Decline and poleward shift in Indian summer monsoon synoptic activity in a warming climate.

Cyclonic atmospheric vortices of varying intensity, collectively known as low-pressure systems (LPS), travel northwest across central India and produce more than half of the precipitation received by that fertile region and its ∼600 million inhabitants. Yet, future changes in LPS activity are poorly understood, due in part to inadequate representation of these storms in current climate models. Using a high-resolution atmospheric general circulation model that realistically simulates the genesis distribution of LPS, here we show that Indian monsoon LPS activity declines about 45% by the late 21st century in simulations of a business-as-usual emission scenario. The distribution of LPS genesis shifts poleward as it weakens, with oceanic genesis decreasing by ∼60% and continental genesis increasing by ∼10%; over land the increase in storm counts is accompanied by a shift toward lower storm wind speeds. The weakening and poleward shift of the genesis distribution in a warmer climate are confirmed and attributed, via a statistical model, to the reduction and poleward shift of low-level absolute vorticity over the monsoon region, which in turn are robust features of most coupled model projections. The poleward shift in LPS activity results in an increased frequency of extreme precipitation events over northern India.

Although precipitating atmospheric vortices known as low-pressure systems (LPS) are found in all monsoon regions, their presence is most prominent over India, where an average of 13 (± 2.5) storms develop each boreal summer, with most originating over the Bay of Bengal (BoB) and adjoining land (1–5). The intensification and propagation of these storms are linked to the strength of the larger-scale monsoon circulation and interactions with precipitating convection (6, 7). The mean monsoon circulation is argued to have weakened in recent decades, with a variety of reasons advanced for this slowdown of winds (8, 9). Some studies have also reported a reduction in LPS activity since the mid-20th century (10–12), but the reliability of this trend has been questioned due to inconsistencies among observational datasets (13).

A lack of understanding of the mechanism of LPS development has hindered projections of future monsoon synoptic activity, leaving us reliant on simulations of future scenarios by comprehensive climate models. However, the unrealistic representation of these systems in global climate models (GCMs) used in the fifth phase of the Coupled Model Intercomparison Project (CMIP5; ref. 14) is an obstacle to the reliable estimation of future LPS activity (4). Similar issues in estimating future changes in global tropical cyclone (TC) activity were addressed using a high-resolution atmospheric GCM; the same high-resolution modeling strategy was also found reliable in simulating the distribution and structure of Indian monsoon LPS (15, 16). Here we use that high-resolution atmospheric GCM, together with a statistical model, to project and understand possible trends in LPS activity over the next century.

Monsoon Synoptic Activity in Current and Future Climates

Two sets of numerical experiments are performed here using the High Resolution Atmospheric Model (HiRAM; ref. 15) with a horizontal grid spacing of 50 km globally. One set of simulations represents the historical (HIST) period and the other a late-21st-century climate scenario based on the strongest Representative Concentration Pathway (RCP8.5). Four ensemble members of these simulations are run, with sea surface temperatures (SSTs) taken from different CMIP5 GCMs selected for their skill in simulating the Indian monsoon. In addition, 30 ensemble members of annual cycle (ANNC) experiments are run for each of the HIST and RCP8.5 scenarios to assess model uncertainty, in which the model is forced with ANNCs of decadal mean SSTs ().

HiRAM simulates a realistic mean monsoon circulation, precipitation, and LPS frequency compared with most CMIP5 models (). The SAI (17), a combined measure of the frequency, intensity, and duration of LPS, shows a strong and well-defined pattern over the monsoon trough region in the HIST ensemble (Fig. 1). The horizontal structure of the SAI in HiRAM compares well with observations (Sikka archive, 1979–2003 period; ref. 18), albeit with an overestimation of the amplitude of the SAI distribution associated with the distribution of storm intensity’s being biased toward stronger systems (). The number of LPS simulated by HiRAM matches observations better than any other CMIP5 model.

Fig. 1.

June–September ensemble mean climatology of Synoptic Activity Index (SAI) from (A) HIST (shaded) and the 1979–2003 Sikka archive (blue contours, ranging from 20 to 140 with an interval of 30), (B) RCP8.5 ensembles, and (C) difference in SAI climatology between RCP8.5 and HIST, where the stippling represents the changes in SAI that are statistically significant at the 5% level for each ensemble, as revealed by a t test. The area averaged difference in the mean of SAI between RCP8.5 and HIST over the box in C is 45%. (D) Frequency distribution of sea-level pressure depth of LPS from ERAI (interim version of European center reanalysis) and HIST and RCP8.5 simulations of HiRAM. Solid (dashed) lines represent systems that form over the BoB (Indian land region). Lines (shading) show ensemble mean (spread) for HiRAM HIST and RCP8.5 experiments. The future change in mean of distribution for LPS over BoB is statistically significant (P < 0.01), as revealed by a Kolmogorov–Smirnov test. Note that the future change in mean distribution for LPS over land region is not statistically significant. (E) The model spread in the annual LPS count for HIST and RCP8.5 simulations from the ANNC experiments forced with decadal mean SST ANNCs, and the observed decadal variability in LPS counts over BoB and land for the decades of 1971–1980, 1981–1990, and 1991–2000, based on the Sikka archive.

