With the use of daily rain gauge data observed at 210 stations in the Indochina Peninsula (ICP) for the 26 yr from 1978 to 2003, this paper describes climatological characteristics of 2 types of intraseasonal variations (ISVs): the 30–60-day variation (30–60DV) and the 10–20-day variation (10–20DV). The authors find that these characteristics are quite different from place to place in the ICP.
During the rainy season, variance of the 30–60DV is generally larger in coastal regions than over inland regions and it has two local maxima: one found in the coastal region of Myanmar (CMY) and the other in the southern Laos and central Vietnam region (SLCV). Wavelet analysis reveals that the 30–60DV in the CMY is active throughout the rainy season (May–October) and exhibits the maximum activity in May–June. In addition, its typical time scale shifts from 40 days in the early half of the rainy season to 50 days in the latter half. Cross-correlation analysis reveals that its signal propagates northward. On the other hand, the 30–60DV in the SLCV is active only during July–October, and its signal propagates northwestward.
The largest variance of the 10–20DV is found in the coastal regions of northern and central Vietnam (CNCV), while the variance in other coastal regions is generally smaller than that in inland regions. In contrast to the 30–60DV, the 10–20DV activity varies significantly over the course of the rainy season. The 10–20DV in the inland regions is active in May and September and inactive in July, while that in the CNCV is active during August–November. The 10–20DV exhibits high spatial coherence over most of the ICP, and its signal propagates west-northwestward.
Relationship of the ISV in the ICP with synoptic-scale ISV structures is also discussed.
In South and Southeast Asia, intraseasonal variations (ISVs) of precipitation and tropospheric circulation are important features during the rainy season and thus have been studied by many researchers. The ISVs can be divided into two types according to their time scales; one has a 30–60-day time scale (30–60-day variation, hereafter 30–60DV) and the other a 10–20-day time scale [10–20-day variation (10–20DV)]. This distinction results from these two types being different in their synoptic characteristics and behavior.
In the boreal summer, the 30–60DV of convection originates over the western and central equatorial Indian Ocean (Kemball-Cook and Wang 2001) and then propagates northward to the Asian monsoon regions (Yasunari 1979; Krishnamurti and Subrahmanyam 1982) and eastward along the equator (Krishnamurti et al. 1985; Murakami and Nakazawa 1985; Lau and Chan 1986). After the eastward-propagating convection reaches the western equatorial Pacific region, it turns to propagate northward or northwestward (Lau and Chan 1986; Knutson and Weickmann 1987; Chen and Murakami 1988) because of the Rossby wave dynamics (Kemball-Cook and Wang 2001; Annamalai and Sperber 2005). Many researchers (e.g., Julian and Madden 1981; Lau and Chan 1986; Lawrence and Webster 2002) have regarded the 30–60DV that occurs in the boreal summer as being part of the Madden–Julian oscillation (MJO; Madden and Julian 1971, 1972) modulated by the monsoon circulation.
The 10–20DV in the Asian monsoon regions has been studied since the early 1970s (e.g., Murakami 1972; Krishnamurti et al. 1973). The 10–20DV of convection and accompanying cyclonic circulation in the lower troposphere originates over the South China Sea or the western Pacific Ocean. It propagates westward and produces rainfall over the Indochina Peninsula (ICP; Yokoi and Satomura 2005). The 10–20DV then amplifies over the Bay of Bengal (Chen and Chen 1993), propagates farther west-northwestward (Yasunari 1979; Chen and Chen 1993; Yokoi and Satomura 2006), and causes rainfall over the Indian subcontinent. The 10–20DV can be regarded as an equatorial Rossby wave modulated by the summer monsoon circulation (Chatterjee and Goswami 2004; Yokoi and Satomura 2005).
Understanding the ISVs of precipitation over land areas, including the ICP and the Indian subcontinent, is of importance because of the serious impacts on various aspects of human life, such as agriculture, disasters, and water resources. It is also of scientific interest since ISV characteristics over land areas seem to be modulated by complex surface conditions and thus are expected to be different from those over the ocean. For these reasons, there have been many studies dealing with the ISVs of precipitation over land. For example, Krishnamurti and Bhalme (1976) and Hartmann and Michelsen (1989, hereafter HM1989) analyzed precipitation data observed in India, Ohsawa et al. (2000) analyzed data in Bangladesh, and Kripalani et al. (1995) and Yokoi and Satomura (2005) analyzed data in Thailand.
