1 Introduction

The central Asia (CA) is one of the largest arid and semi-arid areas in the world, with an area over 4 × 106 km2, across Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, and Uzbekistan and part of northwest China (Huang et al. 2014). In CA, water is particularly scarce and its ecosystems, economy and society are extremely venerable to the variations of rainfall received. Yet, its climate is highly sensitive to global climate change (Huxman and Smith 2001; Whitford 2002). Some studies argued that CA maybe one of the regions which is likely to be strongly affected by global warming (Solomon et al. 2007; Chen et al. 2009).

Indeed, there are studies showing some remarkable regional features of observed climate changes in CA. Chen et al. (2009) showed that the observed warming intensity in CA exceeds the averaged surface temperature changes over the Northern Hemisphere during the past 100 years. Furthermore, recent studies showed that its annual rainfall is enhanced in the last several decades. This is contradictory to the drying trend over the subtropical land regions projected by climate models (Chen et al. 2011; Huang et al. 2012). Several studies documented in details that in the Xinjiang province in northwest China, which is located in the eastern part of CA, its summer rainfall has experienced significant decadal changes and such changes are different from the changes occurred in eastern China (Shi et al. 2002). Nevertheless, studies focusing on questions such as how and why the rainfall changes in the region in past and future climate are still very limited. This has become the purpose of this study which is aimed to explore three questions: (1) Can current climate models reproduce the observed rainfall changes over CA and the associated dynamical processes? (2) How well do the models agree on their projected rainfall changes over CA under warmed climate? (3) How large is the uncertainty in rainfall projections from current climate modeling experiments and what are the likely reasons leading to this?

It is well known that tropical oceans are one of the fundamental drivers of the global climate. Inthe past several decades, observations have shown that global oceans become warmer (Zeng et al. 2001; Gille 2002; Levitus et al. 2005). In particular, the whole Indian Ocean (IO) has experienced significant warming during the last half century (Chambers et al. 1999; Alory et al. 2007; Du and Xie 2008). Some studies even claimed that the SST warming in IO was the strongest and most robust warming signal among the global oceans (Lau and Weng 1999; Hoerling et al. 2004). Many studies have demonstrated that the warming in IO not only has dominant effects in south Asian monsoon regime but also can have far-reaching impacts on regions including Asia, Africa and Australia (Clark et al. 2000; Ashok et al. 2001; Giannini et al. 2003; England et al. 2006).

Furthermore, recent studies showed that the IO SSTs can affect large-scale circulations which are closely related to rainfall over CA. For instance, Kawamura (1994) showed that the SSTs anomalies in IO can influence climate variability in middle latitude of Asia. Correlations analysis found the summer rainfall in east part of CA has a close relationship with the spring and summer SSTs in the IO (Yang et al. 2010; Zhou et al. 2015). Composite analysis showed that summer rainfall in the east part of CA increased during the years of earlier onsets of the south Asian summer monsoon (Xu and Qian 2006). Studies also showed that rainfall over central and southwest Asia is also derived from tropical moisture sources and is well associated with tropical climate variability (Mason and Goddard 2001; Mariotti et al. 2005). Statistical results showed that water vapor from the IO plays an important role in generating summer rainfall over east part of CA, especially in heavy rainfall (Yang and Zhang 2007; Bothe et al. 2012; Zhao et al. 2014).

It is widely acknowledged that large uncertainty exists in current climate model rainfall projection at regional scales. There are a limited number of studies focusing on the potential changes in the summer rainfall over CA under global warming. Based on the outputs from a few selected CMIP5 models, Wu et al. (2013) and Huang et al. (2014) reported significant uncertainties in the model–projected rainfall changes over CA. After selecting a few models with relatively good skills for simulating current rainfall climatology, they reported future rainfall increases across most part of CA in these models. However, it is yet unclear the causes of such rainfall increases and to what extent they are related to the warming in the IO. The current climate projections tend to suggest that the Asian monsoon circulations are to be weakened under global warming (Fan et al. 2010, 2012; Sooraj et al. 2014). Thus, it is not well understood about the combined effects from a warmed IO and a weakened tropical monsoon circulation on the potential rainfall responses over CA. Although observational analysis showed that a weakened south Asian summer monsoon could result in the cooling in middle and upper troposphere over CA and furthermore generate favorable anomalous circulations for generating more summer rainfall in east part of CA (Yang et al. 2009; Zhao et al. 2014), we do not know if the same processes operate in the models.

