Effect of climate and land use change on discharge in the Betwa River basin, India, using SWAT based on CMIP6 and the Land Change Modeler model projections

Water resources are becoming increasingly scarce, particularly in arid, semi-arid, and sub-humid regions of the world under climate change (CC) (Srivastava & Chinnasamy 2024), land use change, and unprecedented population growth (Veldkamp et al. 2017; du Plessis 2023). The 2020 edition of the United Nations World Water Development Report (UNESCO, 2020), entitled Water and Climate Change, states that by 2050, most of the population in the Middle East and South Asia will face severe water stresses. As per the World Development Report 2021 (World Bank 2021), India has the largest freshwater withdrawals (760 billion m³/year) in the world. A report titled Composite Water Management Index, published in 2018 by the National Institution for Transforming India (NITI Aayog 2018), mentioned that the country is facing high to extreme levels of water stress. The Central Water Commission (CWC), New Delhi, India, stated in its 2019 report (CWC 2019) that the water demands in the country will exceed the water supply by 2050.

The increased demand for food and energy by the growing population, urban sprawl, rapid industrialization, and increased socio-economic activities have accelerated the pace, extent, and magnitude of land use and land cover change (LULCC) (Lambin & Meyfroidt 2011; Roy et al. 2024), which in turn influences land surface–energy interactions, the hydrological cycle, ecosystem functioning, global warming, and biogeochemical feedback (IPCC 2021; Taylor & Rising 2021; Singh et al. 2022a, 2022b). LULCC and hydrological processes are interrelated in an intrinsic, complex, interplaying, and nonlinear manner (Lambin et al. 2001; Kumar, N. et al. 2018). LULCC is also coupled with regional temperature patterns, radiative forcing, and energy fluxes, which influence the climate system and thereby the water-cycle components of a region (Foley et al. 2005; Singh et al. 2021).

The Coupled Model Intercomparison Project Phase 6 (CMIP6) has 100 distinct climate models from 40 different modeling groups and remarkably higher climate sensitivity (Tokarska et al. 2020). CMIP6 has equilibrium climate sensitivity (ECS) values higher than any of the CMIP5 models (Kumar, N. et al. 2023; Kumar, N. et al. 2023; Kumar et al. 2024). The CMIP6 has been used to define the physical science basis of the Sixth Assessment Report (AR6) of the IPCC (2021). As per AR6 of the IPCC (2023), the average global mean surface air temperature (GSAT) during the years 2001–2020 has increased by about 0.99 °C compared with the value pre-industrial revolution. The recent increase in GSAT caused by global warming promotes the vapour-holding capacity of the atmosphere, which in turn affects the amount, extent, intensity, and frequency of precipitation (Gupta et al. 2022). The hydrology of the basin, especially runoff mechanisms, is significantly affected by changes in temperature and precipitation (Mahanta et al. 2024).