HM1989 analyzed 70-yr rain gauge data observed in India and described the horizontal distribution of variance of the ISV. HM1989 revealed that the variance of the 30–60DV was statistically significant over most of the Indian subcontinent south of 23°N, while that of the 10–20DV was not significant except in a few small areas in northeastern India. HM1989 therefore concluded that only the 30–60DV was important over the Indian subcontinent.
Characteristics of the ISV over the ICP were reported to differ from those over the Indian subcontinent. Using 20-yr rain gauge data, Kripalani et al. (1995) showed that the variance of the 10–20DV in Thailand was larger than that in India and stressed the importance of the 10–20DV. Their study, however, did not use data observed in other Indochina countries; thus its characteristics in these countries have yet to be clarified. The present study will analyze daily rain gauge data observed not only in Thailand but also in Myanmar, Laos, Cambodia, and Vietnam in order to describe ISV characteristics over the entire ICP.
HM1989 and Kripalani et al. (1995) evaluated the variance of the ISVs over periods that included entire rainy seasons. Yokoi and Satomura (2005), on the other hand, analyzed the ISVs of precipitation during the 1998 rainy season in Thailand and revealed that the 30–60DV dominated during June and July, while the 10–20DV dominated in August and September. Furthermore, Annamalai and Slingo (2001) analyzed principal oscillation patterns of outgoing longwave radiation (OLR) data and demonstrated that the amplitude of the leading principal oscillation pattern with the intraseasonal time scale varied over the course of the rainy season. Although these two studies did not directly reveal the climatological characteristics of the ISV of precipitation in the ICP, their results imply the possibility that its activity tends to vary over the course of the rainy season. The present study will therefore reveal a climatological seasonal march of the activity with the use of wavelet analysis.
2. Data and methods
This study uses daily precipitation data observed by meteorological agencies of the Indochina countries and collected in the Global Energy and Water Cycle Experiment (GEWEX) Asian Monsoon Experiment (GAME) Tropics (GAME International Science Panel 1998) and other related projects. The geographical distribution of the rain gauge stations is represented in Fig. 1 by dots. There are a total of 210 stations: 40 in Myanmar, 98 in Thailand, 11 in Laos, 9 in Cambodia, and 52 in Vietnam. The period analyzed is the 26 yr from 1978 to 2003, with all of the stations having data for more than 13 yr during this period. Figure 1 indicates that the number of stations is larger in the western half of the ICP than in the eastern half. The densest area with stations is the mainland of Thailand.
For the convenience of analysis, daily grid precipitation data were constructed at a 0.5° interval in longitude and latitude. The grid data were obtained by averaging the daily precipitation observed at rain gauge stations where the distance from a given grid point was less than 1°. We used 529 grid points where the grid data were calculated from data of multiple stations (Fig. 2).
The cube roots of the grid precipitation data are used in this study. This is because Stidd (1953) demonstrated that the frequency distribution of the cube roots of daily precipitation data was closer to the normal distribution than that of the original data. The normal distribution is a good condition for the application of statistical methods to the data (e.g., parametric tests).
According to HM1989, the smoothed mean annual cycle of precipitation was then calculated for each grid point. The cube roots of the grid precipitation data for each calendar day were averaged over the 26 yr, and the resulting mean annual cycle was then smoothed 300 times with a 1-2-1 filter to obtain the smoothed mean annual cycle. An anomaly of the precipitation data from the smoothed mean annual cycle, hereafter simply termed a precipitation anomaly, is used for analysis. Note that results shown in this paper do not change qualitatively, even if we use other definitions of smoothed mean annual cycle, such as a sum of the first three, four, or five harmonics of the original mean annual cycle.
