Streamflow projections under climate change framework for the Mahanadi River catchment, India

In light of recent research revealing the escalating environmental challenges and casualties within river basins (Pradhan et al. 2022; Verma et al. 2022b; Sahu et al. 2023b), the projections from the general circulation models (GCMs) play a vital role in understanding future changes in climate (Mishra et al. 2020a; Sahu et al. 2023b). However, the spatial resolution at which GCMs are run is often too coarse to get reliable projections at regional and local scales (Yang et al. 2020; Rajendran et al. 2022). Precipitation and temperature projections at higher spatial resolution are required for climate impact assessments (Behera 2019; Falga & Wang 2022; Sahu et al. 2023c). Moreover, precipitation and temperature from the GCMs have a bias due to their coarse resolution or model parameterizations (Jin et al. 2018). Considering Coupled Model Intercomparison Project-6 (CMIP6) for analysis is paramount due to its advancements in climate modelling. This latest phase incorporates refined representations of Earth system processes, such as clouds, aerosols, and biogeochemical cycles, yielding more accurate climate projections (Kim et al. 2020; Shiru & Chung 2021). CMIP6 offers valuable insights into regional climate variations and extremes, with improved spatial resolution and comprehensive scenarios (Chen et al. 2020; Gusain et al. 2020; Sahu & Mehta 2024). It enables researchers to assess the impacts of different emission pathways on global and local scales, aiding in informed decision-making for adaptation and mitigation strategies (Adib & Harun 2022; Yang et al. 2022; Anil et al. 2024). Thus, utilizing CMIP6 data enhances the reliability and robustness of analyses, contributing significantly to our understanding of present and future climate dynamics.

In the realm of hydrological modelling, the quest for enhanced accuracy and reliability is perpetual. As climate change continues to exert its influence on hydrological systems, the need for robust methodologies to select the most suitable general circulation model ensembles becomes increasingly imperative (Steinschneider et al. 2015; Shiru & Chung 2021; Rajendran et al. 2022). Additionally, methods to refine streamflow predictions within hydrological models present a promising avenue for advancing the precision of water resource assessments (Yang et al. 2022). Synergistically integrating these methodologies enhances the reliability of hydrological predictions, empowers stakeholders with actionable insights, and fortifies resilience against the impacts of climate change on water resources.

The purpose of this research was to determine which global climate models (GCMs) are the most suitable for use in multimodal ensemble aggregation for the purpose of climate projection. Thirteen GCMs that were part of the CMIP6 were evaluated in terms of their ability to replicate precipitation as well as maximum and minimum temperatures over the Mahanadi River basin. In order to calibrate a model used in hydrological modelling, it is typically necessary to have access to hydrometric data (Abbaspour et al. 1997; Chordia et al. 2022). Once the model has been calibrated, it can be used for a variety of purposes, including simulating past catchment flows, studying the effects of climate change, and making predictions for use in water management. Unfortunately, many studies require investigation at sites without nearby gauging facilities (Gadgil & Narayana Iyengar 1980; Gadgil et al. 1993; Azad et al. 2010; Chakraborty & Singhai 2021; Sahu et al. 2021b, 2024). In such situations, so-called ‘regionalization’ approaches can be used to approximate the historic streamflow at the ungauged sites (Carter & Elsner 1997; Rao & Srinivas 2006a, 2006b; Satyanarayana & Srinivas 2008, 2011; Srinivas 2013; Fathian et al. 2020; Sahu et al. 2023c).

Physical similarity, spatial proximity (SP), and multiple linear regression stand out as the methods that are the most flexible and reliable out of those that are now available (Drogue & Khediri 2016; Choubin et al. 2019). This is dependent on the climatic and hydrological regime of the region that is being studied. These three approaches to regionalization may be found in the vast majority of comparative research, and it is generally accepted that they function exceptionally well under specific conditions (Kanishka & Eldho 2020; Wang et al. 2021). Parameter sets from hydrological models calibrated at other sites are transferred using these techniques. The similarity between catchments, proximity, or other describing variables may all be used in the parameter transfer function (Cooley et al. 2007; Yadav et al. 2007; Christensen et al. 2019; Soni et al. 2021).

