The Song River, originating from various small streams in the Dhanolti mountain range and merging with Sahastradhara streams, flows down to the Doon valley basins before eventually joining the Ganga River. Known for its picturesque surroundings, the Song River in Dehradun is particularly renowned for its plentiful natural sulphur springs. These springs emerge from mountain fissures and feed into the main watercourse, enriching the river with sulphur. Visitors flock to immerse themselves in the mineral-rich waters, as sulphur baths are thought to alleviate various health issues, particularly skin conditions. Situated at 30°28′ latitude and 78°8′ longitude, the Song River is vital to the communities of Raiwala, Doiwala, Chiddarwala, and Lacchiwala, serving as their primary water source along its 107 km journey. It converges with the Ganga River at 78° 14′ 54″ longitude and 30° 02′ 02″ latitude, just upstream of Haridwar near the Satyanarayan G&D station maintained by CWC, after passing through the Satyanarayana area (Figure 1).

A significant tributary of the Song River is the Suswa River, which originates in the clayey depression of the Mussoorie range. It drains the eastern part of Dehradun city and joins the Song River southeast of Doiwala. The catchment area includes two major urban settlements: Dehradun and Doiwala. The Rispana and Bindal, two primary drainage networks, carry municipal sewage from these urban areas and discharge into the Song River via the Suswa River. The region experiences an average annual rainfall of approximately 1,451 mm, with about 1,181 mm (81%) occurring during the monsoon season. Consequently, July and August are the wettest months of the year.

The eco-hydrological model Soil and Water Assessment Tool (SWAT) is a versatile hydrological model capable of simulating diverse environmental conditions and scales (Arnold et al., 1998). The establishment of a SWAT model requires spatially distributed data, including Digital Elevation Models (DEMs), soil data, land use land cover maps, and weather data. The model operates exclusively on this data and does not necessitate prior knowledge of catchment behavior or flow processes. SWAT simulates water balance, a fundamental driver of watershed processes, to accurately predict runoff, sediment, and nutrient movement. The model comprises two primary components: the land phase and the routing phase. For surface runoff estimation, the Curve Number method is utilized.

The SWAT model encompasses numerous hydrological parameters whose effectiveness is affected by variables such as soil type, slope, and land cover. To identify the most influential parameters and reduce the number needed for calibration, sensitivity analysis is essential. This research employed the SUFI-2 algorithm within the SWAT-CUP interface to perform sensitivity analysis on 13 crucial parameters, including ALPHA_BF (baseflow alpha factor), CH_K2 (hydraulic conductivity in the main channel alluvium), CH_N2 (Manning’s “n” for the main channel), as well as GWQMN, SOL_AWC, ESCO, and SURLAG. The impact of these parameters on hydrological components like discharge, infiltration, baseflow, groundwater flow, evaporation, and transpiration were examined by methodically altering one parameter at a time while keeping others constant. The relationship between parameters and hydrological responses was quantified using sensitivity coefficients. To enhance the model’s accuracy, both manual and automated calibration techniques were utilized. Manual calibration involved visually adjusting parameters based on observed and simulated flow patterns, considering catchment characteristics. Automated calibration, facilitated by SWAT-CUP, systematically optimized uncertain parameters by comparing model outputs with measured data through an interactive interface. This combined approach ensured effective parameter optimization and improved the SWAT model’s ability to simulate hydrological processes.

There are multiple efficacy measures that can be used to assess the model’s performance. These efficacy metrics show how the model-simulated values and the observed values are reconciled. Nash-Sutcliffe Efficiency (NSE) (Equation 1), coefficient of determination (R²) (Equation 2) and PBIAS (Equation 3) are the most widely utilized metrics among them (Sane et al., 2020; Swain et al., 2022).

Nash-Sutcliffe efficiency (NSE) (Nash and Sutcliffe, 1970),

Percent bias (PBIAS) (Moriasi et al., 2007),

Coefficient of determination (R²) (Suryavanshi et al., 2017),

P_i and O_i are the simulated and observed values of discharge; n is the sample number,

P ¯ and O ¯ are the average of simulated and observed discharge.

