Indian summer monsoon rainfall was examined in two different greenhouse gas emission scenarios: the Special Report on Emissions Scenarios (SRES; B1) and a similar greenhouse gas scenario, the new representative concentration pathways (RCPs; RCP4.5). The rainfall change in the climate model projections through remotely induced changes in precipitation processes and through changes in precipitation efficiency processes was discussed. To that end, two model setups were applied: 1) the regional climate model (RCM) Consortium for Small-Scale Modelling in Climate Mode (COSMO-CLM), nested in the global climate model (GCM) ECHAM5/Max Planck Institute ocean model (ECHAM5/MPIOM), applying the greenhouse gas scenario B1; and 2) the RCM nested in a newer version of the GCM, ECHAM6/MPIOM, incorporating the RCP4.5 scenario. Both GCM simulations showed a slight increase in precipitation over central India toward the end of the twenty-first century. This slight increase was the result of two largely compensating changes: increase of remotely induced precipitation and decrease of precipitation efficiency. The RCM with the scenario RCP4.5 followed this trend, but with smaller changes. However, the RCM with B1 showed a decreasing trend in precipitation because of a slightly larger absolute change of the reduced precipitation efficiency compared to the change caused by the remote processes. Changes of these processes in the scenario simulations were larger than the natural variability, as simulated in an unperturbed preindustrial greenhouse gas control (CTL) climate simulation. Results indicated that the projection of the Indian summer monsoon rainfall is still a key challenge for both the GCM and the RCM.
The evolving uses of global climate (or general circulation) models (GCMs) have given new insight into climate research. Importantly, these models have improved our understanding of the earth’s climate system by focusing attention on the role of synoptic-scale circulation features and land–atmosphere interactions. For instance, efforts have been made to understand the Indian summer monsoon either by perturbing the greenhouse gas (GHG) concentrations or by adjusting or replacing the parameterization modules of the GCM. May (2002, 2004) analyzed present-day conditions and future climates with the help of ECHAM4 GCM–simulated data. The future climate experiments were performed under an enhanced GHG condition. In response to a doubling of the CO2 concentration, May (2004) found an increase in monsoonal rainfall with increases in interannual variability. The warmer sea surface temperature (SST), especially in the tropical Pacific, enhances the evaporation variability, which, in turn, intensifies the variability of Indian monsoon rainfall (Meehl and Arblaster 2003). However, at the same time, a weakening in the monsoonal circulation was found, which has since been investigated and remains a topic of debate (Kitoh et al. 1997; Douville et al. 2000; Ueda et al. 2006; Menon et al. 2013; Ma and Yu 2014).
A major shortcoming of GCMs is their coarse grid spacing, which is generally about 200 km. Hence, the GCMs often miss regional finescale features and are therefore bound to be less effective in representing mesoscale processes. An alternative approach, which is quite popular and efficient nowadays, is the dynamical downscaling technique. In this method, information about large-scale phenomena is taken from a host GCM simulation. This large-scale information provides the boundary conditions with which a high-resolution regional climate model (RCM) is driven. With advanced physical parameterization and adequate representation of local features, such as the orography, land use, and lakes, to provide regionally relevant feedback processes, RCM simulations can add value compared to GCM results (Wei and Fu 1998; Beck et al. 2004; Chang et al. 2009; Asharaf et al. 2012).
South Asia, and particularly the Indian subcontinent, experiences a monsoon season from June to September (JJAS) every year. The monsoon causes most of the rainfall received in India (~75% of the annual amount). The nature of this monsoonal precipitation varies from region to region, and many studies reveal that complex land–atmosphere interactions have a significant impact on the monsoonal characteristics (Medina et al. 2010; Asharaf et al. 2012). The dominant factors behind the variation in monsoonal rainfall are still not very well known.
