On the withdrawal of the Indian summer monsoon

Transcription

1 Q. J. R. Meteorol. Soc. (2004), 130, pp doi: /qj On the withdrawal of the Indian summer monsoon By JOANNA SYROKA and RALF TOUMI Department of Physics, Imperial College, London, UK (Received 27 February 2003; revised 4 September 2003) SUMMARY The withdrawal of the Indian summer monsoon (ISM) is defined in terms of 850 mb daily winds. The withdrawal of the ISM is found to be more variable than the onset. Most of the interannual variability of the total seasonal rainfall is dominated by the variability of the retreat phase. The withdrawal of the ISM follows a period of enhanced convective activity over the Indian subcontinent and is associated with a dry phase of the intraseasonal oscillation. The intraseasonal break dynamics have relatively more hemispheric symmetry than during the main monsoon season. The monthly persistence of interannual anomalies, the correlations with El Niño and remote upper-level temperature signals of a Rossby type are all stronger during the withdrawal than in the summer months. These observations are found to be consistent with a mobile latent heat source migrating towards the west Pacific during the monsoon withdrawal. KEYWORDS: ENSO Intraseasonal variability Monsoon index 1. INTRODUCTION The onset, strength and variability of the summer monsoon have been extensively examined (e.g. Joseph et al. 1994; Webster et al. 1998). However, to date there has been no systematic investigation of the retreat of the monsoon system despite its key contribution to total rainfall variability (Rupa Kumar et al. 1992), the recently observed trends (Syroka and Toumi 2002), its importance for harvesting (Das 1987), water resource management (Mooley and Shukla 1987) and the spread of malaria (Bouma and van der Kaay 1996). The aim of this study is to define the withdrawal and to calculate a notional withdrawal date (WD). We then show that there are different dynamic features in the intraseasonal and interannual variations of the withdrawal compared to the main monsoon season. These differences can be broadly understood in terms of the more equatorward heat source during the retreat phase. 2. THE DATA We primarily use monthly and daily mean reanalysis data from the National Centres for Environmental Pediction/National Center for Atmospheric Research (NCEP/NCAR) (Kalnay et al. 1996) for the period The reanalysis has been used in many studies on the mean evolution and variability of the Asian summer monsoon on both intraseasonal and interannual timescales (e.g. Sperber et al. 2000) and it was found to be robust with respect to European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis and independent observational datasets for the period (Annamalai et al. 1999). The monthly and daily outgoing long-wave radiation (OLR) data used for the period is from the Advanced Very High Resolution Radiometer (AVHRR) instrument (Gruber and Krueger 1984). Precipitation data for this study are obtained from the monthly all-india rainfall (AIR) series (Parthasarathy et al. 1994) and the Climate Prediction Center Merged Analysis of Precipitation dataset, which is a merged product of satellite and rain-gauge data (Xie and Arkin 1997). It is important to note that the NCEP/NCAR reanalysis and the OLR and precipitation Corresponding author: Department of Physics, Imperial College, London SW7 2BZ, UK. r.toumi@ic.ac.uk c Royal Meteorological Society,

2 990 J. SYROKA and R. TOUMI datasets are essentially independent, since no information from them is used to infer the diabatic heating or divergent flow. El Niño Southern Oscillation (ENSO) variability is defined using standardized sea-surface-temperature (SST) anomalies for the NINO3 area (5 N 5 S, 150 W 90 E)(Kaplan et al. 1998). 3. DEFINING THE WITHDRAWAL (a) A daily 850 mb circulation index The withdrawal and gradual equatorward movement and deceleration of the lowlevel westerly flow associated with the retreat of the Indian summer monsoon (ISM) in late September/October is heralded by the seasonal cooling of the Asian continent (Ramage 1971). The domain of highest surface temperature and lowest pressure gradually returns to lower latitudes, with the re-establishment of a trough south of the equator. The upper-level anticyclone, which during June to September (JJAS) is centred over northern India and Pakistan as a result of intense sensible and latent heating in the region, weakens and begins its annual migration southeastward towards Indonesia. Good descriptions of the ISM withdrawal and the march of the monsoon season are found in Ramage (1971) and Rao (1976). Figure 1 shows the seasonal evolution of the precipitation field which acts as a critical heat source and driver of the monsoon circulation. During June, July and August there are two precipitation maxima, one in the western Ghats and the other in the north of the Bay of Bengal. During September and October, the precipitation maximum is only over the Bay of Bengal and retreats towards the south east. This migration and location of the heat source has important implications for the unique characteristics of the withdrawal. Although the mean characteristics of the ISM withdrawal are alluded to in many published texts, the working rules of the India Meteorological Department (IMD) are the only guidelines that exist for defining the WD of the ISM. The IMD classifies the recession of the monsoon with reference to the rather sharp changes in rainfall totals over the regional subdivisions (see Ramage (1971) for a description). However, it is known that, in terms of rainfall, the onset is better defined than the withdrawal, especially over southern India (e.g. Rao 1976; Das 1987) and, although many studies use precipitation amounts to characterize the strength and variability of the monsoon (e.g. Rupa Kumar et al. 1992), they are highly affected by location, particularly regional orography. Rao (1976) defines the withdrawal of the ISM as the southward displacement of the surface trough, the establishment of dry continental air and the development of anticyclonic flow over northern and central India. Motivated by this description, the WD of the ISM is defined in terms of an index, based on the characteristic low-level circulation associated with the ISM. A daily circulation index is defined as the difference in average 850 mb zonal winds between a southern region (5 N 15 N, 50 E 80 E) and a northern region (20 N 30 N, 60 E 90 E), as introduced by Syroka and Toumi (2002). It captures both variability of the position and intensity of the monsoon trough through a first-order approximation of the relative vorticity and the strength of the low-level Somali jet into the ISM domain. The regions are chosen to be dynamically consistent with the convective heating over the summer Indian subcontinent (Wang and Fan 1999) and correspond with the common mode of variability found to dominate the subseasonal and interannual variations of 850 mb June September (JJAS) circulation and rainfall over India by Sperber et al. (2000) (their Figs. 4(d) and 6(c)). They also delimit areas of recent trends observed in the low-level circulation over the south Asian domain during

