Catchment response to climate change under CMIP6 scenarios: a case study of the Krishna River Basin

Climate change is one of the major challenges that must be addressed for sustainable development (Hang 2022). The changing climatic patterns and the increased frequency of extreme events are believed to have major impacts on human activities and ecological systems (IPCC 2013). Among various sectors, the water resources sector is expected to be severely affected by the changing climate. Climate change affects hydrological variables such as temperature and precipitation patterns, thereby altering the global hydrological cycle, which results in frequent floods and droughts (Murray et al. 2012; Zhang & Huang 2013; Faye 2022). The arid and semi-arid regions are most sensitive to climate change, especially to changes in the precipitation pattern (Gong et al. 2004; Vörösmarty et al. 2010; Xing & Wang 2017). Under the climate change scenarios, variations in streamflow have been found to be highly linked with precipitation changes. Quantification of the hydrological effects due to climate change is essential for better design and management of water resources of any region.

The future climate variable projections are obtained using the global climate models (GCMs), which are recognised as credible tools for simulating the long-term future climate in a changing environment (Srinivas et al. 2013; Gouda et al. 2018). Most climate change impact studies integrate the available GCM data into a hydrological model to analyse different scenarios. However, one of the shortcomings of this approach is the uncertainty associated with the predicted climate variables from different GCMs. One of the approaches to overcome this is analysing the climate change impacts using the outputs from multiple GCMs. While this method considers the uncertainty in the climate predictions, it leads to larger plausible impacts of climate change on the water balance components (WBCs). Some studies have reported that the performance of the GCM simulations is region-specific, possibly due to the resolution of the simulation, model structure, parameterisation and boundary conditions (IPCC 2013; McSweeney et al. 2015). Therefore, selecting the GCMs that best represent the climate of the study region is a pre-requisite for climate change studies, but it is seldom followed (Chiew et al. 2009; McMahon et al. 2015; Pour et al. 2018). Further, with the release of new climate change projections, ranking of GCMs for hydrological studies is required for analysing the impacts of climate change.

The evaluation of a climate model is typically conducted by comparing its simulation of historical data with a reference dataset. The evaluation of models was categorised into two groups: time-domain- and frequency-domain-based metrics, as discussed by Rathinasamy et al. (2014). The time-domain category metrics, such as coefficient of determination, Nash–Sutcliffe coefficient (NSE), root-mean-square error, fourth root-mean-square error, Kling–Gupta efficiency, etc., consolidate the error at each time step into a singular metric. The majority of these metrics are highly susceptible to extreme events and outliers. The second category evaluates the similarity between the frequency distributions of observed and simulated variables to assess the goodness of fit. This overcomes previous limitations and allows a more comprehensive assessment of the match between observed and simulated data. Some of the studies used these frequency-domain metrics such as skill score (Perkins et al. 2007; Anandhi & Nanjundiah 2015; Raju et al. 2017) and Brier score (Ruan et al. 2018) in climate model selection. Symmetric uncertainty (SU) is one of the most used frequency-based metrics in recent studies for GCM selection (Khan et al. 2018; Hassan et al. 2020). The SU computes similarities and dissimilarities between simulated and observed time series in terms of common information entropies. The advantage of SU is that it does not depend on the distribution and conditional dependences of the data and can be used in place of conventional statistics (Khan et al. 2018; Salman et al. 2018). The present study used the SU technique to select suitable GCMs due to their advantage compared to other metrics.

Progressive efforts have been made to assess the impact of climate change on large river basins across the world using the outputs of different GCMs from Coupled Model Intercomparison Project (CMIP) Phases 3 and 5 (Githui et al. 2009; Chen et al. 2012; Gao et al. 2018; Reshmidevi et al. 2018; Bhatta et al. 2019; Ndhlovu & Woyessa 2020; Mann & Gupta 2022). The results of these studies indicate that the changes in climatic patterns and their impacts on the hydrology of watersheds vary with location and catchments in addition to climate models and their projection scenarios. The water yield increased in some basins, while it decreased in other basins. Indian river basins (IRBs) are exceptionally susceptible to extreme events such as floods and droughts (Suman & Maity 2020; Poonia & Azad 2022). In the context of IRBs, Reshmidevi et al. (2018) used an ensemble of CMIP3-based GCMs under the A2 scenario to investigate the impact of climate change on the Malaprabha River Basin and observed that marginal changes occurred in annual future precipitation and insignificant changes happened in the future water budget components. In another study, Pandey et al. (2019) investigated the effect of climate change on the Narmada River Basin using CMIP5 projections and found a significant decrease in the annual precipitation in the future under both representative concentration pathway (RCP)4.5 and RCP8.5 scenarios. They also observed that other WBCs were also significantly affected.

