Journal of Geophysical Research: Atmospheres

Four climate change scenarios for the Indian summer monsoon by the regional climate model COSMO-CLM



[1] This paper discusses projections of the Indian summer monsoon (ISM) by the regional climate model COSMO-CLM, highlighting similarities to and differences from its driving model, the global atmosphere–ocean model ECHAM5/MPIOM. The ISM is quantified using the all-Indian monsoon rainfall (AIMR) index and two vertical wind shear indices. To investigate the impacts of greenhouse gas emissions on the ISM, four emission scenarios for the time period 1960–2100 (Special Report on Emissions Scenarios A2, A1B, B1, and commitment) are considered. The COSMO-CLM simulations show significantly weakening ISM trends in all indices for emission scenarios A2, A1B, and B1. Parts of northwestern India are projected to face a decrease in the monsoon rainfall amount of over 70% within this century. For the wind shear indices, the projected decreases are mainly due to changes in the upper troposphere winds. The weakening of the dynamics in the COSMO-CLM is in agreement with the weakening in the driving ECHAM5/MPIOM model. The two models further agree in significantly positive trends of atmospheric water vapor contents and rain day intensities. However, ECHAM5/MPIOM shows no decrease in AIMR. The different AIMR trends in the two models are found to be due to different changes in the residence time of water in the atmosphere: In the COSMO-CLM projections, the residence time is more prolonged than in ECHAM5/MPIOM. This again is the consequence of a decrease in the number of depressions moving toward the northwestern parts of India.

1. Introduction

[2] Daily live, agriculture yields and profitability in many countries in the South Asian region are highly influenced by the Indian summer monsoon (ISM). About 75% of the yearly amount of rainfall falls during the monsoon season (that is, from June to September) providing water that is necessary for irrigation, electric power production and drinking water. The amount of rainfall during the monsoon season has a strong influence on the economy of the South Asian region, for instance the severe drought of 2009 in India [Francis and Gadgil, 2010]. A variety of indices have thus been defined to measure and to predict the yearly variations and future developments of the monsoon strength. The most commonly used indices are based on rainfall [Parthasarathy et al., 1992; Goswami et al., 1999], the vertical wind shear over certain pressure levels [Webster and Yang, 1992; Chen et al., 2007; Goswami et al., 1999] or the estimation of convection [Wang and Fan, 1999] in certain areas. While all of these indices are correlated by some degree to each other, there is no single best index in estimating the ISM strength [Wang and Fan, 1999; Goswami, 2000; Wang, 2000].

[3] Most earlier studies on the topic of ISM projections [e.g., Intergovernmental Panel on Climate Change (IPCC), 2007; Annamalai et al., 2007] are based on global circulation model (GCM) simulations. While the GCMs already provide some insight in the large-scale trends, the spatial distribution of the fields involved can be better resolved by the application of regional climate models (RCMs). Furthermore, there are indications that RCMs are able to resolve climate extremes better than GCMs [e.g.,Duffy et al., 2003]. Thus, because of the increasing availability of computational power, RCM projections have become a popular tool for investigating the fine-scale behavior of the climate system in reaction to enhanced greenhouse gas (GHG) emissions [e.g.,Giorgi, 2006; Kumar et al., 2006; Ashfaq et al., 2009].

[4] The different GCM and RCM projections agree in increasing temperatures over the Indian subcontinent and a weakening of the ISM dynamics. However, for the projected changes in precipitation there is much less agreement. While, for instance, Annamalai et al. [2007] and Kumar et al. [2006] found increasing precipitation amounts, Ashfaq et al. [2009] found a suppression not only in the monsoon dynamics but also in the South Asian summer precipitation due to enhanced GHG emissions. However, most of these studies are limited to time slice experiments, simulating about 30 years in the late 20th and 21st centuries, or include one emission scenario only.