The future projections show a strong weakening of LPS activity in the main genesis region over the BoB and over land immediately to the northwest (Fig. 1 ). The coarse-resolution CMIP5 model ensemble also shows a weakening of synoptic activity over central India in the RCP8.5 simulations (), but the poor representation of LPS structure in those models raised questions about their validity for use in projections.

The reduction in SAI is associated with declines in both storm intensity and genesis frequency. The LPS intensity is estimated as the difference between the central minimum sea-level pressure (SLP) and the outermost closed isobar, which is termed the pressure depth () of LPS. The distribution shows overestimation of LPS intensity over the BoB in the HiRAM HIST simulations, compared with an atmospheric reanalysis (Fig. 1). The frequency distribution of exhibits a shift toward more low-intensity storm days in the RCP8.5 simulations over land. A significant decline in LPS days of nearly all intensity categories occurs over the BoB. The median number of storms per summer also decreases from ∼4.7 to ∼1 over the BoB in the future projections, while the land storm count remains roughly unchanged at ∼7.3 in the ANNC experiments (Fig. 1). LPS frequency in the Sikka dataset (18) also decreases over the BoB and increases over land in the last three decades of the 20th century, but these trends are modest compared with interannual variability within each decade (Fig. 1; shows a similar analysis of individual decades in HiRAM). The climatological difference in the Sikka-based SAI between the last two quarters of the 20th century also shows increased LPS activity over many land regions, but the BoB changes are more modest and inhomogeneous in sign (). Trends in LPS counts and SAI in the HiRAM projections rise above the level of internal variability much more than observed (Fig. 1) or simulated () changes in recent decades; these ensemble means of the HIST and RCP8.5 simulations are not expected to be influenced by internal decadal variability because the SST boundary conditions are drawn from 10 different CMIP5 coupled models.

Observations suggest that genesis frequency over land is less than that over ocean, with about 139 and 176 storms formed over continental India and the BoB, respectively, during 1979–2003. Although HiRAM simulates nearly the same total number of LPS as observed, it produces more frequent genesis over land (180 ) than ocean (128 ) during the 25 y of the historical simulation (Fig. 2 and ). In the RCP8.5 future projection the number of BoB systems declines by ∼63% (to 47 ; Fig. 2 and ). Genesis points over the BoB are more scattered across the basin in the future projection but are clustered around the head of the BoB in the HIST simulations. Over land the LPS count increases slightly in the future projection to 197 ± 6 systems in 25 y. The relatively small intraensemble variance of these changes suggests that LPS frequency trends are relatively insensitive to details of the SST boundary condition and are mostly controlled by common features of a warming climate, such as the weakening and poleward shift of the large-scale monsoon flow.

Fig. 2.

Genesis locations of LPS formed during monsoon season (June–September) from (A) HIST and (B) RCP8.5 simulations of HiRAM. The red (blue) color indicates the genesis location over land (ocean). (C) The Monsoon Disturbance Genesis Index (MDGI) computed from HIST and (D) difference in MDGI between RCP8.5 and HIST simulations of HiRAM. The MDGI has units of number of storm genesis points per 0.5° 0.5° grid cell in the 25-y period. The HiRAM simulations driven by GFDL-CM3 (Geophysical Fluid Dynamics Laboratory Climate Model version 3) SST are shown.

Statistical Projection of Synoptic Activity.

These changes in genesis frequency can be understood using an existing statistical model of the observed spatial distribution and seasonal cycle of global LPS genesis. That model, the MDGI (19), predicts the likelihood of genesis from monthly mean climatologies of precipitable water (PW), low-level absolute vorticity (), an estimate of convective available potential energy (ECAPE), and midtropospheric relative humidity (RH). These four variables were objectively selected using observed LPS counts, but when the MDGI is calculated using the same four climatological variables from HiRAM it also successfully represents the distribution of LPS genesis explicitly simulated in HIST (Fig. 2). Furthermore, the MDGI predicts a strong reduction of genesis frequency over the BoB in the future (Fig. 2). That is, given only the simulated change in the climatological monthly mean state, the MDGI predicts a roughly 50% decrease in genesis frequency over the BoB. It also predicts a roughly 50% increase in genesis over land—much larger than the increase in LPS genesis explicitly simulated by HiRAM—but much of the MDGI increase is along the foothills of the Himalaya where genesis would likely be inhibited by orography.

