1 Introduction

According to Intergovernmental Panel on Climate Change (IPCC) AR5 (IPCC 2013), the global mean surface temperature of the earth has been shown to be increased by about 0.85 °C from 1880 to 2012 with an increase of about 0.72 °C from 1951 to 2012. More specifically, each of the last three decades has successively been the warmest on record. They also have very likely been the warmest in the last 800 years. Based on different observed data sets, there is a general increase in annual global precipitation over the period 1901–2008. However, in recent five decades, the precipitation is decreased marginally on global scale.

In recent years, atmosphere–ocean general circulation models (AOGCMs) have been used to project the climatic consequences of increasing atmospheric concentrations of some greenhouse gases (McGuffie and Henderson-Sellers 1997; IPCC 2001). In areas where coasts and mountains have significant effects on local weather, scenarios based on global models are unable to capture the local level details needed for assessing impacts at national and regional scales. A GCM downscaled to regional climate model (RCM), therefore, is the best tool for obtaining detailed information on a particular region (Giorgi and Hewitson 2001; Jones et al. 2004). The RCMs provide scientifically sound method to add fine-scale details to simulated patterns of climate variability and change. They resolve the local land surface properties such as orography, coastal regions, and vegetation better than the parent GCMs. GCM projections can be used for description of large-scale climate features both over lands and oceans and do not provide necessarily good results over flat terrains (e.g., Miklos et al. 2010). For Indian subcontinent, the narrow coastline over western parts and the immediate north–south-oriented Western Ghats are very crucial. Many studies have shown that GCMs fail to simulate the climate over these regions (e.g., Kripalani et al. 2007). RCMs also resolve internal regional climate variability through better resolution of atmospheric dynamics and processes.

Using the RCMs for downscaling climate change projections from GCMs is an area of active research since 1989 (Giorgi and Bates 1989). The climate change signal over regions characterized by complex topographical and land–ocean features is likely to exhibit fine-scale structure that can be captured only by high-resolution climate models. The number of studies that attempt to characterize expected climate change has been increasing substantially due to a large progress in both global and regional modeling at medium to high resolution. Therefore, the projection on future climate using RCMs leads to substantially improved assessments of a country’s vulnerability to climate change, which enables policy makers to design adaptation strategies.

Many recent studies have used the Providing REgional Climate for Climate Studies (PRECIS) simulations to generate climate change scenarios over India. Using PRECIS simulations, Yadav et al. (2010) reported a general increase in seasonal precipitation (June to September and December to March) over India towards the end of twenty-first century. They have also shown warming of more than 3.5 °C (5 °C) under B2 (A2) scenario in annual average temperatures over India. Climate change scenarios for the period 2021–2050 (referred as 2030s) indicate an overall warming over the entire Indian land mass (Krishna Kumar et al. 2011). The net increase in annual temperatures in 2030s with respect to the present, i.e., 1961–1990 (referred as 1970s), ranges between 1.7 and 2.2 °C, with highest maximum temperatures increasing by 1–4 °C, with maximum increase over coastal regions, and most of the regions are projected to experience an increase in summer monsoon precipitation in 2030s with respect to 1970s, and the increase may be maximum in the Himalayan region and minimum over the northeastern region. The duration of heavy precipitation events are likely to increase by 5–10 days over the Indian land mass. Krishna Kumar et al. (2011) have shown that substantial rise is expected in the mean annual surface air temperature (∼4 °C) and summer monsoon precipitation (∼15%) over India towards the end of the twenty-first century relative to the baseline climate of 1970s. They identified that the intensity of rainfall on a rainy day is likely to be more in the future. The temperature extremes indicate that both the daily maximum and minimum temperatures may be intense in the future under global warming conditions. Changes in the nighttime temperature extremes may be more intense than those of daytime temperatures. Rao Koteswara et al. (2014) have examined projected changes in some extreme precipitation indices using PRECIS simulations. They have indicated an increase in the intensity of rainfall on wet days towards 2071–2100 (referred as 2080s) under A1B scenario. The analysis shows that the precipitation per wet day might be more intense by 10–40% over most of India except southern parts of peninsular India. In recent study, Rajendran et al. (2013) have used time slice simulations of Meteorological Research Institute, Japan-Atmospheric General Circulation model version 3.2S (MRI-AGCM3.2S) under global warming scenario and showed the likely widespread but spatially varying increase in rainfall over interior regions of peninsular, west central, central northeast, and northeast India (∼5–20% of seasonal mean).