[5] In this study, we analyze transient projections of different ISM indices by the RCM COSMO-CLM for the time period of 1960 to 2100. The influence of increased GHG emissions on the ISM strength is investigated by the use of four simulations according to the scenarios A1B, B1, A2, and the commitment scenario as given in the IPCC Special Report on Emissions Scenarios (SRES) [Nakicenovic and Swart, 2000]. To provide further insights in the uncertainties inherent in the model projections, the trends projected by the driving GCM are included in our analysis. The objective of this paper is not only to give an overview on the projected ISM trends but also to find reasons why the trends occur, especially in the case of a disagreement between the RCM and the driving GCM. The ability of the COSMO-CLM to represent the ISM and its capability to provide an added value to the driving GCM during the time period of 1960 to 2000 has been evaluated in a previous study [Dobler and Ahrens, 2010]. However, this study also includes a more detailed evaluation of the added value with respect to the temporal variability of the model index correlations.

[6] The remainder of this paper is structured as follows. In section 2, the COSMO-CLM model and its setup are briefly presented.Section 3 describes the different ISM indices, and section 4 is devoted to the trends projected by the two models based on the four emission scenarios. In section 5 possible reasons for the differences in the precipitation trends among the two models are discussed. Finally, the main findings are summarized in section 6.

2. Model and Model Setup

[7] The COSMO-CLM is a nonhydrostatic RCM based on the Consortium for Small-scale Modeling (COSMO) model (, which is currently used by seven European weather services for their operational numerical weather prediction (NWP). In this work, we applied the COSMO-CLM (version 2.4.11) in a South Asian domain (Figure 1) to simulate regional climate projections within the time period from 1960 to 2100. Details on the model can be found on the CLM web page The main differences between the NWP and RCM version are given by Böhm et al. [2006].

Figure 1.

Simulation domain and the areas used for the different indices. The gray region denotes the area where the 850 hPa surface is below the model orography.

[8] The lateral boundary conditions were provided by the atmosphere–ocean general circulation model ECHAM5/MPIOM [Roeckner et al., 2003] which has a grid spacing of 1.875° (T63). The COSMO-CLM simulations were carried out on a 0.44° rotated grid with 20 vertical layers. Because of the transient simulation of four different scenarios from 1960 to 2100, a higher resolution was not affordable.

[9] The choice of the ECHAM5/MPIOM to provide the lateral boundary data has been based on the work of Kripalani et al. [2007] where 22 GCMs used in the Fourth Assessment Report of the Intergovernmental Panel on Climate Change ( are tested for their performance in the South Asian region. They have shown that the ECHAM5/MPIOM is among the six models which simulate the most realistic 20th century monsoon climate over South Asia. Amongst these six models, they found no best model. Together with the broad experience in driving the COSMO-CLM with the ECHAM5/MPIOM model, the model was the best option as driving model for the regional projections in this study.

[10] Although the COSMO-CLM configuration includes physical parametrizations that are mainly tested in European domains [Hollweg et al., 2008; Jaeger et al., 2008; Kothe et al., 2011], Dobler and Ahrens [2010] showed that by using the same set of parametrizations, the model is still able to improve the spatial distribution of precipitation and wind shear as compared to the ECHAM5/MPIOM driving data in the South Asian region. The model parametrizations include a radiation scheme following [Ritter and Geleyn, 1992], a Kessler-type [Kessler, 1969] microphysics scheme with ice phase processes for cloud water, rain and snow, the Tiedtke convection scheme [Tiedtke, 1989] and a multilayer soil model [Schrodin and Heise, 2002]. Numerical integration was done by a leapfrog scheme using a time step of 240 s.

3. Indian Summer Monsoon Indices

[11] The analysis of trends in the projections of the ISM was carried out for a set of indices. These indices were obtained by averaging the model data over the areas shown in Figure 1 and the monsoon months from June to September. To make the single indices comparable, we standardized the time series with respect to the reference period of 1971–2000. The index approach provides good information on the projected ISM strength but masks the spatial distribution of the projected changes. Therefore, our analysis also includes the linear trends for the fields involved in the index calculations at the model grid points.

