a. Introduction

Observationally, monsoon depressions form at the head of the Bay of Bengal and move westward. They are associated with heavy rainfall that can reach a rate of 300 mm in 24 h. The composite structure of a monsoon depression has been explored by Godbole (1977). He observed several general characteristics, namely that a cyclonic circulation is well defined in the lower and middle troposphere, which is more vigorous at 800 hPa than at the surface. The relative vorticity has a maximum value of the order of 5 × 10⁻⁵ s⁻¹ at the surface, strengthening at 800 hPa to values larger than 12 × 10⁻⁵ s⁻¹ (Sikka 1977). The southwest sector of the depression is the location of the strongest activity where the maximum of the precipitation rate (100 mm day⁻¹) and of the 800-hPa moisture convergence are observed. Saha et al. (1981) studied the distribution of the daily changes of sea level pressure rather than the pressure itself, finding that most of the depressions that form at the head of the Bay of Bengal were associated with precursor disturbances coming from the east.

In this study, the simulated monsoon disturbances are detected using two criteria based on the daily precipitation rate and sea level pressure for each year of the two runs. The region studied in this subjective analysis is restricted to longitudes between 0° and 35°N and latitudes between 65° and 100°E, which encompasses India, Southeast Asia, the Arabian Sea, and the Bay of Bengal. The time period encompasses the months of June, July, and August when the simulated rainfall is observed to be greatest over India. The evolution of the precipitation rate over each grid box covering India and surrounded areas has been examined. This shows that very high rainfall maxima over a grid box (much larger than 40 mm day⁻¹) occur, lasting a few days, followed by much lower values (less than 20 mm day⁻¹). To follow the depressions coming from Southeast Asia, where the mean sea level pressure is higher than in the region of the monsoon trough, we choose to consider the departure of the sea level pressure from the seasonal mean rather than the pressure itself. It would have been more appropriate to use the long-term seasonal mean. However, for the region considered in this study, the change between the long-term seasonal mean and the seasonal mean of each year is less than 1.5 hPa in the 9 yr of the ctrl run and 2 hPa in 7 yr of the Ti 05 run. For the two other years, we have verified that even with the long-term seasonal mean, the criteria we have used to detect the monsoon depressions were still valid. A monsoon disturbance is identified in a grid box when the rainfall rate exceeds 30 mm day⁻¹ and the sea level pressure anomaly is less than −5 hPa. There are few situations with a high precipitation rate that are not correlated with a low pressure system; those that were not correlated were discarded, as well as the few cases of depressions moving to the south. It should be noted that these conditions do not exactly follow those that are observed for monsoon depressions; they are dictated by the mean patterns simulated for the monsoon season by the GCM.

b. Description of monsoon disturbances in the ctrl run

For the 9 yr of data from this integration, 17 monsoon disturbances that lasted between 4 and 8 days with a mean of 6 days were found. This result is consistent with observations (Godbole 1977; Krishnamurti et al. 1981; Saha and Chang 1983; Rajamani and Sikdar 1989). Of these, 10 events occurred in June, 6 in July, and only 1 in August. It is obvious that the number of disturbances that are simulated during the 9 yr seems weak compared to the number observed (Sikka 1977), which is rather six per season. This discrepancy can be due to the severe conditions that we have prescribed. Nevertheless, this distribution is obviously related to the evolution of the Indian rainfall in the ctrl simulation. In August, the Indian rainfall is very weak whereas it is high over the foothills of the Himalayas as if a long break episode was established during this month (K. Laval et al. 2000, unpublished manuscript). But in June, the rainfall is at its strongest and this is linked to a larger number of depressions generated during this month. The systems identified by these thresholds are found to have the characteristics of the monsoon depressions or midtropospheric cyclones, as will be seen in the next paragraph.

