The nature of the increasing frequency of extreme rainfall events (ERE) in central India is investigated by relating their occurrence to synoptic activity. Using a long record of the paths and intensities of monsoon synoptic disturbances, a synoptic activity index (SAI) is defined whose interannual variation correlates strongly with that in the number of ERE, demonstrating a strong connection between these phenomena. SAI furthermore shows a rising trend that is statistically indistinguishable from that in ERE, indicating that the increasing frequency of ERE is likely attributable to a rising trend in synoptic activity. This synoptic activity increase results from a rising trend in relatively weak low pressure systems (LPS), and it outweighs a declining trend in stronger LPS.
Floods are the most frequent natural disaster in India, with the central Indian Ganges–Mahanadi basin being the most vulnerable region (Dilley et al. 2005). A large fraction of these floods are associated with the passage of synoptic disturbances (Sharma and Kaushik 2009).
The impact of a warming climate on flood occurrence is of great practical concern, as more than 400 million people inhabit the central Indian region, of which it is estimated that an average of 40 million are evacuated every year because of floods (Central Water Commission 2009). The effect of rising global surface temperature on the Indian monsoon climate is not yet clear. Although seasonal mean summer monsoon rainfall has shown no clear trend over the past century (Kumar and Dash 2001, and references therein), recent studies indicate that the past few decades have seen a rising trend in extreme rainfall events (ERE; Easterling et al. 2000; Goswami et al. 2006; Rajeevan et al. 2008), the relative constancy of seasonal mean rainfall being maintained by a decreased frequency of moderate rainfall events.
We consider here the nature of the observed ERE increase by examining its relation to changes in synoptic activity. The potential for such a connection is evident, because the synoptic-scale systems collectively referred to as low pressure systems (LPS) bring copious rain to the Indian continent, especially in the central Indian region (Mooley 1973; Sikka 1977, and references therein), and also play a role in moisture transport (Simmonds et al. 1999). Typical time and length scales are 3–5 days and on the order of 1000 km (Krishnamurti et al. 1975). Most of these systems originate in the Bay of Bengal (BoB) and move northwestward. The India Meteorological Department (IMD) classifies LPS as lows, depressions, and so forth, based on peak surface wind speeds estimated from maps of surface pressure (Raghavan and Rajesh 2003; Sikka 2006), as described in more detail later. The ranges of estimated wind speed associated with each such category are summarized in the first row of Table 1.
Reported trends in LPS include an increase in the number of severe cyclonic storms over the north Indian Ocean (Webster et al. 2005); however, these severe cyclonic storms generally appear in the pre- and postmonsoon seasons. A decreasing trend in the frequency of depressions and cyclonic storms since 1970 has also been noted (Kumar and Dash 2001; Sikka 2006, and references therein), which is puzzling in view of the increasing trend in ERE. Here, using 53 years of gridded LPS and daily rainfall data, we show that there is indeed an increasing trend in an index measuring the aggregate impact of LPS, accounted for by an increase in the frequency of the weakest “low” category of LPS, and that this trend is consistent with the rising trend of extreme rainfall events in central India.
A high-resolution (1° × 1°) gridded Indian rainfall dataset (Rajeevan et al. 2006) is used to identify extreme rainfall events. The gridded rainfall dataset is developed using rain gauge data from 1803 stations distributed over the Indian continent with at least 90% data availability for the period 1951–2003. These gridded data more accurately represent rainfall over the Indian region, especially over the west coast and northeast India, compared to other global datasets (Rajeevan et al. 2006).
The LPS data considered here consist of daily records of genesis, position, and intensities of all LPS formed during June–September over the Indian monsoon domain (both Arabian Sea and BoB) from 1951 to 2003. The first 30 years were compiled by Mooley and Shukla (1987) based on a careful examination of daily weather reports and annual summaries of storms and depressions published by IMD. Data for 1984–2003 were compiled using an identical procedure by Sikka (2006). This procedure for categorizing LPS is based on the sea level pressure distribution and is described in detail in the appendix.