Apart from India, PRECIS simulations have also been used to develop climate change scenarios for neighboring regions like Bangladesh (Md. Nazrul Islam 2009) and Pakistan (Siraj ul Islam et al. 2009). Rahman et al. (2012) have used RegCM3 and developed the rainfall and temperature scenarios for Bangladesh for the middle of twenty-first century. The study indicates the likely increase of about 35% in the monsoon rainfall and 0.5 to 2.1 °C warming of mean surface air temperature for different months over Bangladesh towards 2050s.

The possible change in the rainfall and temperature characteristics under global warming scenario will definitely have impact over Indian agriculture as well as health sector. Naresh Kumar et al. (2014) have analyzed the climate change simulations using GCM-MIROC3.2.HI as well as RCM-PRECIS to study the impacts on wheat production over India. The study projects likely reduction in the wheat yield in the range of 6 to 23% by 2050s and 15 to 25% by 2080s. Same models are used to study the regional variability of rice production over India (Naresh Kumar et al. 2013), and the study indicates the reduction in the rain-fed rice yields in India by ∼6% in the 2020s and only marginal decrease (<2.5%) towards 2050s and 2080s. The increase in maximum temperature may increase the malaria incidences in northern states of India and Himalayan region (Bhattacharya et al. 2006).

In the present study, we assess the projected changes in temperature and precipitation from three Quantifying Model Uncertainties in Model Predictions (QUMP) simulations (designated as Q0, Q1, and Q14) of regional climate model (PRECIS) under IPCC-AR4 A1B emission scenario over five homogeneous regions of India (Parthasarathy et al. 1994) (Fig. 1). A continuous simulation of PRECIS over the period 1961–2098 has been used to develop climate change scenarios over time slices 2011–2044 (designated as 2020s—near future), 2041–2070 (designated as 2050s—middle of the future), and 2071–2098 (designated as 2080s—far future) for A1B scenario. The model and the observed data are described in Sect. 2. Section 3 gives an evaluation of the model skills and biases. Variability of the regional climate under global warming scenario is presented in Sect. 4. Results are summarized in Sect. 5.

Fig. 1
figure 1

Map showing homogeneous regions over India viz. northwest (NW) region, central northeast (CNE) region, northeast (NE) region, west central (WC) region, and peninsular (PEN) region. The hilly region is not considered in the present study

Full size image

2 Data

2.1 PRECIS model data

PRECIS is a state-of-art regional climate modeling system developed at the Hadley Centre for Climate Prediction and Research, UK. It is an atmospheric and land surface model of limited area and can be locatable over any part of the globe. The PRECIS RCM is based on the atmospheric component of HadCM3 with substantial modifications to the model physics (Gordon et al. 2000). The atmospheric component of PRECIS model is a hydrostatic version of the full primitive equations. There are 19 vertical levels. The model equations are solved in spherical polar coordinates. PRECIS has a horizontal resolution of 0.44° × 0.44° longitude/latitude. Due to its fine resolution, the model requires a time step of 5 min to maintain numerical stability. PRECIS has 360 days calendar, i.e., each month has 30 days.

A single model simulation provides one realization of climate; however, it does not indicate the uncertainty in model projections. A range of different model simulations provides a better understanding of how the differences in model formulations can lead to uncertainty in the projections. This can be achieved with two possible approaches; One is to use a multimodel ensemble (MME) where different modeling centers contribute their GCM simulations, to generate a range of future climate projections. Another approach is to perturb physical parameters and produce a range of future climates based only on one parent climate model; this is called a perturbed physics ensemble (PPE) approach (Murphy et al. 2004; Stainforth et al. 2005). Parameters are identified that are known to be uncertain and important for the model response to changing greenhouse gas concentration levels.