An example of a monsoon disturbance as identified by our analysis is displayed in Fig. 4. A low pressure anomaly forms over the east coast of Vietnam on day 174 (24 Jun), year 5. The low center moves northwest toward the south of Burma and then reaches the Bay of Bengal where it intensifies on day 176 (26 Jun). The streamlines at 850 hPa show a corresponding center over the east coast of Vietnam and its northwestward displacement toward the Bay of Bengal. The tilt of the trough is from the north-northeast to the south-southwest as observed (Sikka 1977). The cyclonic circulation extends up to 300 hPa with a maximum at 700 hPa (not shown) as observed by Rajamani and Sikdar (1989). The 850-hPa wind anomaly is westerly south of India and easterly to the north. A strong positive 850-hPa relative vorticity maximum is associated with this low system and reaches a peak of 12 × 10⁻⁵ s⁻¹ on day 176 (26 Jun). All of these features are in agreement with observed conditions (Krishnamurti et al. 1975; Sikka 1977). The rainfall is predominantly in the south or southwest sector of the disturbance as also observed (Godbole 1977). The depression moves rather fast at the beginning following formation and then slows down when the disturbance reaches the Bay of Bengal. The displacement between days 174 and 177 (24 and 27 Jun) is approximately 20° in longitude, which, again, is close to observations (Krishnamurti et al. 1975) but obviously during the last two days the displacement was less than one grid point per day as the system stalls.

Another interesting monsoon depression develops on day 150 (1 Jun), year 3, and is related to the monsoon onset (Fig. 5). This type of system has been observed by Krishnamurti et al. (1981) and its characteristics are very similar to the one displayed in this example. Here, a low pressure anomaly appears over the Arabian Sea and deepens during the first few days as it moves north. The streamlines at 850 hPa show a corresponding cyclonic system that moves from 14° to 22°N, reaching the Gujarat region. The strongest rainfall occurs in the southwestern sector of the depression. The evolution of the system shows some spreading into the Indian peninsula rather than a definite displacement. The 850-hPa relative vorticity, which is weak at the beginning, is also enhanced during the growth phase. The 850-hPa wind anomaly strengthens and strong southerlies develop over India during the nature stage.

The 17 identified events illustrate several particular characteristics; for example, the system core generally has rainfall rates higher than 60 mm day⁻¹ as it develops with the low pressure center deepening as the rainfall rate increases. The disturbances are generated in three main regions with most of them in the Bay of Bengal, some over central India, and three over the Arabian Sea. Those that are located over the Bay of Bengal often originate in Southeast Asia (as in Fig. 4). When they arrive at the head of the Bay of Bengal, they become stationary for a few days. The three disturbances over the Arabian Sea occur at the beginning of the monsoon season, during the first days of June. They move north, then northeast, reaching northeast India.

The 850-hPa wind generally reaches 30 m s⁻¹ over the depressions (south of India and/or the Bay of Bengal and/or the Arabian Sea). At this level the wind anomaly has a cyclonic circulation. The relative vorticity at 850 hPa is positive over the perturbed area. For 11 events, it is larger than 8 × 10⁻⁵ s⁻¹. For the six other cases, the relative vorticity is positive and weaker at 850 hPa, but reaches a value of 6 × 10⁻⁵ s⁻¹ and higher values at 700 hPa. The streamlines at 850 hPa show a vortex associated with the sea level pressure anomaly for 11 cases with a strong positive relative vorticity. For the six other cases, and in particular for two of the three cases over the Arabian Sea, the streamlines show a cyclonic circulation only at 700–500 hPa. These last six disturbances can be considered as midtropospheric cyclones and are also associated with high precipitation rates.

c. Description of monsoon disturbances in the Ti 05 run

The same subjective analysis has been performed on data from the Ti 05 integration to detect the monsoon disturbances. There were 31 events found that generally lasted between 3 and 6 days with 5 of them lasting between 8 and 12 days. The mean duration is 5 days. These events are more uniformally distributed during the monsoon season than in the ctrl run, with 9 events in June, 10 in July, and 12 in August: as many depressions have been reported to occur in July and August (Sikka 1977; Godbole 1977; Saha et al. 1981; Rajamani and Sikdar 1989). This distribution appears closer to those observed than for the ctrl run, which simulates only one depression in August during the 9 yr.

