Do CMIP5 models hint at a warmer and wetter India in the 21st century?

India exhibits a wide diversity of climates; the presence of the Himalayas in the north and the Thar Desert in the west play a key role in controlling the climate of India. The Himalayas act as a barrier to the cold winds from Central Asia, thereby making India warmer compared to its latitude, and the Thar Desert attracts the summer monsoon winds, which are responsible for the distinct monsoon season observed in India. However, over the past couple of centuries, anthropogenic activities such as deforestation and urbanization have led to changes in the earth's climate. Meteorological data over the past century suggest significant warming of the earth (Kumar et al. 2006); future projections of Asian monsoon are quite uncertain due to varied reasons (Turner & Slingo 2009; Lee & Wang 2012). According to the report by the Intergovernmental Panel on Climate Change (IPCC, 2014), climate change is already having substantial impacts ‘on all the continents and across the oceans’ and the worst is yet to come. As carbon emissions have accumulated over recent years, scientists are considering global warming as an alarming threat (Strauss 2013). Also, summer sea ice in the Arctic continues to dwindle and the movement of coastal Atlantic glaciers has resulted in an uncontrollable rise of the sea level (Proshutinsky et al. 2001; Khedun & Singh 2014). With the land–sea temperature difference projected to increase in coming years, the amount of water available as moisture in the atmosphere is going to increase further, leading to a more intense monsoon (Kitoh et al. 2013). According to the Indian Network for Climate Change Assessment (INCCA) report #2 (2010), increase in temperature, melting of ice and rise in sea level are some of the paradigm shifts that have been observed in recent years owing to the effects of global warming. The report concludes that India is expected to experience a 2 °C rise in temperature by 2030, along with increased annual precipitation with a lower frequency of rainy days. Such changes could have disastrous effects in a country like India, which is an agricultural economy (a climate sensitive sector).

Impacts of climate change have been studied by several researchers (Dhorde et al. 2009; Xu et al. 2009; Schewe et al. 2011; Chaturvedi et al. 2012). Kumar & Hingane (1988) have studied the past 86–112 years' data of long-term seasonal and annual temperatures for major Indian cities such as Calcutta (name changed to Kolkata), Bombay (name changed to Mumbai), Madras (name changed to Chennai), Pune, Bangalore (name changed to Bengaluru) and Delhi. The study reveals significant warming at some cities, namely Calcutta, Madras and Bombay, while a cooling trend is observed at Delhi. These significant changes could be due to the fact that these are major industrial cities of India, which are subjected to the continuous addition of gaseous pollutants and waste heat, which alter the components of radiation balance. On similar lines, Dhorde et al. (2009) studied the trends in annual and seasonal temperature using linear regression and the Mann-Kendall test, and emphasized the effect of urbanization, leading to a significant shift in temperature during the post-monsoon and winter seasons at four major Indian cities. Xu et al. (2009) raised specific concerns regarding cascading effects of climate change on water, biodiversity and livelihood in the Himalayas. The increased warming in the Himalayan region, resulting in the loss of ice and snow, would lead to huge effects on global sea level rise. In fact, estimates show a minimum average sea level rise of 80 cm by 2100 (Cruz et al. 2007) in comparison to the IPCC's conservative estimate of 40 cm. Such a phenomenon would have far-reaching effects such as the submergence of coastlines in the deltas of southern Asia and loss of agricultural land due to increased salinity.

The extent of change in the precipitation regime over India is still a matter of debate. There have been abnormalities or inconsistencies between different models in predicting the change in Indian monsoon rainfall. While Fu et al. (1999) show an increase in Indian monsoonal rainfall, Guhathakurta & Rajeevan (2008) show an overall decrease in annual rainfall in central and northern parts of India due to a decreased contribution from major monsoon months such as July and September. Yali et al. (2013) show the uncertainty in rainfall behavior, with an increase in mean rainfall for a short run period (1979–2005) and a decrease in rainfall for most parts for a long run period (1950–2005). With the advent of the Coupled Model Intercomparision Project 5 (CMIP5), which is one of the most coordinated multi-model change experiments ever attempted, some significant climate studies have been conducted in recent times. Levine et al. (2013) examined the relationship between Arabian Sea cold sea surface temperature (SST) bias and weak monsoon rainfall in CMIP5 models. The study is an attempt to unveil future projections by CMIP5 which suggest that Arabian Sea SST has the potential to weaken future monsoon rainfall in India by limiting flux acceleration through the non linearity of the Clausius–Clapeyron relationship. According to Ashfaq et al. (2009), based on the high resolution nested model, enhanced greenhouse gas forcing is the prime reason behind decreasing summer precipitation and the delay of monsoon onset in the South Asian regions. Further, the reduced runoff may severely affect agricultural production, water availability and hydroelectric power generation (Mall et al. 2006). The future evolution of Indian summer monsoon rainfall (ISMR) has been studied using 20 CMIP5 models (Menon et al. 2013) for both historical and RCP8.5 scenarios. Nineteen out of 20 models simulated a positive trend in mean monsoonal rainfall under RCP8.5. Also, 17 out of 20 models showed an increase in interannual variability. The study asserts that CMIP5 models are showing a northward shift in the monsoon circulation under global warming. It shows a strengthening of monsoon circulation in the northern parts of India and a weakening of circulation in the southern parts.