Mechanisms of Weakening of Synoptic Activity

The MDGI is based on a log-linear model, so its changes can be linearly attributed to changes in its constituent environmental variables. Most of the MDGI change is produced by a reduction and poleward shift in the low-level absolute vorticity, . The HIST climatology of is maximum over the core LPS genesis area due mostly to shear vorticity of the low-level monsoon westerlies, and this maximum weakens and shifts poleward in the RCP8.5 projection, producing a nearly 100% decrease and a poleward shift in the component of the MDGI associated with (Fig. 3). The changes in PW and RH make small contributions to the MDGI change, while an increase in ECAPE strengthens the MDGI everywhere, especially over India’s east coast and around Pakistan, offsetting some of the reduction due to ().

Fig. 3.

June–September ensemble mean climatology of 850-hPa wind (vectors, ms−1) and absolute vorticity (shaded) from (A) HIST. (B) Difference between RCP8.5 and HIST simulations. (C) Difference in MDGI due to absolute vorticity between RCP8.5 and HIST simulations. All based on HiRAM. Stippling in B represents statistically significant (at 5% level) changes in absolute vorticity for all ensemble members.

This suggests that the large reduction in future LPS genesis simulated by HiRAM is caused by a weakening of the large-scale monsoon circulation and a corresponding drop in the ambient vorticity from which LPS form. Although the genesis mechanism of LPS is still the subject of active research (20), the presence of large low-level vorticity has been long held to be essential for genesis (21). Recently, monsoon LPS have also been shown to have structural similarities to weak TCs (20) and to have a statistical association with the climatological mean state similar to that of TCs (19), which suggests that LPS genesis may also be favored by environments rich in cyclonic vorticity. This reasoning contrasts with a previous argument relating the possible recent decline in LPS activity to a midtropospheric drying (12); HiRAM simulates such a drying, but it produces a comparatively minor effect on the MDGI (). The projected weakening of in HiRAM is in broad agreement with that in the CMIP5 ensemble mean, although the HiRAM weakening is stronger (). Weakening of the large-scale monsoon circulation over the last half-century has also been observed, although its underlying cause is debated (8, 9, 22). A slowing of tropical circulations in general may be expected from the increase in static stability that occurs in a warming troposphere (23), but the degree to which changes in diabatic heating might compensate in monsoon circulations is unclear.

In summary, the hypothesis that future LPS genesis will decrease in frequency over the BoB is supported by numerous future projections of a weakening of the large-scale monsoon circulation and by a statistical model that links the large-scale circulation strength to LPS genesis frequency; in addition, HiRAM explicitly simulates a reduction in LPS genesis frequency for all four SST forcings. It has been argued that monsoon LPS over the BoB may result from the amplification of disturbances originating over the West Pacific (24, 25), which might link LPS frequency to the distribution of West Pacific storms, but those studies were based on relatively small samples of LPS. Automated tracking of low-level 850-hPa vorticity anomalies has not found a large number of BoB LPS to originate from the West Pacific (5), so we leave investigation of the connection between the monsoon LPS and West Pacific disturbances in a warming climate for future work.

The projected reduction in synoptic activity is due not only to reduced genesis frequency but also to the general decrease in LPS intensity (Fig. 1). We now discuss possible reasons for this intensity decrease. Although the mechanism of LPS intensification is not understood, LPS have multiple dynamical similarities to weak TCs (20), which have been projected to decrease in number in a warming climate (26). Those projections of TC counts typically only consider storms with sustained surface wind speeds of 20 m⋅s−1 or greater (26, 27), which is stronger than nearly all Indian summer monsoon LPS; a general reduction in LPS intensity is thus consistent with projections of reduced TC counts, as fewer LPS achieve sufficient intensities to be counted as TCs. The reduction of TC frequency has been attributed to an increase in the midtropospheric saturation deficit (27) (amount of moisture needed to achieve saturation) and to a weakening of the large-scale tropical circulation, both of which are simulated to occur in the HiRAM future projections (). In the HiRAM projections, the troposphere over the BoB and central India exhibits a stronger increase in saturation deficit than that over the equatorial Indian Ocean, possibly causing a reduction in LPS intensity by strengthening unsaturated convective downdrafts and weakening the upward convective mass flux in those storms.