In addition to considerable interdecadal variations (Kumar and Dash 2001), total monsoon season [June–September (JJAS)] LPS within the monsoon region show a statistically significant increasing trend (p < 0.05) after 1950 (Sikka 2006; Fig. 1a). Such a trend is even more evident in the time series of total monsoon season LPS days (Fig. 1b). This occurs despite a decrease in the four strongest classifications in Table 1 as measured by depression and storm days (Fig. 1c), primarily after 1976 or so (Rajeevan et al. 2000; Rao et al. 2004; Sikka 2006). The apparent discrepancy is accounted for by a strong increase in the number of systems bearing the weakest low classification (not shown) and in the number of low days in a season (Fig. 1d).
A low pressure system is characterized by its genesis date, location, intensity, track, and number of days it stays active. All of these affect rainfall in the central Indian region; hence, some aggregate index is necessary to accurately measure synoptic activity. Such an index is constructed here by the following method:
(i) The number of LPS is counted within each 3° × 3° grid box for each day during the summer monsoon season.
(ii) Each count is then weighted by a measure of intensity, which is taken to be the centroid of the range of wind speeds associated with the category of the LPS (as assigned by the IMD; second row of Table 1).
(iii) This quantity is summed over each monsoon season to obtain the annual synoptic activity index (SAI) for each 3° × 3° grid box (this grid size was chosen to enable the spatial variation of SAI to be represented while maintaining adequate samples of LPS within individual grid boxes).
The correlation of SAI averaged over a central Indian region (18.5°–26.5°N, 77.5°–86.5°E) with the number of BoB LPS is 0.41; with BoB LPS days, it is 0.74. The spatial structure of SAI averaged over 53 years is shown in Fig. 2a. The characteristic features of LPS, such as their genesis over the head of the BoB and their paths through the Ganges–Mahanadi basin, are well represented.
The spatial distribution of average ERE intensity is elevated over the central Indian region (Fig. 2b), where ERE are defined here as exceeding the 98.3th percentile of the daily rainfall distribution within each 1° × 1° grid box (this corresponds to 2 events per monsoon season on average). The most intense ERE occur over the west coast of India (Fig. 2b) and are generally associated with local orography, synoptic activity, and offshore vortices over the Arabian Sea (Francis and Gadgil 2006).
The frequency of ERE averaged over the central Indian region shows a statistically significant (p < 0.05) rising trend (Fig. 3), which is consistent with the study of Goswami et al. (2006). SAI summed over the same region shows a similar rising trend, and there is a strong positive correlation (r = 0.68, p < 0.01) between their annual values, demonstrating the strong physical association between central Indian LPS and ERE. This, together with the statistical indistinguishability of the two trends, strongly suggests that the ERE increase is associated with increased synoptic activity. This result is insensitive to the percentile cutoff chosen in defining ERE within the range 95.8–99.8, which correspond to averages of 5 events per monsoon season and 1 event per 5 monsoon seasons, respectively. If only lows and their corresponding numbers, intensities, and duration are considered for constructing the aggregate index (i.e., excluding high-intensity systems), the resulting time series correlates with ERE at r = 0.54. If central Indian SAI is defined purely in terms of cumulative LPS days (i.e., without weighting by LPS intensity), the resulting time series still correlates at r = 0.48 with that of ERE, indicating the robustness of the relationship between ERE and LPS. By contrast, the total number of central Indian LPS in a season shows a much weaker correlation (r = 0.29).
To further illustrate the relationship between ERE and LPS, Figs. 4a,b show the average probability of occurrence per day of ERE in the central Indian region as a function of position relative to LPS centers for the first half of the dataset, 1951–76 (Fig. 4a), and the second half, 1977–2003 (Fig. 4b).1 These daily ERE probabilities, expressed on the 1° × 1° grid used for the rainfall data, exceed 0.1 and are strongly peaked close to the LPS centers, while remaining elevated within ∼500 km of the centers. The concentration southward and westward of the LPS centers and the elongation in longitude of this pattern are very similar to that for 24-h LPS rainfall determined by Mooley (1973). As an approximate measure of the extent to which ERE are attributable to LPS, we categorize ERE as LPS related if they occur within ±5° latitude and longitude of an LPS, and we categorize them as non-LPS related if they do not. By this measure, 62% of ERE in the central Indian region between 1951 and 2003 are LPS related.