The Krishna River Basin (KRB) is one of the major river basins in India and a fast-developing region in the Indian peninsula. The water needs in the KRB are increasing day by day for agriculture, domestic and industrial purposes due to the rapid increase in population and economic growth, which may lead to water shortage and interstate conflict in the future. A study estimates that the water demand is more than the supply in half of the KRB (Lower Krishna, Paleru, Vedavathi, Munneru, Musi and Middle Krishna) (Yee et al. 2009). In addition, the KRB is highly sensitive to climate variations due to its semi-arid nature, which may affect water demand, especially in the agricultural sector (Yee et al. 2009). Therefore, analysing the climate change impacts on the KRB for developing prudent water management strategies is essential. Only a few studies are available that evaluate the climate change impacts on the water resources of the KRB. For instance, it is predicted that the decrease in the precipitation and water yield will lead to seasonal or regular water stress and drought conditions in the future under greenhouse gas (GHG) scenarios and special report on emission scenarios (SRES) (Gosain et al. 2006, 2011). Kulkarni et al. (2014) predicted an increase in annual precipitation, water yield, surface runoff, and actual evapotranspiration for the future period of 2041–2070 under the SRES. However, they found no notable variations in these parameters during the early century (2011–2040). The study did not consider the water storage structures, and it utilised data from a single GCM without any bias correction and performed calibration at only one gauge station. However, the aforementioned studies rely on utilising GHG and SRES, which inherently contain significant uncertainties due to assumptions made regarding economic development, population growth and technological advancements (Soro et al. 2017). In another study on KRB, an increase in the discharge and hydrological extremes in the future under RCP4.5 and RCP8.5 scenarios using only a single regional climate model IITMRegCM4-4 was observed (Nikam et al. 2018). Similarly, the increasing trend of WBC is projected in some of the studies under CMIP5 projections within RCP4.5 and RCP8.5 scenarios using randomly selected GCMs (Mishra & Lilhare 2016; Chanapathi et al. 2018). Considering water storage structures in hydrological modelling will also improve the accuracy of the future projections of WBCs. Relying on a single GCM or random selection of GCMs may incorporate large uncertainty in climate change studies. Hence, the simulations of the multi-model ensemble (MME) mean are identified as a better estimate (Gleckler et al. 2008; Knutti et al. 2010). It is suggested that 50% of top-performing GCMs are optimum for the climate change impact studies to reduce the uncertainty in the model projections (Ahmed et al. 2020; Raju & Nagesh Kumar 2020). However, at present, CMIP6 comprises a more complex atmospheric process and has an improvised parameterisation that can efficiently simulate model ability (Eyring et al. 2016; O'Neill et al. 2016). It is reported that CMIP6 GCMs are more efficient in simulating the Indian summer monsoon than CMIP5 GCMs (Gusain et al. 2020). The climate change impact studies using an improved version of the CMIP5 project, i.e., the CMIP6 climate model, are lacking in the KRB. Therefore, the objectives of the present study are (i) to evaluate the performance of climate models to identify the most suitable GCMs from CMIP6 using the SU concept for KRB, (ii) to assess the impact of projected climate change on WBCs including streamflow using the SWAT model for different bias-corrected socio-economic pathways (SSPs) scenarios and (iii) to analyse the streamflow extremes in KRB.

The basin's climate varies from humid to arid, where most parts (inland areas) fall under semi-arid nature. Approximately 80% of the basin is composed of Archaean and crystalline rocks, with the remaining 20% comprising basalts (Ramesh & Subramanian 1988); additionally, laterite and black soils are prevalent in the area (Das et al. 2005). The land use/land cover is dominated by agriculture, which covers nearly 76% of the area. Average annual rainfall varies from 429 mm in the northern part to 3,080 mm in the southern part, with an average of 745 mm. Average daily maximum and minimum temperatures vary from 27.7 to 40.4 °C and from 20.6 to 27.2 °C, respectively. The southwest monsoon (June–September) period has the predominant rainfall distribution (nearly 90%). The uneven distribution of rainfall occurring in the basin affects the availability of water. Some parts of the basin experience droughts (Singh et al. 2019), whereas delta regions are prone to flooding (http://india-wris.nrsc.gov.in).