Nowadays, hydrological models with geographic information system integration capability have become an indispensable tool for quantifying different components of hydrologic systems (Refsgaard et al. 2022). Hydrological models provide imperfectly simplified representations of real hydrological systems using mathematical equations and are widely used by researchers for analyzing the influence of LULCC and CC on water resources (Yoosefdoost et al. 2022; Kumar, A. et al. 2023; Kumar, N. et al. 2023; Kumar et al. 2024). Many studies have examined the individual and interactive impacts of land use–land cover (LULC) and CC on water-cycle components using hydrological modeling approaches (Kumar, N. et al. 2018, 2022; Kumar, A. et al. 2022). The Soil and Water Assessment Tool (SWAT) model has been the most extensively used open-source semi-distributed hydrological model worldwide (Arnold et al. 2012; CARD 2020). SWAT is widely employed by researchers for hydrological studies in India (Kumar, A. et al. 2022; Kumar, N. et al. 2022; Trivedi et al. 2024). Sahoo et al. (2018) applied SWAT to evaluate the combined impacts of LULC and CC and reported a seasonal increasing trend of discharge of the Gandherswari River basin, West Bengal, India. Chanapathi & Thatikonda (2020) have investigated the impacts of LULC and CC over the Krishna River basin, India, based on SWAT and concluded that both LULC and CC affect hydrology. Sharannya et al. (2021), to assess the long-term effects of LULC and CC on the hydrology of two rivers in the Western Ghats of India, found that the stream flow in the rivers might increase in the future. Rani & Sreekesh (2021) reported that hydrology is more sensitive to CC than land-cover changes. Santra Mitra et al. (2021) quantified the impact of climate and land use change on runoff in the Kangshabati River basin of West Bengal, India. Their findings indicate that surface runoff increases as precipitation patterns change due to CC. Sharma et al. (2022) applied SWAT over the Dharoi catchment in the Sabarmati River basin, India, to assess the effect of climate and land use changes and reported that CC is the major driver for increasing streamflow. Other hydrological models, either semi-distributed or fully distributed, such as HEC-RAS (Hydrologic Engineering Center River Analysis System), HEC-HMS (Hydrologic Engineering Center Hydrologic Modeling System), HBV (Hydrologiska Byrans VattenbalansavdeIning) (Bhattarai et al. 2018), SPHY (Spatial Processes in Hydrology), and VIC (Variable Infiltration Capacity), are also available and have been extensively applied for the different river basins. Before CMIP6, representative concentration pathways (RCPs) were widely analyzed for the impact of CC (Iltas et al. 2024) on the river basin.

Central India, where the rainfall regime is uncertain (Danodia et al. 2021; Som & Dey 2022), erratic, and comparatively short due to the late onset and early withdrawal of monsoons, is facing a more severe situation of water stresses compared with other regions of the country (Roxy et al. 2017; Kumar & Mishra 2020; Saharwardi et al. 2021). Jeet et al. (2022) used SWAT for site selection for building a water-harvesting structure in the Betwa River basin (BRB), central India, and found that the upper region of the basin has the highest potential for water harvesting. According to Niranjannaik et al. (2022), the upper portion of the BRB has experienced a reduction in the magnitude of rainfall (2.07 mm/year) in the last 25 years (1993–2018), resulting in a significant decrease in the level of groundwater, especially in the lower basin. Kumar, A. et al. (2022) and (Kumar, N. et al. 2022) have used SWAT to simulate the combined effects of LULC and CC (CMIP6) on the surface runoff in the Upper Betwa River Catchment, central India. They reported an expected decrease in surface runoff. Hence, assessment of the spatio-temporal effects of LULC and CC on hydrological processes and future water-dynamics is vital for ensuring regional water security and devising sustainable management strategies (Pal et al. 2022; Dolgorsuren et al. 2024).

The effects of LULC and CC on the discharge over the entire Betwa River basin (BRB) under CMIP6 need to be studied. Therefore, we assess discharge behavior under both CC and LULCC using CMIP6. It will improve the understanding of hydrological processes in the basin under CMIP6.

The input data were Shuttle Radar Topography Mission (SRTM), digital elevation model (DEM), LULC, soil data, daily T_max, T_min, daily precipitation (meteorological data), station discharge data, and other auxiliary datasets (soil type and topography). Table 1 provides a detailed description of the datasets used in the study.