In addition to the rain gauge data, we use the 40-yr European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA-40; Uppala 2002) in order to discuss the relationship between the ISV over the ICP and that of the large-scale atmospheric circulation field. The variables used are zonal and meridional wind components at the 850-hPa level during the years from 1978 to 2002. We also analyze streamfunction, which is calculated from these two components.
b. Fourier analysis
Following HM1989, this study performed a Fourier analysis to measure the variance of the ISVs during the rainy season. HM1989 applied a Fourier analysis to precipitation data for the 200-day period from 10 April to 26 October of every year, which included the rainy season in India. Over the ICP, the climatological onset and withdrawal dates of the rainy season differ from place to place by more than one month at the utmost; the rainy season starts on dates from late April to middle May and withdraws on dates from early October to late November (Matsumoto 1997). Therefore, prior to the Fourier analysis, the 200-day period at each grid point was determined by minimizing the difference of the values of the smoothed mean annual cycle between the first and last days of the period. The smoothed mean annual cycle and the 200-day period determined for the grid point at 17°N, 100°E are shown in Fig. 3a as an example. The first date of the 200-day period at this grid point was determined to be 19 April.
After the elimination of linear trends and application of bell tapering (Fig. 3a) to the 200-day precipitation anomaly data, the Fourier power spectrum was calculated for each year and then averaged over the 26 yr. An example of the averaged Fourier power spectrum is presented in Fig. 3b. The variances of the 30–60DV and the 10–20DV were then defined as the summation of the averaged Fourier power with frequencies from 3 to 6 (200 days)−1 (the dark shading in Fig. 3b) and from 10 to 20 (200 days)−1 (the light shading in Fig. 3b), respectively. The statistical significances of these variances were assessed with the use of the χ2 test. The null hypothesis used is the variance of red noise.
As will be shown later, the variance of the 10–20DV is generally larger than that of the 30–60DV. However, note that this fact does not directly imply that the 10–20DV is more important than the 30–60DV. It is difficult to evaluate the relative importance between these two ISVs from this variance alone. Therefore, this study will focus mainly on their horizontal distributions.
c. Wavelet analysis
To examine the hypothesis that ISV activity varies over the course of the rainy season, a wavelet analysis was performed. The Morlet wavelet (Torrence and Compo 1998) was used as the mother wavelet, since the wavelet power spectrum calculated with the use of this wavelet can be easily compared with the Fourier power spectrum (Torrence and Compo 1998; Yokoi and Satomura 2005). Figure 4a shows the wavelet power spectrum at 17°N, 100°E in 1980 as an example. Three maxima are found with intraseasonal time scales: one in May with a 13-day time scale, one in July with a 34-day time scale, and the other in September with a 14-day time scale.
For consistency with the definition of the variance, we then defined 30–60DV wavelet power as the summation of the wavelet power with time scales of 200/3, 200/4, 200/5, and 200/6 days. Definition of 10–20DV wavelet power is the same as the 30–60DV power except for time scales of 200/10, 200/11, . . . , and 200/20 days. Time series of these wavelet powers for the example in Fig. 4a are shown in Fig. 4b. Two maxima of the 10–20DV wavelet power and one maximum of the 30–60DV wavelet power are observed, all of which correspond to the three maxima found in Fig. 4a.
The statistical significance of the ISV wavelet power was then assessed with the use of a Monte Carlo simulation at the 95% confidence level. The null hypothesis used is the wavelet power of red noise. The procedure of the Monte Carlo simulation performed is briefly explained in the appendix. The 95% confidence levels of the ISV wavelet power and the periods with statistically significant power for the example case are shown in Fig. 4b. We can find that all of the three maxima mentioned above are statistically significant.
Finally we define an index of the ISV activity in a certain period as the ratio of the number of days when the ISV wavelet power is statistically significant to the number of all days in the period during the course of 26 yr. The index is referred to as the active day ratio (ADR) in this paper. For example, if there are 156 days out of 26 yr of June days (780 days in total) when the 10–20DV wavelet power is statistically significant, the ADR of the 10–20DV in June is calculated as 0.2. The ADR is considered the climatological frequency of the occurrence of each ISV with significant wavelet power during the period. In consideration of their time scales, we calculated the ADR on a monthly basis for the 10–20DV and on a bimonthly basis for the 30–60DV.
d. Cross-correlation analysis
To reveal propagation features of the ISV in the ICP and their relationship with synoptic-scale ISV structures, a cross-correlation analysis was performed. For the calculation of the cross-correlation coefficient, the 30–60- and 10–20-day components were retrieved with the use of a Lanczos bandpass filter (Duchon 1979) having cutoff periods of 25 and 70 days and a filter length of 201 days, and that having cutoff periods of 10 and 25 days and a filter length of 101 days, respectively. The statistical significance was assessed with the use of the Student’s t test with a null hypothesis of no correlation.