3 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 991 Figure 1. Climate Prediction Center Merged Analysis of Precipitation climatology ( ) for (a) May, (b) June August, (c) September and (d) October in mm day 1. Regions with precipitation greater than 10 mm day 1 are shaded. the withdrawal of the ISM (Syroka and Toumi 2002). A similar index was also recently defined by Wang et al. (2001) to characterize ISM variability. The mean daily circulation index, as defined above, exhibits a strong annual cycle (Fig. 2(a)) with a maximum at the height of the ISM in July and August and a minimum in the boreal winter months. The index changes sign, reflecting both the changing intensity of the low-level westerly monsoon flow and the vorticity associated with the monsoon trough and synoptic activity. There is a clear asymmetry to the annual cycle with a steep increase in the index at the onset of the monsoon and a slower decrease during the retreat. A difference in the variability of the index at the beginning and end of the monsoon season is also observed, as seen by the standard deviation for each day in Fig. 2(a). To further substantiate this choice of index, the interannual variations in monthly mean AIR and monthly circulation index are correlated. They correlate very well for all months from May to October (Table 1), and correlations are particularly large towards the end of the ISM season. The circulation index is therefore both a physically sensible and a practical tool to study the withdrawal of the monsoon.

4 992 J. SYROKA and R. TOUMI Figure 2. (a) Mean annual cycle of daily 850 mb circulation index (see text for definition) (solid line) with ±1 standard deviation for each day (dashed lines). (b) Withdrawal dates (WD) for and available India Meteorological Department dates (squares). TABLE 1. S IMULTANEOUS CORRELATION COEFFICIENT (CC) OF MEAN MONTHLY 850 MB CIRCULATION INDEX AND OBSERVED MONTHLY ALL-INDIA RAINFALL FOR MAY TO OCTOBER, Month May Jun Jul Aug Sep Oct CC All CCs are significant at the 99% confidence level. (b) Defining the withdrawal date The daily circulation index can be used to define both dates of onset and retreat of the ISM from the region. More explicitly, the change of sign of the index in the boreal autumn (Fig. 2) indicates the displacement of the monsoon air by a continental air mass and development of anticyclonic flow over north and central India. Rao (1976) remarks that this change in circulation should be rightly regarded as the end of the southwest, or summer, monsoon over India and it is not appropriate to identify the much later decrease in rainfall in southern India as the withdrawal of the ISM. The southern Indian states receive increased rainfall from October onwards associated with the winter monsoon over the Indian peninsula. Thus, the circulation index can be used to define the withdrawal of the ISM from northern and central India (approximately 15 N), classified by Rao (1976) as the end of the southwest monsoon. The daily circulation index exhibits substantial noise, so a centred 7-day running average is taken. The date of withdrawal of the monsoon is defined as the first of seven consecutive days for which the index becomes negative. Similarly, the date of onset is defined as the first day of seven consecutive days of positive index. The 7-day period was found to be the smallest time interval which smoothed synoptic noise sufficiently to define the dates more easily. The WDs are shown in Fig. 2(b) and listed in Table 2.

5 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 993 TABLE 2. W ITHDRAWAL DATES OF THE INDIAN SUMMER MONSOON (ISM), AS DEFINED BY THE DAILY 850 MB CIRCULATION INDEX, THE INDIA METEOROLOGICAL DEPARTMENT S DATES FOR THE END OF THE ISM SEASON, PUB- LISHED IN Mausam, ARE SHOWN IN BRACKETS WHEN AVAILABLE. Year Date Year Date Nov Oct Nov Oct Oct Oct Oct (11) Oct Nov (19) Oct Oct (3) Oct Oct (25) Oct Oct (21) Oct Oct (24) Oct Oct (12) Oct Oct (13) Oct Sep (17) Oct Oct (18) Oct Nov Oct Oct Oct Nov Sep Oct Oct Nov Oct Oct (8) Oct Oct Nov Oct Nov Oct The mean date of withdrawal for is 19 October with a standard deviation of 14 days. The earliest retreat date is 23 September 1994 and the latest 23 November Closer inspection of this latest withdrawal indicates that the daily circulation index became negative on 5 November for five days before reverting back to a positive phase, only to become negative again for good on 23 November, hence this was chosen as the withdrawal date given the criteria above. This ambiguity highlights that the dates of withdrawal of the ISM cannot be defined with mathematical precision and one cannot overlook that the cessation of the monsoon is a gradual process. However, other years in the 43-year record were less difficult to define than The IMD define the WD at 15 N as the end of the ISM season (Dey et al. 1985). These dates are a good comparison for the index-derived WD, however only 11 such dates, published in Mausam, are available. These are marked on Fig. 2(b) and given in Table 2. All fall within October. For the years and 1997, the average date of withdrawal is 16 October with a standard deviation of 7 days. The correlation coefficient of 0.47 between index-derived and IMD WDs is not significant at the 95% confidence level, but there are only 11 years for comparison. Although undoubtedly there are other possible ways to define the end of the ISM, the method proposed above is robust and consistent with respect to the convection and circulation fields which intrinsically characterize the summer monsoon system over the Indian domain. Thus the indexderived dates for are used to define the WDs of the ISM for the 43-year period. It is instructive to note that, from the circulation index, the mean date of onset is 27 May with a standard deviation of only 10 days. The withdrawal date is thus about 40% more variable than the onset. An explanation for this can be found in the difference