The Coupled Model Intercomparison Project (CMIP) has provided new insights into missing feedbacks associated with the biosphere and the atmospheric cycle. The current framework, phase 5 of CMIP (CMIP5; Taylor et al. 2012), has been conceived as a strategic approach to improve its predecessor experiment, CMIP3 (Meehl et al. 2007). For instance, CMIP5 incorporates a large set of state-of-the-art models, which run at higher resolutions (both in horizontal and vertical model grids) and with more complete representation of external forcing than the CMIP3 models. Additionally, it contains a number of earth system models (ESMs), incorporating carbon cycle feedbacks (Taylor et al. 2012). These feedbacks are believed to be an important part of climate projections (Giorgetta et al. 2013). The incorporation of new aerosols and the revised scenarios [i.e., representative concentration pathways (RCPs)] in CMIP5 is assumed to provide more reasonable insights than those of the former experiment. Note that the CMIP3 experiment was based on the Special Report on Emissions Scenarios (SRES; i.e., A1, A2, B1, and B2 scenario families), which were well documented by the Intergovernmental Panel on Climate Change in the Fourth Assessment Report (IPCC AR4; Solomon et al. 2007).
Previous transient simulations with RCMs give mixed results for the Indian summer monsoon season. For example, Kumar et al. (2013) found an increasing trend of monsoonal rainfall in the future as a result of warming of the climate. In contrast, Ashfaq et al. (2009) found suppression of monsoonal dynamics and rainfall with increasing GHGs in an A1B scenario experiment. Recently, Dobler and Ahrens (2011), in their projection simulations with the RCM Consortium for Small-Scale Modelling in Climate Mode [COSMO-CLM (CCLM)], found significantly decreasing trends for all-Indian monsoon rainfall, in contrast to the driving ECHAM5/Max Planck Institute ocean model (ECHAM5/MPIOM) projections. They claimed that a change in residence time of water in the atmosphere (which is more prolonged in CCLM than in ECHAM5/MPIOM) alters the trend of rainfall.
Owing to an inconsistency in model projections—for instance, the aforementioned suppression or intensification in future monsoonal rainfall—climate projections have been the subject of continuing scientific interest. The inconsistency in model outputs leads one to wonder about the representation of physical processes and associated feedbacks, which can be crucial on regional scales. For instance, based on coupled atmospheric–ocean climate model experiments, Douville et al. (2002) found that the precipitation efficiency and the water vapor recycling rate over land are key parameters that limit monsoonal rainfall. Recently, Asharaf et al. (2012) showed in their sensitivity simulation that different land–atmosphere feedback processes influence the efficiency of precipitation regionally and that the efficiency process often prevails over direct recycling processes on the Indian subcontinent.
Because of major advances in the fidelity of CMIP5 models, their products have begun to appear in an increasing number of studies in recent years (Lee and Wang 2014; Hsu et al. 2013; Menon et al. 2013). Although both CMIP3 and CMIP5 models have a systematic error—for example, in simulating the June to September time-mean rainfall over the monsoonal region—the amplitude of the error is smaller in CMIP5 relative to CMIP3 models (Sperber et al. 2013). In comparison with observation data, it was observed that the multimodel mean of CMIP5 models had an improved JJAS time-mean rainfall rate climatology than the CMIP3 multimodel mean (Lee and Wang 2014; Sperber et al. 2013). The improvements (i.e., the more realistic rainfall magnitudes) were evident along the Western Ghats and near to foothills of the Himalayas in India. Additionally, the spatial pattern of monsoon onset date and the peak monsoon time in the Indian subcontinent were better simulated in CMIP5 than in the CMIP3 multimodel mean data.
Given the mixed results, it is worthwhile to address how the water cycle and the precipitation efficiency vary and how the resulting monsoonal precipitation trends originate within the framework of a modeled future climate. Possible causes behind the changes in the monsoonal precipitation on the basis of regional and global climate model projections will be addressed in the current analysis. The structure of this paper is as follows. Section 2 presents the setup of the RCM that was applied in our climate projection experiments, along with the experimental design and a description of the methods used. The results and the discussion are presented in section 3, which is followed by the conclusions in section 4.
a. Model and model setup
The simulations were performed with the nonhydrostatic, limited-area climate model CCLM (Steppeler et al. 2003; Dobler and Ahrens 2008; Rockel et al. 2008; Kothe et al. 2010). It includes the multilayer soil model TERRA_ML (Schrodin and Heise 2002). A Kessler-type (Kessler 1969) microphysics scheme including ice-phase processes for cloud water, rain, and snow was employed. The Tiedtke (1989) mass flux approach was used as the convection parameterization scheme. At the lateral boundaries of the model domain, a relaxation scheme according to Davies (1976) was used. More details on the model are given on the community website (www.clm-community.eu).