3.1. All-India Monsoon Rainfall

[12] The all-India monsoon rainfall (AIMR) index was defined byParthasarathy et al. [1992]as the total rainfall amount from June to September over India excluding four hilly meteorological subdivisions. Its interannual standard deviation is about ten per cent of the long-term average only. Nevertheless, severe floods or droughts have been observed in years with high (low) values [Webster et al., 1998; Krishnan et al., 2003]. A long-time series of observational data for the homogeneous all-India monsoon rainfall [Parthasarathy et al., 1994] is available for the years 1871 to 2009 by the Indian Institute of Tropical Meteorology (IITM) at

[13] As a measure of convective activity, outgoing longwave radiation (OLR) is often used [Wang and Fan, 1999; Dobler and Ahrens, 2010]. However, as both the spatial patterns and the average of the OLR trends over the index region 10°N–25°N, 70°E–100°E (Figure 1) agree well with those of the monsoon rainfall, the results for OLR are not presented in this work for the sake of brevity.

3.2. Meridional and Zonal Wind Shear

[14] The wind shear indices were calculated as the difference between the lower troposphere winds at 850 hPa and the upper troposphere winds at 200 hPa. The meridional wind shear index (MWSI) was calculated over 10°N–30°N, 70°E–100°E. This area includes almost all of India, the Bay of Bengal and a part of the Indian Ocean close to the west coast of India. The zonal wind shear index (ZWSI) was obtained over 5°N–20°N, 45°E–80°E including the region of the Somali Jet and a large part of the Arabian Sea.

[15] Because of the model's domain size, the two wind shear averaging domains (Figure 1) were slightly smaller than the original domains given by Goswami et al. [1999] and Wang and Fan [1999]. To reduce boundary effects, no data within 3.5° (i.e., eight grid points) distance from the boundary was used. Furthermore, winds extrapolated to pressure levels below the ground should be handled carefully. The region where the COSMO-CLM orography is higher than the model's reference atmosphere at 850 hPa is shown inFigure 1. However, only about nine per cent of the grid points within the MWSI area are in this region, and the effects on the overall MWSI are negligible (not shown). Thus, these points were not treated specially.

[16] The reference wind data for the time period of 1948–2009 at 200 and 850 hPa were taken from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis 1 [Kalnay et al., 1996]. The wind fields in NCEP are highly influenced by observations and are in the most reliable class of output variables [Kalnay et al., 1996]. NCEP reanalysis data were obtained from the NOAA/OAR/ESRL PSD (Boulder, Colorado) Web site ( at 2.5° grid spacing.

4. Results

[17] In sections 4.14.4, we show the results of climate projections over the time period of 1960–2100. For a detailed analysis of the different ISM indices as observed and simulated by the COSMO-CLM and the ECHAM5/MPIOM for the current climate (that is,1960–2000), we refer toDobler and Ahrens [2010]where an evaluation of the two models has been carried out to determine whether the regional model improves the GCM simulation. Although there are some biases in the COSMO-CLM data, the application of the COSMO-CLM to the ECHAM5/MPIOM data has been shown to improve the spatial patterns and the temporal correlations of the modeled ISM indices [Dobler and Ahrens, 2010].

4.1. Rainfall

[18] Figure 2shows the 21 year running means of the standardized AIMR. There are large long-time variations in the observations and projections, and the negative trend observed in the past 50–60 years is within the natural variability. However, there is clear evidence of a low AIMR by the end of the 21st century in the higher-emission scenarios of A1B and A2 in the COSMO-CLM. In the B1 scenario, the projected decrease at the end of the time series is less pronounced.

Figure 2.

Centered 21 year running means of the standardized all-Indian monsoon rainfall (AIMR) index for observation data and COSMO-CLM projections according to different Special Report on Emissions Scenarios (SRES) scenarios.