Datasets necessary for setting up the soil and water assessment tool (SWAT) hydrological model included spatial data such as the digital elevation model (DEM), land use land cover (LULC) maps, soil maps and meteorological information, including precipitation levels and minimum and maximum temperatures. The discharge at multiple locations within the basin was used to calibrate the SWAT model.

In India, the hydrometeorological stations are maintained by the CWC. Daily discharge data observed at the Vijayawada gauging station (mouth of KRB) from 1970 to 2003 and information regarding monthly storage, volume, area, effective and gross storage capacity of the hydraulic structures, and spillway designed capacities were downloaded from the Water Resources Information System (India-WRIS) (https://indiawris.gov.in/wris/). Daily gridded precipitation of 0.25° × 0.25° and minimum and maximum temperatures of 0.5° × 0.5° were obtained from the Indian Metrological Department (IMD; https://www.imdpune.gov.in/). The precipitation dataset had no missing values and was developed from 6,955 gauging stations distributed all over India (Pai et al. 2014). These data performed better than other global gridded datasets such as the National Centers for Environmental Prediction, Coordinated Regional Climate Downscaling Experiment, and Global Precipitation Climatology Project (Bandyopadhyay et al. 2018).

Daily precipitation data from GCMs of CMIP6 repositories were considered in this study. This study used scenarios from Tier-1 Shared Socio-economic Pathways (SSPs) that include SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 to provide a full range of forcing targets similar in both magnitude and distribution to the RCPs used in CMIP5 (Gidden et al. 2019), as presented in Table 1. The GCM data can be accessed from the Earth System Grid Federation portal (https:/esgf-node.llnl.gov/search/cmip6).

The future climate change impact on the hydrology of KRB is categorised into four phases: (1) selection of the suitable GCMs from CMIP6, (2) SWAT model setup and performance evaluation using calibration and validation, (3) future climate change assessment based on WBCs and (4) analysis of streamflow extremes in the KRB.

Selection of suitable GCMs for a given study is necessary to reduce uncertainty induced in the GCMs. Initially, 18 GCM precipitation outputs were considered from the first ensemble member (‘r1i1p1f1’) experiment family and re-gridded to the observed data resolution of 0.25° × 0.25° using the bilinear interpolation technique (Pour et al. 2018; Anil et al. 2021). Due to the presence of biases in the GCMs, their projections are often unreliable. Many studies have used various bias correction methods to reduce biases (Eisner et al. 2012; Pierce et al. 2015; Maraun et al. 2017). In this study, empirical quantile mapping, one of the most useful and robust non-parametric techniques, was employed to reduce the biases associated with the historical and future scenarios of GCM outputs (Piani et al. 2010; Mishra et al. 2020).

The widely used concept called SU (Homsi et al. 2020) was applied to precipitation for the 1951–2014 period to screen the suitable GCMs. The SU was computed based on mutual information (MI) to find similarities and dissimilarities between simulated and observed time series in terms of common information entropies. MI tends to be biased towards higher values, and this bias can be mitigated using SU by normalising the MI in the range of 0–1 (Salman et al. 2018). An SU value of 1 indicates perfect agreement, while a value of 0 indicates no agreement between the observed and simulated time series. It has been reported that there is no rule for selecting the number of GCMs, and most have considered the top three to five GCMs (Tian et al. 2016; Wang et al. 2016; Latif et al. 2018; Ahmed et al. 2019; Noor et al. 2019; Shiru et al. 2019; Iqbal et al. 2020) to develop MME average to reduce the uncertainty. However, some studies have reported that the optimum performance of MME is achieved when 50% of the top-ranked GCMs are considered (Ahmed et al. 2020; Raju & Nagesh Kumar 2020). In the present study, nine GCMs are selected for further climate impact studies based on the total ranking weight (TRW) obtained using Equation (1) after applying SU. The following steps are involved in the ranking of GCMs:

• (i)

  By using the SU score, GCMs are ranked first, second, third, and so on at each grid point.