3. Characteristics of the 30–60DV
Figure 5 presents horizontal distribution of the percent variance of the 30–60DV (the variance normalized by the total variance of precipitation anomaly). The variance is statistically significant at 276 grid points (about a half of all the grid points), most of which are located near the coast rather than in inland regions. The maximum is found in the coastal region of Myanmar (hereafter CMY) at 16°N, 98°E, with a value of nearly 0.2. In this region, the variance in excess of 0.125 is confined to the narrow area west of 98.5°E and south of 17.5°N, while it is below 0.1 in western Thailand just east of the high-variance area. Such horizontal distribution results in a notable zonal gradient across the border between Myanmar and Thailand. Mountain ranges along the border (Fig. 1) possibly play roles in this gradient. Another high-variance area in excess of 0.125 is in the southern Laos and central Vietnam region (SLCV) with a local maximum at 14.5°N, 107.5°E. The horizontal gradient around the SLCV seems weaker than that around the CMY, although data southwest of the SLCV are not available.
Characteristics of the horizontal distribution of the variance in Thailand are generally consistent with the results of Kripalani et al. (1995). For example, both Kripalani et al. (1995) and our Fig. 5 indicate a larger variance in the Malay peninsular region than in central and northern Thailand. The present study further reveals that the 30–60DV in Thailand has a generally smaller variance than in the surrounding regions including the CMY and SLCV.
Then we will describe the seasonal march of the 30–60DV activity. Figure 6 presents the ADR of the 30–60DV on a bimonthly basis, which was introduced in section 2c. In addition, the ADR time series averaged over the area 14.5°–17°N, 96°–98.5°E and over the area 13.5°–16°N, 105°–107.5°E are shown in Fig. 7 in order to focus on the CMY and SLCV, respectively (see Fig. 2).
In the January–February period (Fig. 6a), the ADR at most grid points, except for south of 9°N, is less than 0.05. Consequently, the 30–60DV of precipitation anomaly is inactive and cannot be distinguished from red noise.1 This inactiveness is probably due to the background conditions unfavorable for convection and the lack of precipitation during the period. That is, even if the 30–60DV of circulation anomalies reaches the ICP, it probably cannot bring rainfall. Over the Indian subcontinent, HM1989 revealed that the 30–60DV of precipitation anomaly was also inactive during the dry season. In the March–April period (Fig. 6b), the ADR is still small compared to that during the rainy season and lower than 0.05 in some areas, while it slightly increases in western Thailand.
In the May–June period (Fig. 6c), the ADR is generally larger in the western half of the ICP (west of 101°E) than in the eastern half, with a maximum in the CMY. In addition, the ADR in the CMY has the temporal maximum during this period (Fig. 7). Although a local maximum is also found in the SLCV, its value is about a half smaller than that in the CMY.
The ADR in the SLCV then increases in the July–August period, while that in the CMY slightly decreases (Fig. 7). The horizontal distribution in this period (Fig. 6d) exhibits two local maxima found in the CMY and SLCV, and their values are comparable to each other.
In the September–October period, the ADR in both the CMY and SLCV slightly decreases (Fig. 7), while that in the surrounding areas increases (Fig. 6e), especially in central Myanmar, northeastern Thailand, and the coastal region of central Vietnam. As a result, the horizontal contrast is weak compared to that in the May–June and July–August periods. In the November–December period (Fig. 6f), the ADR decreases over most parts of the ICP, while relatively high values are still found in the coastal region of central Vietnam.
Figure 7 indicates that the 30–60DV in the CMY maintains high ADR values in excess of 0.3 throughout the rainy season (May–October). On the other hand, the ADR values in the SLCV exceed 0.3 only during July–October. We can therefore conclude that the seasonal march of the activity differs between the CMY and SLCV, especially in the early stage of the rainy season.