At the Mahanadi River basin level, the spatial transferability of watershed model parameters was evaluated. We evaluated whether the rainfall–runoff estimating approach based on the closest neighbour catchments is sensitive to the set of rain gauges used to compute rainfall input at gauged and ungauged catchments in the simulation experiment. The goal is to discover the optimum technique for estimating streamflow time-series at Kalma gauging-sites maintained by the Central Water Commission using hydro-meteorological data from neighbouring catchments. In the present study, the principal objectives are to justify the above aspect through certain innovative considerations: (i) the incorporation of refined representations of Earth system processes (CMIP6) using multi-criteria decision-making techniques and (ii) the incorporation of parameter regionalization based on SP to improve the streamflow simulation of the Soil and Water Assessment Tool (SWAT) model.

The Mahanadi, an interstate river basin, is currently facing challenges posed by hydrological variability and climate change (Gupta et al. 2017; Sahu et al. 2021a, 2022a, 2023b; Lee et al. 2023). In past decades, basin yields and river discharge encountered high annual variability, which included floods in the lower Mahanadi sub-basin during the monsoon season and water scarcity across the basin during the non-monsoon season (Sahu et al. 2020; Singh et al. 2020; Pandey et al. 2022; Verma et al. 2022a, 2023). The Mahanadi's riparian states, including Chhattisgarh and Odisha, are in a stage of rapid urbanization and industrialization, accomplishing high economic activities with an annual population growth of 3.3%. The increasing trends of the suggested characteristics would bring the water resources of the basin under stress, with maximum stress during the dry season (Kumar & Bassi 2021). The Mahanadi has changed dramatically in the last two decades, not only due to increased water use but also due to other factors such as inter-sectorial and interstate allocation (Jin et al. 2018). The riparian states’ ongoing water allocation issues would be put under enormous strain, and in the face of climate change, the challenge would exacerbate extreme events of great magnitude.

The study uses 0.25° × 0.25° high-resolution grid-based precipitation data over the Mahanadi River basin (MRB) for the period 1948–2017 from the Indian Meteorological Department (IMD), Pune. IMD data covers the period from 1901 to 2017, i.e., 117 years, and contains data from 6,329 available on-site stations in India. Also, this dataset reflects variability in rainfall station data over time and is prepared using an interpolation scheme by Shepard (1968). The 2,140 selected on-site stations were found suitable and of sufficient record length to be used for preparing a high-resolution grid database (Rajeevan et al. 2006). Recent past studies concerning the functionality of grid-based high-resolution precipitation datasets for investigating variability and change over the MRB (Sahu et al. 2021a, 2021c, 2021d, 2022a, 2022b, 2022c, 2023a, 2023b) have recently been published. The existing results of the said IMD dataset were studied with the CMIP6 models. As a means of ensuring uniformity between CMIP6 and IMD variables, the CMIP6 models were regridded to a spatial resolution of 0.25°. The effect of regridding with bilinear interpolation was nevertheless tested by comparing the gridded datasets with the raw data for the mean of precipitation and maximum and minimum temperatures across India. No significant difference in the average precipitation or temperature was observed. Hence, the study uses a bias-corrected climate projection for South Asia from the CMIP6, prepared by Mishra et al. (2020a) and available as scientific data at nature.com: https://doi.org/10.6084/m9.figshare.12963008.

The 13 GCMs were chosen based on the availability of daily precipitation and maximum and minimum temperatures for the historical case and two scenarios (SSP245 and SSP585). Table 1 illustrates the detailed description of the data used in this study. Table 2 illustrates the model and source description. The study calculated multivariate statistical features for hydro-climatic parameters, including precipitation, maximum and minimum temperature, river discharge, relative humidity, wind speed, and solar radiation.

The methodology integrates multiple steps to optimize and project streamflow using climate model data and hydrological modelling. First, a compromise programming (CP) approach was employed for the optimal selection of climate models from the CMIP6 ensemble. This ensured that the chosen models best represented relevant climatic variables. Second, SP techniques were used to regionalize the calibrated SWAT model parameters at a gauged site, facilitating the derivation of regionalized parameters for an ungauged site. Finally, the optimal climate model outputs and regionalized parameters were incorporated into a calibrated SWAT model to project future streamflow, providing insight into potential hydrological changes under varying climate conditions.