Influence of Changes in Synoptic Activity on Precipitation

A change in LPS activity is expected to alter the distribution of monsoon precipitation. The climatological precipitation is strong over the BoB and into central India along the typical LPS trajectory (Fig. 4). The climatology of precipitation related to LPS originating in the BoB shows that those storms contribute to both continental and oceanic rainfall (Fig. 4). The LPS originating over land contribute more to continental rainfall, especially in the northern parts of India (Fig. 4). These patterns of LPS-related rainfall illustrate the importance of these storms for the hydrology of India and the Himalaya. The projected changes in seasonal mean rainfall show a significant drying over the BoB (Fig. 4). The component of precipitation associated with the BoB storms declines substantially, consistent with the decrease in LPS genesis over the BoB (Fig. 4). The contribution of LPS of continental origin to seasonal mean rainfall increases in the future projection (Fig. 4).

Fig. 4.

June–September ensemble mean HIST climatology (contours) and RCP8.5 minus HIST climatology (shading) of (A) total seasonal mean precipitation, (B) BoB LPS-related precipitation, (C) land-based LPS-related precipitation, and (D) 95th-percentile precipitation, all based on HiRAM. Stippling represents the changes that are statistically significant (at 5% level) for all ensemble members. The color bar at the bottom left corresponds to A–C.

These changes in synoptic activity have consequences not only for the mean precipitation but also for the extremes. Flooding over central and northern India is often associated with LPS activity (3, 28). Also, a recent increase in rainfall extremes over central India has been attributed to increased synoptic activity over that continental region (17, 29, 30). Fig. 4 shows the climatology and the difference between RCP8.5 and HIST simulations of the 95th percentile of daily precipitation over the Indian region. The nonorographic precipitation extremes occur most frequently slightly south of the peak swath of synoptic activity (Figs. 1 and 2 ), consistent with the fact that the peak precipitation in LPS typically occurs southwest of the vortex center (1). The future changes in precipitation extremes over India reflect the changes in LPS activity, with a poleward shift clearly evident in the 95th percentile of precipitation (compare Figs. 4 and 2 ). One caveat is that GCMs with parameterized convection typically do not simulate extreme precipitation accurately, so such projections should be treated with caution. Also, we lack a long-term record of observed rainfall over the BoB that could be used to assess the fidelity of historical model simulations in that oceanic region.

Furthermore, the northward shift of the genesis distribution may reduce the amount of warning India has for flood events. In the HiRAM HIST simulations it takes about 3.7 d for an LPS to make landfall after its genesis over the BoB. This time between genesis and landfall is helpful in preparing mitigation strategies for impending floods. As the LPS genesis shifts from ocean to land the Indian region may face shorter preparation times for mitigating the societal impacts of these storms.

Summary and Discussion

Despite the importance of monsoon LPS, future changes in these systems remain poorly characterized and incompletely understood. We have shown that one of the few GCMs to successfully simulate the genesis distribution of Indian monsoon LPS projects a strong weakening and poleward shift of LPS activity in a warming environment. This occurs for all four ensemble members, despite their differing SSTs. An existing statistical model that successfully reproduces the observed global distribution of LPS also predicts a decrease in LPS genesis and furthermore attributes this decrease to a weakening and poleward shift of the large-scale monsoon flow that is widespread in CMIP5 projections. This important consequence of the weakening and poleward shift of the low-level monsoon flow has not been previously appreciated.

The poleward shift in LPS activity was shown to be associated with increased precipitation extremes over northern India in HiRAM. Given the poor skill with which most other climate models simulate LPS activity and associated precipitation, the HiRAM projections may have greater relevance for future changes in regional monsoon climates (31–33). Whether the positive bias in the intensity of LPS simulated by HiRAM compromises the validity of these projections remains unclear, but this bias in LPS properties is substantially smaller than the LPS biases seen in nearly all other CMIP5 models (4). Central and northern India are some of the most intensively irrigated regions worldwide and have water resources that are already stressed (34) and growing populations that are highly vulnerable to hydrological extremes (35). A shift in synoptic activity that dries central India and increases the likelihood of extreme rainfall in northern India (e.g., Fig. 4) would have major societal impact.

Materials and Methods

Monthly varying SSTs from four climate models (Community Climate System Model 4, GFDL-CM3, GFDL-Earth System Model 2G, and Model for Interdisciplinary Research on Climate 5) participating in CMIP5 are used to force the HiRAM model. These models are chosen based on two criteria: (i) their skill in simulating June–September seasonal mean monsoon precipitation over India (22) and (ii) the SST bias over the Arabian Sea, which is found to introduce a dry bias over India (36, 37) (models with weaker bias are preferred). Biases in climate model SSTs can lead to errors when atmospheric models are forced with these SSTs (38). Hence, we applied an SST bias correction (described below). The HiRAM model is configured with a horizontal resolution of 50 km and 32 vertical hybrid pressure levels.