Therefore, in this study, we first evaluate the skill of current climate models in reproducing the dynamical processes of how the tropical IO SSTs affect the rainfall over CA. Then we assess if the models can reproduce the linkage between the warming in the IO and the rainfall increases over CA. Following these evaluations, we then select models to assess if the models with reasonable skill in capturing these fundamental observed features can offer reasonable agreement in their projected changes of rainfall over CA. Accordingly, the manuscript is arranged as follows: the descriptions of observational and modeling data used in this study are given in Sect. 2. The model-simulated relationship between summer rainfall across CA and SSTs in the IO is evaluated in Sect. 3. The projected changes in summer rainfall over CA and its response to the IO warming are shown in Sect. 4. Results from a climate model’s SST sensitivity experiments are also included in the analysis to examine the processes identified in the CMIP5 model results. Section 5 summarizes the main conclusions from this study and discusses the weakness and further work needed to strengthen our further understanding of the climate and its projects in CA.

2 Datasets used in the analysis

The datasets used in this study are shown as follows:

  1. (1)

    Observational data: As high quality observed rainfall and temperature station data over CA are not readily available, in this study we use the Climate Research Unit (CRU) version 3.10 monthly data which covers global land with a horizontal resolution of 0.5° × 0.5° (Mitchell and Jones 2005). This data is widely used in climate and climate change studies (Houghton et al. 2001; Jone et al. 2001). In this analysis, SST data for the period of 1959–2004 come from the monthly 2° × 2° SSTs by Reynolds et al. (2002) of the US National Oceanic and Atmospheric Administration (NOAA). The ERA-40 reanalysis (Uppala et al. 2005) during 1959–2001is used to reveal the large-scale atmospheric circulation patterns responsible for the impacts of IO SSTs on CA climate. We have re-examined the results using the National Centers for Environment Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis dataset (Kalnay et al. 1996) and results show very similar features and therefore not reported here.

  2. (2)

    Climate model data: Monthly precipitation, surface temperature, zonal and meridional winds from historical and RCP8.5 experiments of 25 CMIP5 coupled climate models (Table 1) are used in this study. These are the models used in the analysis of Dong et al. (2015) and Zhang et al. (2015) in which they assessed the CMIP5 model-simulated changes in the onset/retreat of the Australia–Asian monsoon. When designing the study, we used the same models for allowing us to potentially link changes in tropical monsoon climate with rainfall changes over CA. The model outputs from the historical and RCP8.5 experiments for the periods of 1961–2004 and 2051–2100 are used in this analysis. For all models and experiments, only one ensemble run is used and the model outputs are interpolated onto the same horizontal resolution of 1.5° × 1.5° using a bilinear interpolation method. More details about the experimental designs of the historical and RCP8.5 experiments can be found in Taylor et al. (2012).

    Table 1 The 25 CMIP5 models used in the analysis
    Full size table

To support our analysis of the CMIP5 model results, we have conducted a set of SST sensitivity experiments using one of the CMIP5 model (ACCESS1.3, Bi et al. 2013). The model details are readily available in the CMIP5 model documentation. Here we only brief three AMIP-type SST experiments conducted using the atmospheric model of ACCESS1.3: ACCESS-his uses the global SSTs simulated from the coupled ACCESS1.3 CMIP5 historical experiments for the period of 1971–2000; in ACCESS-rcp85 we super-impose the ACCESS1.3 fully coupled model-simulated changes in global SSTs by its RCP8.5 experiment for the period of 2071–2100 onto the ACCESS-his, together with changes in atmospheric greenhouse gases as in RCP8.5; ACCESS-no IO uses the same SST warming as in ACCESS-rcp85 but with no warming over the tropical IO (with a domain of 20°S–30°N, 40–100°E). ACCESS-only IO uses the SST warming as in ACCESS-rcp85 but SST warming only occurring over the IO. These experiments help us to examine whether IO warming contributes to the rainfall increases in the CA.

Summer is the main rainy season in most part of the CA. It accounts for a large proportion of its annual rainfall totals. For example, in the east part of CA the summer rainfall can accounts for 40–50 % of annual rainfall received (Zhang and Deng 1987). So in current study, we only focus on analyzingresults in the boreal summer season June–July–August (JJA).