Evaporation is another important component of the hydrological cycle to significantly affect the water balance of the earth. Understanding the role of evaporation is very important for a country like India, which hugely depends on agriculture. Penman (1948) mentioned that evaporation can take place from three kinds of surfaces – wet bare soil, turf (grass or vegetation) and from open water. The last is called free surface evaporation, and is defined as water evaporated from an extensive open water surface per unit area under existing atmospheric conditions. Most of the studies refer to evaporation as free surface evaporation with respect to water availability to crops. For example, Chattopadhyay & Hulme (1997) investigated the historical variation of free surface evaporation and evapotranspiration for five decades, ranging from 1940 to 1990, for India. Three seasons, namely pre-monsoon, monsoon and post-monsoon, were analyzed in the study. The results demonstrate a decrease in evaporation for monsoon and pre-monsoon seasons in spite of an increase in temperature. A similar reduction in evaporation in the Volga river basin instead of an expected rise was observed earlier by Golubev et al. (2003). In the present CMIP5 model outputs, evaporation datasets are available at the surface and are defined as the flux (the unit being kg/m^–2 s^–1) of water into the atmosphere due to conversion of both liquid and solid phases to vapour (from the underlying surface and vegetation). Thus evaporation from all sources is considered.

The Fifth Assessment Report of the IPCC made climate projections for the future using the newly developed Representative Concentration Pathways (RCPs) under CMIP5. The periodic assessment by the IPCC has relied heavily on general circulation model (GCM) simulations of future climate driven by various emission scenarios outlined by the RCPs. In this study, the CMIP5 model simulations of three climate variables: temperature, precipitation and evaporation, are utilized for an assessment of the changes in these variables at 48 locations across India for both historical and RCP8.5 scenarios. The study area and dataset are explained in the next section (Materials and Methods). The methodology adopted to regrid the variables and assess their changes is also discussed. The results and discussions section is presented afterwards. This section is divided into two subsections – Twentieth century climate change and Twenty-first century climate change (under the RCP8.5 scenario). Conclusions are presented at the end.

In this study, the three major climate variables are considered, namely temperature, precipitation and evaporation. Four different climate models are selected from the CMIP5 project. These are CanESM2, CSIRO-Mk3.6.0, IPSL-CM5A-LR and IPSL-CM5A-MR. The different GCMs consider different forcing conditions. The second generation Canadian Earth System Model (CanESM2) combines the CanCM4 model and the terrestrial carbon cycle based on the Canadian Terrestrial Ecosystem Model (CTEM) (Chylek et al. 2011), which models the land-atmosphere carbon exchange. The 10 CTEM models are all primary terrestrial ecosystem processes including land use change based on historical changes in crop areas. The concentrations of greenhouse gases and solar variability are based on the CMIP5 recommendations. In addition, the effects of volcanic eruptions are included. The CanESM2 provides daily data at 2.813 ° × 2.79 ° spatial resolution. CSIRO-Mk3.6.0 is developed by the Commonwealth Scientific and Industrial Research Organisation in collaboration with the Queensland Climate Change Centre of Excellence (QCCCE) (Rotstayn et al. 2012; Jeffrey et al. 2013). The horizontal resolution of this model is approximately 1.875 ° × 1.875 ° with 18 vertical levels. As per Jeffrey et al. (2013), CSIRO-Mk3.6.0 includes a suite of historical experiments, and is suitable for detection and attribution studies, and a sufficient number (with respect to the CMIP5 minimum requirement) of the ensembles are available. The IPSL-CM5A-LR and IPSL-CM5A-MR are both provided by the Institut Pierre Simon Laplace (IPSL) Climate Modelling Centre, France. The IPSL CM5 corresponds to the last version of the IPSL model and is a full earth system model. Its primary strengths are that, in addition to the physical atmosphere-land-ocean-sea ice model, it also includes a representation of the carbon cycle, the stratospheric chemistry and the tropospheric chemistry with aerosols. The IPSL-CM5A is available at two standard resolutions – (i) LR stands for low resolution (1.875 ° × 3.75 °) while (ii) MR stands for mid-resolution (1.25 ° × 2.5 °).

