Response of climate change and land use land cover change on catchment-scale water balance components: a multi-site calibration approach

Climate change (CC) and land use land cover (LU/LC) change are the two most important factors influencing hydrological behaviour vis-à-vis the water balance of any basin (Dey & Mishra 2017; Zhao et al. 2019). The effects of CC have been widely acknowledged from a scientific perspective that includes fluctuations in rainfall patterns (Meshram et al. 2017), rise in global temperature (Abbass et al. 2022), sea-level rise (Durand et al. 2022), and increased occurrence of extreme events (Mishra et al. 2019). Likewise, changes in LU/LC have a direct impact on the water balance components of any basin (Lang et al. 2018) and biodiversity fatalities (Liang et al. 2019).

Hydrologists focus on quantifying the response of CC and LU/LC change for a better understanding of hydrological processes using elasticity analysis (Moussa & Lhomme 2016), statistical analysis (Dey & Mishra 2017), and hydrological modelling (Arnold et al. 1995) in different river basins. Among these hydrological models, the Soil and Water Assessment Tool (SWAT), variable infiltration capacity, and MIKE SHE modelling system are mostly preferred (Arnold et al. 1995; Gao et al. 2013). The SWAT was chosen for this study due to its ability to simulate hydrological outputs while considering diverse LU/LC conditions, even in agricultural-dominated catchments, and future climate scenarios across multiple spatio-temporal scales (Lang et al. 2018; Dash et al. 2020).

Calibration and validation processes are essential steps for reliable representation of the basin-scale hydrological processes in any model. Based on the characteristics of the basin, single-site calibration (SSC) and multi-site calibration (MSC) may be applied to reproduce the desired time series of the model. The MSC is increasingly being used to calibrate complex hydrological models of medium to larger river basins to achieve enhanced model performance by representing the basin's spatial variability. Santhi et al. (2008) found that the use of SSC may not accurately account for the spatial variability within the basin and may lead to unreliable simulation outcomes. Instead, they suggested using MSC to better represent the heterogeneity of the basin and improve model performance. Anderton et al. (2002) studied the spatial heterogeneity of large river basins and concluded that the SSC approach is ineffective and incapable of expressing it. Similarly, Malik et al. (2022) compared the effectiveness of the SSC and MSC approaches in the Laddar and Bharathpuzha catchment of India and concluded that if spatial heterogeneity is significant such as in medium- to large-sized river basins, the MSC approach should be used to minimize uncertainty. Several studies have compared the performance of SSC and MSC techniques, and most of the studies have shown the robustness of MSC over SSC in reflecting basin spatio-temporal heterogeneity to improve the model's predictive capacity (Anderton et al. 2002; Santhi et al. 2008; Malik et al. 2022; Serur & Adi 2022).

Multiple studies have been performed to examine the response of CC and LU/LC change on regional water balance components such as streamflow, water yield (WYLD), groundwater recharge (GWR), and evapotranspiration (ET) (Lang et al. 2018; Zhao et al. 2019; Dash et al. 2020). Chaturvedi et al. (2012) studied the response of CC on the water balance of the Upper Bhīma River basin in India and reported significant changes in runoff, soil moisture, and GWR. Similarly, Setyorini et al. (2017) studied the response of LU/LC change on hydrological processes in the Upper Citarum watershed in Indonesia and found that the forest to agricultural land conversion may lead to a reduction in GWR and an increase in runoff. Furthermore, Wang et al. (2012) studied the response of CC and LU/LC on streamflow in the Chaohu Lake in China and found a significant decrease in streamflow due to these factors.

Most of the hydrological modelling studies used primarily SSC approaches to assess the response of LU/LC and CC on water balance components (Lang et al. 2018; Han et al. 2019; Dash et al. 2020) that might result in errors because of the spatial heterogeneity and complex behaviour of the basin. These reviewed studies identify some research gaps such as (i) there is a dearth of comparative analysis of the SSC and MSC approach in a medium-sized river basin, (ii) no past studies have examined the response of CC and LU/LC change on the catchment-scale water balance components using the MSC approach, and (iii) no past studies have quantified the relative contribution of LU/LC and CC of the water balance components using the MSC approach. Keeping these research gaps in view, the present study is planned to evaluate the streamflow, GWR, WYLD, and ET of a medium-sized river basin under LU/LC and CC using the MSC approach. The specific objectives of the study are (1) to analyse the SWAT performance in simulating the water balance components of a medium-sized river basin under SSC and MSC approaches; (2) to examine the response of isolated and combined LU/LC and CC response on streamflow, GWR, WYLD, and ET considering the MSC approach; and (3) to examine the relative contributions of LU/LC and CC in alteration of water balance in the SRB.