Climate researchers use CP, a multi-criteria decision-making technique. The primary idea of CP is to locate the ideal spot, or the place where all relevant attributes have their best possible values (Zeleny 1973). Therefore, the best answer is the one that comes closest to the ideal. CP's ability to pinpoint the optimal option helps keep decision-making from getting in the way. CP has been used in climate research to find a middle ground between competing performances, determine the most accurate gridded precipitation data for distinct regions, and rank GCMs (Fattahi & Fayyaz 2010; Tarebari et al. 2018; Salehie et al. 2021; Shiru & Chung 2021; Rudraswamy & Umamahesh 2024). In this research, CP was used to rank GCMs based on the four statistical performance measures: NRMSE (normalized root mean square error), Pbias, NSE (Nash–Sutcliffe efficiency), and R².

For SP to be put into practice, there must be a few catchments in the surrounding area that have been successfully predicted and calibrated by the hydrological model (Steinschneider et al. 2015). The catchment that is geographically closest to the ungauged site is designated as the ‘de facto donor,’ and the parameters of the hydrological model are moved there (Fung et al. 2022).

The SP method is the one that can be put into action with the least amount of difficulty. It is built on the concept that surrounding (or neighbouring) catchments must share physical qualities, such as soil type, slope, land cover, climate data, elevation, and so on. It does not require any information regarding catchment attributes; instead, it operates on the assumption that this is the case. According to this idea, there is no need to search for the catchment that is the most comparable because the ones that are nearby may be ‘similar enough’ merely because of the fact that they are located in the same general area. The simple Euclidean distance between the catchment centroids is used to determine the distance between the ungauged basin and the donor candidates.

SWAT predicts and forecasts the impression of land and water as well as climate change on agricultural chemical yields, sediments, and water in complex catchments under varying land use and soil conditions. The application of SWAT has been widely accepted throughout the world for dealing with different aspects related to agricultural conservation, water management, hydrological processes, best management practices, climate change impacts, water quality and sedimentation, and land-use impacts. A complete review and application of the SWAT model is available in Gassman et al. (2007), while detailed documentation can be found in Neitsch et al. (2011). A short brief on the SWAT model is presented here.

The SWAT model examines the hydrology in two phases: (1) the land phase that controls the quantity of water, sediment, nutrient, and pesticide loading in each catchment; and (2) the water phase, i.e., water routing, that defines the movement of water, sediment, nutrient, and pesticide in the network channel. In the SWAT model, a basin is processed into sub-catchments, and each sub-catchment is represented by at least one stream order. These sub-catchments are further processed into lumped areas, which are hydrological response units within the catchment forming unique combinations of slope, land use, and soil.

SWAT-CUP (SWAT-Calibration and Uncertainty Procedures) produces output results at each station as 95PPU as well as showing the best fit (e.g., the simulation run with the best objective function value) (Arsenault et al. 2019), but for simplicity and clarity of presentation, we only show the calibration and validation results for the best simulation as a continuous graph and report the overall statistics. The river discharge in the upper and middle Mahanadi River basin, i.e., the portion of the catchment falling in Chhattisgarh state, covers 52.9% (74,970 km²) of the total catchment area. Sensitivity analysis identifies which model parameters have the most influence on the outputs of interest. The study region was calibrated using parameters such as CN2, GWQMN, REVAPMN, SOL_AWC, ESCO, GW_REVAP, GW_DELAY, SLSUBBASN, ALPHA_BF, ALPHA_BNK, SOL_BD, SOL_K, CH_N2, and CH_K2. Parameters were then ranked based on their impact on model outputs.

Tables 3, 4 and 5 describe the best model selection. Despite the fact that the ideal values for each GCM are different from one another, the overall ranking shows that a GCM that may have its value as the most ideal for more metrics does not necessarily rank highest. This is because the ideal values vary from GCM to GCM. For instance, MPI-ESM1-2-HR is placed fourth overall despite having three of its statistical magnitudes as the most ideal value (Table 5). The four GCMs with the highest ranking for precipitation are the MPI-ESM1-2-HR, NorESM2-MM, EC-Earth3-Veg, and MRI-ESM2-0 models (Table 3). The model with the lowest ranking for precipitation is the CanESM5 model. The rankings of the GCMs determined from CP are provided in Table 4, and the performance metrics for maximum temperature were calculated for each and every GCM. MPI-ESM1-2-HR, MPI-ESM1-2-LR, INM-CM4-8, and EC-Earth3-Veg are, in descending order, the GCMs with the highest ranks possible. CanESM5 was the GCM with the lowest maximum temperature rating when utilizing the CP technique.

