25–80-day bands, with greater amplitude during the peak of the intraseasonal event (see Fig. 7b). Similarly, Goswami et al. (2003) showed that the genesis of the monsoon low pressure systems, with a typical time scale of 3–5 days, is strongly modulated by the MISO, being 3 times as likely in the active phase than during the suppressed phase.

5. Composite analysis of summer intraseasonal variability

The relatively regular behavior of the summer MISO (e.g., Webster et al. 1998; Lawrence and Webster 2002; Waliser et al. 2003a, b) has been used greatly in forecasting the discharge of the Ganges and Brahmaputra Rivers and regional rainfall over India (e.g., Goswami and Xavier 2003; Webster and Hoyos 2004). To study in detail the structure of the MISO over the Indian Ocean and Southeast Asia, the dates of maximum rainfall associated with intraseasonal activity (25–80-day bandpassed data) with magnitudes greater than 1 standard deviation in the band were selected for the regions shown in Fig. 1. The composite evolution of several fields was studied. It was found that SST, surface pressure, lower-tropospheric winds, and rainfall itself (or OLR) emerge as the most coherent fields to explain the fundamental features of the intraseasonal activity.

For the BoB, a total of 24 cases were selected in the period 1997–2004. From those, 16 cases are classified as what may be referred to as the canonical MISO, 4 are classified as having developed in situ, and 4 as having developed through orographic interactions under break conditions over the Bay of Bengal.

6. Conclusions

A variety of data sources were used to study the contribution of rainfall variability over a range of time scales to the long-term mean seasonal summer (JJAS) monsoon rainfall (Fig. 2a). The role of the MISO in generating internal interannual variability of the monsoon system was also studied.

It was shown that both the amplitude and the distribution of the cumulative MISO-related rainfall strongly vary from year to year, constituting one of the most important sources of internal monsoon interannual variability. To a large degree, this internal variability is required to explain many important features of the seasonal rainfall distribution. In addition, the low values of the interregional correlation of rainfall within the monsoon system (Table 1) suggest large internal monsoon variability. Collectively, these factors introduce a scientific challenge for all teleconnection studies. To evaluate the effect and/or mutual influence among the monsoon and different external forcing phenomena (e.g., ENSO), it is necessary to design and evaluate tools that separate internal from external interannual variability. Statistically, it is difficult to separate internal and external sources of interannual variability from datasets because of the relatively short high-resolution and spatially distributed data available; however, ensemble model experiments could be used for that purpose (e.g., Quan et al. 2004).

To obtain a better understanding of the nature of the intraseasonal activity over the Southeast Asian monsoon region and its effect on the distribution of rainfall, a detailed composite analysis was undertaken based on strong MISO events over six key regions for the monsoon. Emerging from the composite analysis is the most coherent temporal and spatial structure of the MISO instability. The structure was referred to as the canonical MISO and is shown schematically in Figs. 15a,b. This mode starts in the equatorial Indian Ocean following warmer than average surface waters. Anomalies then generally propagate eastward to the EIO where they bifurcate, generating sequential active and break periods over all the regions considered, even as far east as CMD. The northward propagation of the convective anomalies, accompanied by a low pressure system, lags the propagation of the SST anomalies. The variability within the convective MISO envelope, especially over the BoB, shows individual southward-propagating storms with life spans less than a day (Fig. 10).

It was noted that large-scale ocean–land temperature gradients cannot explain the mean seasonal precipitation patterns and are especially unrelated to the BoB precipitation maximum. However, the strongest canonical MISO signal occurs over the BoB, following the bifurcation of the near-equatorial convective anomalies to the west of Sumatra. After the northward propagation of the low pressure system, the center of the latter is positioned over India, generating cyclonic circulation that tends to drive moist air toward the Burma mountain range, which possesses an average elevation of about 1800 m. Upstream effects of mountain ranges considerably enhance rainfall over the ocean in the northeast corner of the Bay of Bengal, explaining much of the observed seasonal maximum parallel to the mountains. The mountains of Burma do not seem to play a defining role in the MISO evolution itself, nor in the northward propagations, but it would seem that they directly influence the cumulative MISO-associated rainfall and, hence, the magnitude of the observed seasonal rainfall.

Other studies (Grossman and Durran 1984; Grossman and Garcia 1990; Xie et al. 2006) have noted the importance of orographic features in the spatial organization of the monsoon rainfall. However, where this study differs in particular is that it considers the interaction of orography with a variety of circulation patterns of different time scales. Xie et al. (2006), for example, used monthly precipitation and excluded by this choice important intraseasonal time scales. In this work, we specifically identified the interaction of the MISO during both active and suppressed conditions with orography in order to recognize the specific cause of precipitation over the Bay of Bengal.

The determination of the important role of the interaction between the MISO and orography is an important result since it suggests that in order for the climate models to reproduce the observed seasonal monsoon rainfall structure, they need to simulate better the intraseasonal activity observed and also be able to include the important effects of tall narrow mountains.

The study identified why there is regional precipitation occurring from time to time over the bay during break conditions over northern India. This occurs when a high pressure system is located in the south of the bay, inducing an anticyclonic circulation that drives moist air toward the mountains near the Bangladesh–Burma border. This sequence is described schematically in Figs. 16a,b. Under the proper conditions of atmospheric stability and moisture, orographically induced rainfall develops, augmenting the seasonal rainfall parallel to the mountains.

In addition to the canonical MISO, in situ activity can develop over CI, the GP, and the BoB. Over CMD, westward-propagating disturbances developing to the east of the Philippines also generate rainfall both in the 25–80- and 4–20-day bands. The main difference in these 2 cases is that in the 25–80-day band, the SST variability is higher, suggesting stronger ocean–atmosphere coupling in this band compared to the 4–20-day band. Over CI, activity on the 4–20-day time scale also comes from the east with propagating disturbances generated over the Gulf of Thailand and South China Sea.

It is quite possible that due to the sensitivity of models to initial conditions and the nonlinear nature of a MISO event, models will not be able to forecast more than one MISO event ahead of time. As a consequence, the full MISO activity during the summer season cannot be completely anticipated. However, because of the regular and coherent nature of the MISO, in order to estimate better the spatial structure of the seasonal rainfall, a summary of the statistical nature of intraseasonal activity prior to the summer season may help. Such a statistical quantification of the MISO would have to come from a model that is able to sustain intraseasonal variance with a magnitude similar to that observed. Wang and LinHo (2002) reached a similar conclusion when studying the features of the rainy season of the entire Asian–Pacific domain, noting that “an adequate description of the spatial–temporal structure of the mean Asian–Pacific monsoon rainy season should take the intrinsic subseasonal components into account.”

This study of the subseasonal rainfall variability has provided some key aspects that help in understanding the processes that determine the seasonal distribution of summer monsoon rainfall. Much work is still required to understand the role of the background state on individual intraseasonal event propagation. While the large-scale dynamical features of a MISO appear to be very robust, the background state may affect rainfall in one region and not the others. Collectively, a deeper understanding is needed for both empirical forecasting of monsoon rainfall and for determining model standards for numerical prediction.