The Met Office Hadley Centre has run a 17-member perturbed physics ensemble called QUMP (the name of the project under which they were developed) based on the HadCM3 global model (Collins et al. 2006). The individual ensemble members are referred to as Q0–Q16, where Q0 has the same parameter values used in the standard HadCM3 global model (Murphy et al. 2009). The perturbed members Q1–Q16 are numbered according to the value of their global climate sensitivity; thus, Q1 has the lowest global average temperature response to a given increase in atmospheric CO2, and Q16 has the highest (Jones et al. 2012). Based on a preliminary evaluation of these 17 global runs (viz. Q0, Q1,....., Q16) (Kamala 2008), for their ability to simulate the gross features of rainfall and temperatures over India, the LBCs of three QUMP simulations viz. Q0, Q1, and Q14 were made available by Hadley Centre, UK, as these three ensembles give comparatively better simulation of the present summer monsoon climate over India. Hence for the present study, these three PRECIS simulations with LBCs from global QUMP runs were carried out at Indian Institute of Tropical Meteorology (IITM), Pune, India, for the continuous period of 1961–2098 for a domain extending from about 1.5° N–38° N and 56° E–103° E and are utilized to generate an ensemble of future climate change scenarios for the Indian region.

2.2 Observed data

The daily gridded (1° × 1° longitude/latitude) rainfall data set (1951–2007) prepared by India Meteorological Department (IMD) (Rajeevan et al. 2006) based on well-distributed 1803 stations over Indian region is used for the evaluation of model. The daily gridded (1° × 1° longitude/latitude) temperature data for the period 1969–2007, prepared by IMD (Srivastava et al. 2008) using 395 well-distributed stations, are used for validation of the temperature simulations.

3 Validation of precipitation and temperature

Parthasarathy et al. (1994) analyzed the summer monsoon rainfall series for the period 1871–1990 for 29 subdivisions over India to construct five homogeneous microregions. These regions have been identified according to the similarity in rainfall characteristics and the association with 12 regional/global circulation parameters. The same five subregions have been considered in the present study. The subdivisions combined in the homogeneous regions are given in Parthasarathy et al. (1995). Figure 1 gives the five homogeneous regions viz. northwest (NW) region, central northeast (CNE) region, northeast (NE) region, west central (WC) region, and peninsular (PEN) region. The hilly areas in north and northeast India are excluded from the study due to the sparse network of reporting stations over this region (Rajeevan et al. 2006).

The model simulations are validated with observed gridded data sets from India Meteorological Department for different parameters viz. the rainfall, the mean maximum temperature (Tmax), and the mean minimum temperature (Tmin). Daily maximum temperatures are considered for the month of March–April–May (MAM), being the hottest months for India. Daily minimum temperatures are considered for January–February (JF), being the coldest months for India. The PRECIS simulations corresponding to 1961–1990 represent present climate. The regional model simulations viz. Q0, Q1, and Q14 show reasonable skill in representing the spatial and temporal patterns of summer monsoon precipitation and annual average temperature over Indian subcontinent (Krishna Kumar et al. 2011).

Table 1 gives the summer monsoon rainfall as well as the seasonal maximum and minimum temperatures for baseline period of 1961–1990 as simulated by Q0, Q1, and Q14 and their interannual variability given by standard deviation. The PRECIS rainfall simulations based on the period 1961–1990, representing the present climate, show that the summer monsoon rainfall for the months of June through September is reasonably well simulated by the models over all the five homogeneous regions as shown in Fig. 2. Rain shadow region in PEN region, low rainfall zone in NW region, and highest rainfall zones over NE region as well as coastal parts of WC and PEN regions are very well represented in the model.