The historical gridded meteorological data of the period (1984–2018) of the Indian Meteorological Department (IMD), Climate Hazards Group Infrared Precipitation with Stations (CHIRPS), CMIP6, and National Centers for Environmental Prediction (NCEP) reanalysis (bilinear interpolation) data on a daily time-scale were collected. We used the CHIRPS dataset, which has quasi-global coverage spanning 50°N to 50°S and 180°W to 180°E and provides rainfall data at a high spatial precision of 0.05° latitude–longitude. Compared with the IMD gridded rainfall data, which has a resolution of 0.25° × 0.25°, this resolution is finer. In our previous work (Singh et al. 2022a, 2022b), we compared the CHIRPS and IMD data and concluded that both datasets have significant levels of correlation. That being said, CHIRPS has a higher resolution, and that is why we choose to work with the CHIRPS data above the IMD data for the better performance of the model. CHIRPS data can provide a good substitute for steep watersheds where rain gauge stations are limited or nonexistent (Varghese et al. 2024). Even though IMD gridded datasets have many applications and are deemed suitable for numerous hydrometeorological research projects (e.g. Saikrishna et al. 2022), it is advisable to use data collected from at least two to three grids. Where sub-kilometre spatial resolution may be crucial, CHIRPS is therefore less ambiguous than coarser IMD-derived products. The employ of CHIRPS data which has earlier been validated using cross-validation of empirical tests for the Indian subcontinent (Saicharan & Rangaswamy 2023) would lead to a higher degree of authenticity of the results and analysis.

CMIP6 provides climate projections based on a new scenario framework, i.e. socio-economic pathways (SSPs) in conjunction with the RCPs of CMIP5 (O'Neill et al. 2016; Eyring et al. 2019; Kumar, A. et al. 2022; Kumar, N. et al. 2022). CMIP6 provides historical simulations from 1850 to 2014 (the 20th century) based on observed natural and anthropogenic forcings. CMIP6 provides future simulations from 2015 to 2100 (the 21st century) based on SSP emission scenarios. The CMIP6 framework includes nearly 300 climate models developed by 49 modeling groups, which are set for integration into the AR6. Numerous studies have shown the strong performance of these models based on observed scenarios (Aadhar & Mishra 2020; Almazroui et al. 2020; Gusain et al. 2020; Mishra et al. 2020) over the Indian region. For this study, we selected two CMIP6 models ACCESS-ESM1-5 and MRI-ESM2-0 due to their fine spatial resolution and ECS values above 3.0 °C. ECS, the projected long-term warming response to a doubling of atmospheric CO₂ (Meehl et al. 2020), is a critical indicator of how the climate system may respond to substantial future warming. Additionally, model selection was prioritized based on the availability of datasets (2015–2100) and relevant predictor variables across various SSPs.

Statistical Downscaling Model (SDSM) is public-domain software that generates fine-resolution CC scenarios (Wilby et al. 2002) by downscaling general circulation model (GCM) outputs. It operates on combined downscaling methods, namely multiple regressions and stochastic weather generators, to establish a statistical relationship between the local predictor(s) (climate variables) and the predictors (large-scale climate variables) derived from NCEP reanalysis data and GCM output. SDSM performs downscaling of climate projections using five basic steps. For more details readers can consult the website: https://www.sdsm.org.uk/. We did not apply bias correction to the GCM models used in this study. Rather, to minimize variation and achieve comparability, the findings were scaled against the observed data (IMD). As per the developed methodology used in SDSM for statistical downscaling of GCM models, only data standardization/normalization is required for normalizing the GCM predictors and observed predictands before predicting the future scenarios (Wilby et al. 2002).

The NCEP and GCM predictors with significant statistical relationships to in situ climate data were selected (Table 2). The SDSM model was calibrated (1984–2014) and validated (2015–2018) using these predictors and observed IMD variables. To optimize SDSM, the ordinary least squares method was employed (Huang et al. 2011). A statistical comparison was conducted between synthetically generated climate variables and observed data for accuracy assessment. Finally, future climate projections (2018–2100) were made using NCEP predictors and CMIP6 models.

The maximum likelihood classification algorithm was applied to Landsat (5, 7, and 8) images to classify the data of the years 1990, 2000, 2010, and 2020 (pre-monsoon seasons). The modified classification scheme of the National Remote Sensing Agency, India, and the Anderson classification system (Anderson et al. 1976) have opted to classify the study area into seven different LULC classes, namely built-up land, agricultural land, water bodies, barren land, shrub land, open forest, and dense forest.