• (ii)

  Weights (⁠⁠) are assigned to each GCM in such a way that their ranking is inversely proportional.

• (iii)
  For each rank, the frequency (⁠⁠) of each GCM is calculated. Finally, the total ranking weight of each GCM is computed using the following equation:

  

  (1)

After sorting the TRW in descending order, the GCM's final ranking is determined. For a detailed explanation regarding the concept of SU and the procedure involved in selecting the ranking of GCMs, readers can refer to Anil & Anand Raj (2022).

The SWAT model can simulate the impacts of the long-term climate and LULC variations on the hydrology in large complex catchments (Wang et al. 2008; Neitsch et al. 2011; Arnold et al. 2012). The SWAT model divides the entire catchment into sub-catchments based on topography and then segregates these sub-catchments into hydrological response units (HRUs) using the unique combinations of slope, LULC and soil classes. The HRUs are vertically divided into different control volumes like surface layer, root zone, shallow aquifer and deep aquifer. Using climate data and LULC patterns, the SWAT model can estimate anticipated watershed scenarios. Moreover, it can evaluate streamflow variability by taking into account forecasted climate variables for the future.

The entire KRB was divided into 85 sub-basins after the delineation process, and a 5% overlap of LULC, soil and slope was provided to define HRUs, resulting in 1,490 HRUs. The water storage structures in any basin will significantly influence the performance of the model, which shows an impact on the WBC (Sahoo et al. 2018; Chanapathi & Thatikonda 2020). To improve the model performance and reduce the uncertainty in the output, 13 major reservoirs–Bhadra, Tungabhadra, Nagarjunasagar, Srisailam, Almatti, Koyana, Narayanapura, P.D. Jurala, Ujjaini, Malaprabha, Pulichintala, Hidkal and Vaanivilas– were considered based on the data availability from CWC for setting up the model. Various details of reservoirs are provided, such as area, volume, principal, gross storage, live capacity, principle volume, emergency, principle surface area, design flood, maximum and minimum monthly outflows, target storage capacity and consumptive storage capacity. Observed data for 34 years (1970–2003) were considered for calibrating and validating the SWAT model. The first 3-year (1970–1972) data were taken as a warm-up period.

The SWAT model was calibrated and validated using the sequential uncertainty fitting (SUFI-2) algorithm in the SWAT-Calibration and Uncertainty Program (CUP) (Abbaspour et al. 2015). Prior to calibration, the sensitivity of the model parameters influencing streamflow was estimated in the SWAT-CUP. The p-values and t-test values were used to identify the sensitive parameters that can influence the model output. Smaller p-values with larger t-test values represent higher sensitivity for the optimisation function and thus the observed variable, i.e., discharge. The t-test represents the relative significance, while the p-value represents the significance of sensitivity (Sinha et al. 2020).

A comparison of the future projected precipitation from all GCMs with the historical GCM observed precipitation indicates that the mean precipitation in the basin is increasing in most future scenarios. Only a few years in all the GCM simulations have shown lower precipitation than the historical GCM precipitation within the basin. It is also observed that the variability in the future projected mean annual precipitation is increasing, as evident from the box plots of Figure 7. The MPI-ESM1-2-HR and MPI-ESM1-2-LR GCMs, which occupied first and second ranks in GCM ranking, predicted less increment in the future compared to other GCMs. It can be seen that NorESM2-LM and NorESM2-MM overestimated the higher extremes and underestimated the lower extremes. Although NorESM2-LM ranked fourth in GCM ranking, it showed more extremes compared to other GCMs in all periods except FF under SSP5-8.5. This could be due to the inherent fundamental uncertainties in the climate change projection of GCMs (Li & Jin 2017; Ndhlovu & Woyessa 2020). The Ec-Earth3 and Ec-Earth3-Veg outputs predicted more precipitation in FF under SSP5-8.5. In the FF, the precipitation magnitude increase was more than that in the NF and MF under SSP3-7.0 and SSP5-8.5. The high variability in the precipitation for the future projected GCMs suggests that the basin will experience climate extremes more frequently in the future. The increase in the total projected precipitation is also evident from the MME average value for the three future periods. The prediction bounds and percentage changes in the future projections are evident from the difference in the projected values from different GCMs (see Tables S6 and S7, Supplementary material). The ensemble mean of projected annual precipitation is projected to increase by 24.1–70.3% in future time frames compared to the observed IMD data in the baseline period. The maximum average annual precipitation is predicted to increase by 30.56, 39.31, 56.96 and 70.30% by the end of the 21st century under SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5, respectively, compared to the observed IMD data in the baseline period.