6 994 J. SYROKA and R. TOUMI between the primary heat sources during the two periods. The onset of the monsoon is largely driven by the solar insolation of the Eurasian landmass and sensible heating of the Tibetan plateau in particular. This relatively regular heating can be contrasted with the inherently more variable and spatially mobile latent heat source over the Bay of Bengal during the withdrawal phase. It is the difference in the variability of the heat sources which most likely explains the difference in the onset and withdrawal date variability. The impact of ENSO in this context may also affect the ISM system differently throughout the monsoon season. This issue is addressed in section 5. (c) Intraseasonal variations To gain a deeper insight into the withdrawal of the ISM, it is necessary to understand the evolution of the convection and circulation anomalies before, during and after the withdrawal. For this purpose, Figs. 3 and 4 show the sequence of composites for daily OLR and 850 mb vector wind anomalies, to examine the propagation characteristics associated with the end of the ISM. The composites are constructed in two steps. Firstly, a daily time series of OLR and 850 mb vector wind anomalies is constructed by subtracting the daily climatology for from each day at each grid point. Secondly, composites are constructed from these records by fixing the day of ISM withdrawal, as defined by WD, as the reference time (i.e. day 0). The process of compositing serves as an efficient method for filtering out possible noise that can contaminate the regional-scale features. The sequence of smaller-scale transient features associated with the final days of the ISM may be lost in straightforward averages for calendar dates. Significance levels of the composites are calculated using a one-sample t-test, to examine the null hypothesis that the amplitude of the anomalies at each time step has been drawn from a population characterized by a zero mean. If the critical value for the 95% confidence interval is exceeded, we can reject the null hypothesis above, implying that mean departure for that particular composite day should not be considered as an ordinary sampling fluctuation. For the 850 mb wind anomalies, if either or both of the zonal and meridional wind components at each grid point exceed the critical t-value, the wind field at that grid point is considered to be significantly different from zero. Regions which exceed the 95% confidence level are highlighted in Figs. 3 and 4. It is acknowledged that there may be limitations in using the above method for testing statistical significance, particularly for wind fields and because of the small sample size used and the assumption of independence between components. However, the significance implied is a good indicator for aiding the interpretation of the composite fields and should be treated in a qualitative rather than quantitative sense. Furthermore, only data beginning in 1979 are used for consistency between the AVHRR OLR and the NCEP/NCAR reanalysis datasets (Kalnay et al. 1996). (i) Outgoing long-wave radiation. Prior to the withdrawal of the ISM, it is clear from Fig. 3 that there is enhanced convection over much of the Indian subcontinent, the Bay of Bengal, the Indochina peninsula and the western Pacific and suppressed convection over the central equatorial Indian Ocean. It can been seen that the suppressed convection moves steadily poleward, sweeping over the Indian peninsula, squeezing the convective activity progressively northeast. On the day of the notional WD, anomalous convective activity is primarily focused on Myanmar and southern China as convection over the Indian subcontinent, particularly the north, has been suppressed.

7 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 995 Figure 3. Composites of daily outgoing long-wave radiation (OLR) anomalies from 12 to +3 days (in steps of 3 days) from withdrawal day WD (defined as day 0) for Contours are at 10 W m 2 intervals, with negative (positive) OLR anomalies denoted by dashed (solid) lines. Shading denotes the 95% confidence level by a t-test statistic.

8 996 J. SYROKA and R. TOUMI There is interesting daily structure in this progression. From day 9, we notice the region of suppressed convection gradually begins to grow and spread northward over the Arabian Sea and further eastward over the equatorial Indian Ocean. From day 6, we notice the suppressed convection begins to intensify further and spread northward to envelop the Indian peninsula and southward into the southern Indian Ocean. Simultaneously, a region of positive OLR also begins to develop over Indonesia and the eastern South China Sea. From day 3 the suppressed anomaly continues to intensify and spread northward. The western part of the anomaly over the Arabian Sea and Indian subcontinent spreads northward more rapidly than that over the eastern Indian Ocean and Bay of Bengal, giving the impression of a northeast sweep across the Indian domain. We note the Bay of Bengal is an area of intense heating and convection during the withdrawal of the ISM, thus the poleward progression of convectively stable anomalies may favour regions of the Indian domain where heating is not as intense, i.e. to the west of the Bay of Bengal. A combination of a more interactive land surface, friction or weaker convection over land may favour faster propagation over land than over the oceans. The positive OLR anomaly over the Indian domain steadily merges with that over Indonesia as the convectively stable anomaly appears to spread rapidly eastward from the equatorial Indian Ocean (day 3). By day 0, convectively stable anomalies dominate the entire Indian subcontinent, Arabian Sea, Bay of Bengal, northern Indonesia and the South China Sea. Convective activity is restricted to the northern Indochina peninsula and western Pacific. A region of enhanced convection is also evident just south of the equator over the Indian Ocean. Indeed, it appears as if the convection change over the equatorial Indian Ocean behaves in the opposite sense to that over the Indian subcontinent. The influence of both these regions has long been recognized as the basis of low-frequency oscillations in the monsoon system (e.g. Sperber et al. 2000) and the competing influence of the oceanic and continental positions of the Intertropical Convergence Zone on the active/break cycle of the ISM. In accord with the notion of end of the ISM and our definition, the convectively stable anomaly over the Indian subcontinent persists past the withdrawal date. (ii) 850 mb winds. Consistent features are noted in the evolution of the composited wind anomalies at 850 mb (Fig. 4). Day 12 shows a strong cyclonic anomaly over central India and the Indochina peninsula, in agreement with the regions of enhanced convection in Fig. 3. The cyclonic anomaly over India remains, but is seen to propagate northeastward with time, in accord with the observed squeezing of negative OLR anomalies. By day 6, enhanced cyclonic circulation is seen over northeast India, northern Bay of Bengal and the Himalayan foothills, with strengthened westerlies to the south over the Indian subcontinent. It is not yet clear that the withdrawal of the ISM has begun. By day 3 however, an anticyclone is beginning to form over the Indian peninsula at 70 E, in agreement with the suppressed convection over the region. Concomitantly, in the southern hemisphere, the anticyclonic anomaly is manifest just south of the equator at 90 E. These two anomalous anticyclones are clearly evident by day 0, when the anticyclone in the northern hemisphere has moved north and is centred over the Indian subcontinent. In the southern hemisphere, an anticyclonic anomaly has moved slightly west and has intensified, now evident just south of the equator at 80 E. By day +3, the anticyclone over India has intensified further and anomalous easterly/north-easterly flow is established over the subcontinent. The anticyclone in the southern hemisphere has also moved further west. In particular, it is interesting to note this anticyclone has also intensified, and by day +3 itswesternflankeffectively reverses the direction of the anomalous cross-equatorial flow off the East African coast,