3 Observed and model-simulated linkages between IO SSTs and CA rainfall

Complex terrains occupy the CA region, including high mountain ranges and flat plains. Prior to the analysis, we firstly assess the spatial homogeneity of summer rainfall variations in the CA domain. Figure 1a shows the correlations of area-averaged rainfall over eastern part of the CA (with the domain of 73–95°E, 35–50°N) and rainfall over the whole CA during 1961–2004. Positive correlations occupy most of the region, and high correlations are largely located in the domain of 65–95°E, 35–50°N where the zonal westerly dominates (Lioubimtseva et al. 2005). Above result suggests the summer rainfall variations are by and large homogeneous in space across most of the CA despite of regional contrasts in their long term means. Based on the results above, we now define a summer precipitation index (SPI) over CA as regionally averaged summer rainfall over the domain of 65–95°E, 35–50°N.

Fig. 1
figure 1

a Correlations between regionally averaged rainfall over the domain of 73–95°E, 35–50°N and rainfall over central Asia in summer, b correlations between regionally averaged rainfall from the domain of 65–95°E, 35–50°N and SSTs in Indian Ocean in summer, c interannual variations of summer rainfall averaged over the domain of 65–95°E, 35–50°N during 1961–2004 and SSTs averaged over the domain of 50–75°E, 5–25°N during 1961–2004, together with correlations between the two with/without detrended data

Full size image

Figure 1b shows the SPI correlations to SSTs in the tropical IO, with statistically significant results being marked. It illustrates that the SPI is well correlated to the SSTs in tropical IO. The correlations are higher in the Arabian Sea than in the Bay of Bengal. This is consistent with other observational analyses (e.g., Yang et al. 2010). Furthermore, Fig. 1c shows the times series of the interannual variations of summer rainfall averaged over the domain of 65–95°E, 35–50°N over CA and SSTs averaged over the domain of 50–75°E, 5–25°N in IO during 1961–2004. Clearly, summer rainfall over CA is closely linked to the SSTs in the tropical IO, with the two having a correlation of 0.45. Although the figure exhibits rainfall increasing trend since the late 1970s when the IO experiences substantial warming, the correlation does not reduce much after removing the long term trends in these time series (with correlation coefficient being 0.43 with detrended data). This suggests the linkage between the two is strong at both interannual and decadal time scales.

How do the SST anomalies in the IO influence the summer rainfall over CA? Previous studies suggested that this is largely through the atmospheric circulation responses to the SST anomalies in the tropics (Zhao et al. 2014; Zhou et al. 2015). Here, we examine the likely underlying processes by asking two questions: how IO SSTs affect the regional circulations which determine the dynamical conditions for rainfall generation; the other is how IO SSTs affect the atmospheric moisture condition which is the thermodynamic condition for rainfall generation. First of all, Fig. 2a displays the correlations of regionally averaged SSTs of Reynolds et al. (2002) over the domain of 50–75°E, 5–25°N in the IOas in Fig. 1c and the horizontal wind averaged between 850 and 700 hPa (representing the lower tropospheric circulation) in the ERA-40 reanalysis data. It also shows the horizontal convergence calculated from the wind correlations with the SSTs. It is obvious that there are significant atmospheric responses to the SSTs in the Arabian Sea region. Warmer SSTs there are favorable for the formation of a strong cyclonic circulation, enhanced low-level convergence and therefore strengthened upward motion in the tropical IO region. This suggests SSTs are the driving force for the atmospheric dynamics in the domain, rather than being passive responses to atmospheric forcing. To its north, there is a giant anti-cyclonic flow anomaly over the domain around 20–30°N and 35–70°E. The anti-cyclonic system can largely be viewed as the corresponding descending branch of the enhanced ascending motion in the tropical IO. This anomalous pattern corresponds to a significantly enhanced southerly flow over the longitudinal band of 35–45°E. In the eastern part of the tropical domain, one can be seen a strong anti-cyclonic pattern occupying most of the Indian subcontinent, the Bay of Bengal and to the south of the Tibetan Plateau. Associated with this pattern, a strong southerly flow dominates the subcontinent and extending further north up to the low-land region between the Iranian Plateau and Tibetan Plateau. Together, these two enhanced southerly branches are important for conveying tropical moisture into the subtropical region. Although big terrains block the moist air in the east of 70°E, Zhao et al. (2014) showed that in between the Tibetan Plateau and the Iranian Plateau most part of valley is below 1500 m and the southerly anomalies can transport the moist air into CA through this valley. To the north of these two anti-cyclonic patterns, there is a cyclonic pattern around the domain of 30–50°N and 40–65°E. The anomalous southwesterly wind associated with the trough pattern over CA can further transport the moist air originated from the IO into CA for generating more summer rainfall. This is the so-called two-step process as revealed in Zhao et al. (2014).