This difference between the CMY and SLCV may be interpreted from the viewpoint of seasonality of synoptic-scale 30–60DV features revealed by Kemball-Cook and Wang (2001). They compared propagation features of the 30–60DV between May–June and August–October periods. They found that to the west of the ICP, OLR signals propagated northeastward to reach the Bay of Bengal (adjacent to the CMY) in both periods. On the other hand, they revealed that northwestward-propagating signals emanated over the equatorial western Pacific could reach the eastern part of the ICP (which includes the SLCV) only during the latter period, while those during the former period dissipated over the western Pacific. The seasonality of the synoptic-scale features to the west and east of the ICP seems consistent with the ADR seasonal march in the CMY and SLCV, respectively.
The 30–60DV in the CMY has a seasonal change in its typical time scales. Figure 8a shows the 26-yr-mean annual cycle of the wavelet power spectrum averaged over the CMY (14.5°–17°N, 96°–98.5°E). The 40-day time scale has the largest wavelet power among intraseasonal time scales in May, June, and early July. In late July, August, and September, the wavelet power with the 50-day time scale is the largest. Therefore, the typical time scale shifts from 40 to 50 days in mid-July.
Hartmann et al. (1992) also studied seasonality of the typical time scale in India and the western Pacific. From their figures, we can find that a spectral peak shifted from a 35-day period in May to a 50-day period in August. Therefore, the shift of the typical time scale is not a local feature confined to the CMY, but can be observed in other Asian and western Pacific monsoon regions.
The 26-yr-mean annual cycle of the wavelet power spectrum averaged over the SLCV (13.5°–16°N, 105°–107.5°E) is shown in Fig. 8b. In contrast to the CMY, we cannot find a spectral peak at the 40-day time scale in the early half of the rainy season, while the wavelet power at the 50-day time scale is largest during June–September. We cannot find a shift of the typical time scale as well.
Figure 9a presents the result of the cross-correlation analysis between the 30–60-day component of precipitation anomaly at 16°N, 98°E, which is in the CMY, and that at the other grid points. The analysis was performed for periods from 1 May to 31 October every year. This figure shows the lag at which the maximum cross-correlation coefficient is obtained, and grid points without statistically significant coefficients have been blanked out. Regions where the coefficient is statistically significant are the CMY, SLCV, a part of northeastern Thailand, and the coastal region of Cambodia.
The lags of the maximum coefficient around the CMY indicate northward propagation of the signal at least up to 16°N. The lags at the base of the Malay Peninsula along 12°N are −2 days (indicated in Fig. 9a by moderate shading with a cross), those south of the reference grid point are −1 day (light shading with a cross), those in its neighbors are 0 days (an open circle), and those to its northwest are +1 day (light shading without a cross). We can estimate its speed at about 1° of latitude per day. In addition, some indication of westward propagation is found between 16° and 19°N, although this study will not deal with this feature in detail.
We can regard the 30–60DV in the CMY as a part of the synoptic-scale 30–60DV structure in the Asian monsoon region. Figure 10a shows the 30–60-day component of horizontal wind at 850 hPa simultaneously regressed on that of the precipitation anomaly at 16°N, 98°E. Anomalous westerly wind prevails over India, the Bay of Bengal, the ICP, and the South China Sea. This anomaly originates over the equatorial region and then propagates northward over the Bay of Bengal at a similar speed to the precipitation signals in the CMY (Fig. 10b). These horizontal and temporal structures are consistent with well-known synoptic-scale 30–60DV features revealed by many researches (Lau and Chan 1986; Kemball-Cook and Wang 2001; Jiang et al. 2004, and others).
The cross-correlation analysis with the reference grid point at 15°N, 107°E reveals propagation features around the SLCV. We performed this analysis for the period from 1 July to 31 October every year, the active period of the 30–60DV in this region. The lags of the maximum cross-correlation coefficient are shown in Fig. 9b. In addition to the northward propagation around the CMY, a clear west-northwestward propagation feature around the SLCV is discernible. The lags southeast of the reference grid point are −2 or −1 days, those in its neighbors are 0 days, and those to its west are +1 or +2 days. This propagation feature may be associated with the northwestward propagation of the 30–60DV over the western Pacific (Lau and Chan 1986; Kemball-Cook and Wang 2001; Annamalai and Sperber 2005).