9 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 997 Figure 4. Composites of daily 850 mb vector wind anomalies (m s 1 ) from 12 to +3 days (in steps of 3 days) from withdrawal date (defined as day 0) for Shaded regions are considered to be significant at the 95% confidence level. The reference arrow in the top right hand corner of each box represents 5 m s 1.

10 998 J. SYROKA and R. TOUMI inhibiting the southerly flow of the Somali jet into the Indian domain and thus cutting off this moisture flux into the monsoon region. Anticyclonic circulation and/or enhanced easterlies persist over the Indian subcontinent for the remaining days, as does the anomalous northerly cross-equatorial flow, although the statistical significance of the anomalous wind field falls below the 95% confidence level by day +9 (not shown). Thus it is clear that the arrival of a dry anomaly into the domain from the equatorial Indian Ocean is associated with anticylonic circulation over the Indian subcontinent, signalling the end of the ISM as defined by the circulation index. Furthermore, the weakening or even termination of cross-equatorial flow into the Indian domain further amplifies the final withdrawal phase. Thus, the evolution of the composited wind anomalies from the NCEP/NCAR reanalysis data is generally dynamically consistent with that of the observed OLR anomalies. (d) Evidence of Rossby wave propagation The anomalous anticyclones on each side of the equator have a striking resemblance to a Rossby wave (Matsuno 1966; Gill 1980) and can be interpreted as a response induced by the convectively stable anomaly just north of the equator in the eastern Indian Ocean observed on day 3 (Fig. 4). However, the Rossby wave observed is not symmetric about the equator. The northern anticyclone is observed to move northward (from day 3 to day 0) from the equatorial Indian Ocean to the Indian subcontinent, initiating the end of the ISM. The southern anticyclone moves only westward and, although always weaker than its northern counterpart, begins to intensify once the anticyclonic circulation is established over the Indian subcontinent. The description outlined above of the mean withdrawal centred on the WD has very similar characteristics to those observed for the northward propagation of dry anomalies associated with the intraseasonal oscillation (ISO) from the equatorial Indian Ocean (see the review in Annamalai and Slingo 2001). A Fourier harmonic analysis of daily OLR variation in the Indian domain throughout the ISM season confirms that intraseasonal variability is governed by oscillations with periods between 20 and 73 days, and the withdrawal of the ISM is associated with the arrival of a dry phase of this ISO into the region. The analysis is not repeated here to avoid repetition of previous work (e.g. Wang and Xie 1997), however many studies have used similar timescales to isolate ISO activity (e.g. Lawrence and Webster 2001). In particular, the mean features of Figs. 3 and 4 are similar to the break dynamics described by Krishnan et al. (2000). Although their study specifically did not focus on the withdrawal, the subsequent northwest propagation of the anticyclonic anomaly excited by the convectively stable anomalies in the eastern Indian Ocean is very similar to those observed in Figs. 3 and 4 (compare with Figs. 4 and 5 of Krishnan et al. 2000), resembling an asymmetric Rossby wave response about the equator. This suggests Rossby wave dynamics may contribute not only to the poleward propagation of high-pressure anomalies during monsoon breaks, but also during the final days of the ISM. However, there are some key differences between the evolution of the withdrawal depicted above and the monsoon breaks described by Krishnan et al. (2000). In particular, although the southern hemisphere anticyclone is always weaker than its northern counterpart, it does appear to be stronger during the withdrawal than during a normal monsoon break. This is particularly important as the anticyclonic circulation is instrumental in inhibiting the southerly cross-equatorial flow of the Somali jet into the Indian domain. The asymmetry between the Rossby wave emanation in the northern and southern hemispheres associated with the ISO has been noted before in observational (e.g. Kemball- Cook and Wang 2001; Lawrence and Webster 2002) and modelling (e.g. Wang and