The general features of the disturbances are the same as in the ctrl run because the same criteria are used to identify them. But two important points should be noted:first they are more numerous, almost twice the number as in the ctrl run, and second, the streamlines at 850 hPa in this run reveal a persistent vortex that follows the pressure low and that characterizes the monsoon depressions for the 31 cases, whereas in the ctrl run, we mention that 6 cases were associated with a cyclonic circulation only at the higher levels of 700–500 hPa. Two examples of monsoon depressions are presented from this run, one occurring over the south of India with a displacement that has a strong northward component and one that occurs over the Bay of Bengal, which moves westward.

The first example shows a monsoon disturbance that occurs during days 171–174 (21–24 Jun), year 3 (Fig. 6). A maximum in rainfall appears first at the southern tip of India and moves along the western coast. The surface pressure anomaly deepens and moves in the same direction. The 850-hPa streamlines display a vortex corresponding to the low pressure anomaly that moves northward into the Arabian Sea, at 23°N on day 176 (26 Jun) (not shown). Because the height of the plateau was reduced, the incursion of the streamlines is more inland over the Tibetan region. The rainfall occurs in the southern or southwestern sector of the vortex. During these four days, the 850-hPa relative vorticity is positive up to the 300-hPa level and the southwesterly wind is enhanced over the south of India and the Bay of Bengal.

The second example is taken from year 6, during days 198–201 (18–21 Jul) (Fig. 7). A strong core of precipitation appears at 15°N, over the east of India, and moves west. Simultaneously to the north, a deep sea level pressure anomaly develops and reaches the west of India. At 850 hPa, the streamlines show a vortex over the Bay of Bengal (18°N, 87°E), which travels from Orissa to west Madhya Pradesh, reaching the position 22°N, 80°E on day 201 (21 Jul). The vortex follows quite well the surface pressure anomaly. The wind anomaly and relative vorticity patterns show the usual features already noticed in the previous examples.

d. Subjective analysis conclusions

A subjective analysis of the monsoon disturbances that occur in two numerical integrations with a different height for the Tibetan Plateau has been performed. These integrations simulate very different summer rainfall patterns over India (Figs. 2 and 3). Using criteria based on the precipitation rate and sea level pressure anomaly of 30 mm day⁻¹ and −5 hPa, respectively, the monsoon depressions are suitably identified and are found to have the following features that agree with observed depressions.

1. They are associated with strong rainfall that reaches values higher than 100 mm day⁻¹ at the mature stage.

2. The relative vorticity at 850 hPa reaches values higher than 8 × 10⁻⁵ s⁻¹ except for the six cases (in the ctrl run) that we identify as midtropospheric cyclones. Generally, the relative vorticity is a maximum at 700 hPa.

3. A cyclonic circulation at 850 hPa is identified for all 31 events in the Ti 05 run and for 11 cases (out of the 17 cases) in the ctrl run. The wind in the southern sector generally reaches values of 30 m s⁻¹.

4. The strongest rainfall occurs in the southern or southwestern sector of the vortex at 850 hPa for many cases and under the vortex for the few remaining cases.