The rankings of the GCMs determined from CP are provided in Table 5, and the performance metrics for minimum temperature were calculated for each and every GCM. MPI-ESM1-2-LR, MRI-ESM2-0, EC-Earth3-Veg, and MPI-ESM1-2-HR are, in descending order, the GCMs with the highest ranks possible. CanESM5 was the GCM with the lowest minimum temperature rating when utilizing the CP technique.

Table 6 illustrates the scores and overall rankings achieved by the various GCMs in their attempts to replicate the observed precipitation as well as the maximum and minimum temperatures reported from CP. The highest GCM rankings for precipitation are MPI-ESM1-2-HR, NorESM2-MM, EC-Earth3-Veg, and MRI-ESM2-0. MPI-ESM1-2-HR, MPI-ESM1-2-LR, INM-CM4-8, and EC-Earth3-Veg retain the first four slots for maximum temperature, while MPI-ESM1-2-LR, MRI-ESM2-0, EC-Earth3-Veg, and MPI-ESM1-2-HR retain their position order from the top for minimum temperature. The lowest ranking in descending order (bottom four) for precipitation are INM-CM4-8, ACCESS-CM2, ACCESS-ESM1-5, and CanESM5. Similarly, for maximum temperature, the models are MRI-ESM2-0, ACCESS-ESM1-5, ACCESS-CM2, and CanESM5, while ACCESS-CM2, INM-CM4-8, NorESM2-LM, and CanESM5 retain the lowest ranking for minimum temperature.

SWAT application incorporates a detailed assessment of hydrological system processes to study the impact of land management practices, water management interventions, and climate change on precipitation, sediments, pesticides, chemical fertilizer (agriculture) yields, and other applications (Gassman et al. 2007; Arnold et al. 2012; Abbaspour et al. 2015). The SWAT model was applied to each of the eight-gauge catchments and calibrated using monthly climatic and streamflow data from January 1985 to December 2017. The data were split into calibration (January 1985 to December 2003) and validation (January 2004–December 2017) periods. Before calibration, a warm-up period of four years was used for initialization so that model parameters attained appropriate initial values. Each catchment was divided into several elevation zones, and this interval was selected to balance the total number of elevation bands that could be accommodated in the SWAT modelling setup. This threshold was also appropriate to avoid having too many or too few divisions of the study catchments. Each elevation zone was divided into three vegetation zones, namely forest (zone 1), cropland (zone 2), and range/bare lands (zone 3). The elevation is known to have major impacts on the distribution of rainfall and temperature, which have already been studied in the region.

The calibration results showing the comparison of observed and simulated streamflow are provided in Table 7, summarizing the monthly NSE, R², P-factor, R-factor, RSR (root mean square error–observations standard deviation ratio), and bR² estimates. The NSE values were quite good for most of the catchments (i.e., >0.6), excluding one station (Kotni) in validation whose value is 0.35. Similar patterns were indicated by R², P-factor, and R-factor, depicting reasonably good model performance in most cases. Although during the validation period, NSE and R² values were congruent with their corresponding values during the calibration period, the values were reasonably good in most cases (i.e., NSE >0.6). The calibration and validation results suggest that the optimized parameter sets could simulate the rainfall–runoff relationships reasonably well in most cases. However, it should be noted that the models are not perfect and may involve uncertainties resulting from model structure, input data, and parameter values. Therefore, the results should be interpreted cautiously.

The discharge of Jondhra*, the sub-tributary (Jonk and Pairi rivers), and Rajim (gauging site, Mahanadi) were very well simulated at Seorinarayan* (GDSQ, area = 48,050 km², average annual discharge rate = 6,254.73 cumec), suggesting a P-factor of 0.89, i.e., 89% of the observed data lie within the 95PPU band with a large R-factor (0.70). The model accuracy was 0.79 (R²) and 0.76 (NSE), suggesting the model had performed above the satisfactory mark. Further to this, the cumulative discharge of Seorinarayan* and Bamnidih (Hasdeo) was simulated at a P-factor of 0.97, i.e., 97% of the observed data lie within the 95PPU band with large uncertainty (R-factor = 0.89) at Basantpur* (GDSQ, area = 57,780 km², average annual discharge rate = 7,442.33 cumec). The model performed exceptionally, with R² = 0.93 and NSE = 0.93. The lowest uncertainty was observed at Simga (R-factor = 0.44) and Kurubhata (R-factor = 0.45), and the largest at Jondhra (R-factor = 1.0). The model was unable to simulate Kotni during validation, with only 35% of observed data encountered in the 95PPU band and an overall accuracy of R² = 0.35 and NSE = 0.35 (Figure 5 and Table 7).