The SST from GCM can be decomposed into monthly climatological mean in the present climate () and anomaly () asThe observed SSTs can also be decomposed and written as anomalies about a present-day climatological mean:Here we used Hadley Center Sea Ice and Sea Surface Temperatures (HadISST1.1; ref. 39). We assume that the bias is in the mean SST field of the model. The bias-corrected SST is obtained by adding the model anomalies to the observed climatology:The SST anomalies from RCP8.5 simulations are added to the observed climatology to get bias-corrected future SST fields. The observed climatology is calculated for 1979–2005. Variability in the SSTs is characterized in and discussed in the . Two types of experiments (transient time slice and ANNC) are performed using HiRAM, as follows.

Time-Slice Simulations.

Historical time-slice experiment (HIST).

In this experiment, the HiRAM model is forced with bias-corrected SSTs for the period 1979–2005. In addition to the SST and sea ice concentrations (SIC) from the CMIP5 coupled models, the standard CMIP5 forcings such as monthly varying greenhouse gases, solar irradiance, aerosol, ozone, and so on are used to drive the model. An ensemble of four simulations is produced, with each simulation corresponding to the SST and SIC from a CMIP5 coupled model.

Late-21st-century time-slice experiment (RCP8.5).

Bias-corrected future SSTs from the four CMIP5 models along with other CMIP5 atmospheric forcings for RCP8.5 are used to drive the HiRAM model to simulate four ensemble members for the 2069–2095 period. To avoid analyzing the model spin-up period the analyses are restricted to 1981–2005 and 2071–2095 for HIST and RCP8.5, respectively.

ANNC Simulations.

Since performing this ensemble of HiRAM simulations is computationally intensive, another set of experiments is run in ANNC mode to quantify model uncertainty. These simulations are forced with ANNCs of 10-y mean SSTs and other model forcings. SSTs from 10 different CMIP5 coupled models (historical and RCP8.5 experiments; see ) are used for these simulations. Three sets of ANNC simulations are carried out with each coupled model SST, where the SST ANNCs correspond to the decades 1971–1980, 1981–1990, and 1991–2000 for the HIST and the decades 2071–2080, 2081–2090, and 2091–2100 for the RCP8.5 scenario. In these experiments HiRAM is integrated for 24 mo, with the first 12 mo treated as a spin-up period.

LPS Tracking.

LPS tracks from HiRAM simulations and reanalysis data are extracted using a tracking algorithm that mimics the conventional detection and tracking of LPS by identifying closed isobars (4). The LPS activity from HIST simulations are compared with those from ERA-Interim reanalysis as well as from the Sikka archive. See for details of the LPS tracking algorithm. The storm intensity is measured by the maximum pressure depth () achieved by the storm during its life cycle. is defined as the difference between the value of outermost closed isobar and the central minimum pressure of the storm. It is also to be noted that here we only consider storms that lasted 3 d or more. The LPS activity is quantitatively represented by an SAI which is track density-weighted by wind speed (17) aswhere x and y are the longitude and latitude of the center of LPS, = = 1.5°, and U is the wind speed magnitude based on the intensity category of the LPS. The values of U are 4.25, 11, 15, 20, and 27.5, respectively, for LPS categories of Low (1 hPa 2 hPa), Depression (2 hPa 4 hPa), Deep Depression (4 hPa 10 hPa), Cyclonic Storm (10 hPa 16 hPa), and Severe Cyclonic Storm ( 16 hPa).

MDGI.

The MDGI is a statistical model that predicts the likelihood of LPS genesis as a function of climatological monthly mean variables (19). The expected number of LPS genesis points is written as a log-linear model,where b is a vector of coefficients multiplying the climatological variables in the vector x, and are the longitude and latitude grid spacing, respectively, is the number of years in the storm count record, and is latitude. The variables in were objectively selected in the original MDGI using cross-validation to avoid overfitting. The functional form of the MDGI permits fractional changes in genesis frequency to be expressed as a linear function of changes in the climatological variables,Using Poisson regression we refit the log-linear model to HiRAM storm counts in the HIST simulations for June–September, giving one set of MDGI coefficients for each of the four ensemble members. All four sets of coefficients were similar to each other and to the previously derived coefficients obtained using global observations of LPS counts for the full calendar year (19), but the HiRAM coefficients had larger standard errors due to the smaller number of storms occurring in the Indian region.