[19] The very low AIMR values at the end of the COSMO-CLM projections are a result of a decrease in monsoon precipitation throughout most of India.Figure 3gives the distribution of the linear trends of monsoon precipitation in the South Asian region according to the A2, A1B, and B1 scenarios. In the northwestern parts of India, the Indo-Gangetic belt and the adjacent regions, a significant decrease is projected, which would have a strong impact on future water availability in that region. For A1B and B1, the drying pattern is similar to A2 but with reduced amplitudes in accordance with the overall AIMR trend. Furthermore, there is some increase in monsoon precipitation at the southeastern edge of the Indian peninsula. The commitment scenario shows almost no significant trends (not shown).

Figure 3.

Linear trends in monsoon rainfall (%/century) during the time period 1960–2100 in (a) the A2, (b) the A1B, and (c) the B1 COSMO-CLM runs. Colored areas show significant trends (at the 5% significance level).

[20] Note that locally, some larger trends appear in A1B than in A2. This is due to the combination of the decadal variabilities and the effects of higher GHG emissions in the A1B scenario at the beginning of the future projections (nearly up to the year 2050). As can be concluded from Figure 2, the linear fit to the A1B projection can result in a larger slope than in the A2 scenario because of the decadal variabilities. Although it is also clear from Figure 2that the assumption of linear trends in the monsoon precipitation is a simplification, it is helpful to summarize the projected trends and their statistical significance. The method also allows to include long-term variations and makes the trends more independent of the choice of a control and scenario period.

4.2. Wind Shear

[21] The 21 year running means of the standardized MWSI and ZWSI are shown in Figures 4 and 5, respectively. For the MWSI, the long-time variations and trends are similar to the AIMR index (Figure 2). For the ZWSI, the variations are much smaller, and the decreasing trends in the A2 and A1B scenario are striking. In the single scenarios, there is a high agreement in the temporal evolution of the wind shear indices and the AIMR index. While the observations show a negative trend for MWSI (Figure 4) in accordance with the observed AIMR trend over the past 60 years, no trend can be seen in the corresponding ZWSI (Figure 5).

Figure 4.

Same as Figure 2 but for the meridional wind shear index.

Figure 5.

Same as Figure 2 but for the zonal wind shear index.

[22] The spatial distributions of the horizontal wind fields in the COSMO-CLM, averaged over the monsoon season from 1971 to 2000, are shown inFigures 6a and 6b for 850 and 200 hPa, respectively. Within the ZWSI area, there is a large shear of the zonal winds visible from 850 to 200 hPa, while the shear of the meridional winds in the MWSI area is small. Furthermore, a counterclockwise rotation can be seen to the west of Bangladesh at 850 hPa, and a clockwise rotation at 200 hPa over the Himalayan ridge.

Figure 6.

Divergence (colored areas) and directions (arrows) of wind fields at (a) 850 hPa and (b) 200 hPa for the monsoon season from 1971 to 2000 as simulated by COSMO-CLM. The meridional wind shear index (MWSI) and zonal wind shear index (ZWSI) areas are given by solid and dotted rectangles, respectively. The white area in Figure 6a shows the region where the 850 hPa surface is below the model orography.

[23] Figures 7a and 7bshow the changes in the wind fields from the COSMO-CLM A1B projection as linear trends over the time period 1960–2100. The major wind changes at 850 hPa within the index areas are a decrease of eastward winds in the southern part of the ZWSI area. These changes have a size of about 2–3 m/s in A2 and A1B (Figure 7), and about 2 m/s in B1 (not shown). The reduced MWSI and ZWSI projected at the end of the 21st century are thus mainly due to changes in the upper tropospheric winds. They result in a decrease of the southward shear in the eastern part of the MWSI area and a decrease of westward shears in the ZWSI area (Figure 7). At 200 hPa, there is also an increased convergence visible over the Bay of Bengal (Figure 7). In all four scenarios, the distribution of the trends is again similar, with the generally highest amplitudes in A2 followed by A1B, B1 and almost no trends in the commitment scenario (not shown).

Figure 7.

Linear trends (per century) in the divergence and directions of the winds at (a) 850 hPa and (b) 200 hPa during the time period 1960 to 2100 in the COSMO-CLM A1B simulation. Colored areas and arrows show significant trends (at the 5% significance level). In Figure 7a, the area where the 850 hPa surface is below the model orography is white.