Table 1 Simulated seasonal rainfall (JJAS), maximum temperature for pre-monsoon season (Tmax) and minimum temperatures for winter season (Tmin) for the baseline period of 1961–1990 in all the five homogeneous regions and their standard deviations (shown in brackets) for three members Q0, Q1, and Q14
Full size table
Fig. 2
figure 2

The observed mean spatial pattern of summer monsoon rainfall (mm) using IMD gridded data set compared with the three baseline simulations viz. Q0, Q1, and Q14 of PRECIS based on 1961–1990

Full size image

The maximum temperatures for the MAM as simulated by the model are validated using the daily gridded data set prepared by IMD. The seasonal maximum temperature simulations for the period 1961–1990 show good agreement with the observed pattern as shown in Fig. 3. The highest maximum temperatures over WC region are well captured by the model. The model, however, simulates the colder seasonal maximum temperatures over NE region.

Fig. 3
figure 3

The composite of observed mean maximum surface air temperature (°C) for the pre-monsoon (MAM) season using IMD gridded data set compared with the three baseline simulations viz. Q0, Q1, and Q14 of PRECIS for the period 1961–1990

Full size image

The minimum temperatures for JF as simulated by the model are validated using the daily gridded data set by IMD. Figure 4 shows that the highest minimum temperatures in PEN region and lowest in the NE as well as NW region are very well represented in the model. It can be seen from Figs. 3 and 4 that the seasonal Tmin is better simulated than the Tmax.

Fig. 4
figure 4

The observed composite of mean minimum surface air temperature (°C) for the winter (JF) season using IMD gridded data set compared with the three baseline simulations viz. Q0, Q1, and Q14 of PRECIS for the period 1961–1990

Full size image

Though the large-scale climatological features are well simulated by the model, there are some biases noticed in rainfall as well as seasonal maximum and minimum temperatures as shown in Table 2. It is the average of model bias taken over all the grid points falling in the particular homogeneous region. Model bias in the JJAS rainfall is expressed as % bias, whereas the bias in Tmax and Tmin is just the deviation from observed value.

Table 2 Model bias (observed value minus model simulated value) as seen in the baseline simulations for 1961–1990 in all the five homogeneous regions, for summer monsoon rainfall, Tmax, and Tmin
Full size table

The bias in seasonal monsoon rainfall shows large variations among the three members as well as for different homogeneous regions. The Q1 simulates dry bias over all the five homogeneous regions that can also be seen from Fig. 5 (first column) that gives the spatial distribution of bias (not expressed as %) in seasonal rainfall. Dry bias of more than 50 mm can be seen over some parts of CNE region and south PEN region in all the three simulations and hence reflected in the ensemble bias. The wet bias is simulated over east NW region and western parts of NE region in all the three simulations.

Fig. 5
figure 5

Difference between observed data and model simulations for baseline period 1961–1990 for summer monsoon rainfall (mm, first column), Tmax (°C, second column), and Tmin (°C, third column). First three rows show the biases for Q0, Q1, and Q14, respectively. Last row gives the ensemble bias for the three parameters

Full size image

Tmax is overestimated in all the three simulations over all regions showing warm bias as seen from Table 2. The spatial pattern of bias in Tmax shown in Fig. 5 (second column) indicates that there are some parts in the eastern NE region with cold bias. Southern parts of PNE region, eastern part of CNE region, and southern coastal region of NW region have warm bias in all the three simulations that is also seen in the ensemble bias given in the last row of Fig. 5.

Figure 5 (last column) gives the model bias in the Tmin. Warm bias in all the three simulations over southern parts of PEN region is remarkable. All the three simulations show cold bias for NW, CNE, and NE regions that can be seen from Table 2. The bias is small (−1 to +1) for WC region. The spatial pattern of bias in Tmax as well as Tmin is coherent among different simulations. However, the bias is less in Tmin as compared to Tmax and can be seen from Table 2.