The increased precipitation in the historical GCMs compared to the observed data is translated to other WBCs such as water yield, surface runoff and streamflow at the catchment outlet. Similarly, the future scenarios of all GCMs are also following a similar increasing trend in WBC to that of historical GCMs. Water yield, surface runoff and streamflow showed an increase in years with increasing rainfall, while in some years, these values were lower than the historical GCM. The increase in the surface runoff, water yield and streamflow at the outlet follows a similar trend to the precipitation data. Precipitation change has a significant influence on future streamflow, with projections showing a potential increase of 31–114.5% (see Tables S6 and S7, Supplementary material). The maximum ensemble mean changes in average annual streamflow are projected to increase 40.5, 61, 91 and 114.5% under SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5, respectively, compared to the historical GCM ensemble average of the baseline period by the end of the 21st century. The variation of mean annual future streamflow shows significantly increasing trends from NF to FF under all SSPs (Figure 10) compared to the baseline period of historical GCMs. Among the four SSP scenarios, the highest increase in projected streamflow is projected to occur under SSP3-7.0 and SSP5-8.5 compared to the historical GCM streamflow of the baseline period. In the NF and MF, NorESM1-2-LM is predicting more future streamflow under SSP1-2.6 and SSP2-4.5. For the scenarios of SSP3-7.0 and SSP5-8.5, the maximum average streamflow values are observed for EC-EARTH3 output during the MF. The FF is experiencing higher magnitudes of streamflow, especially under the SSP5-8.5 scenario, compared to other periods. It is also observed that a maximum flow of 8,856 m³/s has occurred in the FF under the SSP5-8.5 scenario for EC-EARHT3 GCM output, which is 1.96 times the maximum flow value (4,500 m³/s) of the GCM historical. In contrast, the peak flow of the ensemble average is predicted in the future as 5,381 m³/s, which is two times the peak flow (2,692.7 m³/s) of the ensemble average during the baseline period. The highest precipitation, water yield, surface runoff and streamflow are projected to increase by 54, 125, 124 and 114.5%, respectively, in FF under the SSP5-8.5 scenario compared to the ensemble average of the baseline period. The higher increase in the water yield, surface runoff and streamflow components compared to the precipitation may suggest that the level of precipitation is increasing during the monsoon period, leading to saturated conditions, which could result in higher runoff generation. The percentage increase in the evapotranspiration (ET) is observed to be, in general, low for future periods compared to the other WBCs (see Table S7). The ET is projected to increase in the range of 4–18% for future periods under different SSP scenarios compared to the ensemble average. One of the reasons for the lower ET increase is that although the future rise in temperature may accelerate evapotranspiration, increased CO₂ concentration and increased humidity (caused by increased rainfall intensity) may inhibit transpiration by altering leaf stomata, potentially offsetting the increased rate of evapotranspiration, which is also reported in other studies (Snyder et al. 2011; Chanapathi et al. 2018). The lower ET demand in the future period, combined with the higher precipitation, will lead to higher runoff generation from the basin. The percentage change of WBC is continuously increasing from NF to FF under all scenarios except for SSP1-2.6. In the case of SSP1-2.6, the percentage change of WBC (except streamflow) is higher than MF and FF.

The relative change calculated between the historical GCM ensemble average and future GCM ensemble average was applied to IMD observed values of the baseline period to obtain the expected future WBCs (see Table S8, Supplementary material). The average annual precipitation is expected to increase in the range of 781–1,072 mm for future periods under different SSP scenarios based on IMD data. In a similar way, future predictions indicate that surface runoff, water yield, evapotranspiration and streamflow range from 104 to 184 mm, 155 to 268 mm, 540 to 610 mm, and 920 to 1,506 m³/s, respectively, for different scenarios.

A further analysis was performed to evaluate the climate change impacts during the Indian summer monsoon rainfall (ISMR: from June to October) period. The prediction ranges of various GCMs and their ensemble averages compared with the GCM historical baseline period are given in Tables 5 and 6, respectively.