We used data from three different climate simulations: two with the GHG scenarios SRES B1 and RCP4.5 and one from an unforced control (CTL) experiment. Herein, the downscaled and the driving model data are referred to with suffixes attached to the names of the experiments CCLM and ECHAM/MPIOM (EC), respectively (CCLM_B1, CCLM_RCP45, and CCLM_CTL for the downscaled; EC_B1, EC_RCP, and EC_CTL for their driving data). The initial and boundary conditions were taken from simulations with the coupled atmospheric–ocean general circulation model ECHAM5/MPIOM and the new Max Planck Institute Earth System Model, low resolution (MPI-ESM-LR). The MPI-ESM-LR is a new version of the coupled atmospheric–ocean model (ECHAM6/MPIOM), which includes additionally the carbon cycle feedbacks (Giorgetta et al. 2013). Physical parameters, such as, land use, orography, soil type, and vegetation fraction were used as RCM lower boundary conditions [see Smiatek et al. (2008) for more details about the physical parameters].
For the CCLM_B1 simulation, the driving data were taken from the experiment IPCC-AR4 MPI-ECHAM5 T63L31 MPI-OM GR1.5L40 SRESA1B run no.1: Atmosphere 6-hour values (IPCC-AR4 EH5-T63L31 OM-GR1.5L40 20C 1 6H) in the twentieth century with anthropogenic forcing B1 (EC_B1; Roeckner et al. 2006b). Over South Asia, this data product was successfully tested and downscaled to 0.44° horizontal grid spacing with the CCLM model (Dobler and Ahrens 2010, 2011). It was observed that the COSMO-CLM forced by ECHAM5/MPIOM added value to the spatial precipitation pattern and to the precipitation variability of the ECHAM5/MPIOM simulations. However, in their SRES scenario experiments, they found that the Indian monsoon rainfall trend behaved differently from the driving data. To investigate whether the simulated trend was the result of climate change, natural variability, or model drift, a long (150 yr) RCM simulation (CCLM_CTL) was carried out with constant preindustrial GHG concentrations. This simulation was downscaled from the driving global model ECHAM5/MPIOM (IPCC-AR4 EH5-T63L31 OM-GR1.5L40 CTL 6H; Roeckner et al. 2006a) of the preindustrial control experiment (EC_CTL).
Furthermore, we performed a CCLM simulation (CCLM_RCP45) driven by the MPI-ESM-LR (ECHAM6/MPIOM) model under the revised scenario RCP4.5 (EC_RCP45). Here, ECHAM6 is the sixth generation of the atmospheric GCM of ECHAM. The major changes implemented in ECHAM6 were in the representation of shortwave radiation transfer, the inclusion of dynamic vegetation, a new surface albedo representation, and additional model layers that better represent the middle atmosphere. It was found that the simulated circulations in the extratropics with ECHAM6 were remarkably improved from those generated with the predecessor ECHAM5 (Stevens et al. 2013). However, over the tropics, ECHAM6/MPIOM showed only marginal improvements over the former model performance (Stevens et al. 2013). Note that the motivation behind the choice of the RCP4.5 in the current analysis was the similar GHG emission paths used with respect to the B1 scenario (Knutti and Sedláček 2013).