The annual cycles of precipitation as well as Tmax and Tmin are reasonably well simulated by the PRECIS simulations as shown in Fig. 6. Q1 simulates dry bias of around 3 mm/day during June and July months over all the homogeneous regions except NE region (Fig. 6, first column). However, the shape of the annual cycle, with highest rainfall in summer monsoon season, is very well captured by the model. The annual cycles for Tmax and Tmin (Fig. 6, columns 2 and 3, respectively) are well simulated by the model with the highest temperatures in the pre-monsoon months of March–April–May and the lowest temperatures in the winter months of January–February. The model simulated maximum temperatures are warmer than the observed temperatures. The highest warm bias of more than 2.5 °C is seen over NE and CNE regions. The model simulates colder temperatures (less than 1 °C) in January–February over all the regions except PEN region where the model simulated values are warmer than the observed temperatures by around 1.3 °C.

Fig. 6
figure 6

Annual cycles of precipitation (first column), Tmax (second column), and Tmin (third column) as simulated by PRECIS simulations for the baseline period of 1961–1990 compared with the observed annual cycles using IMD gridded data set

Full size image

4 Projected changes

4.1 Seasonal rainfall

The time series of seasonal mean monsoon precipitation for all simulations during the period 1961–2098 for the five homogeneous regions are shown in the Fig. 7 (first column). All the three simulations depict interannual variability in the time series for summer monsoon precipitation over NE, WC, and PEN regions. However, the interannual variability of summer monsoon precipitation is more in Q0 and Q14 as compared to Q1. Q0 and Q14 show slight increase in seasonal mean precipitation, especially after mid-century, where as Q1, that shows dry bias in terms of seasonal mean quantum of rainfall, does not show any increase in the seasonal rainfall over NW, PEN, and WC regions.

Fig. 7
figure 7

Time series of model simulated summer monsoon rainfall (left), Tmax (middle), and Tmin (right) over five homogeneous regions

Full size image

Table 3 gives the average projected percentage change in the seasonal monsoon precipitation in three time slices viz. 2020s, 2050s, and 2080s with respect to model baseline 1970s. The summer monsoon rainfall may increase over all the homogeneous regions, in all the three time slices. The last column in Table 3 shows that all the regions except PEN may experience 15 to 19% increase in the summer monsoon precipitation, towards 2080s, as shown by the ensemble mean of all the three QUMP simulations. However, PEN region may have marginal increase as shown in Table 3 (4.7%) towards 2080s.

Table 3 Average % change simulated in summer monsoon (JJAS) precipitation for 2020s, 2050s, and 2080s with respect to 1970s
Full size table

4.2 Seasonal temperature

Unlike precipitations, the temperature changes are consistent among the three simulations that can be seen from Fig. 7. Tmax (Fig. 7, second column) as well as Tmin (Fig. 7, third column) show monotonous increase towards the end of the present century. This feature is seen in all the three simulations.

Tables 4 and 5 give the average changes in Tmax and Tmin, respectively, for the five homogeneous regions towards 2020s, 2050s, and 2080s. Range of the temperature is given in brackets. Both the temperatures show consistent rise in all the three simulations, over all the five homogeneous regions. It can be seen from the tables that around 4 °C rise is likely towards the end of century in pre-monsoon season Tmax as well as winter season Tmin. However, towards the end of the present century, i.e., 2080s, the rise in Tmin is likely to be more than the rise in Tmax for all the homogeneous regions except the PEN region, i.e., rise in nighttime temperatures may be more than that for daytime temperatures.

Table 4 Average change (°C) simulated in maximum temperature in pre-monsoon season (MAM) for 2020s, 2050s, and 2080s with respect to 1970s
Full size table
Table 5 Average change (°C) simulated in minimum temperature in winter season (JF) for 2020s, 2050s, and 2080s with respect to 1970s
Full size table

4.3 Probability density functions

Probability density functions (PDFs) of daily maximum (MAM) and minimum (JF) temperature and summer monsoon (JJAS) precipitation have been computed for observed and model simulations based on 1961–1990, and the projected distributions are computed for three time slices 2020s, 2050s, and 2080s over five homogeneous regions. The PDF of ensemble daily precipitation in the left panels of Fig. 8 is based on 120 days in JJAS season for 30-year period, i.e., 120 days × 30 years (3600 days) of the three-member simulations, over five homogeneous regions of India. The PDF of ensemble model baseline simulations fails to capture the observed shape of PDF. NW being the dry region, the maximum probable rainfall is around 1 mm/day. The observed probability of 1 mm/day rainfall is around 0.15, whereas model simulates more such events with probability around 0.5. The model projects similar shape of PDFs in the future also over this region. Over CNE, WC, and PEN, the model captures the shape of the PDFs correctly. In future projections, over CNE, the events of 5–10 mm/day rainfall may be more probable. Over WC events with 1–2 mm/day and over PEN 3–5 mm/day rainfall may be more frequent in future.