4.3. Index Correlations

[24] A teleconnection between the ISM and the El Niño Southern Oscillation (ENSO) is well documented [Walker, 1923; Rasmusson and Carpenter, 1982; Ju and Slingo, 1995] and has often been reported to have weakened in recent decades [e.g., Kripalani and Kulkarni, 1997; Kumar et al., 1999]. The weakening is evident in the observation data (Figure 8). As can be seen, the 21 year sliding explained variance (R212) of AIMR by the NINO3.4 index drops below 0.1 at the starting year of 1989. During the whole time period of observation (starting in 1871), no value of R212 below 0.1 can be found before 1989 to 2009 (not shown). For the ten to 14 year sliding R2, however, the data show values above 0.1 for the last few years (Figure 8), indicating a new strengthening of the teleconnection. Note that sliding windows below 10 years are excluded.

Figure 8.

Explained variance (R2) in the observations of AIMR by Niño 3.4 (top right) and by MWSI and ZWSI (bottom left) for different sliding window sizes (left and right axes) over the time period from 1948 to 2009. Top and bottom axes denote the starting year of the sliding window. The black arrow illustrates the 21 year running correlation between AIMR and Niño 3.4 starting in 1948.

[25] To create Figure 8, the NINO3.4 data were obtained from the Climate Prediction Center, NOAA (United States), Web site ( The data set starts in 1871 and is updated continuously. To derive the values of the explained variance, a linear regression has been carried out [see, e.g., Ahrens, 2003].

[26] Figure 8 also includes the explained variance in the observed AIMR by ZWSI and MWSI for different sliding window sizes using a linear regression. Here too, a clear decrease of R212 during the last decades is evident, but no further decrease is apparent in the last few years.

[27] The sliding explained variances of the COSMO-CLM A1B run are given inFigure 9for the years of 1960–2100. Like the observations, the COSMO-CLM simulation shows values ofR212below 0.1 for AIMR explained by NINO3.4. Note that the NINO3.4 index for the COSMO-CLM projections has been calculated using the data from the ECHAM5/MPIOM because the NINO3.4 averaging region is outside the regional simulation domain. Thus, the correlation between the AIMR and the NINO3.4 index in the COSMO-CLM is expected to be smaller than that in the driving model. This biases this evaluation to some extent.

Figure 9.

Same as Figure 8but for the COSMO-CLM A1B simulation over the time period from 1960 to 2100.

[28] For the explained variance of AIMR by ZWSI and MWSI only, there are no such limitations. During the time period common with the observations (1960–2009), the resulting R2is 0.63 for the COSMO-CLM and 0.55 for the observations. Over the complete time series, the 21 year sliding explained variance shows a mean value of 0.64 for COSMO-CLM and 0.57 for the observations with a minimum (maximum) value of 0.44 (0.83) for COSMO-CLM and 0.27 (0.79) for the observations.

[29] Including the MWSI, ZWSI and Niño 3.4 in the linear model to explain the total variance in AIMR was also tested but provided no significant increase. The results for the 21 year variances are about 0.01 to 0.03 higher than in the model using only ZWSI and MWSI.

4.4. GCM Projections

[30] Contrary to the COSMO-CLM, the ECHAM5/MPIOM projections show an increase in AIMR for all scenarios by the end of the 21st century (Figure 10). Figure 11 shows the linear trends of monsoon precipitation in the GCM A1B simulation over the time period 1960–2100. Comparing the GCM and RCM results (Figure 3), one can see that in parts of Pakistan the GCM also projects a decrease in precipitation, but no significant trends appear in most of northwestern India. For Bangladesh and the eastern parts of India, the two models even disagree in the sign of the trends. However, they agree in an increase over the southern tip of the Indian peninsula and the ocean.

Figure 10.

Centered 21 year running means of the standardized AIMR index for observation data and ECHAM5/MPIOM projections according to different SRES scenarios.