Fig. 2
figure 2

a Low-level wind (averaged between 850 and 700 hPa) from ERA-40 reanalysis data (Uppala et al. 2005) for the period of 1959–2001 correlating to the same IO SST index defined in Fig. 1c. Shaded colours are the horizontal convergence calculated from the wind and SST correlations, b correlations between the IO SST index with the volumetrically integrated precipitable water from ERA-40 reanalysis for the same period

Full size image

Figure 2b further shows the correlations between volumetrically integrated atmospheric precipitable water from the ERA-40 reanalysis to the same area-averaged SSTs used in Fig. 2a. Associated with the SST warming in the Arabian Sea region, the atmosphere is further moistened in the CA region. The agreement between circulation patterns in Fig. 2a and the moisture conditions in Fig. 2b is reasonably well, suggesting the moistening process can be attributed to direct consequences of the atmospheric circulation responses to the SST forcing in the IO. Furthermore, over CA, not only is the atmospheric moisture condition much improved (Fig. 2b), but also the low-level convergence is strengthened (Fig. 2a) corresponding to the cyclonic circulation pattern over CA. Such a nice combination of improved dynamical and moisture conditions leads to high correlations between IO SSTs and CA rainfall as revealed in Fig. 1.

Our analysis is consistent with many previous studies on the important role of IO SSTs on CA rainfall. Synoptic analysis from Zhang et al. (2012) showed that two classic synoptic situations corresponding to Xinjiang rainfall generation: either (1) the trough embedded in the middle latitude westerly is deep enough extending far southward to bring tropical moist air into the CA region; or (2) the middle latitude trough is supported by another subtropical system which is able to transport the tropical moist air into the middle latitude trough region. There are plenty of trough/ridge developments along the middle latitude westerly due to the baroclinic nature of the atmosphere in the region, but it is moisture condition limiting the summer rainfall received. This is the reason why throughout this analysis, we focus on the moisture source in explaining rainfall variations and changes in CA. Following the observational analysis, we now use such information to assess whether CMIP5 models can reproduce such a connection between IO SSTs and summer rainfall over CA and the associated processes shown in Figs. 1 and 2. Figure 3 shows the correlations of regionally averaged rainfall from the eastern part of CA (the domain of 73–95°E, 35–50°N) and rainfall across a much larger CA domain in the 25 models in their historical runs (1961–2004). Positive correlations occupy most part of the CA in a majority of the models, which are similar to Fig. 1a. High correlations are roughly located in the domain of 65–95°E, 35–50°N as observed (Fig. 1a).

Fig. 3
figure 3

Correlations between regionally averaged rainfall from the domain of 73–95°E, 35–50°N and rainfall over central Asia in JJA in 25 CMIP5 models’ historical run for the period of 1961–2004. Statistically significant results at 95 % confidence level are marked by dot points

Full size image

Furthermore, in order to assess the model-simulated relationship between summer rainfall over CA and IO SSTs, Fig. 4 gives the correlations of regionally averaged summer rainfall over the domain of 65–95°E, 35–50°N and SSTs in IO. Again, statistically significant correlations are highlighted in the figures. Although a large number of the models show some positive correlations, the magnitude of such correlations varies significantly across the models. Such cross-model variations could have significant implications on their projections of the changes in CA rainfall under global warming which will be discussed in the following section. In addition, the correlation is generally higher in the Arabian Sea than in the Bay of Bengal. This is also seen in Fig. 1b. Therefore, in the following analysis, we will mainly focus on the analysis of SSTs in the Arabian Sea.

Fig. 4
figure 4

As Fig. 3, but for correlations between regionally averaged rainfall from the domain of 65–95°E, 35–50°N and SSTs in Indian Ocean

Full size image

Above results only give the correlations of regionally averaged summer rainfall over the domain of 65–95°E, 35–50°N and SSTs in IO. It is likely that some models may have some skill in simulating such a linkage but the location of high rainfall-SST correlations may be away from the domain selected in Fig. 4. Thus, to give a more comprehensive picture about the connections between SSTs in the IO and rainfall over CA, Fig. 5 displays the correlations of regionally averaged SSTs from the domain of 50–75°E, 5–25°E in the Arabian Sea and summer rainfall over a much larger domain over CA. Indeed, in most models, high correlations are all located in the domain of 65–95°E, 35–50°N which are quite consistent with the distribution in Fig. 1a. Nevertheless, the extent of such correlations varies among the models, with some models giving much higher correlations than others.