Fig. 8
figure 8

PDFs of ensemble mean of rainfall during summer monsoon season (left panel), Tmax during pre-monsoon season (middle panel), and Tmin during winter season (right panel) over five homogeneous regions

Full size image

The middle panels depict PDFs for ensemble daily Tmax during pre-monsoon season (MAM, 90 days × 30 years = 2700 days) over five homogeneous regions. The shape of the PDF is reasonably captured in the ensemble baseline simulation. In general, except NE region, the model simulates warmer mean maximum temperature by approximately 4–5 °C. The maximum temperature may be most probably in the range of 40–45 °C. Over NE region, cooler Tmax are simulated by the model. In the future, the mean Tmax may mostly remain around 30–35 °C. CNE and NW regions project more variable maximum temperatures. The PDFs depict positive shift in the future indicating average warmer mean maximum temperatures towards the end of the present century.

The PDFs of ensemble daily Tmin during winter season (JF, 60 × 30 days) are shown in the right panel of Fig. 8. Over all regions except NE, the shape of the PDF of mean minimum temperature is simulated reasonably well. In general, over all the regions, the minimum temperatures may increase in the future. Over NW, CNE, and NE, the mean minimum temperature is projected to be around 10 °C on maximum number of days, whereas it may be around 15 and 20 °C over WC and PEN regions, respectively. The Tmin is projected to have larger variability in the future as compared to those of the present century. The PDFs in the future time epochs indicate warming in Tmax as well as Tmin which will have profound impact on various sectors like agriculture, public health, etc.

5 Summary

The present study examines the potential impacts of climate change on the climate of five homogeneous regions of India. The regional climate model simulations form PRECIS are used to project the future climate under global warming scenario. PRECIS simulates the present climate over the homogeneous regions of India reasonably well. The climate change simulations indicate positive change in the mean monsoon rainfall in all the three simulations over all the five homogeneous regions, the change being minimum over peninsular region. The events with rainfall 10 mm or less are projected to increase in the future, over all homogeneous regions; however, the analysis does not indicate any conspicuous change in the frequency of medium and heavy rainfall events in future.

Temperature simulations show that the maximum as well as minimum temperatures are likely to be more under global warming scenario. However, the rise in winter time (January–February) minimum temperatures is likely to be more than those in pre-monsoon time (March–April–May) maximum temperatures for all the homogeneous regions except the peninsular region. The highest probable temperatures, both minimum as well as maximum, may be warmer towards the end of the present century by around 4–5 °C over all the five homogeneous regions.

The scenarios presented in this article are indicative of the expected range of changes in the climate over the five homogeneous regions of India. Since the projections are based on limited number of simulations, they still have large quantitative uncertainties. Also, the model simulations used in this study are carried out with the LBCs from the IPCC-AR4 A1B emission scenarios. Till now, PRECIS simulations are the only high-resolution regional climate model simulations available over Indian region. Recently, the simulations over South Asian region are made available for three models under CORDEX-SA program of WCRP and have been archived at the data portal of Center for Climate Change Research (CCCR) of Indian Institute of Tropical Meteorology, Pune, India. These models are of 0.5 × 0.5 lat/long resolution and have been derived using the lateral boundary conditions from CMIP5; however, this set does not include PRECIS. Such type of analysis has not been documented before using PRECIS simulations. Since CORDEX-SA provides the multimodel outputs for historical and RCP4.5 experiments and gives the range of uncertainty of model simulations, one can have more confidence in these projections. This will allow us to quantify the uncertainties better and also gain more confidence in the projected future climate over the country.