We conclude that the GCM is able to simulate the depressions that have characteristic features of observed disturbances (Godbole 1977). The comparison between the two runs has shown that more disturbances were found in the Ti 05 run. The number of depressions was 31, compared to 17 in the ctrl run, or even to 11 cases when we restrict the selection to the ones that descend to lower levels. The number of depressions in August is only 1 in the 9-yr ctrl run whereas 12 are detected in the Ti 05 run. The number of heavy rainfall days in the 9 yr in the Ti 05 run is more then 1.5 times the number found in the ctrl run. The increase in the total number of heavy rainfall days in the Ti 05 run is mainly through the increase in the total number of disturbances identified. Moreover, the disturbances for the ctrl run arrive from Southeast Asia and often stall over the head of the Bay of Bengal for several days. On the contrary, the disturbances occurring in the Ti 05 run travel from the Bay of Bengal toward India with a definite displacement over land. To emphasize the role these depressions have on the Indian rainfall, we show in Fig. 8 the total precipitation rate accumulated during only the days with monsoon disturbances. The shift of the pattern between the two runs that was obvious in the JJA average (see Fig. 2) also appears here. It is striking how the change in Fig. 8c looks similar to the JJA average difference shown in Fig. 3. Over central India, the precipitation variation is about 3000 mm for the 9 yr: this is a substantial fraction of the JJA difference, which occurs during the nine summer seasons (Fig. 3) of the integration. Therefore it is obvious that the rainfall difference between the two runs simulated by the GCM is related to the way each run represents monsoon depressions. This study is complemented by an objective analysis performed in the next section.

a. The method

In this section, we use the objective analysis program developed by Hodges (1995, 1996) to track the monsoon disturbances. This objective method identifies minima or maxima in the chosen field. In this case minima in the sea level pressure anomaly field are determined. The minima at consecutive time steps are linked together by first initializing a set of tracks based on a nearest neighbor search. These tracks are improved by the minimization of a cost function based on measures of the track smoothness in terms of changes in direction and speed (for more details see Hodges 1994, 1995). Constraints on the motion are also adaptively applied during the cost function minimization (Hodges 1999) for the displacement distance and the track smoothness. The tracking parameters have been chosen and validated against the manually analyzed systems previously described.

The analysis was performed on the daily values of the sea level pressure anomaly field, to complement the subjective analysis already performed. In a very detailed analysis, it has been verified that, for each month of June, July, and August of each year, there is a good correspondence between the monsoon trajectories obtained by this method and the location of the events determined previously. Two approaches have been explored: one using the pressure departure from the monthly average and the other one using the pressure departure from the JJA average. The second option appears to enable the tracking code to better follow the disturbances that were also detected by the subjective analysis. To eliminate inconsistent features, the postprocessing parameters of TRACK are carefully chosen. Three conditions are prescribed: for the lifetime of the depression, which must be longer than 2 days; for the attained value of the sea level pressure anomaly, which is set at −8 hPa because we have noticed in our subjective analysis that the disturbances have values ranging between −6 and −12 hPa; the displacement from initiation to disappearance must be at least 8° (∼800 km), which corresponds to almost the longitudinal scale of two grid boxes in the model. It is important to notice that the distributions show little change when we modify the intensity threshold to −5 hPa or the distance criterion in a range from 5° to 10°. We have applied the same criteria in the ctrl and Ti 05 runs to compute the statistical diagnostic fields.

The track ensembles are used to compute the statistics for the distribution of systems as well as their mean attributes such as mean intensity, growth/decay rates, and mean velocity. The statistics are computed using the spherical nonparametric estimators described by Hodges (1996) with adaptive smoothing. The distributions for such variables as feature density, track density, and genesis and lysis densities have been scaled to number density per season where the unit area is equivalent to a 5° circle (∼1 × 10⁶ km²) instead of the probability density functions originally described by Hodges (1996). Also, where the feature density is weak, the mean attribute statistics are suppressed so that only mean attributes from high density data areas are displayed. Although one might argue that the length of the integrations is insufficient to provide statistically robust results, the computed statistics do provide a means of comparing the spatial variation of the systems in terms of their distribution and mean attributes.