The SWAT model requires the digital elevation model (DEM), LU/LC map, soil map, rainfall, and temperature information as the primary inputs. For this study, the required DEM and Landsat satellite images with spatial resolutions of 90 and 30 m, respectively, were acquired from the United States Geological Survey-Earth Explorer (USGS-EE) web portal (https://earthexplorer.usgs.gov/.in). Four Landsat tiles with a spatial resolution of 30 m and track numbers 139/44, 139/45, 140/44, and 140/45 for the years 1987, 2002, and 2018 were collected to classify LU/LC in the area using the supervised classification method. The ERDAS Imagine 5.1 and Arc GIS 10.1 were used for atmospheric and geometric corrections, satellite image processing, and land use classification. The soil map of the SRB was collected from the Food and Agriculture Organization (FAO) of the United Nations (UN) at a spatial resolution of 1 × 1 km (Nachtergaele et al. 2010).

Data for gridded rainfall (0.25 × 0.25°), maximum, and minimum temperature (1 × 1°) from 1987 to 2013 were acquired from the India Meteorological Department (IMD) in Pune. To achieve a finer resolution, the bilinear interpolation technique was used for resampling 1° gridded temperature data into 0.25° gridded data (Gusain et al. 2020). Daily streamflow information for the Muri (up-stream), Jamshedpur (mid-stream), and Ghatshila (down-stream) gauging stations for the years 1987–2013 were obtained from the Central Water Commission (CWC), Bhubaneswar, and utilized in this study (Swain et al. 2023). The required drainage network, rail, and road networks for developing future LU/LC scenarios were downloaded from the Open Street Map website (https://www.openstreetmap.org). The climate projections of rainfall and temperature were obtained from five General Circulation Models (GCMs) as part of the Coupled Model Intercomparison Project Phase-5 (CMIP5), as mentioned in Table 1. These projections, which underwent two crucial processing stages, Kernel regression-based statistical downscaling (from various respective-spatial scales to 0.25 ° scale) and a quantile mapping technique for bias correction, were acquired from the web portal (http://www.regclimindia.in) for two scenarios: representative concentration pathway (RCP) 4.5 and 8.5 covering the temporal range from 2020 to 2033. These bias-corrected future climate projections have been used in previous research studies and were established as the most appropriate information (Gusain et al. 2020; Navarro-Racines et al. 2020). These refined and corrected datasets were then directly incorporated as input variables within the SWAT framework. In the study, from 1987 to 2013 is considered as the base period (historical), whereas the years 2020–2050 are taken as the future period, as mentioned in Table 1.

The SWAT estimates the Potential Evapotranspiration (PET) using three basic approaches, namely the Priestley–Taylor method (Priestley & Taylor 1972) suitable for areas in humid regions experiencing low advective conditions, the Hargreaves equation (Hargreaves & Samani 1982) under scarce data conditions, and the Penman–Monteith equation (PM) (Monteith 1965), when ample data are available. The PM method is chosen over the other two methods for estimating PET in this study due to its recognized accuracy and versatility. The PM method incorporates comprehensive parameters, including meteorological data such as temperature, wind speed, solar radiation, and humidity, offering a better representation of the complex processes governing ET, ensuring more accurate estimations under various environmental conditions, and data availability scenarios (Monteith 1965; Kadkhodazadeh et al. 2022; Anaraki et al. 2023a, 2023b).

Surface runoff depth estimation can be done using either the Green-Ampt method or the Soil Conservation Services (SCS) curve number method. For this study, the SCS curve number method is chosen, which predicts the runoff volume based on specific HRU characteristics and antecedent soil moisture conditions. It is chosen because of its simplicity, requiring minimal data inputs such as soil type, land use, and hydrologic soil group conditions. The SWAT also allows the revised SCS conceptualization for high slope conditions in the catchment, making it suitable under undulated topography conditions (Anaraki et al. 2023a, 2023b).