It can be found that nearly 90% of the climate change impact on WBC is occurring in the ISMR period. This study mainly focused on precipitation, water yield and surface runoff. It can be observed that there are significant variations in all WBCs from the prediction bound values. The ensemble mean of precipitation varies from a minimum of 129 mm (under GCM: NorESM1-LM) to a maximum of 2,569 mm (under GCM: IPSL-CM6A-LR) in the future (Table 5). From the ensemble average, it can be observed that almost all the scenarios show a similar kind of increase in precipitation in the near future. The ensemble average of ISMR precipitation is projected to increase in the range of 13.7–55% for the future period compared to the historical GCM ensemble average of the baseline period. The maximum increases in the ensemble average precipitation are observed as 19, 28.3, 43 and 55%, respectively, in the far future under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 (Table 6). The increases in precipitation are larger under SSP5-8.5 and SSP3-7.0 than under the other scenarios. The direct response of water yield and surface runoff to the precipitation also follows the same trend as the precipitation. The surface runoff and water yield components are anticipated to increase up to 175 and 245 mm for the ISMR period and 184 and 268 mm for the annual period, respectively, by the end of the 21st century under the SSP 5-8.5 scenario after applying the relative percentage of GCMs to the IMD observed values (see Table S9, Supplementary material). Almost 95% of the annual surface runoff and 92% of annual water yield occur during the ISMR period. This is due to a higher occurrence of precipitation during the monsoon period. Generally, the rise in the WBC may indicate excessive availability of water and high groundwater storage. The streamflow is expected to increase significantly in the future under all the scenarios. The individual streamflow for one of the GCMs is projected to experience a maximum increase of up to 20,382 m³/s under the SSP5-8.5 scenario in the FF period. The selected SSP scenarios showed an increase in surface runoff, water yield and streamflow due to increasing rainfall. The evapotranspiration also follows an increasing trend but a lesser increase than the other WBCs under all SSP scenarios. These results are similar to those reported in earlier studies (Kulkarni et al. 2014; Chanapathi et al. 2018). The ensemble mean of ISMR future streamflow ranges from 4,929 to 8,009 m³/s under different SSP scenarios, which is 22.5–99% more than the historical GCM ensemble average. The results suggest that more water will be available in the basin in the future and the basin may experience more floods.

It can be observed that there is a shift in the occurrence of peak flow value in the future period compared to the observed period 1973–2003. In the observed period, the highest recorded flow rate was 2,540 m³/s, which took place during August. However, in the future periods, the peak flow values for most of the individual GCMs and MME averages were observed in September. For the EC-EARTH3 model, the peak flow value is projected to occur in October in FF under the SSP5-8.5 scenario. This shift in peak streamflow in future periods compared to the observed period can be related to the extension of ISMR even up to October, resulting in an increase in surface runoff. This significant surface runoff in September and October negatively impacts the Kharif crops in the KRB since most of the crops are in the harvesting stage, particularly rice, maize and cotton. It can be observed that the streamflow increment is more in MF and FF than in NF under all SSP scenarios except the SSP1-2.6 scenario. Under the SSP1-2.6 scenario, the highest MME average streamflow in the NF is predicted to rise about 3.17 times (i.e., 7,942 m³/s) compared to that in the historical period (i.e., 2,540 m³/s). Likewise, the projected increases in maximum MME average streamflow during the MF and FF under SSP5-8.5 are estimated to be 3.52 (8,835.9 m³/s) and 4.38 (10,973 m³/s) times, respectively, compared to the baseline period. These findings align with previous studies (Nikam et al. 2018; Chanapathi & Thatikonda 2020).