When detecting the disturbances, a large region has been analyzed, from 40° to 120°E and from 0° to 40°N. But to focus on the changes occurring over the Indian monsoon region the statistical analysis is performed over a much smaller domain, from 60° to 110°E and from 0° to 30°N. In particular, this enables us to exclude the strong activity over the Tibetan Plateau, which is spurious because it is related to the extrapolation to sea level.

b. The track results

As mentioned before, we have checked for each month of each year that the low pressure systems detected by TRACK are consistent with those determined previously. A high variability is found from year to year:one month, they can be located over southern India, whereas another month, they are found over the monsoon trough or northward. The direction of their displacement also shows some variabilities. They often move from west to north but some of them have a northeastern displacement. Overall, this analysis, based only on the sea level pressure anomalies, identifies more low systems than the subjective analysis, which also prescribes a condition for rainfall rate.

It is interesting to show trajectories detected by the TRACK program to check that they agree with what is observed. A similar study has been conducted on the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis (ERA) and the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Annamalai et al. 1999). This work was based on the 6-hourly 850-hPa relative vorticity. Our comparison should ideally have been performed on the same fields (relative vorticity). However, as shown in Figs. 4–7, the comparison between the relative vorticity and the anomaly of the sea level pressure demonstrates that the maxima of the fields are occurring in the same location and therefore this has less impact than the errors in the monsoon simulation. Tracking vorticity maxima in daily data is more problematic due to the larger number of spurious centers generated.

Figure 9 illustrates two examples of the low pressure anomaly trajectories for the ctrl run. The result presented is for June (the month showing the highest frequency of monsoon disturbances) of years 4 and 5. The diamond-shaped symbols indicate the first day of the disturbance and the crosses correspond to the position of the depression center at each sampling time. The color gives the strength of the sea level pressure anomaly.

During year 4 of the ctrl run (Fig. 9a), a depression forms over the eastern Arabian Sea and reaches the Bay of Bengal by an eastward displacement, describing a northerly curve. This depression is an important incursion since it crossed the whole landmass of central India and corresponds to a monsoon depression noticed in the subjective analysis (not shown). Another system is generated in the center of the Indian landmass and moves straight to the north. A third one appears over the southern tip of India and move northward, along the east coast of India. These three systems are moving preferably north and northeast rather than westward. Their intensity increases as they propagate north. The last system located at 35° latitude is rather odd: it moves north, then west toward the Tibetan Plateau. It corresponds to a flaw due to the fact that we use an extrapolation of sea level pressure for these points over mountains. During year 5 of the ctrl run (Fig. 9b) the depressions are generated preferentially over Burma, the Bay of Bengal, and India. Their displacement is mainly north and northwest. The depression generated over the Bay of Bengal that crosses it corresponds to an example displayed previously in the subjective analysis (see Fig. 4). It has been observed (Saha et al. 1981) that many monsoon depressions are initiated by a low coming from the east and it appears that this depression is associated with a precursor disturbances coming from Southeast Asia: the southern system, which is generated over the China Sea and moves northwest, disappears one day after the monsoon depression over the Bay of Bengal is generated. The transition between the two systems is also shown in Fig. 4b where the two cyclonic lows exist simultaneously on day 175. We also notice one situation where the depression appears tracked along the monsoon trough and corresponds to another monsoon depression detected by the subjective analysis. This behavior has been widely observed and studied by several authors (Godbole 1977; Saha and Chang 1983; Rajamani and Sikdar 1989; Douglas 1992a).

Overall, the trajectories of simulated depressions are in agreement with the ones found in the ERA reanalysis of 1988 (Annamalai et al. 1999) where many depressions coming from south Arabian Sea or the Bay of Bengal cross the Indian subcontinent.