Cellular automaton–Markov chain (MC) was used to predict the plausible future LULC. Land Change Modeler (LCM) assesses the influence of various factors on future LULC changes, quantifies the extent of land cover transformation between preceding and subsequent LULC states, and subsequently computes the relative magnitude of transitions across distinct LULCs (Singh et al. 2022a, 2022b). Multilayer perceptron (MLP) performance stands out when multiple transition types are modeled, demonstrating greater efficacy in modeling the nonlinear relationship between land change and the associated explanatory variables. MLP accurately predicts the land areas that are anticipated to undergo a change from the later date image to the specified simulation date. The receiver operating characteristic for traits was applied to validate the predicted LULC of 2020. Other details pertinent to the predicted LULC of the entire study area are available (Singh et al. 2022a, 2022b).

SWAT is a conceptual, semi-distributed, time-continuous, physically based, comprehensive, process-oriented model. It simulates water-cycle components and the effects of management practices on fluxes of energy and matter at daily time-steps but can aggregate the results to monthly or annual output (Arnold et al. 1998, 2012). SWAT lumps areas having homogeneous soil, topography, management, and land use characteristics into unique hydrologic response units (HRUs) within the sub-watersheds of a basin. These HRUs have no interconnection among them and are routed individually following the law of water balance toward the outlet of the sub-basins (Arnold et al. 1998; Neitsch et al. 2011). The SWAT model computes surface runoff based on the soil conservation services (SCS) curve number method (Anand et al. 2018).

SWAT is a semi-distributed model that simulates watershed processes; reliable hydrometeorological data are necessary for model calibration. They include climate data, land use and land cover maps, and soil texture information. Daily gridded climate data for the period (1984–2018), with three years (1984–1987) as a warm-up period, including maximum and minimum air temperature, wind speed, solar radiation, and relative humidity, were obtained from the Global Weather Data for SWAT (available at http://globalweather.tamu.edu) with a spatial resolution of 0.30° × 0.30°. Precipitation data for the same period, with a finer spatial resolution of 0.25° × 0.25°, were sourced from the IMD (http://www.imdpune.gov.in). The valley was represented by two gauging stations located in Shahijina where recorded river discharge values were employed in model calibration and validation to simulate the hydrologic processes.

In Sequential Uncertainty Fitting 2 (SUFI-2), the only source of variability is parameter uncertainty, which includes driving variables, the conceptual model, parameters, and measured data (Abbaspour et al. 2015). Confidence intervals of 2.5% and 97.5% are estimated using an output variable sampled at Latin hypercube 95% percent prediction uncertainty (PPU) (Abbaspour et al. 2007). The ideal value of the p-factor equals 1, which suggests that the model performs well on every set of data while predicting that 100% of the data are incorporated by the predictive uncertainty bounds. However, this may also mean high uncertainty in models with lower-quality outputs, which most of the time would be true (Arnold et al. 2012; Abbaspour et al. 2015). The r-factor as well as the p-factor are necessary for evaluating model calibration and uncertainty (Arnold et al. 2012; Abbaspour et al. 2015). The p-factor expresses the proportion of real-life data taken by models while the r-factor estimates the range of the uncertainty of predictions. An r-factor that is close to zero, where it is coupled with survey data, indicates the correct calibration of the model (Yang et al. 2008). Potential buyers also need a lower r-factor because it can signify smaller uncertainty, even though to achieve a high p-factor one often has to bear more r-factor. Therefore, moderation of the p-factor and minimizing the r-negative factor serve as reasonable strategies for minimizing the volatility of the model. When the acceptable values of both factors are reached, the parameter uncertainties are in the calibrated parameter range (Abbaspour et al. 2004, 2009).

In this work, the impact of land use change and CC was assessed. Further, the combined effects of future land use and climate scenarios were also assessed. The LCM and SWAT models were applied to understand the streamflow behavior in the basin.