b. Description of the domains
The CCLM_B1 simulated the climate at a resolution of 0.44° (~50 km) for 147 × 121 grid points on a rotated grid with 20 vertical levels. The CCLM_CTL and CCLM_RCP used the same resolution and rotated coordinates, but a relatively bigger domain (209 × 162 grid points) with 35 vertical levels was used in the CCLM_RCP experiment. Although the number of vertical levels was different in the experiments, the lower-tropospheric layers (especially in the planetary boundary layer) were similar. Therefore, we expect that the different level numbers did not affect the current analysis results, as the moisture during the monsoon season is mainly concentrated in the lower part of the atmosphere. The simulation domain for the CCLM_B1 experiment encompassed approximately 8°S–45°N and 31°–114°E (Dobler and Ahrens 2008), whereas CCLM_RCP and CCLM_CTL used the Coordinated Regional Climate Downscaling Experiment (CORDEX) South Asia domain for the model integrations. These two domains entirely covered the whole Indian subcontinent. While the different domain sizes may have had an effect, such effects are likely not pronounced in cases of mean bias and variability, as was found by the domain size experiments over South Asia of Bhaskaran et al. (1996). A similar study by Beck et al. (2004), but over Europe, revealed that the domain size has a minor effect on simulated precipitation.
c. Model diagnostics
1) Atmospheric water budget
The mass conservation for water vapor in an atmospheric column can be written as
where W is the amount of water vapor stored in the atmospheric column. In the case of long-time averages, changes in the water storage term can be assumed to be zero. The term is the convergence or net inflow (incoming − outgoing) of water vapor into the region. The terms ET and P represent evapotranspiration and precipitation, respectively. With the flux integral approach (Bosilovich and Chern 2006), we calculated the incoming (IN) and outgoing (OUT) water vapor fluxes.
2) Changes in precipitation (the feedback effects)
where the Δ terms indicate differences between the future and present-day time slice simulations, and is the precipitation efficiency. The precipitation efficiency follows that described in Schär et al. (1999), and it is defined as in the analysis regions. The prime denotes the future state.
Following Asharaf et al. (2012), the first term on the right-hand side of Eq. (2) is the changes in precipitation through changed atmospheric precipitation efficiency. This term depicts the indirect contribution, as it does not explicitly depend on the direct water availability (either influx or evapotranspiration process). However, the second and third terms illustrate the precipitation change by direct processes through changed water availability. These are named as surface and remote effects, as the changes mainly depend upon the changes in evapotranspiration and influx of water vapor, respectively.
The analysis was performed on a common analysis grid of the driving model (1.875° × 1.875°), and the selection of the subregion was motivated by the studies of Goswami et al. (2006) and Bhaskaran et al. (2012). As shown in Fig. 1, the subregion was defined over the area 18°–27°N, 75°–88°E, which contained total 48 grid points. This subregion was used for the water budget and the changes in rainfall analyses. The current analysis focused on the JJAS monsoon season, ranging from the year 1971 to 2100. The analysis built on monthly data, though the adopted averaging for moisture fluxes was based on model daily output data.
3. Results and discussion
a. Spatial and temporal analyses
To assess the model simulated JJAS precipitation and its spatial variability for the present-day climate period (1971–2000), we used observation-based gridded data from the Global Precipitation Climatology Centre (GPCC; full data reanalysis product version 6; Schneider et al. 2011) at a 0.5° grid spacing. This product dataset consists of monthly global land surface precipitation, which is based on ground measurements.
Figure 1 shows the monthly mean rainfall of the present-day climate monsoon season. The simulated precipitation pattern generally agreed well with GPCC data. Both the GCM and the RCM showed similar spatial distributions of precipitation to the observation. However, there were significant differences, especially over the mountainous and plain regions. For instance, the RCM showed an overestimation in the Western Ghats (west coast of India) and an underestimation over the central part of India. This is consistent with the results from previous modeling studies (e.g., Dobler and Ahrens 2010; Lucas-Picher et al. 2011). Although heavy rainfall areas were evident in both the GCM and RCM simulations, their magnitudes were amplified in the CCLM simulations. Comparing the CCLM results with their driving GCM simulations, CCLM_B1 failed to reproduce the rainfall pattern over the northern region of Bangladesh. Over the same region, the CCLM_RCP-simulated rainfall was quite similar to its driving GCM data.