The SWAT model shows uniformity among HRUs within sub-basins concerning soil, land use, topography, and climate. It computes watershed response by aggregating HRUs' response linearly, without considering interactions or nonlinear effects among them. Moreover, the model assumes relatively constant catchment characteristics and parameters, assuming a stationary hydrological system. For a comprehensive explanation of the SWAT model, refer to Arnold et al. (1998).

The Sequential Uncertainty Fitting (SUFI-2) technique, integrated with the SWAT, is utilized in the Soil and Water Assessment Tool-Calibration and Uncertainty Programs (SWAT-CUP) to evaluate the model parameter's uncertainty and sensitivity analysis in a streamflow simulation (Abbaspour et al. 2018). In this study, initially, 25 parameters were chosen based on the existing research (Mandal et al. 2021). Subsequently, following the sensitivity analysis, the 15 most sensitive parameters were identified for SSC and MSC approaches, as listed in Table 3. The sensitivity analysis helps to select the sensitive calibration parameters that significantly affect specific model outputs and are characterized based on the t-stat value and p-value. A lower p-value and higher magnitude of the t-stat value indicate a more sensitive model parameter (Anaraki et al. 2021). Uncertainty analysis is evaluated from all input and output sources based on p-factor and r-factor. The p-factor near one and a lower value (closer to zero) of the r-factor indicate reduced uncertainty (Abbaspour et al. 2018; Anaraki et al. 2021).

The calibration of the SWAT model was performed at a monthly time scale for streamflow using three gauging locations within the SRB. These locations were chosen to represent different sections of the river, namely the up-stream site (Muri), mid-stream site (Jamshedpur), and down-stream site (Ghatshila) of the SRB. Here, the SSC and MSC approaches were executed, and a better-performing MSC approach was used for the response of CC and LU/LC change analysis. The results of the calibrated parameters obtained through the SSC and MSC approaches are presented in Table 3. In the SSC approach, the basin was separated into three catchments, and each one was set up to get the best calibration results. In contrast, the MSC approach utilized a spatially distributed calibration strategy in which the up-stream gauging location was calibrated using the observed data from that specific location and the resulting model parameters were fixed. Then, the following sub-basin or gauging location was calibrated, and so on, until the calibration reached the outlet of the entire basin (Anderton et al. 2002; Malik et al. 2022). The MSC approach followed in this research involved initially calibrating Muri as the up-stream gauging location. The calibrated parameter set obtained for Muri was then fixed for the calibration of the Jamshedpur gauging location. Finally, the calibrated parameter set of Muri and Jamshedpur was fixed for the calibration of the Ghatshila gauging location. This MSC approach aligns with similar approaches employed in previous research (Santhi et al. 2008; Malik et al. 2022; Serur & Adi 2022).

The CA-Markov model, a hybrid model of Cellular Automata and the Markov model, is used to anticipate future LU/LC changes through the IDRISI software (Ghalehteimouri et al. 2022). Unlike many other models that may require extensive data, the CA-Markov model performs well even with limited data availability, which is particularly advantageous in agricultural-dominated catchments like the study area. Many studies have employed this model for LU/LC prediction (Mathanraj et al. 2021; Ghalehteimouri et al. 2022).

Here, assuming that the current state ⁠, changes to state in the subsequent step based on transition probabilities, represented as ⁠. Hence, the state within the system can be determined by the preceding state in the Markov chain. The transition probability matrix and land use area transfer matrix were obtained for 1987 to 2002 and 2002 to 2018 with the help of the Markov chain.

The model utilizes primary inputs, including current year LU/LC (image), transitional area matrix, and the suitability map. The multi-criteria evaluation module is commonly used to assess and generate suitability maps for different land use classes. The transition area matrix was derived using the Markov model by utilizing two years (1987 and 2002) LU/LC of the order 8 × 8 since there were eight land use classes in the categorization. To assess the model's efficiency and validate the observed LU/LC, the Kappa index was employed, specifically ⁠, ⁠, and ⁠. A Kappa index value greater than 0.75 is generally considered a satisfactory performance of the model, as indicated by previous research (Mathanraj et al. 2021).

The CA-Markov model assumes spatial homogeneity within cells or regions, maintaining a stationary transition probability matrix over time, and suggesting fixed transition probabilities for land use changes. Additionally, this model often assumes static drivers of land use change, such as socioeconomic factors or policies, throughout the simulation period.