SWAT-CUP has a SUFI-2 algorithm-based hydrological parameter sensitivity and uncertainty analysis framework (Abbaspour et al. 2007). SUFI-2 incorporates Latin hypercube techniques that facilitate the Latin hypercube–one-factor-at-a-time approach to ascertaining the most sensitive parameters. A set of sensitive parameters comprised in SUFI-2 has been used for calibration and validation purposes (Table 3).

Streamflow plays a pivotal role in the global hydrological cycle and constitutes a critical variable for enhancing our comprehension of flood and drought risks, water resource management, and the impacts of CC (Asadieh & Krakauer 2017; Wang & Liu (2023)). The study attempted to quantify the synergetic effects of LULC and CC on the streamflow. Desai et al. (2021a) and (2021b) employed SWAT for the analysis of the average annual water balance of the BRB, central India, and observed that of the total received annual precipitation of 878 mm, 61% was lost as evapotranspiration, 8% flowed as baseflow, and only 31% contributed to the surface runoff in the basin. The Betwa catchment is undergoing extensive and intensive changes due to hydrometeorological variability and anthropogenic disturbances (Pandey & Palmate 2019; Palmate et al. 2021, 2022; Singh et al. 2022a, 2022b). Jeet et al. (2017) have observed uneven rainfall patterns over the BRB and found that the upper portion of the basin receives more rainfall compared with the middle and lower parts of the basin. This study revealed that the magnitude of the rainfall in the basin is consistently decreasing. Suryavanshi et al. (2017) have used SWAT to quantify the water balance components of the BRB in central India. Their study showed an increasing trend in rainfall and a decreasing trend in ET from the basin during the monsoon and winter, while the opposite trend occurs during the summer. Desai et al. (2021a) and (2021b) have used SWAT to evaluate the changes in surface runoff from the BRB under the CMIP5 scenario and observed an expected increase of 4%–29% and 12%–48% for the periods 2040–2069 and 2070–2099, respectively.

Previous work (Singh et al. 2022a, 2022b) quantitatively analyzed classified LULC maps of BRB and revealed a reduction in the area occupied by agricultural land, dense forest, and shrubland, accompanied by an increase in open forest and built-up areas. Palmate et al. (2017a, 2017b) also concluded that agricultural land within the BRB is expected to decrease by 2100. Kumar, A. et al. (2023), Kumar, N. et al. (2023), and Kumar et al. (2024) attributed the expansion of built-up areas and reduction in forest cover in the BRB to population growth.