The monthly study of the tracks of the monsoon depressions in the Ti 05 run basically shows the same features (not shown). However, in this run, there is no initiation of depressions detected over the Indian subcontinent.

c. Statistics of the disturbances

The statistics have been computed as described above for the filtered track ensembles derived from the 9-yr runs of the ctrl and Ti 05. These statistics are displayed in Figs. 10 and 11 for the ctrl and Ti 05 runs, respectively, and show the feature density, genesis density, lysis density, mean growth/decay rates, mean intensity, and the mean speed/velocity. This provides a useful means of displaying the mean properties of the systems and their spatial variation, although the statistical robustness of the results requires longer integations. Note, because contour smoothing has been used for the mean attributes, where the mean attributes have been supressed in low data density regions, a smooth transition is apparent at the edge of the distribution that is spurious.

In the ctrl run, there is a strong maximum in genesis of systems over the Burma region (Fig. 10f). A secondary and weaker maximum also appears over central India. These are both associated with growth as indicated in the growth/decay rate statistic (Fig. 10d). The disturbances that are generated over south Asia move west (or slightly northwest) with a speed between 4 and 5 m s⁻¹ (Fig. 10b) before slowing down over the Bay of Bengal where the speed is very weak in accordance with the systems becoming stalled. This slowing of the system speed over the Bay of Bengal can be seen to be associated with a decay (Fig. 10d) and with a lysis (Fig. 10c) maximum related to the disappearance of systems. The mean intensity (Fig. 10a) is high in the Bay of Bengal indicating that the systems here are mature. The feature density (Fig. 10e) is also high in the Bay of Bengal associated with the very slow propagation of disturbances in the area. The systems that are generated over central India move north (or slightly northeast) before disappearing over the foothills of the Himalayas. Overall, these results are in agreement with our subjective analysis, which found that the disturbances coming from the east move toward the Bay of Bengal where they become stationary. There were also three systems that originated in the Arabian Sea and moved toward the northern region, but due to the scarcity of these types of systems their contribution to the statistics is minimal.

There are several changes in the statistical distributions of the Ti 05 run compared to the ctrl run. The distributions of mean speed and velocity (Fig. 11b) show that the lows generated over Burma (Fig. 11f) move northwest along the trough with a speed of about 6 m s⁻¹, higher than in the ctrl run. There is no strong northward displacement over the Indian continent (Fig. 11b) as observed in the ctrl run, which can induce breaks. The feature density (Fig. 11e) shows maximum values that are shifted north as for the rainfall distribution. The strong maximum of genesis density (Fig. 11f), although still over Burma, is shifted toward the north and also is more extended with smaller values, invading eastern India and western China, indicating that a wider region generates these instabilities. This genesis is still associated with growth (Fig. 11d) although with lower mean values. The secondary maximum, over western India, has disappeared in the Ti 05 run and, indeed, it is noticed in the subjective analysis that less disturbances (in a relative sense) occurred over the Arabian Sea than in the ctrl run. Lysis still occurs in the Bay of Bengal (Fig. 11c) associated with decay (Fig. 11d) and low system speeds (Fig. 11b) associated with the systems becoming stationary. The mean intensities (Fig. 11a) show less variation although it is still apparent that the systems in the Bay of Bengal are mature systems. Ideally the statistical significance of these comparisons should be tested. However, due to the short integration length and the fact that significance testing is more difficult for multivariate distributions (requires a multivariate Kolmogorov–Smirnov test for densities), this has not been done.

The comparison between the results of the LMD6 GCM and the reanalyzed data (Annamalai et al. 1999), although not on the same field, can give some insight into the performance of the LMD6 model. In the reanalyses, the maximum in genesis density is located over the head of the Bay of Bengal and the lysis density maximum over north (ERA) or northwest (NCEP) of India. For the LMD6 model, the two patterns are found more to the east as the genesis density maximum is located over Burma and the maximum of lysis density is over the head of the Bay of Bengal. Accordingly the growth rate maximum found over the Bay of Bengal in the reanalyses is shifted over Burma in the LMD6 model.

This comparison shows that the model generates too many disturbances over Burma to the east of the reanalyses and that these systems move westward toward the monsoon trough as in the ERA analysis.