Three cases were considered to analyse the CC impacts: base period (historical), RCP 4.5, and RCP 8.5. The base period consisted of observed climate information, including temperature and rainfall, spanning 27 years from 1987 to 2013. RCP 4.5 and RCP 8.5 represent divergent future climate scenarios with varied greenhouse gas emissions. RCP 4.5 signifies a future with moderated emissions and mitigation efforts, leading to a more moderate CC trajectory. Conversely, RCP 8.5 indicates a future with persistently high greenhouse gas emissions, following current trends without significant mitigation, resulting in amplified CCs and higher global temperatures. These scenarios help assess the possible impacts of different variations of CC on the SRB (Kadkhodazadeh et al. 2022).

For future climatic projections, five GCMs' (Table 1) outputs under RCP 4.5 and 8.5 were downloaded for 2020–2050. Furthermore, the ensemble means of respective RCP from the five GCMs were developed to analyse changes in temperature and rainfall compared to the base period. These data provide a more robust and reliable estimate of how climate variables may change corresponding to the base period. The multi-model ensemble analyses, used in previous research studies (Chaturvedi et al. 2012; Navarro-Racines et al. 2020), are widely used for CC analysis.

In the evaluation of hydrologic model performance, it is necessary to compare observed information with that of simulated counterparts. This comparison typically involves the use of common statistical indicators, such as Nash–Sutcliffe efficiency (NSE), percent bias (PBIAS), and coefficient of determination (⁠⁠), during the calibration and validation (Abbaspour et al. 2018). NSE varies from − to 1, while NSE closer to one is considered a better model performance. elaborates the goodness of fit among simulated and observed values, and its value closer to one indicates the better performance of the model (Gupta et al. 1999). Positive PBIAS values suggest that the observed value is underestimated, while negative values show the range of the observed variables is overestimated although the ±15% range is considered acceptable (Gupta et al. 1999). So, for calibration, the parameters that were identified by statistical analysis and graphical interpretation are employed.

The SWAT model was calibrated and validated following SSC and MSC approaches using observed streamflow at three gauging locations in the SRB, namely Muri (up-stream), Jamshedpur (mid-stream), and Ghatshila (down-stream). The sensitivity analysis of model parameters showed CN₂ (SCS runoff curve number) as the most sensitive parameter at all three gauging sites which indicates the direct effects of initial soil moisture, land use, and soil permeability on streamflow. Recent studies analysed that as CN₂ increases, surface runoff increases while base flow decreases (Abbaspour et al. 2018). In this study area, the Ghatshila (down-stream) gauging sub-basin has relatively higher CN₂ values than the other sub-basins, indicating that Ghatshila surface runoff will be higher. The ALPHA BNK (base flow alpha factor for bank storage) was the second most sensitive parameter, suggesting that bank storage contributes to the flow in the main channel and the flow in the unsaturated zone surrounding the main channel. The parameter values with a substantial difference across the three gauging locations are listed in Table 3.

The SWAT model was executed using a monthly time step for the SSC and MSC approaches. Under the SSC approach, the NSE, R², and PBIAS vary, respectively, from 0.76 to 0.91, 0.83 to 0.88, and −37.9 to 8.2 during calibration and from 0.62 to 0.84, 0.58 to 0.85, and −30.2 to 13.3 during validation, as listed in Table 4. Low p-factor values (60–66% for calibration and 57–63% for validation) and high r-factor values (0.71–0.89 for calibration and 0.77–0.91 for validation) further confirmed the high uncertainty. Under the MSC approach, the NSE and R² values improved at the Ghatshila and Jamshedpur gauging stations compared to the SSC approach, resulting in better matching between observed and simulated monthly streamflow. The uncertainty analysis revealed substantial uncertainty in the model parameters with high p-factor (ranging from 89 to 91% during calibration and 86 to 88% during validation) and low r-factor (varying from 0.45 to 0.59 during calibration and from 0.51 to 0.62 during validation).

The model's performance throughout the calibration and validation periods was satisfactory using an MSC approach, except at the Muri gauging station, where the streamflow was overestimated during calibration and validation (PBIAS < −15), and the R² values were below desirable levels during validation (R² < 0.60). However, the model was developed without considering the influence of the Getalsud reservoir situated up-stream near the Muri gauging location. The reservoir has a greater impact on the up-stream regions of Muri compared to Ghatshila and Jamshedpur in the lower and middle streams, respectively. Consequently, the performance of the Muri gauging station falls below satisfactory levels.