The spatial correlation and spatial variability of the models simulated JJAS rainfall with GPCC data for the all-India monsoon region (AIMR) were shown by the Taylor diagram in Fig. 1f. Before the comparison (as shown in the Taylor plot), the model-simulated rainfalls were interpolated to the spatial grid of the observation. Results showed that the standard deviations of the simulations were comparable to the observation data, though the models underestimated the spatial variability. The lower spatial variability in models relative to GPCC can be linked to the rainfall biases over the Ghats and plain regions. However, the rainfall biases were slightly reduced with the CCLM_RCP45, which simulated a standard deviation value close to the observation’s standard deviation, although all models underestimated the precipitation in the water budget analysis region. In comparison with the GCM experiments, the RCM simulations provided additional information: the spatial variability and spatial correlation especially were remarkably improved in CCLM_RCP45 for the AIMR.
It is important to mention here that the emphasis of this study is on the investigation of projected rainfall trends and their physical processes, rather than on model evaluation. Hence, the results, along with the available observation data, are not investigated in depth in the present paper. This was already done by Dobler and Ahrens (2010) and Lucas-Picher et al. (2011) with the same model and at the same grid spacing, and those studies demonstrated that the CCLM reproduces the basic monsoonal features, such as rainfall distribution, wind patterns, monsoon onset, and withdrawal date, reasonably well.
Figure 2 shows the summer monsoon rainfall trends that were computed from the linear fitting method. Both the CCLM_CTL and EC_CTL experiments showed no significant trends (data not shown). However, significant decreasing rainfall trends were detected in some areas, such as in northwestern and central India, with the CCLM_B1 data in contrast to their driving EC_B1 model data. It was explained in a previous study that the attenuated number of depressions, which propagated westward and frequent (compared to the driving model) heavy rainfall events, were the major factors behind the decreasing rainfall trend in CCLM_B1 (Dobler and Ahrens 2011). Over land and especially near the very southern part of India, CCLM_RCP yielded somewhat similar positive trends to those from EC_RCP and EC_B1. However, this agreement was not consistent over the eastern part of India and near Bangladesh, where EC_RCP and EC_B1, respectively, showed significant positive trends.
Figure 3 shows the simulated JJAS monthly mean evapotranspiration values for the present-day climate. The CCLM predominantly estimated lower evapotranspiration values than the EC model. Over land, the differences between CCLM- and EC-simulated evapotranspiration were mostly seen in the central and the northwestern parts of India, which can be attributed to less soil water availability over the regions in the CCLM than in the EC. Although the simulated evapotranspiration trend patterns (Fig. 4) were similar to the rainfall trend patterns (Fig. 2), their values were reduced. A small but significant negative trend emerged in B1 over the central and northern parts of India. The southern part of India showed an increasing trend (about 5%–15% century−1), which covered a larger portion in CCLM_B1 than in EC_B1. In contrast, RCP simulations showed no significant trends almost over the whole AIMR.
Figure 5 shows a combination of the vertically integrated moisture flux and the difference between evapotranspiration and precipitation (ET − P) for the reference period. One can see here that the moisture was transported from the Arabian Sea to the Indian subcontinent. The ET − P generally shows the sources (positive values) or sinks (negative values) of atmospheric water vapor at the surface of the earth. Our simulations illustrated that the Arabian Sea (along the Somali coast) provides a large source of the moisture, whereas surface water sinks were located in the southern Indian Ocean, along with the Western Ghats, the Bay of Bengal, and the region near eastern India (associated with the upper part of Bangladesh). The distributions of sources and sinks were similar in the CCLM and EC simulations. An anticlockwise-rotated moisture flux was evident at the eastern (near to the Bay of Bengal) part of India in both B1 and RCP runs. However, the position of this cyclonic flux in CCLM_B1 differed from its position in RCP and EC_B1, which resulted in changes in the moisture flow to the analysis region. A general explanation for this pattern could be related to the vegetation and leaf area index parameters, which were unalike over the area north of Bangladesh in the CCLM_B1 and CCLM_RCP experiments. As a result of the different biometeorological variables, the surface heat fluxes changed (data not shown), potentially locally affecting the convective activity over the region. This might have contributed to the altered position of the cyclonic structure.