Figure 11.

Linear trends in monsoon rainfall (%/century) during the time period 1960 to 2100 in the ECHAM5/MPIOM A1B run. Colored areas show significant trends (at the 5% significance level).

[31] These differences also appear in the A2 and B1 scenario, resulting in a slight increase of the AIMR in the GCM simulations, while the COSMO-CLM runs show a clear decrease (Figure 2). For the wind shear indices, the trends in the GCM projections generally agree in sign and amplitude with the corresponding RCM runs (Figures 12 and 13).

Figure 12.

Same as Figure 10 but for the meridional wind shear index.

Figure 13.

Same as Figure 10 but for the zonal wind shear index.

[32] Considering the explained variances in AIMR, the values in the ECHAM5/MPIOM A1B run are generally higher than in the COSMO-CLM A1B run (Figure 14), especially for small sliding window sizes. For instance, R212 of AIMR by ZWSI and MWSI shows a mean value of 0.66 with a maximum value of 0.93 from the year 1960 to 2100. For the current climate period (1960–2009), the resulting R2is 0.84 for the ECHAM5/MPIOM. Thus, while both models show an overestimation of the explained variance, the overestimation in the COSMO-CLM is smaller than in the ECHAM5/MPIOM. For the other SRES scenarios, and for the inclusion of the NINO3.4 index, the reduction in the index correlations is comparable (not shown). This is also in agreement with the findings ofDobler and Ahrens [2010] for the present climate where it has been shown that the COSMO-CLM improves the temporal correlations of the modeled ISM indices compared to the driving ECHAM5/MPIOM.

Figure 14.

Same as Figure 9 but for the ECHAM5/MPIOM A1B simulation.

5. Discussion

[33] Because of an increase in the AIMR and a simultaneous decrease in the other indices at the end of the GCM simulations, very low values of R2 appear in the future climate projections, when using a sliding window size of over 80 years (Figure 14). This phenomenon of simultaneously increasing precipitation and weakening monsoon dynamics is often called the wind-precipitation paradox and has been observed and discussed in GCM projections before, for instance, byKitoh et al. [1997], Ueda et al. [2006], and Fan et al. [2010]. Basically, the reduced circulation over India is compensated by an enhanced water holding capacity of the warmer atmosphere, resulting in an overall higher moisture transport and a larger amount of precipitable water over India. Thus, while the COSMO-CLM and the ECHAM5/MPIOM differ in the changes of the AIMR, the models agree in the increase of the precipitable water (Figure 15). This is a common behavior in global warming experiments [Sabade et al., 2011] where, following the Clausius-Clapeyron relation, a warming results in an increased atmospheric water vapor content.

Figure 15.

Changes in precipitable water (%) during the monsoon season from the time period 1971–2000 to 2071–2100 in (a) the COSMO-CLM and (b) the ECHAM5/MPIOM A1B scenario.

[34] Meehl and Arblaster [2003] and Emori and Brown [2005] have shown that because of the increased atmospheric water vapor content a generally higher precipitation intensity will occur. Figure 16shows the changes in precipitation intensity from the time period 1970–2000 to 2071–2100 in the COSMO-CLM and ECHAM5/MPIOM A1B runs. As expected, both models project an increase in the precipitation intensity over large parts of India, including the eastern parts. Thus, although we do not find the wind-precipitation paradox in the COSMO-CLM projections, the same changing relationship between the dynamics and the intensity of precipitation over India is evident.

Figure 16.

Same as Figure 15 but for the rain day intensity.

[35] The disagreement of the two models in the AIMR trends and the general agreement in the rain day intensity changes over India point to different changes in the rain day frequency. As can be seen in Figure 17, the patterns of the rain day frequency changes highly resemble the monsoon rainfall trends. The main differences between the two A1B projections are located over the eastern parts of India, as for the monsoon rainfall trends.

Figure 17.

Same as Figure 15 but for the rain day frequency.