4. Characteristics of the 10–20DV
This section will describe climatological characteristics of the 10–20DV. Its percent variance (Fig. 11) is statistically significant at 60 grid points (11% of all the grid points), which are located at a few inland regions and the coastal region of northern and central Vietnam (CNCV) at latitudes between 14° and 20°N. The largest variance is found in the CNCV at 14.5°N, 109°E. In addition, we can find weak maxima in northern and northeastern Thailand. The variance in the inland regions is generally larger than that in the coastal regions except in the CNCV and westernmost Myanmar. The most inactive region is southernmost Vietnam.
Horizontal distribution of the ADR of the 10–20DV on a monthly basis is shown in Fig. 12. Similar to the 30–60DV, the 10–20DV from January to March (mid dry season) is inactive, which is probably due to background conditions unfavorable for precipitation during the period.
During the rainy season (May–October), the ADR in the inland regions, including central Myanmar, northern Thailand, and northeastern Thailand, possesses a qualitatively similar seasonal march and has two temporal maxima found in May and September. These two maxima exhibit different characteristics in their magnitude and horizontal distribution. The 10–20DV is much more active in September than in May; about 20% of all the grid points have ADR values larger than 0.3 in September, while only 4% have it in May. In addition, while high-ADR regions in both months expand in a northwest–southeast direction in the inland regions, they are located a few more degrees of latitude north in September than in May.
Between these maxima, an ADR minimum is found in July. In particular, the ADR in central Myanmar and the western half of Thailand is below 0.05, suggesting that the 10–20DV is as inactive as that in the dry season.
The CNCV is quite different from the inland regions in the ADR seasonal march. The 10–20DV in this region is active during the August–November period and exhibits maximum activity in October. During this 4-month active period, the position of the ADR local maximum moves southward from 22°N to 12°N. Note that the ADR in August and September in the southern part of the region is below 0.05, resulting in a significant north–south contrast of activity. Fudeyasu et al. (2006) reported that an average path of westward-propagating vorticity disturbances crossing over the CNCV migrated southward during this period. Since Yokoi and Satomura (2005) revealed that such vorticity disturbances could cause the 10–20DV of precipitation over the ICP, the southward migration of the path is possibly responsible for the southward movement of the ADR local maximum.
The above examination of the ADR reveals that the 10–20DV in the inland regions and in the CNCV is active at least in some periods of the rainy season, although its variance calculated for the 200-day rainy season is insignificant at many grid points. This small variance is probably due partly to the existence of inactive periods in the rainy season.
To reveal spatial coherence and propagation features of the 10–20DV, we performed the cross-correlation analysis between the 10–20-day component of the precipitation anomaly at 16°N, 103°E in northeastern Thailand and that at the other grid points for periods from 1 May to 31 October for every year. Figure 13a shows the lags at which the maximum correlation coefficient is obtained. This figure reveals that the 10–20-day components at most grid points (88% of all the grid points) are significantly correlated with that at the reference grid point. It also shows a systematic shift of the lags. The lags in southern Vietnam are −1 day, those around the reference grid point are 0 days, those in the western half of Thailand are +1 day, and those in central and northern Myanmar are +2 or +3 days. That is, the 10–20DV signal propagates west-northwestward with time. The westward component of propagation velocity can be estimated at about 3° of longitude per day. The widespread region of significance and the clear propagation feature further suggest the importance of the 10–20DV. Moreover, the cross-correlation analysis with the reference grid point at 19°N, 98°E in northern Thailand also demonstrates these two characteristics (Fig. 13b), suggesting strong robustness of the results.
To discuss the relationship between the 10–20DV in the ICP with that of the synoptic-scale circulation field, we examine the 10–20-day components of streamfunction and horizontal wind at the 850-hPa level regressed on that of the precipitation anomaly at 16°N, 103°E (Fig. 14). At lag 0 (Fig. 14b), there exists a cyclonic circulation (negative streamfunction) anomaly centered at 15°N, 95°E that is accompanied by southerly wind anomaly over the ICP. The cyclonic circulation anomaly propagates west-northwestward from 15°N, 110°E at a lag of −4 days (Fig. 14a) to 17.5°N, 80°E at a lag of +4 days (Fig. 14c). Estimated westward propagation speed is nearly 4° of longitude per day, which is a little bit faster but in fairly good accordance with the speed of the precipitation anomaly over the ICP. Furthermore, the horizontal and temporal structures in Fig. 14 are consistent with those revealed by previous studies (Chen and Chen 1993; Chatterjee and Goswami 2004; Yokoi and Satomura 2005). These results imply that the 10–20DV of the precipitation anomaly analyzed here is a part of synoptic-scale 10–20DV in the Asian monsoon region.