The vertically integrated moisture fluxes increased in the future period for the B1 and RCP4.5 scenario experiments because of a corresponding increase of the atmospheric moisture content (Fig. 6). In CCLM_B1, the strengthened moisture fluxes from the Arabian Sea were deflected near to the analysis region by low pressure systems. The flux patterns were not much changed in the RCP experiments, although the CCLM flux magnitudes differed regionally from the EC moisture fluxes. The ET − P trend, as illustrated in Fig. 7, demonstrated that the moisture source over the Arabian Sea (along the Somali coast) increased, while the trend decreased near the Bay of Bengal region and the southern Indian Ocean region. These behaviors were similar in EC_B1 and EC_RCP45. However, the trend’s intensity and pattern were somewhat different in the CCLM projections. For instance, the increased water source over the Arabian Sea (as evident in the present-day climate period) was smaller in CCLM_B1 than in CCLM_RCP45. Additionally, the ET − P trend was not significant in CCLM_B1 over the Bay of Bengal and associated Bangladesh regions, which was in contrast to the driving EC_B1 model trend.
Figure 8 illustrates the 21-yr running mean of the individual water components over the whole 130-yr time period (here, B1 and RCP45 covered from 1971 to 2100) for the subregion in eastern India. For the sake of comprehensibility, each 130-yr series was divided by a normalized influx factor. The normalized influx factor was computed by taking into account a 30-yr mean EC_CTL influx (i.e., the 30-yr mean influx series of B1 and RCP4.5 were divided by the 30-yr mean EC_CTL influx, respectively). It can be seen from the figure that the simulated precipitation and evapotranspiration showed a substantial difference between EC and CCLM, whereby CCLM underestimated the diagnostic values with respect to the driving model’s values. Interestingly, rainfall increased somewhat at the end of the century in the projections EC_B1, EC_RCP, and CCLM_RCP, but not in the CCLM_B1 simulation, although no significant trends (against a 5% significance level) were detected in the simulations. The influx and outflux of the B1 and RCP simulations showed an intensification at the end of century, which is likely to happen in a warm climate (May 2004).
b. Changes in precipitation and process analysis
Changes in precipitation through the feedback processes [as defined in Eq. (2)] over the subregion (as used for the analysis of the water components) in eastern India for three time slices are illustrated in Fig. 9. To get a robust range of natural variability, a total of six nonoverlapping time slices were selected from the CTL simulations. Each dot shown in the figure represents a difference value. This was taken from arbitrarily chosen 30-yr time slices, whereas the maximum of the absolute value among the six dots expresses the range of the model’s natural variability. The selected reference period for the present analysis was 1971–2000 for the two projections (B1 and RCP45). While the preindustrial CTL simulation has arbitrary calendrical years (here, 2171–2300), it has no actual reference period as in the two scenario experiments. To calculate the differences for the CTL simulation, we first selected four arbitrary nonoverlapping sets (ni, where i = 1, 2, 3, and 4) of each with 30 years (say n1 = 2171–2200, n2 = 2206–35, n3 = 2241–70, and n4 = 2271–2300) of the simulation. In this way, six distinct pairs were produced: pair 1 (n1, n2), pair 2 (n1, n3), pair 3 (n1, n4), pair 4 (n2, n3), pair 5 (n3, n4), and pair 6 (n2, n4). Once the sets were prepared, we simply took the difference from each pair (i.e., n1 − n2, n1 − n3, n1 − n4, n2 − n3, n3 − n4, and n2 − n4). The subtraction could have been performed in the reverse direction. In that case, the sign would have changed, but the magnitude would have remained the same. Nevertheless, in the present analysis we were comparing maximum absolute values of the magnitudes obtained from the six CTL differences. The sign of the resulting magnitude was insignificant in the present context.
Results from the CTL simulations showed a large multidecadal natural variability, which was larger in the EC simulations compared to the CCLM simulations. One could see a mutual change in remote and efficiency effects (Fig. 9). As mentioned in section 2c, here the term “remote effect” is the change in precipitation by direct processes through changed influx, whereas the latter is the indirect contribution through changes in atmospheric precipitation efficiency.