[36] However, the increased amount of precipitable water (Figure 15) also increases the midtropospheric stability [Fan et al., 2010], preventing moist air from rising to levels where it may rain out [Cook et al., 2006]. As shown by Ojo [1970] for Nigeria, a lack of precipitation formation can be caused by the absence of mechanisms to release the precipitable water, even within a moistened atmosphere.

[37] Recently, Francis and Gadgil [2010]analyzed highly convective events over South Asia and showed that the generation of low-pressure systems over the head of the Bay of Bengal and their westward propagation did not occur in 2009, which lead to a deficit in the AIMR by 23% compared to the long-term mean. They suggested that the sea surface temperatures (SST) of the Bay of Bengal and the eastern equatorial Indian Ocean lead to the phenomenon that convection was only maintained for three to four days over the Bay of Bengal and the usual movements of the convective systems did not occur.

[38] As the ISM precipitation trends are varying drastically between the COSMO-CLM and the ECHAM5/MPIOM, we have investigated the lack of precipitation formation mechanisms in the COSMO-CLM projections.Figures 18 and 19 show the convective events (OLR < 200 W m−2) at 85°E for the COSMO-CLM and ECHAM5/MPIOM A1B run during the months May to August in the time periods 1971–2000 and 2071–2100. In the COSMO-CLM projections, decreases over the head of the Bay and between the coastline and Himalayan foothills occur, similar to the observations of 2009 [Francis and Gadgil, 2010]. The ECHAM5/MPIOM shows a more or less constant number of convective events.

Figure 18.

Daily variations of outgoing longwave radiation (W m−2) at 85E for the months May to August during the time periods (top) 1971–2000 and (bottom) 2071–2100 in the COSMO-CLM A1B run. The gray lines indicate the location of the coastline and the beginning of the Himalayan foothills in the model.

Figure 19.

Same as Figure 18 but for ECHAM5/MPIOM.

[39] As most of the precipitation during the ISM is convective, the mean break period between two convective events is directly proportional to the average residence time of water inside the atmosphere (τ). Over India, τ (calculated as the ratio between precipitable water and precipitation according to the work by Douville et al. [2002]) shows an increase of 45% in the COSMO-CLM and an increase of 21% in the ECHAM5/MPIOM. In the COSMO-CLM, the increase in the northwestern part of India is about 100%, while in the ECHAM5/MPIOM, the increase is around 50%. This suggests that the enhanced moisture in the northwestern part of India in the COSMO-CLM is less frequently released than in the ECHAM5/MPIOM model because of the absence of disturbing systems. Note thatτ is increasing all over India in both models, with the exception of a small area at the southern tip of the peninsula (not shown).

[40] Looking at Figure 7a, it can further be seen that in the COSMO-CLM A1B run there is a significant decrease of westward winds at the foothills close to Nepal and Bhutan, where a cyclonic circulation is visible in the 20th century (Figure 6). This confirms the assumption that fewer low-pressure disturbances moving westward from the Bay of Bengal are entering the Indian continent than in the current climate, reducing the release of precipitable water and thus increasingτ.

[41] Interestingly, the six dynamically weakest monsoon years in the NCEP reanalysis 1 data (1957, 1972, 1979, 1992, 2002, and 2009) yield an average amount of precipitable water over India of only 5% below the 1971 to 2000 mean, while the AIMR (based on IITM data) is 16% below normal and τ is increased by 14% (about 1 day). For the four years with severe droughts (1972, 1979, 2002, and 2009), τ is increased by 20% and for the year 2009 only, even by about 23%.

[42] These results show that τ is an essential parameter for the annual rainfall amount in India, and that it is highly affected by the ISM dynamics. Further investigations in this direction, including for instance more model simulations, reanalysis or observational data, would be desirable to provide a more certain estimate on the magnitude of the ISM dynamics effect on τ and yield more insights into the physics behind it. However, this is out of the scope of this paper.