5. Summary and discussion
The characteristics of the intraseasonal variations (ISVs) of precipitation over the Indochina Peninsula (ICP) were studied with the use of daily rain gauge data observed at 210 stations in the Indochina countries for the 26 yr from 1978 to 2003. The use of the station data with good spatial density revealed the fine spatial distribution of the variance, the seasonal march of activity, and propagation features of two types of ISVs: the 30–60-day variation (30–60DV) and the 10–20-day variation (10–20DV). We found that these characteristics were quite different from place to place.
The variance of the 30–60DV was statistically significant over a wide area of the ICP, except for the inland regions. The maximum variance was observed in the coastal region of Myanmar (CMY) and the secondary maximum was found in the southern Laos and central Vietnam region (SLCV).
With the use of the active day ratio (ADR) defined based on the wavelet power spectrum, we revealed that the seasonal march of the 30–60DV activity in the CMY and SLCV during the rainy season (May–October) differed from each other. The 30–60DV in the CMY was active throughout the rainy season and most active in the May–June period, while that in the SLCV was relatively inactive during May–June and active only during July–October.
Typical time scales of the 30–60DV in the CMY varied over the course of the rainy season. The 40-day time scale had the largest wavelet power among the intraseasonal time scales in the first half of the rainy season, while the power with the 50-day time scale was the largest in the latter half.
Cross-correlation analysis revealed that the 30–60DV in the CMY propagated northward with a speed of 1° of latitude per day and could be regarded as a part of the synoptic-scale 30–60DV over the Bay of Bengal. The 30–60DV in the SLCV propagated west-northwestward.
In contrast to the 30–60DV, the variance of the 10–20DV was statistically significant only in the coastal region of northern and central Vietnam (CNCV) and some inland regions. The 10–20DVs were spatially coherent over most of the ICP. They propagated west-northwestward with a zonal speed of about 3° of longitude per day, which was in good accordance with the propagation speed of the synoptic-scale 10–20DV cyclonic circulation anomaly in the lower troposphere passing over the ICP. The 10–20DV of the precipitation anomaly can be regarded as a part of the synoptic-scale 10–20DV structure.
The 10–20DV activity in the inland regions exhibited two temporal maxima that were found in May and in September. The activity is higher in September than that in May. On the other hand, the 10–20DV in July was inactive, and in some areas it was as inactive as that in the dry season. In the CNCV, the 10–20DV was active only during August–November, and the ADR temporal maximum was found in October.
The geographical contrast in the variance of the ISV of precipitation anomaly revealed in this study implies that the synoptic-scale ISV disturbances can only bring substantial precipitation to specific regions of the ICP despite their large horizontal scales. This is possibly attributed to topographical effects, upon which further studies are necessary for understanding a causal mechanism for the contrast.
One important finding about the 10–20DV is that its activity varies drastically over the course of the rainy season compared to the 30–60DV. Although the 10–20DV in most regions exhibited statistically insignificant variance calculated over the entire rainy season, it was active at least during some periods of the rainy season. In addition, the high spatial coherence revealed with the cross-correlation analysis also suggests the importance of the 10–20DV.
Determination of monsoon onset is another important subject on precipitation variability. Many studies (e.g., Cadet 1986; Lau and Chan 1986) have stressed a relationship between the monsoon onset and the ISVs. Zhang et al. (2002) averaged observed precipitation data over inland and coastal regions of Thailand and insisted that the onset was accompanied by the arrival of the 10–20DV from the east. However, our results imply a possibility that the ISVs that modulate the monsoon onset differ in their time scales from place to place. At least in the CMY, where the climatological onset date is around mid-May (Matsumoto 1997), the 30–60DV is a major candidate because of its high activity in the May–June period (Fig. 6). The locality of the monsoon onset processes is an important subject and remains for our future study.