In this section, we aim to see if a better description of the rainfall input in calibration catchments leads to a more accurate estimate of the flow in the target catchment. According to our regionalization approach, we moved the best parameter sets (i.e., parameter sets optimized with rain gauges to maximize the model's validation performance) from the three nearby catchments to the recipient catchment. The model was then run on the receiver catchment for the given validation periods using rainfall input derived using all rain gauges. We evaluate the model effectiveness of the calibrated model in validation using rainfall input derived with subsets of rain gauges, i.e., at-site calibration, to examine the influence of the regionalized model on the streamflow error at the outflow of a target catchment. Application of an SP-based regionalization approach to model parameter values can take advantage of neighbour-catchment-based knowledge of optimal streamflow estimation. Table 8 illustrates the ranking of the SP metric (centroid distance) between Kalma and other gauged sites. Kalma's ungauged site was modelled based on the averaged parameter values of Basantpur, Kurubhata, and Bamnidih (Table 9).

The average annual rainfall for the period 1985–2017 was 1189.5 mm, which comprises components such as surface runoff (156.23 mm), baseflow (74.60 mm), shallow aquifer recharge (166.67 mm), deep aquifer recharge (8.33 mm), and actual evapotranspiration of 67.44% (i.e., 802.2 mm). When the SWAT model was re-simulated with the CMIP6 model data and the parameterization and optimization output, the water-balance components significantly changed. The proportion of the water balance for the re-simulation was as follows: annual precipitation (1,250.9 mm) increased by 5.16%, surface runoff (192.57 mm) increased by 23.26%, baseflow (110.15 mm) increased by 47.65%, shallow aquifer recharge (474.1 mm) increased by 184.85%, deep aquifer recharge (23.7 mm) increased by 184.51%, and actual evapotranspiration of 38.0% (i.e., 474.6 mm) decreased by 40.84%. Since evapotranspiration is the major factor in water losses, it is justified by the fact that the dominant land use and land cover in the study area is green, at 95.22% (agricultural land at 64.92%, forest at 30.30%).

The components, including shallow aquifer recharge, lateral flow, and surface runoff, are the major contributors to water yield in the streamflow at the whole catchment outlet. An amount of 776.82 mm of water yield is the difference between total water yield (800.52 mm) and deep aquifer recharge (23.7 mm). The water yield obtained here for the CMIP6 model data is 96% higher compared with the historical baseline of 396.26 mm.

The precipitation history is as follows: the trends of the Mahanadi River basin based on the gauged and grid data portray a different picture than what we actually obtained in this study, which is supported by many results of previous research findings (Jin et al. 2018; Sahu et al. 2020,, 2021d, 2022a, 2023b, 2024; Kumar & Bassi 2021; Kumar et al. 2023). Since the study area is located near the equator, several climate phenomena such as Hadley and regional Walker circulation, ITCZ (inter-tropical convergence zone), and El Niño–Southern Oscillation (ENSO) are predominant. The Mahanadi River basin, which is just beneath the Gangetic Plains, is equally influenced by the low-pressure belt created by the trade winds (ITCZ), also called the Indian Monsoon Trough. Interaction between Hadley and regional Walker circulation and ENSO influences Indian summer monsoon rainfall (ISMR) (Ashok et al. 2001; Cherchi et al. 2021; Feba et al. 2021), and ‘SST sea surface temperature anomalies during ENSO events significantly draw the Pacific ITCZ equatorward, causing erratic weather patterns and drastically affecting rainfall’ (Torrence & Webster 1999; Pausata et al. 2020; Lee et al. 2023). Furthermore, the monsoon indices are a dynamic construction based on the interaction between convection and circulation (Zhang et al. 2022). The research has validated ENSO's effect on ISMR and also augmented the variability of monsoon indices in the same domain (Das et al. 2020; Reshma et al. 2021; Hussain et al. 2022; Rehana et al. 2022; Roy et al. 2022). The present study considers IMD grid data to study the hydroclimatology of the basin, and the findings of the same are studied with the 13 CMIP6 GCMs. The study simulated the future streamflow at the outfall of the Mahanadi River catchment in Chhattisgarh. The simulated characteristics of the Mahanadi River catchment as portrayed by the statistical analysis herein create further scope for research on the hydroclimatology of a basin subjected to mixed climate phenomena.