The evaluation of future CC, from 2020 to 2050 with respect to the base period of 1987–2013, was conducted utilizing the ensemble means derived from five GCMs, as listed in Table 1 under both the RCP scenarios. Comparison analysis shows −18 to 21% change in annual rainfall under RCP 4.5 whereas the same is projected to be −19 to 23% following RCP 8.5 (Figure 5). Muri and Ghatshila stations showed an expected increase in rainfall, respectively, by 21 and 17% under RCP 4.5, and 23 and 10% under RCP 8.5. Conversely, Jamshedpur showed a decrease in rainfall by 18% under RCP 4.5 and 19% under RCP 8.5. The results indicate a projected increase in rainfall at Muri and Ghatshila, while Jamshedpur is anticipated to experience a reduction in rainfall under both the RCP scenarios. The observed variability compared to the base period showed an increase in rainfall from March to May and October to November, but a decrease during monsoon months. A reduction in rainfall might directly impact streamflow and GWR. These findings are consistent with the study conducted by Op de Hipt et al. (2019) in which they utilized a multi-GCM ensemble technique to study the effects of CC on hydrological variables. Their research revealed a 50% increase in rainfall under the RCP 4.5 scenario but a 10.9% decrease in rainfall projection under RCP 8.5.

The transition probability matrix, presented in Appendix B, provides insights into the conversion probabilities from one LU/LC class to another different land use class between 1987 and 2002. These probabilities were evaluated by the CA-Markov model, which was utilized to predict LU/LC changes for 2018. For instance, a dense forest has a probability of 24.90% of transitioning into an open forest and a 26% probability of transforming into a mixed forest (Sang et al. 2011). The agricultural land has a 14 and 5.50% probability of converting to mixed forest and settlement, respectively, and a 66.90% probability of remaining agricultural land. Mixed forest has a 13.80% probability of converting into dense forest and an 18.90% probability of converting into agricultural land.

Figure 8 also illustrates the distribution of ET in the two scenarios. At the Jamshedpur region, there is minimal variation in ET. However, noticeable changes are projected for the Muri and Ghatshila regions. Under both RCP 4.5 and 8.5 scenarios, it is projected that ET may be reduced in all three regions from April to June and may increase in the remaining months throughout the year. The results also indicate a significant change in WYLD across all three locations under the two different scenarios. WYLD may decrease in Jamshedpur and Ghatshila which is consistent with the overall reduction observed in those areas. The hydrological analysis reveals alterations in streamflow, GWR, ET, and WYLD across the SRB under future LU/LC and CC scenarios.

To evaluate the individual response of CC on water balance components, the future CC only (S₁) scenario was subtracted from the base period (S₀) under two different (RCP 4.5 and 8.5) scenarios for the three gauging locations Muri, Jamshedpur, and Ghatshila. Streamflow, WYLD, and GWR got reduced by 15–27%, 7–47%, and 5–30%, respectively, while ET showed an increase of 5–7% under both scenarios (RCP 4.5 and 8.5) at Jamshedpur and Ghatshila gauging locations. Rising temperatures might be a possible reason for an increase in ET (Snyder et al. 2011). In Jamshedpur, annual WYLD, GWR, and streamflow reduced by 46–47%, 29–30%, and 13–15%, respectively, while ET increased by 5–7% under both RCP scenarios (Appendix C). In Ghatshila, a gradual decrease in annual streamflow, GWR, WYLD, and ET was estimated. Jamshedpur may experience the highest reduction in GWR and WYLD, followed by Ghatshila and Muri. The future projections indicate that the SRB will receive less rainfall in comparison with the base period and the considerable reduction in hydrological responses (streamflow, GWR, WYLD, and ET) can be attributed to a decrease in rainfall, with Jamshedpur experiencing the maximum decrease (Chanapathi & Thatikonda 2020). The range of relative changes in GWR is larger than the streamflow, WYLD, and ET across all three gauging locations. Previous studies have also shown that the combination of reduced rainfall and rising temperatures can have an additive impact on streamflow reduction in river basins (Bao et al. 2012). The findings are consistent with other research conducted in different basins, highlighting the collective effects of reduced rainfall and increasing temperatures on water balance components (Bao et al. 2012; Tian et al. 2019).