There exists a possible linkage between the seasonal march of the 10–20DV activity and that of precipitation. Matsumoto (1997) and Takahashi and Yasunari (2006) revealed that in most parts of the inland regions, precipitation was less in July than in May and September. Thus precipitation exhibited a quite similar seasonal march to the 10–20DV activity. Such a relationship is also found in the CNCV. In the coastal region of northern Vietnam, large amounts of precipitation are observed in September and October (Fig. 15a), when the 10–20DV is also active. Precipitation time series in central Vietnam (Fig. 15b) lag about one month behind those in northern Vietnam, which is in good accordance with the southward migration of the position of the ADR local maximum. Furthermore, the 30–60DV in the CNCV is more active during September–December than other months (Fig. 6). These facts suggest the possibility that the seasonal difference in ISV activity determines the climatological seasonal march of precipitation to at least some extent. However, this subject also remains for our future study.
There have been a number of studies that aimed to simulate and predict the large-scale 30–60DV in the Asian monsoon region (e.g., Waliser et al. 2003). These studies focused mainly on the 30–60DV in South Asia and over the western Pacific and did not pay much attention to that over the ICP. However, it seems fruitful to reproduce the geographical distribution of its variance over the ICP described in this study. This is because successful reproduction of the 30–60DV over the ICP will contribute not only to better prediction of local precipitation but also to better reproduction of the horizontal distribution of the heat source, whose complicated impact on the large-scale circulation field was stressed in Annamalai and Sperber (2005). The sharp gradient of the variance over the mountain ranges found in this study suggests that appropriate treatment of topographical effects is necessary for successful reproduction.
The rain gauge data were observed routinely by the Department of Meteorology and Hydrology in Myanmar, the Thai Meteorological Department, the Department of Meteorology and Hydrology in Laos, the Department of Meteorology in Cambodia, and the National Hydrometeorological Service of Vietnam. Data were collected by the Climate Research Laboratory of the Department of Earth and Planetary Science at the University of Tokyo under the GAME-Tropics and other related projects conducted in this laboratory. The authors are grateful to these agencies and laboratory for their provision of the data. This study was supported financially by a Grant-in-Aid for the 21st Century COE Program (Kyoto University), a Grant-in-Aid for Scientific Research B-2-15340157 of the Japanese Ministry of Education, Culture, Sports, Science and Technology, and by Core Research for Evolutional Science and Technology, the Japan Science and Technology Agency (Leader: Prof. Masakazu Suzuki). All figures were produced with the use of the GFD-DENNOU library.
Monte Carlo Simulation
In this study, the ISV activity is measured with the use of the wavelet power averaged over a range of time scales. Unlike the Fourier power spectrum, it is difficult to analytically determine the probability density function (PDF) of the averaged wavelet power (Torrence and Compo 1998). Therefore, we performed a series of Monte Carlo simulations to determine the PDF in order to estimate confidence levels used for the significance test. First, we constructed 100 000 series of univariate lag-1 autoregressive processes with the unity variance with the use of the uniform random numbers, which is a mathematical model of red noise. Then we calculated the ISV wavelet power defined in section 2c for each series and derived the PDF. Finally, we estimated the 95% confidence level of the one-sided test from the PDF. Figure A1 shows an example of the PDF of the 30–60DV wavelet power and the confidence level in the case of the lag-1 autocorrelation coefficient of 0.4. The confidence level can be estimated at 2.19. A number of simulations with autocorrelation coefficients ranging from 0 to 0.5 reveal that the confidence level is independent of the coefficient and the level is 2.19 for the 30–60DV and 2.24 for the 10–20DV. This study adopts these values to assess the significance test.
* Current affiliation: Department of Earth and Planetary Science, Graduate School of Science, University of Tokyo, Tokyo, Japan
+ Current affiliation: Department of Geography, Tokyo Metropolitan University, Tokyo, Japan
Corresponding author address: Satoru Yokoi, Department of Earth and Planetary Science, Graduate School of Science, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan. Email: firstname.lastname@example.org
Since the confidence level used for the ADR calculation is 95%, the ADR of red noise is expected to be 0.05.