Figures 4a,b indicate that, between 1951–76 and 1977–2003, the probability of ERE occurrence near a given LPS decreased by ∼10%, as measured by the integral of the depicted probabilities within ±5° of the LPS centers, and by ∼30%, as measured by their peak values (Table 2). The significance of these differences was tested by randomly drawing 104 subsamples, identical in size to the 1951–76 and 1977–2003 subsamples, from the 1951–2003 combined sample (sample sizes are indicated in the bottom-right corners of Figs. 4a–f). Based on the resulting probability maps, the null hypothesis that the differences between Figs. 4a,b are a consequence of random sampling alone can be rejected at p < 0.03 for the integrated probabilities and at p < 10−4 for the peak probabilities.
The overall increase in ERE thus occurs despite a decrease, on average, in ERE occurrence near individual LPS. To explain this seemingly counterintuitive result, the probability of occurrence of ERE associated with the weaker lows versus all stronger systems (labeled as “storms”) are plotted in Figs. 4c–f. It is seen that ERE frequencies associated with a given LPS classification change little between 1951–76 and 1977–2003. The overall ERE increase, which amounts to 7.5% between these two time periods, is thus attributable to the strong 84% increase in the number of low days, which outweighs a 42% decrease in the less frequent depression and storm days (Table 2), rather than to any changes in ERE occurrence near a given category of LPS. The aggregate SAI, which weighs synoptic activity by LPS number, intensity, and duration, succeeds in characterizing this net increase in synoptic activity.
A further conclusion that can be drawn from Figs. 4c–d is that lows do elevate ERE occurrence, despite their relative weakness as LPS; the differences between Figs. 4c,e and those between Figs. 4d,f are significant at p < 10−4 according to resampling tests similar to the one described earlier.
The trends described here can be further characterized by noting that LPS-related ERE (i.e., those occurring within ±5° latitude and longitude of an LPS) increased by 22% between 1951 and 2003, whereas non-LPS-related ERE increased by only 2.5% (because ERE occurring within ±5° of an LPS include some subsynoptic-scale events that are not LPS related, the former value likely represents a lower limit to the trend in LPS-related ERE). The association of LPS with the ERE over the central Indian region is also evident in the tracks of these systems. LPS genesis and termination locations for two high ERE years and two low ERE years are shown in Fig. 5. In high ERE years, LPS are more frequent, longer lived, and more likely to pass over in the central Indian region, all of which contribute to higher SAI in these years.
This study has found that the observed increase in central Indian ERE is associated with an overall increase in synoptic activity that occurs, despite the declining frequency of stronger systems noted, for example, by Sikka (2006). The accompanying increase in weaker LPS classified as lows implies a marked redistribution in LPS intensity. Such a change is particularly evident in Fig. 6, which depicts yearly values of the fraction of monsoon season LPS that become depressions or stronger systems. Prior to 1983, this ratio was typically 0.5 or greater, whereas from 1983 until 2003 it did not exceed 0.33, except in the strong El Niño year of 1997.
The genesis and development of LPS are governed by multiple environmental factors, which include sufficiently large SST and ocean heat content, Coriolis parameter, cyclonic low-level relative vorticity, low- to midtropospheric lapse rate, midtropospheric relative humidity, and weak vertical shear according to Sikka (1977) and Gray (1979; an index combining these factors is discussed by Camargo et al. 2007, and references therein). It is thus natural to enquire whether some trend or shift in ambient conditions might be responsible for the observed changes in LPS. Several studies have examined the observed decrease in depressions and stronger LPS in the context of environmental factors believed to govern the frequency of LPS formation. It has been found that, although increasing BoB SST should favor LPS formation, other trends, including decreasing cyclonic low-level vorticity, increasing vertical wind shear, and decreasing midtropospheric moisture, appear to be acting oppositely (Rajeevan et al. 2000; Mandke and Bhide 2003; Dash et al. 2004; Pattanaik 2005; Pattanaik and Rajeevan 2007). Some of these changes are consistent with an observed southward shift in the low-level jet that supplies south Asian monsoon rainfall with moisture generated over the southern Indian Ocean and Arabian Sea (Joseph and Simon 2005). A strong connection has also been noted between the frequency of monsoon season BoB storms and the strength of the upper-level tropical easterly jet over southern India (Rao et al. 2004).