Temperature and rainfall are critical climatological factors that intricately influence the hydrology of a region (Gleick 1989; Singh et al. 2022a, 2022b). In this study, average annual temperature across the BRB exhibited a persistent increase throughout the baseline period (1987–2018), whereas average annual rainfall (mm) over the entire basin during the same period showed a slight decline. Kumar, A. et al. (2023), Kumar, N. et al. (2023), and Kumar et al. (2024) also observed a decreasing trend in annual rainfall from 1980 to 2018, along with a significant increase in the annual mean temperature in the BRB. Mondal et al. (2015) documented a significant decline in rainfall (at the 95% confidence level) in East Madhya Pradesh and Chhattisgarh, regions within the West Central India zone. The study also observed a pronounced rise in both minimum and maximum temperatures across much of the country, with the most notable increases occurring in northeast India, north-central India, the east and west coasts, and the interior of peninsular India. Overall, the findings reveal a clear trend of reduced rainfall and rising temperatures across most months and seasons, underscoring the evident impact of CC on both precipitation and temperature patterns throughout the BRB. Guhathakurta et al. (2015) observed a long-term decline in monsoonal rainfall over the Monsoon Core Region of India (MCRI), defined by the coordinates 74.5°–86.5°E and 16.5°–26.5°N. Our findings corroborate this trend, as most of the basins within the MCRI also show a similar reduction in monsoonal rainfall over the same period. Earlier, Suryavanshi et al. (2017) reported satisfactory results in SWAT model calibration and validation using SWAT-CUP at the outlet of the BRB. The SWAT model's performance using the CHIRPS dataset showed favorable results due to the dataset's fine resolution and greater spatial coverage across the river catchment area. CHIRPS precipitation data proved more reliable for estimating total water consumption within the basin. CHIRPS datasets have performed better than other sources, making them well-suited for hydrological modeling and CC studies in similar topographical and climatic regions across India Venkatesh et al. (2020). Researchers, such as Dembélé et al. (2020), Kumar, A. et al. (2022), Kumar, N. et al. (2022), and Singh et al. (2022a), (2022b), have also compared CHIRPS satellite-based precipitation with IMD data for hydrological analysis across various regions in India, reporting a significant positive correlation. The present study shows ALPHA_BF as the most sensitive parameter for the entire BRB, which aligns with Kumar, A. et al. (2022) and Kumar, N. et al. (2022), who also identified ALPHA_BF as the most sensitive parameter in the upper part of the Betwa River. The parameters with higher t-stat values and lower p-values were identified as the most sensitive (Abbaspour et al. 2007). Desai et al. (2021a), (2021b) also achieved a comparable level of success in evaluating model performance with multi-site calibration on a monthly time-step, identifying SOL_AWC as one of the most sensitive parameters in BRB. The effect of land use and CC on river discharge was analyzed using two GCMs (ACCESS-ESM1-5 and MRI-ESM2-0) from CMIP6 under the scenarios SSP245 and LULC 1990 for 2018–2100. The simulation of average annual discharge based on MRI-ESM2-0 (SSP245) using the LULC scenario from 1990 indicated a decreasing trend from 2018 to 2100. In contrast, projected streamflow using ACCESS-ESM1-5 (SSP245) and the same LULC scenario showed a substantial increase in river discharge compared with MRI-ESM2-0, albeit with a slight overall declining trend. Future climate predictions from the ACCESS-ESM1-5 and MRI-ESM2-0 models indicate an approximate 0.4 °C increase in mean annual temperature for SSP245 in the basin. The future projections indicate a decrease in average annual rainfall of approximately 120 mm for both the ACCESS-ESM1-5 and MRI-ESM2-0 models in the basin.

SWAT-CUP-based SUFI-2 algorithm was used for calibration, sensitivity, and uncertainty analysis. The hydrological model's sensitivity analysis identified 17 key parameters, achieving ‘Good’ calibration and ‘Satisfactory’ validation results. The most sensitive factors for the basin, according to sensitivity analysis, were ALPHA_BF, OV_N, CH_K2, CH_N2, ESCO and SOL_AWC. The SUFI-2 methods display the smallest differences between the observed and SWAT-simulated flow; additional iterations and parameter range adjustments are necessary for satisfactory results. In the validation phase, the SWAT model showed minor uncertainties and a good validation result; however, the monthly simulation for the Sahijina station might be adequate for the calibration period. The projected LULC changes for 2030 and 2050 used the TerrSet LCM and MC model. The calibrated model revealed a declining annual discharge trend for MRI-ESM2-0 (SSP245) under all LULC scenarios, with the most significant decline occurring in 2010. Conversely, ACCESS-ESM1-5 (SSP245) displayed a similar trend, with the most significant decline in 2050. It exceeded the magnitude of the decreasing trend of MRI-ESM2-0's predicted discharge. The models also showed an increasing trend in discharge from mid-century to the end of the 21st century. These findings suggest a recent decrease in BRB streamflow, with a potential increase after mid-century. The study highlights the synergistic impacts of CC and LULC changes, which are expected to significantly alter the flow regime of water resources in the region. These changes are likely to reduce water availability and quality while intensifying climate events such as droughts and floods. The resulting impacts extend beyond human existence and livelihoods, also threatening the health of ecosystems.

The authors would like to sincerely thank the Central Water Commission in New Delhi, India, for providing the gauge discharge data. Additionally, authors RPS and SKS would like to acknowledge DST-FIST, the Government of India, and the University of Allahabad for providing the necessary infrastructure and facilities for this work.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.