11 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 999 Xie 1997) studies. Wang and Xie (1997) partly attribute the amplification of the northern Rossby cell to the mean vertical wind shear of the monsoon flow, confining the wave to the lower troposphere, while the westerly shear (upper-level westerlies) in the southern hemisphere does not favour the amplification of the southern Rossby cell in this way. Through modelling experiments, Krishnan et al. (2000) also show that the basic monsoon westerly flow is a primary mechanism for the northwest propagation of the anticyclonic anomaly in the northern hemisphere. We note that because of the more southerlyposition of the primary heatsource (Fig. 1), the backgroundmonsoonflow and vertical wind shear are weaker in the northern hemisphere during the withdrawal than in the main season. This would favour a more symmetrical Rossby wave emanation about the equator. The seasonal cooling may also facilitate more symmetrical equatorial Rossby wave emanation. Lawrence and Webster (2002) and Annamalai and Slingo (2001) suggest that the underlying basic state of the ISM modifies the equatorial eastward-moving Madden Julian Oscillation, giving rise to the observed ISO. In agreement with Kemball-Cook and Wang (2001), they suggest warm ocean surface and moist boundary layer extending north to the ISM domain cause the whole of the northern Indian Ocean to become conditionally unstable, with preferred large-scale surface convergence into the northern Rossby cell. However, during the withdrawal, the available low-level moisture and temperature gradients are smaller than during the main monsoon season, discouraging such bias. Indeed, several studies have noted a seasonal cycle in ISO strength over the Indian domain, observing a weaker ISO amplitude during the later part of the monsoon season (Lawrence and Webster 2001). The background state plays an important role in withdrawal dynamics depicted in Figs. 3 and 4 and appears to be fundamental in controlling the timing of the withdrawal of the ISM and hence its interannual variability. 4. INTERANNUAL VARIABILITY Rupa Kumar et al. (1992) noted that the deficit or excess of seasonal rainfall is realized frequently in the later half of the monsoon season. The retreat phase of the monsoon system during September and October has the highest coefficient of variation when compared to the onset (May June) and the mature phase (July August) of the monsoon cycle (see Table 3). Hence, although September and October contribute only about a quarter to the total seasonal rainfall, these months contribute most to the total rainfall variability. Thus it appears that understanding the retreat phase and the timing of the ISM withdrawal is an important aspect of understanding the interannual variability of the entire Indian summer monsoon season. Table 4 shows the correlation coefficients between monthly AIR totals from May to October. In general, these correlations are weak (Rupa Kumar et al. 1992), however there is greater persistence towards the later part of the monsoon season, with the correlation coefficient between September and October monthly AIR being 0.48 (significant at the 99% confidence level) for There is also statistically significant positive correlation between August and September AIR totals, whereas May and June AIRs, for example, show no such significance. The correlation coefficients between the 43 WD values and the monthly AIR values are also shown in Table 4. It is clear that variations in WD are most strongly correlated with rainfall totals for September and October, although there is a weak positive relationship between WD and August AIR. When September and October rainfall totals for the Indian regional subdivisions are considered (not shown), the interannual variations in WD and rainfall totals correlate highly at the 95% confidence level in all regions except the Indian peninsula. This implies that late-withdrawal years

12 1000 J. SYROKA and R. TOUMI TABLE 3. MEAN, STANDARD DEVIATION (SD) AND COEFFICIENT OF VARIATION (CV) FOR TWO-MONTHLY ALL-INDIA RAINFALL (mm), May Jun Jul Aug Sept Oct Mean SD CV CV = SD/mean. TABLE 4. C ORRELATION COEFFICIENTS (CCS) FOR ALL-INDIA RAINFALL (AIR) FOR MONTHS MAY TO OCTOBER AND JUNE SEPTEMBER (JJAS) TOTALS, AND WITH THE MONSOON WITHDRAWAL DATE (WD), AIR May Jun Jul Aug Sep Oct JJAS WD May * Jun Jul AIR Aug Sep Oct * JJAS 1 CCs significant at the 95% and 99% confidence levels are denoted by * and respectively. are associated with enhanced rainfall over most of the Indian subcontinent, but rainfall over the Indian peninsula varies somewhat independently late in the ISM season, as also noted by Patnaik et al. (1977). Table 4 also indicates that there is a weak positive relationship between the strength of the main monsoon and the withdrawal of the ISM, with a correlation coefficient of 0.25 between WD and JJAS AIR for Following section 3 and the timescale for the mean withdrawal depicted in Figs. 3 and 4, the correlation coefficients above imply September and October are both important in the consideration of the retreat phase of the ISM. However, Table 4 indicates that any variability in the withdrawal date of the ISM will be most evident in October, the traditional transition month heralding the end of the monsoon season (Ramage 1971; Das 1987). The origin of the stronger monthly persistence and interannual variability during the retreat phase can be understood by considering the different position of the heat source driving the ISM circulation. The onset and early stages of the ISM develop in response to sensible heating from the surface of the Tibetan plateau as a consequence of seasonal solar insolation (Yanai et al. 1992). As the monsoon develops, the diabatic heating associated with convection plays an increasingly important role in maintaining the monsoon circulation (Slingo and Annamalai 2000). Figure 1 shows it is during the established phase of the ISM that heating strength reaches its maximum over the Indian subcontinent, with rainfall primarily over land. By contrast, during the withdrawal phase, the main driver of the ISM circulation is the mobile latent heat source of the monsoon itself. The Tibetan plateau reverses from being a geographically fixed heat source to a heat sink in September and October (Yanai et al. 1992). The weakening monsoon flow accompanies the southward progression of the convection during this time. The maximum heating regions are almost entirely over the ocean