The findings at the Kalma outlet were well supported by effective modelling and validation of gauged sites (Figure 8(a) and 8(c)). The results were carefully verified at critical intersections, and their simulations were validated so that they could be effectively carried forward to the next gauged station. The Jondhra site covers the upper Mahanadi area of approximately 30,761 km², which was well simulated, with 91% of the data lying within the 95PPU and uncertainty reaching 1.0 (permissible <1.35). Similarly, Basantpur, covering most of the upper and middle Mahanadi River basins (approximately 48,050 km²) and with an average annual discharge rate of 7,697.91 cumec, was well simulated with 93% and 89% model accuracy and uncertainty, respectively. The spatial transferability of watershed model parameters was evaluated for Kalma using SP-based regionalization. The model based on regionalized parameters for Kalma performed with 96.8% accuracy, and the results were analysed with the CMIP6 (MPI-ESM1-2-HR) model to assess streamflow projection. The projection appeared significant towards an increasing trend with a magnitude of 91.81 mm/year (JJAS) during the wet season (Figure 8(b)).

The above result is in line with the claim that ‘the climate monsoon indices have an effect on transport pathways of atmospheric water vapour and are associated with substantial thermal gradients between low pressure in the north (warm Asian continent) and high pressure in the south (cold water bodies), especially in the Indian Ocean, Arabian Sea, and Bay of Bengal’ (Li & Zeng 2002). This might be due to the Arabian Sea warming, which diminishes the west trade winds across the southern Bay of Bengal and gathers more moisture in the Arabian Sea (Mishra et al. 2020b), worsening the monsoon precipitation drop in the Mahanadi.

Through the above-mentioned studies, a literature review, and strict parameter-estimation procedures, we were able to calibrate, validate, and analyse the uncertainty of the SWAT model for the Mahanadi River basin in a way that was quite good. As a result, the SWAT model could comfortably be used to analyse several scenarios of water use in the basin. Two scenarios from the three best models of CMIP6 climate data were used to simulate future rainfall and predict future streamflow. The output streamflow from these scenarios was compared with the baseline period of 1985–2017. The baseline simulations were run with the model structure that was ultimately chosen and a set of parameters that were derived via the calibration procedure.

• (1) Here, MPI_ESM1_2_HR stands out as a robust and reliable climate model across all three variables, particularly excelling in temperature predictions while also being the best for precipitation in terms of correlation. Its ability to closely align with observed values and maintain low error-rates positions it as a top contender among the models assessed in the Taylor diagram.

• (2) The Kalma site was modelled based on the averaged parameter values of Basantpur, Kurubhata, and Bamnidih. The modelling trend statistics of the core river intersections (Jondhra, Seorinarayan, and Basantpur) suggest a significant impact at the catchment outlet (Kalma) of about 92 mm/year during the wet season (JJAS).

• (3) At Kalma, the proportion of overland flow gets cumulated and is recorded as low compared with that of the other catchments. This is mainly a cumulative change in the percentage of forest cover and the soil properties within the catchment; this suggests higher surface runoff generation with a lower infiltration rate. In addition, there is an average gradient of 1:30–1:35 (steep) towards the south of Mahanadi (Chilika Lake) and 1:50–1:43 (gentle) towards the north of Mahanadi (Maikal Hills), which suggests that perched water flows out easily and generates lateral baseflow.

• (4) The modelling output also suggests that with a changing climate, Jondhra is expected to increase river discharge by 44.67%, and on the other hand, Seorinarayan and Basantpur are expected to increase by 27.88% and 38.10%, respectively, in the near future (2019–2050). The current research work addresses the changes in precipitation patterns and climate change impacts that alter the hydrological regimes of the Mahanadi River basin system.

The research concludes that management of available water resources and sustainable use in the Mahanadi Basin depends primarily on an understanding of the pervasive high level of variability in hydrology and water resources, for which this study has laid a solid foundation. Due to climate change, the current planning of water allocation requires a thorough revision, which will essentially require a balance between human and environmental use.

The precipitation data used in this study were provided by the Indian Meteorological Department (IMD), Pune, and are highly appreciated. Suggestions and comments from reviewers are greatly acknowledged.

R.T.S., S.D.T., and U.R. conceptualized the process; R.T.S. and U.R. supervised and reviewed the work; R.T.S. and S.D.T. developed the methodology; R.T.S. wrote the original draft; R.T.S. wrote and edited the article.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.