The details of data availability and basic model information are shown in Table 2. The four models are chosen out of 24 GCMs based on the availability of data. We have analyzed four CMIP5 models with total ensembles of 22 simulations for historical experiments and four models with an ensemble size of 13 for the RCP8.5 experiments. The historical experiment simulations are focused on the 156-year period from January 1850 through December 2005. The historical experiments account for simulation with natural forcing (a mixture that might include solar and volcanic forcing) as well as anthropogenic forcing (a mixture that might include well-mixed greenhouse gases, aerosols, ozone, and land use changes). The RCP8.5 simulates a future projection that focused on a 95-year run from 2006 through 2100. This refers to a rising radiative forcing pathway leading to 8.5 W/m² in 2100. The RCP8.5 represents a situation where CO₂ equivalent concentrations are greater than 1,370 ppm, almost four times the present day atmospheric CO₂ concentration (Kumar et al. 2013). Similarly, RCP4.5 and RCP6.0 represent the pathways for which radiative forcing reaches 4.5 W/m² and 6 W/m², respectively, in 2100. RCP2.6 peaks in radiative forcing at 3 Wm⁻² before 2100 and declines afterwards, reaching 2.6 Wm⁻² in 2100 (Menon et al. 2013).

All the GCM models have different spatial resolutions. Thus, all of them are regridded or interpolated to a common resolution of 2.5 ° × 2.5 °. Forty-eight grid intersections are found to fall within the Indian territory. For regridding GCM data, different functions in NCAR Command Language (NCL) are available at www.ncl.ucar.edu/Applications/regrid.shtml. Since interpolation of a scalar quantity from one Gaussian grid to another is required here, and the GCM gridded data are available for the whole globe, the function ‘g2gsh_Wrap’ (available at www.ncl.ucar.edu/Document/Functions/Contributed/g2gsh_Wrap.shtml) is found to be suitable and is utilized here.

After obtaining monthly regridded data for all the ensembles of all the models, the ensemble average is computed for each of the variables for each model at all grid intersections. For example, the regridded data are obtained for 48 grid points in India. For each grid point, monthly data are available for both the historical (1850–2005) and projected period for the RCP8.5 scenario (2006–2100). Since each model has multiple ensembles (different runs), data are averaged over a particular month across different ensembles to obtain a single value.

Furthermore, a best fit linear regression line is fitted to the monthwise time series of each variable in each scenario (historical and RCP8.5) and the slope (linear trend over time) is calculated. The slope of the line is further multiplied by the number of years in the respective scenario, which gives the change experienced by the variable over the study period for a particular month. For example, if the slope of the temperature time series is 0.03 °C/per year in January, then the change during 2006–2100 is computed as 0.03 × 95 = 2.85 °C. Such changes are computed for all the 12 months during this period. After computing such changes for each model and each location, the multi-model average is also computed in order to make final deductions. The multi-model average is taken, and necessary results are depicted using boxplots and maps.

Before studying the change of the three variables (temperature, precipitation and evaporation) during the historical period (1850–2005) and future period (2006–2100) using GCM data, it is worthwhile carrying out a preliminary analysis on the association of the GCM datasets with observation during the historical period. For this purpose, daily precipitation and temperature data provided by the India Meteorological Department (IMD) at 1 ° × 1 ° spatial resolution are used. The precipitation data are available for the historical period 1951–2003 and the temperature data are available for the period 1969–2005. These datasets are first converted to monthly scale and then their relationships with the corresponding CanESM datasets of the aforementioned periods are analysed. The scatter plots between IMD rainfall and CanESM rainfall are observed (figures not shown) and their correlation coefficients (r) are determined at various locations. The r value is found to be in the range of 0.05–0.81. In the case of temperature, the correlation coefficient between the IMD dataset and the CanESM dataset is found to be in the range of 0.81–0.98. In general, it is observed that the association of GCM rainfall with observations is not very good; however, the association of GCM temperature with observations is quite good. This observation is somewhat expected since most GCMs are known to have less accuracy in simulating precipitation. Thus, it may be noted that the future projections of temperature are possibly more reliable than that of precipitation. In the case of the latter, the GCM projections may give underestimated values based on their association during the historical period. Since the general tendency of GCM is to underestimate precipitation, when there is a projection of increasing precipitation, it may be accepted with more confidence.