The redistribution of LPS intensity after 1983 implied by Fig. 6 coincides with a period during which tropical SST in the far eastern Indian Ocean and far western Pacific are strongly anticorrelated with SAI (Fig. 7b), whereas earlier in the record no such relation is seen (Fig. 7a). Cool SSTs in this region are associated also with the positive phase of the Indian Ocean Dipole (IOD), anomalous moisture transport from the southeastern tropical Indian Ocean to the BoB, and anomalous low-level cyclonic vorticity over the head of the BoB (Ajayamohan and Rao 2008). Somewhat similar SST anomalies in this region have been noted in association with El Niño events following the apparent change in ENSO properties and in ENSO–Indian summer monsoon linkage that occurred around 1976 but not prior to this period (Annamalai and Liu 2005). According to their modeling study, such cool SSTs favor stronger Indian summer monsoon precipitation and partially counteract the opposing influence of warm SSTs in the central and eastern equatorial Pacific. The correlation patterns in Fig. 7 remain qualitatively similar if periods before and after 1976/77 instead of 1982/83 are considered.
The above studies do not address why a strong increase in weaker LPS has accompanied the decreased frequency of depressions and storms. The causes could, of course, lie in the environmental changes noted earlier: for example, if increased SST caused more LPS to develop but other factors such as reduced cyclonic vorticity and humidity and increased wind shear limited their development. Such questions are perhaps best addressed by regional modeling studies, in part because of the limited resolution of reanalysis and their lesser reliability in the pre-satellite era. A step in this direction has been taken by Stowasser et al. (2009), who examined changes in the distribution of monsoon season storm intensities under increased CO2 using a regional model forced by boundary conditions from a global climate model.2 They noted an increase in the number and intensity of storms, as well as a northward shift in their genesis locations, accompanied by increased low-level westerly winds and tropospheric moisture. These changes, occurring under a strong increase in radiative forcing, are opposite to the recent observed trends, although some model projections suggest that the frequency of ERE in this region will rise as the climate warms (Turner and Slingo 2009). This result points to the key question of whether the recent observed changes are anthropogenically forced or are the result of climatic variations largely unrelated to anthropogenic warming. Further studies will be needed to resolve such questions.
The central Indian region is the focal point of the northwestward-moving monsoon LPS, which are closely linked to the monsoon trough, a planetary-scale low pressure feature that extends across the Indian subcontinent from southeast to northwest. LPS are also linked to the relatively large-scale intraseasonal oscillations (active-break phases) of the monsoon (Ajayamohan 2001; Goswami et al. 2003) and contribute importantly to water resources in this region (Dhar and Bhattacharya 1973; Dhar et al. 1974). This study has found a significant increase in overall central Indian summer monsoon season synoptic activity since the 1950s, even though the frequency of more intense systems has decreased, and it has shown that this trend tracks the increased incidence of ERE. This result, along with high interannual correlation between ERE and SAI, strongly suggests that the rising frequency of central Indian ERE in the summer monsoon season detected by Goswami et al. (2006) is associated with a concurrent increase in synoptic activity.
The connection illustrated here between the frequency of central Indian ERE and synoptic activity during the monsoon season points to a potential capability for predicting ERE risk to the extent that synoptic activity itself can be predicted. This should be so on synoptic time scales (through forecasts of the paths and strength of LPS), on seasonal to interannual time scales (if the level of synoptic activity in a season can be forecast), and with respect to long-term trends associated with global warming. Although the first instance represents a challenge for numerical weather prediction, the second and third rely at present on the ability of seasonal forecast and climate models to predict future climate skillfully. However, such models are limited at present in their ability to represent synoptic activity explicitly because of their relatively coarse resolution. Applying such forecasts to ERE prediction thus could be aided by tying the level of synoptic activity to more readily predicted phenomena such as ENSO on seasonal to interannual time scales and SST warming on multidecadal time scales. However, although ENSO modulates total seasonal monsoon rainfall, it does not strongly correlate with ERE or SAI. Other controlling influences must thus be sought, with a view toward establishing physically based causal relationships.
Steps in this direction have been taken by Ajayamohan and Rao (2008), who note an apparent modulation of summer monsoon ERE over the Ganges–Mahanadi basin by the Indian Ocean Dipole (IOD) mode of SST variability, whereas Vecchi and Harrison (2004) noted such a modulation of central Indian total summer monsoon rainfall. Although both analyses are based on relatively short time series of reliable SST data available since the early 1980s, Ajayamohan and Rao (2008) also note that the frequency of positive IOD years appears to have increased in recent decades. Further exploration of such relationships may enable predictions for ERE to be derived from seasonal to interannual climate predictions and projections of future climate change.