13 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 1001 (Fig. 1) and this preferred location of the heat source during September October may explain the greater persistence observed in Table 4. The slower response times of the ocean compared to the land may allow for stronger coupling of the monsoon system to oceanic conditions. Furthermore, as we will show below, because the heat source is closer to the slowly moving ENSO action centres in the west Pacific, monthly persistence can also be expected for this reason. It is therefore possible that it is the larger sensitivity to external boundary forcing, as well as the internal monsoon variability, which makes the withdrawal relatively more variable than the onset and main monsoon season. Nevertheless, the relatively strong persistence of the withdrawal months implies enhanced potential for predictability toward the end of the monsoon season. 5. INFLUENCE OF ENSO ON THE WITHDRAWAL OF THE MONSOON (a) Correlation analysis Interannual variability of the ISM system has been linked to ENSO boundary forcing (see Webster et al for a review). Figure 5 shows the NINO3 SST anomalies, averaged using a rolling three-month window, correlated with WD for The WD values are held fixed and the NINO3 SST anomalies are lagged to observe the relationship between pre- and post-monsoon SST anomalies and the timing of the withdrawal of the ISM. The NINO3 SST anomalies well before the monsoon season (up to seven months before) have a weak positive correlation with WD, and hence the ISM withdrawal. Starting with the SST anomalies leading WD by seven months, the correlations rapidly become negative with the contemporaneous correlation of However, the strongest negative correlations occur after the monsoon season, particularly the three-month periods centred on November, December and January. A lead/lag correlation analysis by Kirtman and Shukla (2000) between all-india JJAS rainfall anomalies and NINO3 SST anomalies yields very similar results. Ailikun and Yasunari (2001) also found that the mid-late summer Asian monsoon is related to the anomalous state of ENSO in the following rather than the previous winter. Table 5 shows the simultaneous correlation coefficients between monthly AIR totals during the ISM season and NINO3 SST anomalies. The strongest simultaneous correlation occurs in September. The onset rainfall (May June) exhibits a weak relationship with simultaneous NINO3 SST anomalies. Table 5 also shows the correlation coefficients between monthly AIR totals and NINO3 SST anomalies before (March May) and after (November January) each monsoon season (i.e. the AIR totals are held fixed when computing the correlation coefficients). September and October show a strong negative correlation with post-monsoon SSTs, but little correlation with pre-monsoon SST anomalies. However, June and July AIR show a stronger, albeit weak, inverse relationship with pre-monsoon NINO3 SST anomalies, qualitatively agreeing with the findings of Ju and Slingo (1995) and Ailikun and Yasunari (2001). JJAS total seasonal rainfall also exhibits a strong inverse relationship with simultaneous and post-monsoon SST anomalies, as noted by Kirtman and Shukla (2000). However, from Table 5 it is clear that the ISM exhibits stronger correlations with ENSO during the retreat rather than any other part of the monsoon season. While correlations do not necessarily imply a cause and effect relationship, they have led to the suggestion that ISM anomalies can have a significant effect on ENSO variability (e.g. Kirtman and Shukla 2000; Goswami and Jayavelu 2001). In this context, it appears that late ISM withdrawals are associated with negative NINO3 SST anomalies and La Niña conditions in the following winter/spring. This does not necessarily mean a late (early) ISM withdrawal causes a La Niña (El Niño) to develop in the tropical

14 1002 J. SYROKA and R. TOUMI Figure 5. Lead/lag correlation coefficients for withdrawal date and NINO3 sea-surface-temperature anomalies (averaged over a 3-month rolling period) for The 95% (dashed) and 99% (dotted) confidence levels are shown. TABLE 5. CORRELATION COEFFICIENTS (CCS) OF ALL-INDIA RAINFALL (AIR) FOR MONTHS MAY TO OCTOBER AND JUNE SEPTEMBER (JJAS) TOTALS AGAINST SIMULTANEOUS,MARCH MAY AND NOVEMBER JANUARY NINO3 AREA SEA-SURFACE-TEMPERATURE (SST) ANOMALIES OVER THE PERIOD NINO3 SST May Jun Jul Aug Sep Oct JJAS Simultaneous * * 0.47 March May November January * 0.43 CCs significant at the 95% and 99% confidence levels are denoted by * and respectively. AIR Pacific. Since it is well known that the mature phase of ENSO can be locked to the end of the calendar year (e.g. Tziperman et al. 1998), an alternative interpretation may be that there is a preferred occurrence of a late (early) ISM withdrawal during a developing phase of La Niña (El Niño). As discussed in the previous section, a plausible explanation for the relatively closer association of ENSO with the withdrawal phase than the main monsoon season lies in the timing and position of the heat source. The migration of the precipitation maximum towards the ENSO action centre in the west Pacific and the seasonality of ENSO itself ensures both a temporally and spatially stronger monsoon ENSO relationship. The coupling will be an important source of monthly persistence, as noted earlier. The sensitivity to ENSO external forcing suggests enhanced relative predictability during the withdrawal.