Across all models, locations and months, the range of temperature change in the pre-industrial period (1850–2005) varies from 1 to 3.62 °C. The maximum change (3.62 °C) is observed by CanESM2 in the month of April in the southern part of India. Large increments are also observed in the northern and eastern regions.

On observing the monthwise change in temperature (°C) in different models at different locations in the historical scenario, it is found that all the models, namely CanESM2, CSIRO, IPSL-MR and IPSL-LR, show a consistent increase in temperature at all locations over India. However, there is wide inconsistency among the models regarding the locations experiencing maximum change in temperature for a particular month. For instance, in January, according to CanESM2, maximum change is observed at location 9, i.e. Bangalore, whereas location 1, i.e. Jammu, witnesses a maximum change as per CSIRO. On the other hand, IPSL-LR and IPSL-MR indicate a maximum change at location 4, i.e. Gwalior and location 43, i.e. Bhuj, respectively. Similar observations are made for other months too. Again, the month showing the maximum change in temperature also varies across models. For instance, CanESM2 observes a maximum temperature change in the month of April, as 3.62 °C, whereas IPSL-MR observes it to be 2.76 °C and 3.51 °C in May and June, respectively.

Considering the multi-model average monthwise change in temperature (°C) in the historical scenario (1850–2005), it is observed that the changes are not consistent for different regions in India, rather, spatial variation is very prominent. Whereas some regions show an exceptional increase in temperature, others regions do not seem to follow the same trend. Large variations of 1–3 °C change are observed in the month of January in the Northern regions of the country such as Delhi, Jammu, Shimla, Punjab, Kanpur etc. In the month of March, more than five out of nine locations in Southern India have witnessed changes in temperature of more than 1.7 °C.

Across all models, locations and months, the maximum change (2.76 mm/day) is observed in the Himalayas (location 21) by IPSL-LR in the month of August. However, the multi-model average is negative at this location (Figure 2). This points to the inconsistencies evident among the various model results. The monthwise change in precipitation averaged across all models for all the regions shows an interesting feature – all the months in all the locations show a decrease in precipitation. However, the North Eastern region has experienced an increase in precipitation in the months of May and February, with a maximum change of 0–0.9 mm/day.

Among the monsoonal months, the maximum decrease in precipitation is observed in the month of August. The regions – Southern, North Eastern, Himalayan and Western states – show a decrease of about 2 mm/day. Pre-monsoonal and post-monsoonal months show very little or no decrease in precipitation, the maximum value being 0.2 mm/day.

The monthwise change in evaporation (multi-model average) for all regions in India for the historical scenario (1850–2005) is studied. In general, all the months have seen little or no change in evaporation over this period. In the North Eastern region, a very small decrement of −0.03 mm/day is observed. Southern states such as Tamil Nadu and Kerala have observed relatively larger decrements in all the months. A decrease in evaporation was also observed in the months of June, July, August, September and October in Western, Central and North Eastern regions. In January, February and March, the maximum decrease in evaporation is noticed in the Western and Southern regions. Location 44 (Gandhinagar), location 43 (Bhuj) and location 20 (Karnataka) are the regions that are affected the most. Individually, among all the models, the maximum increase (0.354 mm/day) in evaporation in the 20th century is indicated by CanESM2 for the central region, i.e. Chhattisgarh. Again, the maximum decrease (–2.54 mm/day) in evaporation is observed in Tamil Nadu by the CSIRO.

Across all models, locations and months, the maximum change in precipitation for the RCP8.5 scenario is observed in Assam in the month of August by CanESM2, the value being 14.67 mm/day. The multi-model average of precipitation in different regions is discussed below.