The seasonal analysis at the Jamshedpur and Ghatshila stations indicated a decrease in WYLD across all seasons, while at Muri station, an increase was observed in winter and post-monsoon season under both RCP scenarios. Streamflow showed a reduction of 19–40% during the monsoon season and an increase of 1–44% after the monsoon season at all three gauging sites in both scenarios. These findings highlight the influence of CC on the seasonal streamflow regime in the SRB, potentially attributed to seasonal shifting of rainfall patterns. GWR showed a consistent decrease across all seasons. During the monsoon season, GWR decreased by 23–40% under the RCP 4.5 scenario and 5–46% under the RCP 8.5 scenario across all three gauging locations, as Jamshedpur station showed the maximum reduction followed by Ghatshila and Muri stations. Regarding ET, in Muri and Jamshedpur, an increase is observed across all seasons under the RCP 4.5 scenario, while a reduction is observed in the pre-monsoon season under the RCP 8.5 scenario.

To assess the response of LU/LC change, the future climate and LU/LC scenario (S₂) simulation was subtracted from the future climate only (S₁) under two different scenarios, RCP 4.5 and 8.5, for the three gauging locations. The analysis revealed significant changes in LU/LC, particularly in dense and mixed forests, which decreased in extent while agricultural and settlement land expanded. During the monsoon season, GWR gradually decreased by 7–34% under both RCP 4.5 and 8.5 scenarios across all three gauging locations, as shown in Appendix D. Specifically, under the RCP 4.5 scenario, GWR reduction due to LU/LC change was 31, 36, and 9% in Ghatshila, Jamshedpur, and Muri, respectively. The study highlights the loss of agricultural and forest areas due to rapid urbanization, which might be the possible reason for the reduction of GWR (Astuti et al. 2019). The relative contribution of LU/LC change is less than CC for disturbing the water balance component of this research area, as shown in Figure 10. In line with similar findings, Mojid & Mainuddin (2021) highlighted that urbanization can have a detrimental impact on GWR. The growth of urban areas typically involves the construction of impervious surfaces like concrete and asphalt, which prevent rainwater from infiltration into the ground. This reduction in infiltration disrupts the natural process of GWR, ultimately diminishing groundwater availability (Joodaki et al. 2014; Astuti et al. 2019). The primary factors leading to a reduction in runoff were identified as intensive agricultural land use and excessive water usage by Yang & Tian (2009).

During the monsoon season, ET increased across all gauging locations under both RCP scenarios. In terms of WYLD, Jamshedpur and Ghatshila showed a reduction in the monsoon, post-monsoon, and winter seasons. GWR exhibited a gradual reduction of 36–64% during the monsoon season under both RCP scenarios across all three gauging locations. In Jamshedpur, GWR exhibited a decrease throughout all seasons under both RCP scenarios. Similarly, in Ghatshila, the same pattern was observed with the exception of a 24% increase in the pre-monsoon season under the RCP 4.5 scenario. The simulations considering the combined effects of CC and LU/LC changes exhibited patterns similar to those of CC alone, as shown in Figure 10. The corresponding changes in streamflow were more prominent for CC compared to LU/LC changes. The findings of this research are consistent with previous study that emphasizes the complex relationships between LU/LC and CC in influencing hydrological variables (Han et al. 2019).

The relative contributions of CC and LU/LC changes to the water balance components in the SRB are shown in Figure 10. Analysing the streamflow change, the contribution of CC is 98.57% in Ghatshila, 97.94% in Jamshedpur, and 98.41% in Muri, with the remaining contribution attributed to LU/LC change (1.42% in Ghatshila, 2.05% in Jamshedpur, and 1.58% in Muri) under the RCP 4.5 scenario. These findings indicate that CC has a higher influence on streamflow changes compared to LU/LC change. In the Ghatshila gauging location, CC contributed to 40.16% of the change in GWR, while LU/LC change accounted for 59.83% of the change. On the other hand, in the Jamshedpur gauging location, the impact of CC on GWR was slightly higher at 57.61% compared to the impact of LU/LC change, which accounted for 42.38% of the change under RCP 4.5.