15 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 1003 (b) Upper tropospheric temperature It is well known that, during the established phase (July August), the monsoon is very much a self-sustaining system, driven by the local distribution of latent heat release within the monsoon itself (e.g. Slingo and Annamalai 2000). The effects of ENSO operate within this framework. Previous studies have highlighted the importance of tropospheric temperatures over the Tibetan plateau in relation to the onset (Li and Yanai 1996) and variability of the ISM (Sperber et al. 2000; Liu and Yanai 2001). Li and Yanai (1996) found that the onset of the Asian summer monsoon is concurrent with the reversal of the meridional temperature gradient south of the Tibetan plateau. The reversal is the result of large temperature increases in May and June centred on the plateau with no appreciable change over the Indian Ocean. Figure 6 shows the correlation coefficients between October monthly mean mb temperature and WD. The pattern observed shows large-scale highly significant positive correlations over the Indian subcontinent, southern Eurasia, Arabia and northeast Africa, as well as over the western coast of Australia and southern Indian Ocean, and significant negative correlations in both hemispheres over the Pacific Ocean. The mean temperature between the 200 and 500 mb levels is proportional to the thickness of the layer they define. Late withdrawals are associated with greater geopotential height over the Indian subcontinent and southern Indian Ocean, and lower geopotential heights in the cooler equatorial Indian Ocean and Pacific regions. This geopotential height gradient will drive a stronger monsoon circulation. Figure 6 gives equivalent information to the stream function in the upper troposphere (not shown for brevity). Thus the pattern in Fig. 6 can be interpreted as a forced Rossby wave response to a near-equatorial heating anomaly. Such an anomaly is expected to generate upper-level anticyclones upstream and twin cyclones downstream of the anomalous convection, symmetric about the equator (e.g. Matsuno 1966; Gill 1980; Sardeshmukh and Hoskins 1988; Highwood and Hoskins 1998). We note that such a symmetric response in the upper troposphere is only evident during the withdrawal and not the main monsoon season. This symmetry may be due to the heating anomaly being closer to the equator during the retreat phase. It has been previously observed that the highest correlation between JJAS AIR and upper tropospheric temperature over the Tibetan plateau occurs during September and October (Liu and Yanai 2001). However, given the foregoing discussion, it is clear that ENSO may be partly responsible for the quadrupole temperature pattern observed in Fig. 6. In particular, La Niña conditions associated with enhanced convection over the Malaysia/Indonesia area may be responsible for such an atmospheric signal. To quantify this possible relationship, time series of October mean mb temperature at 30 N and also at 5 N, averaged over longitudes 60 E 105 E (the longitudinal extent of the Tibetan plateau), were constructed. The difference between the temperatures at 30 Nand5 N is a measure of the mean meridional upper-level temperature gradient, or thermal contrast, for the October of each year. The interannual variations in October upper tropospheric temperatures calculated at 5 N, 30 N and the gradient were correlated with the October NINO3 SST anomalies from 1958 to The simultaneous correlation coefficients are 0.75 (significant at the 99% confidence level), 0.20 (not significant at the 95% level) and 0.78 (significant at the 99% level) respectively. The September temperature gradient and September NINO3 SST anomalies also correlate well ( 0.58, significantat the 99% level). This implies La Niña (El Niño) conditions may enhance (reduce) the upper-level temperature gradient over the Indian domain, as part of a quasi-stationary Rossby wave response (Matsuno 1966; Gill 1980) to heating anomalies over Malaysia and Indonesia. We speculate that this enhances (reduces) the

16 1004 J. SYROKA and R. TOUMI Figure 6. Correlation coefficients of withdrawal date against October mean temperature averaged over the 500, 400, 300, 250 and 200 mb pressure levels taken from the NCEP/NCAR reanalysis for Negative correlations are shown by dashed contours, positive correlations by solid contours. Regions of correlation significant at the 95% and 99% confidence levels are shaded in dark and light grey, respectively. upper-level thermal contrast and thus the pressure gradient (Fig. 6) over the equatorial ocean and the Asian landmass, facilitating a late (early) ISM withdrawal. Kiladis et al. (2001) observed similar upper-level heating patterns associated with ENSO. The interannual variations in the mean October tropospheric temperature at 5 N, 30 N and the temperature gradient (30 N minus 5 N) (over 60 E-105 E) have correlation coefficients with WD of 0.70 (significant at the 99% level), 0.03 (not significant at the 95% level) and 0.66 (significant at the 99% level), respectively. The September temperature gradient and WD have a correlation coefficient of 0.40 (significant at the 99% level). Late withdrawals of the ISM are associated with years of enhanced thermal contrast in the upper troposphere in the Indian domain. In particular, it appears that it is the temperature in the upper troposphere over the Asian landmass, rather than over the equatorial ocean, that strengthens this correlation. Apart from the possible remote control of the upper troposphere during the withdrawal, local heating could also be important. Liu and Yanai (2001) found strong correlations between September upper-tropospheric temperatures and October rainfall totals. They suggested enhanced late-season rainfall over the Himalayan foothills may warm the upper troposphere locally through the release of latent heat. This heating anomaly persists into October, facilitating the maintenance of the upper-level thermal contrast and thus delaying the retreat of the monsoon. Traditionally, an inverse relationship has been observed between winter and spring Himalayan snow cover and the strength of the subsequent monsoon (e.g. Blanford 1884). However, Dey et al. (1985) reported that a larger than average summer Himalayan snow cover area is likely to be followed by