Table 4 depicts the maximum and minimum values of the multi-model average precipitation change across different regions. North Eastern and Eastern regions (such as Assam, Mizoram, Nagaland, Odisha, Durgapur, Jharkhand, Patna etc.) are expected to receive a heavy rainfall change of 5.45 and 5.05 mm/day, respectively. On the other hand Northern, Western and Central regions are expected to experience much less change in rainfall. The Southern regions that are near to the coast of India like Tamil Nadu, the Western Ghats are likely to receive a considerable increase in precipitation along with the Himalayan belt (Figure 8).

Simulations from four GCM models, namely CanESM2, CSIRO, IPSL-CM5A-MR and IPSL-CM5A-LR are considered to assess monthwise changes in temperature, precipitation and evaporation across India for historical (1850–2005) and RCP8.5 (2006–2100) scenarios. Forty-eight grid intersection points are selected in such a way that all the regions, i.e. northern, western, eastern, north eastern, central, southern and the Himalayas, are well represented. The nature of changes in the climate variables for future scenarios do not seem to be very optimistic in relation to Global Warming. An increase in temperature is observed in all regions across India during the last century and also for future scenarios. All the major cities such as Delhi, Mumbai, Chennai etc. have shown a major increase in temperature. The effects of urbanization and increasing pollution in these regions could be the possible reasons for such observations. The high density of buildings in urban areas inhibits the transfer of heat to the higher atmosphere. This is one of the well known effects of urbanization, which develops when urban cooling rates are slower than rural ones. Apart from that, other possible reasons for this effect include heat emissions by buildings, air conditioning, transportation and industries. The temperature increase in the Northern, North-Eastern and Himalayan regions is a major concern. The Himalayan region, for example, holds the largest mass of ice outside the polar regions and is the source of the 10 largest rivers in Asia (Xu et al. 2009). A 10 °C increase in temperature in the Himalayas in RCP8.5 scenario could lead to serious problems for India as it would immensely impact water availability, monsoonal behaviour and high-elevation ecosystems in that area.

This study also reviews the previous inconsistencies observed in precipitation for the historic period and provides a comparison with the future projected scenario. Outputs from CMIP5 models indicate a decrease in precipitation in the historical scenario, but an increase in precipitation for the RCP8.5 projection. In general, the ISMR is found to increase under future warming. These results are consistent with previous studies (Fu et al. 1999; Menon et al. 2013), which suggest a clear connection between the global temperature increase and Indian monsoonal rainfall. Kharol et al. (2013) cite pollution as one of the major factors for such an increment. Pollutants spewed from factories and vehicles lead to increased cloud formation, which often leads to heavy rainfall in some places. The major increment in precipitation is observed in southern cities like Bangalore, Hyderabad etc. These cities have seen urban expansion and extensive changes in land use and land cover over the last 50 years, in addition to increased vegetation (Narayanan & Hanjagi 2009) and accelerated industrial activities, which could be the major cause for increased precipitation in these cities. Meanwhile, positive trends for winter monsoonal rainfall have been observed in the Western Ghats. In the Himalayan region, summer precipitation is projected to increase whilst very little change is projected for the winter months. The projected increase in the frequency and magnitude of high intensity rainfall events could affect infiltration and groundwater recharge, with further implications for water availability for people as well as vegetation (Sharma et al. 2009).

It is well established that emissions of carbon dioxide, aerosols and other pollutants have led to an increase in global temperatures over the last century; however, evaporation does not seem to follow the same trend. It may be expected that an increase in temperature would lead to an increase in precipitation and evaporation, and thus intensify the hydrological cycle. Contrarily, the present study shows a decrease in evaporation between 1850 and 2005 at most of the locations in India. The Northern and Himalayan regions have experienced relatively less change, about 0–0.2 mm/day compared to Southern, Western and Central regions where the change is 0–1.0 mm/day approximately. The result matches with the findings of Chattopadhyay & Hulme (1997), which state that temperature alone does not provide a satisfactory indication of changes in evaporation as is commonly presumed. For the RCP8.5 scenario, pre-monsoonal months do not show any significant change in evaporation at most of the locations, but the same cannot be said about monsoonal and post-monsoonal seasons. Southern, Central and Western regions are predicted to experience a change of about 0.9 mm/day during JJAS.

This study provides an insight into India's changing climate in the twentieth and twenty-first centuries. However, it needs to be appreciated that there are substantial disagreements in the outputs of the four models considered in this study. This is not unexpected, since the forces that drive climate change are rather complicated and uncertain. Hence, in addition to the adoption of sustainable policies globally, it is essential to devise climate resilient infrastructure and strategies for various sectors at the regional scale.