[43] Although an RCM approach with a tested setup [Dobler and Ahrens, 2010] and four different SRES scenarios have been used in this study, there are some limitations. First, a broader ensemble of different GCMs and more runs per GCM and scenario would provide more information on the uncertainties in the projected changes and their variability. However, more simulations with the forcing ECHAM5/MPIOM have shown that a higher number of runs from each scenario does not produce a notable difference in the spread of the different ISM index projections and the internal variability (not shown). Second, the tested setup shows some nonnegligible biases [Dobler and Ahrens, 2008, 2010; Lucas-Picher et al., 2010] of which we can only assume, that they are constant in the model projections and thus removed by the standardization of the indices. However, a similar study using the RCM RegCM3 [Ashfaq et al., 2009] suggests that the simulated change of the ISM rainfall is robust to changes in the driving GCM, vertical resolution and initial conditions. This is further supported by the similarity of the results found therein and in this study, using very different models and setups.

6. Summary and Conclusions

[44] The present study explores possible changes in the ISM on the basis of regional and global climate model projections, using three ISM indices. For the 21st century, all three indices show decreasing trends for the SRES scenarios A2, A1B, and B1 in the regional climate model COSMO-CLM. Generally, the trends are most negative in A2, followed by A1B and B1. Almost no trends can be found in the commitment scenario, suggesting a negative influence of GHG emissions on the ISM strength. In the global coupled atmosphere–ocean model ECHAM5/MPIOM the trends are similar to the COSMO-CLM simulations with exception of the AIMR. Here, the ECHAM5/MPIOM shows small positive trends in all scenarios.

[45] The use of transient climate simulations from 1960 to 2100 allows us to include long-term variations in the analysis. Although there are large variabilities in all time series, the trends in rainfall and meridional and zonal wind shear are statistically significant in many regions of the simulation domain. For the northwestern part of India, with a seasonal mean rainfall of only about 1–2 mm d−1in the present-day climate, the COSMO-CLM simulations show negative trends in the ISM precipitation amount of up to more than −70% per century.

[46] An investigation of the explained variance in the AIMR by the meridional and zonal wind shear and by the ENSO has been carried out. It shows that the currently observed low predictability of the AIMR [e.g., Kumar et al., 1999] is below the simulated minimum values, and that the average explained variances in both models are generally higher than in the observations. However, the values in the COSMO-CLM are lower than those in the driving ECHAM5/MPIOM model. This shows that for the current climate, the index correlations are improved by the regional model. Nevertheless, this also shows that in both models essential parts in the interactions of dynamics and physics which affect the ISM are missing. In the COSMO-CLM this may partly be the lack of an atmosphere–ocean coupling: The negative feedback between low precipitation over northwestern India and enhanced sea surface temperatures over the Bay of Bengal, which then would lead to more low-pressure disturbances and thus enhance the precipitation over northwestern India again, is not included. This may also contribute to the diverging trends of the AIMR projections in the two models.

[47] Despite the fact that the two models investigated in this work differ in the AIMR projections, they highly agree in the projection of increased precipitable water and precipitation intensity. The different AIMR projections are thus mainly due to different changes in rain day frequency. The positive trends in rain day intensity with increasing GHG emissions is a robust signal among the two models.

[48] Our investigations further show that the residence time of water inside the atmosphere is a key factor for the ISM rainfall amount. Thus, a steady rain day frequency is crucial for the area's hydrology and economy. This should be kept in mind when analyzing projections of precipitation over India. Looking at trends of precipitation rates only, different changes in the residence time of water inside the atmosphere (and thus in the precipitation efficiency) are partly masked by a common increase in precipitable water.


[49] This work was partly funded by the EC project BRAHMATWINN, contract 036592 (GOCE). The authors also acknowledge funding from the Hessian initiative for the development of scientific and economic excellence (LOEWE) through the Biodiversity and Climate Research Centre (BiK-F), Frankfurt am Main. NCAR is sponsored by the National Science Foundation (NSF). Access to and support in using the COSMO-CLM was supplied by the COSMO-CLM community. Computational time was provided by the German Climate Computing Centre (DKRZ) and the Center for Scientific Computing (CSC) of Goethe University Frankfurt.