Fig. 5
figure 5

As Fig. 3, but for correlations between regionally averaged SSTs from the domain of 50–75°E, 5–25°N and rainfall over central Asia

Full size image

To further examine if the rainfall-SSTs correlations in the models are caused by the same processes identified in the observational results in Fig. 2, we have calculated the correlations between regionally averaged SSTs from the domain of 50–75°E, 5–25°E in the Arabian Sea and the horizontal wind averaged between 850 and 700 hPa as used in Fig. 2a. We found (not shown) that in the 25 models, their skills in reproducing the dynamical processes identified in Fig. 2 vary significantly. Some of them show reasonable skills, while others give large differences from observed results.

To objectively quantify the model skills in reproducing the connections between IO SSTs and CA rainfalls and the underlying processes, we have combined two measurements in our model selecting exercise: one is the correlation between CA rainfall and SSTs over 50–75°E, 5–25°N as shown in Fig. 1c, and the other is the correlation between the same SSTs data with the meridional wind component in the area of 15–30°N, 35–45°E where in Fig. 2a we can see significant circulation signal responding to the IO SSTs conditions. Figure 6 shows the correspondences between the two correlation coefficients in each of the 25 models. We have found out of the 25, seven models are able to reproduce statistically significant correlations as observed in Figs. 1 and 2. These models are BNU-ESM, CanESM2, CMCC-CMS, GFDL-ESM2M, IPSL-CM5B-LR, MIROC5 and MPI-ESM-MR. Some of the models in our selection have also been selected in Huang et al. (2014) in which they selected CMIP5 models based on their skills of reproducing observed rainfall mean climatology, rather than on the processes governing the rainfall variations as applied in this analysis.

Fig. 6
figure 6

Solid line shows the correlations between summer rainfall averaged over the domain of 65–95°E, 35–50°N and SSTs averaged over the domain of 50–75°E, 5–25°N in the 25 CMIP5 models’ historical experiment for the period of 1961–2004, Dashed line shows the correlations between SSTs averaged over the domain of 50–75°E, 5–25°N and the meridional wind component in the area of 15–30°N, 35–45°E. Seven models marked by dots are the ones showing both positive correlations passing 95 % confidence level

Full size image

In Fig. 7, it is obvious that these seven models, by and large, show overall circulation patterns as observed (Fig. 2a). Anti-cyclonic flow patterns seen in Fig. 2a over north Arabian Sea and its northwest and over the Bay of Bengal region, to a varying extent, are presented in these model results. The ensemble averages from the seven models offer satisfactory agreement with the patterns shown in Fig. 2a. Nevertheless, there are some notable inter-model differences. In the CanESM2 model, the cyclonic flow over CA region is not as clear as seen in other models. The results from GFDL-ESM2M to MPI-ESM-MR are largely similar to others but over the Bay of Bengal, there is an anomalous cyclonic pattern instead. The MPI-ESM-MR also shows relatively weak cyclonic flow patterns over CA. In the MIROC5 model anomalous anticyclones present over north Arabian Sea and Bay of Bengal, but its location is further northward. Despite of these inter-model differences, the 7-model ensemble results in Fig. 7 illustrate the overall following common features: over tropical IO, there is an enhanced cyclonic flow patterns. To its north, anti-cyclonic flow patterns correspond to enhanced southerly flow which carries warm and moist air north of the IO. At the same time, there is a cyclonic anomaly over CA which can further transport such tropically originated moisture into the CA region and favor rainfall generation. They are in good agreement with observational results presented in Fig. 2 and other studies (e.g., Zhao et al. 2014).

Fig. 7
figure 7

Correlations between regionally averaged SSTs over the domain of 50–75°E, 5–25°N and horizontal wind averaged between 850 and 700 hPa in seven selected models (see text for details). Shaded regions present over 95 % significant level. The last panel is the 7-model ensemble averages

Full size image

4 Projected change in summer rainfall over CA

In Sect. 3, we have focused on assessing the CMIP5 model skills in reproducing the observed influence of SSTs in tropical IO (in particular over the Arabian Sea) on the CA summer rainfall and the underlying processes. In this section, we analyze the model-projected changes in future and uncertainties in these projections. Several recent studies suggested that rainfall over CA may continue to increase in the twenty-first century based on the model outputs from CMIP5 (Wu et al. 2013; Huang et al. 2014), but less attention has been paid to explore the mechanisms leading to such projected changes. Besides, none of these studies have investigated the likely causes of uncertainty in these models. Here, we concentrate our analysis on the linkages between IO SST warming, tropical monsoon circulation changes and rainfall responses over CA under greenhouse warmed climate.