Meteorological anomalies within the basin contribute to the hydrological extremes such as floods and droughts over a catchment. It can be seen that inter-annual variations in streamflow (see Figure 10) are increasing in the future periods under all scenarios compared to observed flows, indicating that extremes may be increasing. The projected changes in the hydrological extremes were analysed by flood frequency analysis of the annual maximum discharge for the flow at the watershed outlet. The floods of different return periods (T_p: 25, 50, 75 and 100 years) were estimated using General Extreme Value type-1 Gumbel distribution (Gumbel 1941). Hydrological extremes are defined as floods having magnitudes equal to or greater than these return period events. Analysis was carried out to count the number of extreme events during future periods of chosen GCMs and their magnitudes for 25-, 50-, 75- and 100-year return periods under four scenarios (Table 7). It can be observed that the frequency of extreme events significantly increases from NF to FF and the magnitude of these events also increases. The maximum number of extremes can be seen for NorESM2-LM, followed by IPSLCM-6A-LR, EC-EARTH3, EC-EARHT3-VEG, GFDL-ESM4, NorEM2-MM, BCC-ESM-MR and MPI-ESM1-2-LR under different scenarios. MPI-ESM1-2-HR does not show any extreme events in the future. The highest number of hydrological extremes of the 100-year return period is observed as 88 in 25 years for IPSL-CM6A-LR under SSP5-8.5 in FF. The maximum heavy floods of future periods for different return periods vary from 47,700 to 85,690 m³/s. These values are very high compared to previous floods (1998, 2005, 2006 and 2009 year floods) in the KRB, and the increment ranges from 1.72 to 3.1 times the highest flood (27,660 m³/s) that occurred in the year 2009 (Central Water Commission 2019). The analysis of the extreme flood events would help manage the water resources and is very useful for various stakeholders in planning and designing different hydraulic structures.

The research on the KRB is of utmost importance due to its semi-arid environment and susceptibility to climate change, resulting from an uneven spread of rainfall. Previous research has suggested that choosing appropriate GCMs can help to decrease uncertainties in climate change projections (Raju & Nagesh kumar 2014; McSweeney et al. 2015). The selected GCMs using the SU technique agree with the previous studies of different CMIP phases over the KRB. The NorESM and BCCCSM1.1 families of GCMs secured the top rank in evaluating ISMR (Babar et al. 2015; Parth Sarthi et al. 2016). The GFDL and BCCR families of GCMs secured the top rank in evaluating the precipitation in the KRB (Raju & Nagesh Kumar 2015), and the IPSL family performs well for the temperature over India (Raju & Nagesh Kumar 2016). The NorESM, BCCCSM1.1, GFDL, BCCR and IPSL families GCMs are ranked in the top 50% in the present study. However, for the CMIP6 phase climate models, the MPI family performs than other GCMs, which agrees with models for the whole Indian peninsular (Anandhi & Nanjundiah 2015). Therefore, the selected top 50% GCMs were utilised to compel the hydrologic SWAT model. The average annual precipitation of the MME average, along with individual GCM, significantly increases from IMD precipitation in all three future periods; this is in contrast with a previous study, which reported decreasing precipitation projections in the NF using the GHG (Gosain et al. 2006), CNRM-CM5 and GFDL-ESM2M GCM (Chanapathi & Thatikonda 2020) outputs.

The increase in the precipitation resulted in an increase of other WBCs, such as surface runoff, water yield and streamflow, and a decrease in evapotranspiration in the future can be observed in the present study. The results align with previous studies (Kulkarni et al. 2014; Kundu et al. 2017; Nikam et al. 2018; Chanapathi & Thatikonda 2020). For example, the maximum amount of annual average precipitation was projected to increase by about 33.4% under CNRM-CM5 output in FF under the high-emission RCP8.5 scenario (Chanapathi & Thatikonda 2020) compared to the baseline period, whereas in the present study, the MME average is projected to increase about 54% in FF under the SSP5-8.5 scenario. Ensemble average precipitation was projected to increase up to 1,073 mm under the RCP8.5 scenario, whereas in the present study, it is projected to increase up to 1,187 mm under the SSP5-8.5 scenario in the FF. Similarly, the ensemble average values of water yield, surface runoff and streamflow were projected to increase up to 499 mm, 459 mm and 6,021 m³/s, whereas, in the current study, these are projected to increase up to 609 mm, 354 mm and 4,226 m³/s in the FF, respectively. The lower increase in streamflow and surface runoff could be attributed to the incorporation of the reservoir system into the SWAT model. The results of the study suggest that future policies for cropping patterns, cropping periods and reservoir operations must be updated for sustainable water management within the basin. According to Nikam et al. (2018), the number of extreme streamflow events is expected to increase up to 289 for the 100-year return period over a 20-year period, whereas in the present study, the probability of extreme streamflow events for the same return period is projected to be about 88 times for MF.