17 WITHDRAWAL OF THE INDIAN SUMMER MONSOON 1005 a later withdrawal. In the context of this study, enhanced snowfall is associated with enhanced latent heating. Yanai et al. (1992) also noted a substantial contribution from latent heat release on the eastern plateau after the onset of the summer rains. Sardeshmukh and Hoskins (1988) previously noted the importance of recognizing the correct nature of the tropical Rossby wave source. During the withdrawal of the ISM, the convective maximum driving the circulation in the Indian region is located closer to the equator than during the main monsoon season. Thus, anomalous convection associated with the monsoon itself may also drive a similar upper-level response, enhancing the anomalous heating associated solely with ENSO. Adiabatic warming due to intensified large-scale subsidence to the west of the Indian subcontinent over Arabia and Afghanistan may be the result of a stationary Rossby wave response to the enhanced ISM convection (Rodwell and Hoskins 1996). There is evidence of significant positive correlations between temperature and WD over these regions in Fig. 6. Strong subsidence warming to the west of the major rain area was also shown by He et al. (1987) and Yanai et al. (1992) during the onset of the monsoon in When ENSO years (defined as those with greater than ±0.5 standard deviations from the October mean NINO3 SST for ) are removed, the interannual variations between WD and the meridional temperature gradient between 30 Nand5 N are still strongly correlated. The correlation coefficientis 0.65for the 16years remaining after the ENSO years are removed from the dataset. This is significant at the 99% confidence level. The correlation pattern of WD with upper tropospheric temperature for non-enso years does not exhibit the clear quadrupole observed in Fig. 6, but does show a region of strong positive correlation over the Indian subcontinent and Afghanistan (not shown). This implies that the upper-level meridional temperature gradient still accounts for approximately 43% of WD variability which can be attributed to the impact of the internal dynamics and self-sustaining nature of the ISM during its retreat phase. This may indicate the limits of the predictability of the withdrawal phase. 6. CONCLUSIONS We have defined the withdrawal in terms of a physically based 850 mb wind index, which captures the larger-scale monsoon dynamics and correlates well with rainfall over the Indian subcontinent. Several features mark out the withdrawal as different from other parts of the monsoon season: the withdrawal is slower and has a larger interannual variation than the onset, the intraseasonal break dynamics have more hemispheric symmetry, there is stronger monthly persistence of interannual anomalies, the correlation with ENSO is stronger, a large part of the interannual variability of the total seasonal rainfall is dominated by the variability of the withdrawal, and remote upper-level temperature signals of a Rossby type are stronger. We suggest that a plausible explanation for all these features is the different position and nature of the primary heat source. During the onset, the primary heating source is sensible heating from the Asian landmass and the Tibetan plateau in particular. This fixed geographical location, driven by the regular solar insolation, ensures relatively little interannual variation of the onset. During the main monsoon season, the latent heat sources are also geographically centred on the western Ghats and in the north of the Bay of Bengal. However, during the withdrawal, the heating maximum migrates south over the Bay of Bengal. This mobile latent heat source coupled to the ocean ensures

18 1006 J. SYROKA and R. TOUMI a more gradual demise of the monsoon with significant monthly persistence. However, both the position and strength of the heating can be expected to have relatively large interannual variations dominating the variability of the monsoon circulation and hence rainfall totals over the entire domain. Its more southerly location weakens the monsoon flow and allows more hemispheric symmetry and stronger Rossby wave signals, both on intraseasonal and interannual time scales. The closer geographic proximity to the ENSO action centre in the west Pacific also ensures that the relationship of the monsoon with developing ENSO phases is more prominent than at any other time. The upper-level temperature response associated with WD variability is particularly interesting, and warrants further investigation. We conclude that an improved understanding of the withdrawal of the monsoon is central to understanding monsoon variability and predictability. ACKNOWLEDGEMENTS JS was supported by the Natural Environment Research Council and Accord Energy Ltd. REFERENCES Ailikun, B. and Yasunari, T ENSO and Asian summer monsoon: Persistence and transitivity in the seasonal march. J. Meteorol. Soc. Jpn., 79, Annamalai, H. and Slingo, J. M Active/break cycles: Diagnosis of the intra-seasonal variability of the Asian summer monsoon. Clim. Dyn., 18, Annamalai, H., Slingo, J. M., Sperber, K. R. and Hodges, K The mean evolution and variability of the Asian summer monsoon: Comparison of ECMWF and NCEP-NCAR reanalyses Mon. Weather. Rev., 127, Blanford, H. F On the connection of the Himalayan snowfall with dry winds and Bouma, M. J. and van der Kaay, H. J. the seasons of drought in India. Proc. R. Soc. London, 37, The El Niño Southern Oscillation and the historic malaria epidemics on the Indian subcontinent and Sri Lanka: An early warning system for future epidemics? Trop. Med. Int. Health, 1, Das, P. K Short- and long-range monsoon prediction in India. Pp in Monsoons. Eds. J. S. Fein and P. L. Stephens. John Wiley and Sons, New York Dey, B., Kathuria, S. N. and Kumar, O. B Himalayan summer snow cover and withdrawal of the Indian summer monsoon. J. Clim. Appl. Meteorol., 24, Gill, A. E Some simple solutions for heat-induced tropical circulation. Q. J. R. Meteorol. Soc., 106, Goswami, B. N. and Jayavelu, V On possible impact of the Indian summer monsoon on the ENSO. Geophys. Res. Lett., 28, Gruber, A. and Krueger, A. F The status of the NOAA outgoing long-wave radiation dataset. Bull. Am. Meteorol. Soc., 65, He, H., McGinnis, J. W., Song, Z. and Yanai, M Onset of the Asian summer monsoon in 1979 and the effect of the Tibetan Plateau. Mon. Weather Rev., 115, Highwood, E. J. and Hoskins, B. J The tropical tropopause. Q. J. R. Meteorol. Soc., 124, Joseph, P. V., Eischeid, J. K. and Pyle, R. J Interannual variability of the onset of the Indian summer monsoon and its association with atmospheric features, El Niño, and seas surface temperature anomalies. J. Climate, 7, Ju, J. and Slingo, J. M The Asian summer monsoon and ENSO. Q. J. R. Meteorol. Soc., 121,