As one can expect, Fig. 8 shows that all the models simulate SST warming in IO under the global warming. However, the SST warming is not uniform in the Arabian Sea and Bay of Bengal. The SST warming in Arabian Sea is stronger than in the Bay of Bengal in most models. This is consistent with the observed changes in the IO in the last several decades (Roxy et al. 2014). We now explore whether these can be used to explain the model-projected changes in rainfall.

Fig. 8
figure 8

Changes in SSTs (°C, 2051–2100 minus 1961–2004) in JJA over Indian Ocean by 25 CMIP5 models

Full size image

Figure 9a displays the projected change in summer rainfall over CA (averaged over the domain of 65–95°E, 35–50°N) from the 25 models between the rainfall climatology averaged over 2051–100 in their RCP8.5 runs against the one derived from their historical runs over the period of 1961–2004. Despite of the warmed IO in all the models, changes in summer rainfall show remarkable differences among the models. For the 25 models, 14 of them show increased summer rainfall while the other 11 models suggest rainfall decreases. If one does not separate these models in terms of their skills in simulating the processes affecting summer rainfall over CA, a simple multi-model ensemble average yields a very weak change due to the offsetting results among the models. Such results are very misleading. In contrast, using the seven models selected in Sect. 3, such uncertainty becomes much smaller. Five out of the seven models show the summer rainfall will increase, with only two models (CMCC-CMS and MPI-ESM-MR)) showing much weaker decreases (Fig. 9b). Based on the fact that these models have reasonable skill in reproducing the important process of IO SSTs in modulating the rainfall in CA, results derived from such models are more convincing and reliable.

Fig. 9
figure 9

a Scatter plots of the changes in regionally averaged rainfall (mm day−1, 2051–2100 minus 1961–2004) from the domain of 65–95°E, 35–50°N in x-axis and regionally averaged SSTs (°C, 2051–2100 minus 1961–2004) from the domain of 50–75°E, 5–25°N in y-axis in 25 models, b as a but for seven selected models

Full size image

Figure 10 further shows the spatial features of the changes in summer rainfall over CA in the seven selected models. Five of the seven models illustrate overall rainfall increases over CA region, particularly over a large part of the Xinjiang Province in northwest China (about 73–95°E, 35–50°N). Most of the changes are statistically significant with t test. Note that although the magnitude of these changes is small, they are huge changes in terms of percentage to the very low rainfall climatology received. Most of the changes here are more than 50 % of their corresponding seasonal means (not shown). MPI-ESM-MR is the only model showing overall rainfall reduction in most of the region. Despite of the fact that area-averaged results in Fig. 8 suggest CMCC-CMS simulates rainfall reduction, it actually has rainfall increases over part of the northwest China.

Fig. 10
figure 10

Changes in summer rainfall (mm day−1, 2051–2100 minus 1961–2004) over central Asia by seven selected models. The last panel is the seven-model ensemble averages

Full size image

For the seven models which have shown their reasonable skills in modeling the influence of SSTs in the Arabian Sea on the summer rainfall over CA, the dynamical processes leading to such rainfall increases are also consistent with each other. By and large, they all show (Fig. 11) that SSTs warming in the Arabian Sea and tropical IO is associated with an anti-cyclonic flow pattern in the region. The enhanced southerly flow brings warm and moist air up further north into the CA region. For the MIROC5 model, although it simulates an anti-cyclonic flow pattern in the Arabian Sea and western part of the IO, a cyclonic flow dominates over the Bay of Bengal and this flow prevents the southwest flow over the Arabian Sea penetrating further northeastward. As a result, in this model, summer rainfall increases are limited in lower latitudes than in other models. Note that for the two models (CMCC-CMS and MPI-ESM-MR) with averaged rainfall decreases over CA in Fig. 8b, the enhanced southerly flow can also be seen in part of the region. Therefore, there may be other important factors of rainfall generation in these models over CA which offsets the effect of enhanced moisture source from the tropics and leads to a weak rainfall reduction in the region. Within the context of our analysis, we are not able to explore this further to discuss the exact causes of such detailed inter-model differences.

Fig. 11
figure 11

Changes in horizontal wind averaged between 850 and 700 hPa (m s−1, 2051–2100 minus 1961–2004) in summer by seven selected models. The last panel is the seven-model ensemble averages

Full size image

Nevertheless, based on our detailed process-based analysis by comparing observed and model-simulated processes influencing the summer rainfall in CA, our confidence is enhanced that under global warming, the CA is likely to receive more summer rainfall due to the dynamical and moisture transport processes demonstrated in these skillful models.