In the Ghatshila gauging location, under RCP 8.5, CC accounted for 51.79% of the change in GWR, while land use change (LU/LC) contributed 48.20% to the change. Similarly, in the Jamshedpur gauging location, the impact of CC on GWR was 58.61%, slightly higher than the impact of LU/LC change, which accounted for 41.86% of the change. CC dominates nature to change GWR in both gauging locations under RCP 8.5. Regarding ET variation, under the RCP 4.5 scenario, CC contributed −146% and LU/LC change contributed 46% to the change in Ghatshila. In Jamshedpur, CC accounted for 112.98% of the change, while LU/LC change contributed −12.98%. In Muri, CC contributed 69% and LU/LC change contributed 30.11% to the variation in ET. Under the RCP 8.5 scenario, the contributions were 36.79% (CC) and 63.20% (LU/LC) in Ghatshila, 72.20% (CC) and 27.79% (LU/LC) in Jamshedpur, and 40.58% (CC) in Muri. In terms of WYLD variation, CC showed a dominant influence under both scenarios across all three gauging stations, as shown in Figure 10. The CC and LU/LC change contribute relatively equally to disturbing the GWR and ET, but CC dominates in nature to change the streamflow and WYLD. A previous study by Chawla & Mujumdar (2015) has also shown that CC has a greater effect on streamflow compared to LU/LC changes alone, in various basins. Other findings, including Kim et al. (2013) and Swain et al. (2023), have consistently demonstrated that CC has a more significant effect on streamflow compared to the joint response of CC and LU/LC change or LU/LC change alone.

This research examined the separate and combined impacts of CC and LU/LC change on the water balance components of the SRB, a sub-humid sub-tropical river basin, under RCP 4.5 and 8.5 scenarios using five GCMs. The study tested the SWAT model performance using SSC and MSC approaches to simulate observed streamflow at three gauging locations. It was found that the MSC approach provides more reliable parameter values compared to the SSC approach, making it the preferred calibration approach for model simulations to evaluate the response of LU/LC change and CC on water balance components of the medium-size river basin at three gauging locations. Future LU/LC of the basin for the year 2033 was developed by using the CA-Markov model. The analysis of LU/LC change during 1987–2018 revealed a decrease in dense forest and open forest areas, accompanied by an increase in settlement and agricultural areas in the basin. This trend in LU/LC change is assumed to continue from 2018 to 2033 in developing the future LU/LC of 2033. Rapid urbanization in the SRB resulted in reduced vegetative cover, infiltration, and water storage capacity, leading to diminished GWR.

The findings revealed that annual rainfall may increase in the upper (Muri) and lower (Ghatshila) regions of the basin, while the same is expected to reduce in the middle part (Jamshedpur station) of the basin under both RCP scenarios. The middle part (Jamshedpur station) of the basin is expected to observe the maximum reduction in GWR and WYLD due to the impact of CC, followed by the upper and lower parts (Ghatshila and Muri stations). The reduction in streamflow during the monsoon season across the basin (all gauging locations) was attributed to reduced rainfall. Pre-monsoon rainfall exhibited the highest variability, followed by the monsoon and post-monsoon seasons. It is expected that future seasonal rainfall variability may negatively impact crop production in the agriculture-dominated SRB.

The combined impacts of LU/LC change and CC showed a similar pattern as of CC alone, with CC having a greater influence on streamflow and WYLD compared to LU/LC change. The reduction in GWR and increase in WYLD may lead to water scarcity and floods in the study area, respectively. To ensure sustainable water resources management in the basin, it is crucial to comprehend the response of CC and LU/LC on every aspect of the water balance components. The study may help to develop the policy to reduce the destructive impacts of LU/LC change and CC by the collaboration between planners, policy-makers, and researchers.

This study offers significant insights but acknowledging its limitations is crucial. The CA-Markov model's uncertainties in future LU/LC predictions, arise from unpredictable human interventions such as urban expansion and policy shifts, which may have a substantial impact on reliable inputs for models such as the SWAT. Despite attempts to refine GCM-derived meteorological data, persistent uncertainties affect SWAT's hydrological process simulations. Addressing these limitations remains a critical aspect of enhancing the performance of predictive models in water resource management. Furthermore, for a robust understanding of the water resources perspective in the basin, future studies should consider incorporating artificial interventions such as dam and reservoir operations to enhance model interpretation. Incorporation of dams and reservoirs into the SWAT model may offer robust insights into managing water resources, particularly in assessing flood risks, regulating agricultural water allocation, and enhancing overall water resource resilience amid changing climatic conditions.

The IMD and CWC, India, provided the hydro-meteorological data necessary for this work, which the authors would like to acknowledge. The first author sincerely acknowledges the Ministry of Education, India, for providing a research scholarship.

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

The authors declare there is no conflict.