This study examines the impacts of climate change on the WBCs and hydrology of the KRB for three future time frames. Tier-1 scenarios, SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5, which are similar in magnitude and distribution to the RCPs in CMIP5, were used in the current study. The future scenarios of the selected top 50% GCMs were forced as input to the well-calibrated and validated distributed hydrological model (SWAT) to simulate the WBCs and streamflow. Significant changes were observed for hydrological components in annual and ISMR periods. The future flood frequency analysis of different return periods (25, 50, 75 and 100 years) was carried out using Gumbel distribution over the observed period. Based on these flood extremes, the future hydrological extremes were identified. The following are the key findings of the study.

The MPI-ESM1-2-HR and MPI-ESM1-2-LR GCMs closely matched observed precipitation data during the baseline period, with no significant changes expected in the future except under the SSP5-8.5 scenario during the FF period. The outputs of NorESM2-LM and NorESM2-MM GCMs consistently overestimated future precipitation across all SSP scenarios except in the FF under the SSP5-8.5 scenario. It was observed that the ensemble mean of projected mean annual precipitation significantly increased in the range of 12–54% in the future periods under different SSP scenarios compared to the GCM ensemble. The fluctuation of these predicted precipitation levels will likely have a significant impact on other WBCs in the basin. The ensemble average of annual surface runoff and water yield is projected to increase from 27 to 124% and from 30 to 125%, respectively, under different SSP scenarios for future periods. With the increase in precipitation, ET is also projected to increase under all scenarios in the future periods, but the percentage increase of ET is less than the baseline period ET and the other WBCs in the future. The future streamflow is significantly impacted by changes in future precipitation, and the annual MME average is projected to increase from 31 to 114.5% from NF to FF under four SSP scenarios. The ensemble mean of future flows occurred mostly in September, and the increase in flows can also be observed even in October and November due to monsoon extension. This shows the adverse effects on Kharif crops as these are in the harvesting stage. Policymakers can decide on cropping patterns and cultivation for better agricultural productivity. Future streamflow is significantly impacted by the large increases in precipitation, surface runoff and water yield, which may result in major flood events in the KRB. A significant increase in the monthly flows can be observed for the NF to FF compared to that for the baseline period. Compared to the non-monsoon period, the maximum peak flows are observed during the monsoon period. It is found that the MPI-ESM1-2-HR model will not experience any extreme flow occurrences in the future. The maximum number of streamflow extremes can be observed for NorESM2-LM and IPSL-CM6A-LR outputs. More probable extreme flood events are observed under SSP5-8.5 in the far future for all the return periods. The projected maximum heavy floods in future periods range from 1.72 to 3.1 times the highest recorded flood of 27,660 m³/s in 2009. This information could be very beneficial for reservoir operations in the KRB. Therefore, it may be concluded that there are more water availability chances in the KRB under different CMIP6 climate change projections and also there is a probability of floods in future periods. As a result, this knowledge is beneficial for sustainable development and improved planning and management of water resources.

The climate change impact on the hydrological analysis of the WBC of the catchment is mostly affected by the uncertainties in the climate projections and hydrological model itself. The individual GCM results and their minimum and maximum bounds also show the uncertainties associated with selected projections. The MPI-ESM1-2-HR and MPI-ESM1-2-LR GCMs secured the first and second ranks and had shown less uncertainty range in the future projections compared to other GCMs. The present study also tried to reduce the uncertainty in climate projections by making an ensemble of selected GCMs. Even though the hydrological model is calibrated and validated, the uncertainty may still exist, which can be reduced to some extent by employing an ensemble of hydrological models (Wolock & McCabe 1999; Arnell & Liv 2001). The present study considered historical LULC for future climate change projections; however, alterations in land use, soil composition, management practices and other climate factors will also have a significant impact on both water availability and crop production. The present study results can be improved by employing future land use with respect to the different climate projections. Finally, the results presented here are very helpful to the decision makers for better planning and management of water resources in the basin.

All relevant data are available from an online repository or repositories. For Meteorological data of observed datasets use Indian Metrological Department (IMD; https://www.imdpune.gov.in/). The GCM data can be downloaded from Earth System Grid Federation (ESGF) portal (https://esgf-node.llnl.gov/search/cmip6)

The authors declare there is no conflict.