Overall results in this section demonstrate the importance of analyzing the underlying physical/dynamical processes in the climate models when deriving regional climate change projections. It is possible one model may give a good rainfall climatology or simulate a good rainfall correlations with SSTs or other important climate drivers, but such results do not necessarily mean the underlying processes supporting such results are realistic in these models. Therefore, results from this study question the selection of models based on their rainfall/temperature climatology alone.

As described in Sect. 2, to complement the analysis presented above, in this study, we have further conducted four SST sensitivity experiments using a local climate model named ACCESS1.3 which also participated in CMIP5. The differences between ACCESS-rcp85 and ACCESS-his represent the climate changes with warmed global SSTs and increased atmospheric greenhouse concentration. Similarly, the differences between ACCESS-onlyIO and ACCESS-his represent the changes caused by warmed IO and increased atmospheric greenhouse concentration. In contrast, the differences between ACCESS-no IO and ACCESS-his reflect the changes caused by increased atmospheric greenhouse concentration but no IO SST warming. Finally, the difference between ACCESS-onlyIO and ACCESS-no IO can be attributed to the influence of IO warming on the regional climate projections.

Figure 12a shows the seasonal evolution of monthly rainfall changes averaged over the longitudinal band of 65–95°E between the four sets of experiments. In Fig. 12a, with the increases in greenhouse gas concentrations and globally warmed SSTs, the model displays a reduction in tropical summer monsoon rainfall over the latitudes of 0–20°N during the April to August period. To its north, there is a weak rainfall increase. If only superimposing the warming in the IO (without SST warming in other ocean basins), the model results in Fig. 12b not only suggest a much broader and more enhanced rainfall increases in the tropical domain, but the rainfall increases are expanded further northward into the middle latitudes, up to 35°N in July–August-September. In contrast, without the IO warming (but SST warms over other ocean basins), the whole tropical latitudinal band is dominated by rainfall decreases and the rainfall decreases extending up to the latitudes around 30°N (Fig. 12c). By calculating the differences between ACCESS-onlyIO and ACCESS-no IO, Fig. 12d shows that the IO SST warming leads to the rainfall increases in the tropics and extending further northward to 30–35°N. In addition, over the central Asian domain latitudinal band (35–60°N), along the increase in tropical monsoon rainfall, a rainfall increase is clearly presented in Fig. 12d. This modeling results suggest that IO SST warming is an important factor leading to rainfall increases in CA. Note that rainfall is reduced (along the band of 30–40°N during May to August period) between the increases in tropics and the increases in CA in Fig. 12d. This can be explained by the enhanced subtropical descending motion as discussed in Figs. 2a and 7 in which we see enhanced tropical Indian monsoon tend to lead to enhanced anti-cyclonic circulation to its north in the subtropical region. This is also consistent with the two-step moisture transport mechanism discussed in Sect. 3, with the rainfall increases in CA being an indirect consequence of tropical monsoon variations responding to tropical SST forcing.

Fig. 12
figure 12

a Zonally averaged (65–95°E) changes in monthly rainfall between ACCESS-rcp85 and ACCESS-his SST experiments (mm day−1). See text in Sect. 2 for experimental details, b as a but for ACCESS-only IO and ACCESS-his, c as a but for ACCESS-no IO and ACCESS-his, d as a but for ACCESS-only IO and ACCESS-no IO

Full size image

Furthermore, several pieces of additional information can be obtained from these particular modeling results in Fig. 12. (1) IO SST warming is a driving force for the atmospheric circulation variations, rather than a passive response to regional atmospheric circulation variations caused by the forcing outside of the IO domain (i.e. tropical Pacific SSTs). Warmed IO SSTs enhances the Indian monsoon and its cyclonic circulation pattern in the tropics; (2) Without the IO warming, much warmed SSTs over tropical central Pacific Ocean (with an El Nino-like SST warming pattern) could lead to significant rainfall decreases in the IO basin (Fig. 12c). The IO warming therefore acts to offset a large part of such influence from the tropical Pacific; (3) It is well known Indian monsoon is generally weak in climate models including ACCESS1.3. In particular, they tend to lack of northward penetration of monsoon rainfall into the Indian subcontinent. Several studies have suggested cooling biases in the Arabian Sea (e.g., Levine et al. 2012) contributed to such modeling results. Our experiments here also tend to support this claim as IO warming in this model leads to enhanced tropical monsoon rainfall and its northward penetration. Although ACCESS1.3 is not part of the seven models selected in this analysis, with a weak Indian monsoon in its climate simulations (Bi et al. 2013), the processes identified in this model’s SST experiments here are still valid for complementing our CMIP5 model analysis. In future studies, we plan to further examine the processes revealed here using a different model and with more carefully designed SST experiments (such as separating the SST warming in the Arabian Sea and the Bay of Bengal).