It is likely that the effects in projections will also vary, as large natural variability is inherent in the remote and efficiency effects (see the CTL dots in Fig. 9). The large variations occurred in the B1 and RCP projections, where the remotely driven precipitation increased with time. At the same time, the efficiency of precipitation decreased. The surface effect, which was relatively small, increased as well at the end of the twenty-first century in EC_B1 and EC_RCP45. The increase in the surface effect can be explained by the fact that, as temperature increases, it promotes more surface evapotranspiration. This surplus in evapotranspiration enhanced the surface effect term corresponding to Eq. (2). However, in CCLM experiments, the surface effect was almost negligible at the end of the twenty-first century, which can be attributed to the lessened availability of soil water, leading to less evapotranspiration over the region. Dobler and Ahrens (2011) found that the CCLM simulated more frequent heavy rainfall events than the driving model ECHAM5/MPIOM. The frequent heavy rainfall generally amplifies the runoff process, which, in turn, reduces the soil infiltration rate, yielding less soil water in the ground (Saeed et al. 2013). Obviously, the increase in the remote effect was due to a larger influx entering into the analysis region, which is likely to occur in a warm climate (May 2004). The compensation between remote and efficiency effects, however, was not much affected by the surface effect. For instance, EC_RCP45 at the end of the century yielded a value of +1.5 mm month−1 of the surface effect, whereas the efficiency and remotely induced changes in rainfall were −13.6 and +26 mm month−1, respectively. Here, the magnitude of remotely induced rainfall change was almost 2 times larger than the efficiency effect. Similar fractions and behaviors, but with reduced magnitudes, were also found in CCLM_RCP45. In contrast, the amplitude was reversed in the CCLM_B1 experiment: that is, the absolute value of the decreased efficiency effect (−11.4 mm month−1) was two times smaller than the absolute value of increased remote effect (+5.3 mm month−1). Summation of the aforementioned three effects ultimately yielded a negative value in total precipitation change (−5.8 mm month−1) at the end of the twenty-first century in CCLM_B1. However, this behavior was not seen in the driving EC_B1 model data. The absolute amounts of the remote and surface effects were relatively smaller in EC_B1 than the absolute remote and surface effects of EC_RCP45 (Fig. 9).
The decomposed effects, as shown in Fig. 9, were further compared with the unforced steady-state CTL simulation. Our analysis showed that the simulated increasing trends of the remote effect in the projections B1 and RCP4.5 were statistically significant (tested with a Mann–Kendall test at a 5% significance level) relative to the unperturbed CTL remote effect trends. It was noticed that the remote signal in the projections, with the exception of CCLM_B1, exceeded the range of CTL variability. This was more pronounced in RCP4.5 than in B1 simulations. On the contrary, decreased efficiency effects of the projections consistently tended to stay (except in the second time slice of EC_RCP) within the range of the CTL efficiency effect variability (see data in Fig. 9). This implies that the precipitation efficiency signal was contaminated by internally forced variability. The interpretation of such data was not simple, because the efficiency-generated precipitation was a consequence of natural variability with possible impacts from anthropogenic forcing or climate change.
The decreased efficiency in projection runs may be linked to the shifting of troughs and also to the gradual decline in the number of depressions, as was found in previous modeling studies (Dobler and Ahrens 2011; Menon et al. 2013). The depressions or troughs are generally associated with large-scale processes. To investigate this issue quantitatively, we further computed the Q-vector (analogous to the Quasigeostrophic omega equation; Hoskins et al. 1978; Davies-Jones 1991) convergence. Note that the Q vector is a good diagnostic tool for investigating large-scale processes and stratified precipitation. While the Q-vector analysis showed a reduction in large-scale ascent in the troposphere (data not shown), the reduction was not statistically significant. The Q vector, though, does have some limitations, because it is based on certain assumptions.
Summer monsoon rainfall processes in the Indian subcontinent were analyzed through two different experiments involving the B1 and RCP4.5 scenarios. To investigate these processes regionally, the regional climate model CCLM and the corresponding driving global coupled atmospheric–ocean models ECHAM5/MPIOM and MPI-ESM-LR were used.