This work was supported by the Global Atmosphere-Ocean Prediction and Predictability research network, which is funded by the Canadian Foundation for Climate and Atmospheric Sciences. John Fyfe, Steve Lambert, and several anonymous reviewers are thanked for suggesting improvements to an earlier version of the manuscript. RSAM acknowledges Dr. V. Krishnamurthy (COLA) for discussion and Dr. M. Rajeevan (NARL) and Dr. R. S. Mahesh Kumar (IITM) for data support.
Procedure for Identifying the LPS
The procedure for identifying LPS that is used in constructing the datasets considered here is described in detail by Mooley and Shukla (1987) and Sikka (2006). Because those documents are not readily available online, a summary of the procedure is excerpted here. The morning (0300 UTC) sea level synoptic pressure analysis published as the Indian daily weather report by the India Meteorological Department (IMD) for the months of June, July, August, and September for the study period form the basic data source for the identification of LPS.
The low pressure area is classified as low, depression, deep depression, cyclonic storm, and severe cyclonic storm on the basis of the pressure distribution at sea level around the center of the system, with isobars drawn for even hectopascal (hPa) values at 2-hPa intervals. The standard IMD criteria for classification of LPS based on estimated maximum wind speeds (as tabulated in Table 1) are also mentioned for ease of comparison with other categorization schemes.
(i) Low: A single closed isobar for even millibar value with pressure near center being lower than the isobar value by about 2 hPa. The associated maximum wind speeds of these systems are below 8.5 m s−1.
(ii) Depression: Two closed isobars of even millibar value with pressure near center being about 4 hPa lower than the outer isobar value. Here, the maximum associated winds have speeds between 8.5 and 13.4 m s−1.
(iii) Deep depressions: About 5 or more closed isobars of even millibar values and estimated central pressure being about 10 hPa lower than the outermost isobar value. The associated maximum wind speeds range from 13.5 to 16.4 m s−1.
(iv) Cyclonic storm: About 8 closed isobars of even millibar values and estimated central pressure being 16 hPa lower than the outermost isobar value. The associated maximum wind speeds range from 16.5 to 23.4 m s−1.
(v) Severe cyclonic storm: About 10 closed isobars of even millibar values and estimated central pressure being 20 hPa lower than the outermost isobar value. Here, the maximum associated winds have speeds between 23.5 and 31.5 m s−1.
In establishing this categorization, the following guidelines are adopted:
The position of systems formed over the ocean areas and the adjoining Indian continent are noted from 1 June to 30 September for each day of the life history of an LPS. The daily positions of LPS forming over land are noted if the monsoon has advanced over the area.
Any LPS forming in the dry monsoon trough (heat trough) region over northwest India, prior to the advancement of the monsoon over the region, has not been included.
Only systems that have a life history of at least two days with distinct identity are included to avoid counting very short-duration LPS (ephemeral LPS) that may form within the monsoon trough under break monsoon conditions and show no continuity.
There are days when two or more distinct systems may be present within the Indian monsoon region. In those cases each individual LPS is counted and each LPS is allocated its LPS days.
In classifying each pressure system, the weather charts with the isobars and mean sea level pressures at the observatories located within the innermost isobar of the low pressure system have been considered.
This procedure is undertaken through a careful examination of the daily weather charts. It may be mentioned that, although objectivity in the identification of LPS has been aimed at through application of the detailed criteria listed here, the fact that this is not a fully automated procedure, such as that described in Lim and Simmonds (2007), means that subjectivity cannot be completely eliminated. However, the use of detailed local information provided by the IMD charts provides a more accurate and complete account of LPS in this region than provided by the reanalyses typically used by automated cyclone identification procedures.
Corresponding author address: R. S. Ajayamohan, Canadian Centre for Climate Modelling and Analysis, University of Victoria, P.O. Box 3065, STN CSC, Victoria, BC V8W 3V6, Canada. Email: email@example.com
Similar results are obtained if the data are divided at 1970/71 or 1982/83 instead; the possible significance of 1976 as a climatic change point is discussed in section 5.
A description of regional temperature and precipitation changes simulated by such models under twenty-first-century radiative forcing scenarios is contained in section 11.4.3 of Bernstein et al. (2007).