5 Conclusion and discussions

Rainfall is very low over CA. Thus, small rainfall changes in this region can have significant social, economic, and ecological consequences. Existing studies showed that one of the most important social and political risks associated with climate change pertains to water availability, especially over CA (Dinar and Dinar 2003; Wolf et al. 2003). So understanding the variations of rainfall in the region in future warmed climate became very important. Nevertheless, large uncertainty exists in current climate model projected change in rainfall over CA (Huang et al. 2013). In this study, we have focused on exploring the impacts of IO warming on the summer rainfall over CA.

In recent decades, SSTs over the IO become warmer, especially in the Arabian Sea (Chambers et al. 1999; Roxy et al. 2014). Meanwhile, rainfall also has experienced an increased trend over CA (Chen et al. 2011). Some observational studies suggested that summer rainfall over CA could be closely related to the SST anomalies in IO, with moist air being transported to CA under suitable circulations (Yang et al. 2010; Bothe et al. 2012; Zhao et al. 2014; Zhou et al. 2015). In order to confirm the relationship, we have firstly analyzed the correlations between the summer rainfall over CA and SSTs in IO in 25 CMIP5 models in their historical experiments for the period of 1961–2004. After that, we have explored the dynamical processes leading to the impact of SST warming in the Arabian Sea on the projected change of summer rainfall over CA in the second half of twenty-first century in these models RCP8.5 experiments. Furthermore, we have conducted ACCESS1.3 SST sensitivity experiments to re-examine the processes derived from the CMIP5 models.

Correlation analysis with observational data shows that summer rainfall is well correlated to the SSTs in northern IO. This is largely contributed to the two-step moisture transport process as discussed in Zhao et al. (2014). While a majority of the CMIP5 models show similar correlations as observed, the magnitude of the correlations varies significantly across the models. Furthermore, most models tend to show that the SST anomalies in the Arabian Sea have higher correlations than ones in the Bay of Bengal. Thus in the analysis we have concentrated on the impacts of the SSTs warming in the Arabian Sea.

Based on the correlation analysis, we have selected seven out of the total 25 models with relative better statistical relationships between CA rainfall with IO SSTs and regional circulation connections with IO SSTs as revealed by observational analysis. Most of the models can capture the observed circulation features associated with summer rainfall over CA (Yang and Zhang 2007; Zhao et al. 2014) and SSTs in IO. In these models, the SST warming in the Arabian Sea is favourable for an enhanced southerly flow over the west part of the IO which transports warm and moist air from the IO into the lower lands between the Iranian Plateau and the Tibetan Plateau. At the same time, there is a cyclonic circulation pattern over CA which brings these moist air masses further north into the CA region.

Our results also showed that under global warming, the SSTs in the Arabian Sea increase in all the 25 models, but the changes of summer rainfall over CA present remarkable differences. Of the 25 models, about 14 of them projected rainfall increases in CA while the rest 11 models projected the opposite changes. When using the seven models objectively selected based on their skills in capturing the influence of IO on CA rainfall, the uncertainties are largely reduced: five out of the seven models suggested the rainfall will increase over CA under warmed climate. Above result confirms the importance of exploring and understanding physical mechanisms presented in the climate models when deriving future rainfall changes at region scale. Results from our SST sensitivity experiments further support our claims. The four sets of ACCESS1.3 experiments have shown that IO SST warming is favourable for the enhanced tropical Indian monsoon and its northward penetration.

Yet, there are several weaknesses in our analysis. In the current study we only focused on the Arabian Sea because it was highly correlated to the summer rainfall over CA. We don’t distinguish it from the effects caused by SSTs gradients in the Arabian Sea and Bay of Bengal. In addition, the SST warming is a global phenomenon. Besides the Arabian Sea, the SSTs in north Atlantic and the warm pool of northwest Pacific are also related to the summer rainfall over CA (Yang et al. 2010). Although we have conducted ACCESS1.3 SST experiments as our first attempt to separate such impacts, it is likely that such results may be model dependent. Further work is need to confirm such results using different climate models and with more carefully designed experiments. It is well known land–air coupling is another important processes in the CA climate, with modelling studies from Koster et al. (2004) showing a strong land-air coupling in this region. We have not explored the issue that to what extent the uncertainty in these model projections are also caused by different land-surface processes in current and future climate. All these will be pursued in our further studies.