In comparison with the available observation data for the present-day climate period (1971–2000), the results showed that the models were, in general, able to capture the spatial distribution of the monsoonal rainfall reasonably well. The RCM added detailed information of the regional scale; the spatial rainfall variability and correlation, in particular, were better simulated than its driving model data.
Considering the projection runs, the CCLM with B1 experiment showed a small but significant decreasing trend in precipitation mostly over the western and northwestern parts of India, which was in contrast to the results from its driving model. On the other hand, the CCLM with RCP4.5-simulated trends were consistent with the results from their driving model. To understand the processes behind the increasing or decreasing future rainfall, the precipitation changes in three different time slices (2006–35, 2041–70, and 2071–2100) of the scenario experiments with respect to the present-day climate (1971–2000) were analyzed. We decomposed the changes in precipitation into three different processes: that is, whether the precipitation results were derived primarily from changes in local evapotranspiration (surface effect), were driven by amplifications in external sources (remote effect), or were due to the efficiency effect. It was inferred from the analysis that changes in the efficiency (efficiency effect) of precipitation will show decreasing trends throughout future time periods according to the CCLM with RCP4.5 and B1; this trend was more pronounced in the driving model. In contrast, the remotely driven precipitation increased with time.
The residence time [i.e., the inverse of the water vapor cycling rate (ratio of change in precipitation to the change in precipitable water; Douville et al. 2002)] is a key factor in diagnosing the Indian summer rainfall. Dobler and Ahrens (2011) explained that the residence time of water in the atmosphere in CCLM is more prolonged than that in the forcing ECHAM5/MPIOM data, which alters the trend of rainfall from increase to decrease. However, decrease in the cycling rate, in general, increases the atmospheric water holding capacity, which, in turn, intensifies the water vapor transport. It was found in previous studies that the moisture transport is partly counterbalanced by the precipitation efficiency over the Indian subcontinent (Asharaf et al. 2012; Douville et al. 2002). Our analysis indicated that the combination of these two strong processes with different signs (i.e., the increasing remote effect and the decreasing efficiency) resulted in a weak overall change in the projected monsoonal rainfall. In CCLM_B1, the decreased efficiency was more pronounced than the leading remote effects. As a result, it yielded less rainfall at the end of century than the present-day climate period. In the CCLM_RCP45, EC_ECP45, and EC_B1 experiments, the increased remote effect is slightly larger than the decreased precipitation efficiency. These different signs of the remote and efficiency effects were in line with the increasing greenhouse gases and aerosols study of Douville et al. (2002).
There is, however, a caveat to the current interpretation. For instance, the uncertainties associated with different experimental frameworks in both the RCM and the GCM and greenhouse gas concentrations (e.g., RCP4.5 and B1) cannot be ignored in present analysis. Moreover, it is likely that small changes from large-scale forcing will induce meaningful changes in the RCM-simulated trends. Yang et al. (2012) demonstrated that significant differences can be introduced in a regional dynamical downscaling because of slight changes to large-scale forcing, although the regional downscaling can provide better regional features. Nonetheless, both model (CCLM and EC) projections showed a significant decrease in precipitation efficiency. Our study revealed that interpretations of changes in precipitation efficiency are not straightforward, as the variability in the efficiency was confined to the natural variability range of the models. The projected rainfall trends for the Indian summer monsoon season remain uncertain and ambiguous, since the remote effect and efficiency effect are the leading factors that drive changes in monsoonal precipitation (Asharaf et al. 2012). Of course, the remote effect has consistently shown increasing trends, which possessed a more robust signal in the RCP4.5 scenario compared to the B1 scenario. Yet the question still remains as to what processes and changes were involved in the simulated future decrease in rainfall efficiency.
The authors 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. We sincerely thank three anonymous reviewers for their constructive comments in improving the manuscript. The COSMO-CLM Community supplied access to and support in using COSMO-CLM. The authors thank the Center for Scientific Computing (CSC) of the Goethe University Frankfurt and the German High Performance Computing Centre for Climate and Earth System Research (DKRZ) for supporting part of the calculations.