This research utilized previous studies to guide parameter selection for calibration, examining 13 parameters in total, with 8 identified as sensitive at a 0.05 significance level. Table 3 outlines the range, fitted values, t-statistics, significance levels, and sensitivity rankings for these parameters. The study employed a three-year warm-up period (1971–1973), with calibration and validation periods spanning 1974–1995 and 1996–2004, respectively. The dotty plots generated in SWAT-CUP indicate that the baseflow recession constant (ALPHA_BF) and Channel roughness (CH_N2) are the most sensitive parameters. This conclusion is evident from the noticeable variation in the model’s performance metrics corresponding to changes in their values. The ALPHA_BF was calibrated to 0.05, suggesting a contribution of groundwater to streamflow, during low-flow periods is a critical factor in accurately simulating streamflow. CH_N2 was set at 0.04, indicative of a dredged channel, while effective hydraulic conductivity (CH_K2) was calibrated to 121.15 mm/h, signifying high-permeability conditions. The return flow threshold depth (GWQMN) was set at 735 mm, and soil water capacity (SOL_AWC) showed a 47% decrease, indicating reduced plant-available water. The groundwater “revap” coefficient (GW_REVAP) was calibrated to 0.07, suggesting limited water movement to the root zone. Manning’s coefficient (OV_N) was determined to be 0.83, and deep aquifer percolation (REVAPM) required 404.5 mm of shallow aquifer water. Soil evaporation compensation (ESCO) was set at 0.12, indicating high demand from lower soil layers. Surface runoff lag time (SURLAG) was calibrated to 2.33, and plant uptake compensation (EPCO) decreased to −0.29, demonstrating reduced uptake in lower soil layers. Groundwater delay (GW-DELAY) was established at 472.5 days, indicating a slow aquifer response, while lateral flow travel time (LAT_TTIME) was set at 22.14 days, representing lateral flow dynamics within the hydrological response units (HRUs). These calibrated parameters significantly enhanced model performance and improved the simulation of hydrological processes.

The study employed daily observations from 1974 to 2004, with Figure 5 illustrating the temporal comparison between observed and simulated discharge during both calibration and validation phases. Model effectiveness was assessed using R², NSE, and PBIAS metrics, as presented in Table 4. The observed and simulated flows showed strong correlation, with R² values of 0.79 and 0.66 for calibration and validation, respectively. During calibration, both baseflow and peak flows aligned well between observed and simulated data. Although the validation period showed underestimated goodness of fit, the results remained satisfactory. The NSE reached 0.7 during calibration and 0.62 during validation. PBIAS values fell within acceptable ranges for both periods, measuring 17.06 for calibration and 19.60 for validation.

The weekly average simulated and observed flow for the 1974–2004 period is depicted in Figure 5 as a time series plot. This graph reveals a high degree of similarity between the simulated and observed discharge, with the model successfully capturing overall flow trends, including seasonal fluctuations and high-flow events. The simulated flow closely tracks the observed data during the calibration phase, demonstrating the model’s precision in replicating both low-flow conditions and peak discharge occurrences. Although the validation period exhibits minor underestimations of some peak flows, the general trend remains aligned, indicating satisfactory model performance across the extended timeframe. This sustained consistency underscores the model’s dependability for long-term hydrological flow simulations.

Figure 6 presents a scatter plot depicting the weekly mean simulated versus observed flow during the calibration and validation periods. The plot demonstrates a strong linear relationship between the simulated and observed flow data, with the majority of points clustering in close proximity to the 1:1 line, indicating a high degree of accuracy in the model’s predictions. During the calibration period, the scatter plot exhibits a tight correlation, reflecting the model’s efficacy in capturing the observed flow dynamics. Although the validation period displays a slight dispersion from the 1:1 line, the overall alignment remains satisfactory, confirming the model’s robustness in simulating streamflow across diverse hydrological conditions.

The model’s applicability to the study area was evaluated by comparing its simulated flows with ADCP-measured discharges obtained during a field survey. A strong correlation was observed between the simulated flows and observed discharge data at both upstream (site-1) and downstream (site-2) locations. The model’s performance was quantified using statistical indicators R², NSE, and PBIAS, with results presented in Table 5, demonstrating its robust performance. The R² values, ranging from 0.78 to 0.79, indicated a strong correlation, while NSE values of 0.79 upstream and 0.68 downstream showed high agreement between simulated and observed flows. PBIAS values of 0.04 upstream and − 16.20 downstream fell within acceptable ranges, further confirming the model’s reliability. A time series plot comparing weekly mean simulated flows with observed flows for 2022–2023 at both sites is shown in Figure 7. The plot reveals exceptional alignment at the upstream site, where the model accurately captures flow variations and peak flows. Although minor discrepancies are noted at the downstream site, the overall flow patterns are well-represented, highlighting the model’s ability to accurately simulate flow conditions in the Song River basin under diverse field conditions.

Figure 8 presents a scatter plot depicting the relationship between weekly mean simulated flow and field survey observed flow at both the upstream and downstream sections of the Song River. The plot demonstrates a strong correlation between the simulated and observed data, with data points closely aligned with the line of perfect agreement. The upstream section exhibits a marginally higher degree of concordance, indicating superior model accuracy in this region, while the downstream section, despite displaying some dispersion, still demonstrates robust predictive performance. In aggregate, the scatter plot corroborates the model’s efficacy in replicating observed flow conditions across distinct sections of the river.

The pre-calibrated runoff model was subsequently utilized for sediment load calibration. Four additional parameters Channel erodibility factor (CH_EROD), Peak rate adjustment factor for sediment routing in the subbasin (ADJ_PKR), USLE equation support practice (P) factor (USLE_P) and Linear parameter for calculating the maximum amount of sediment (SPCON) added in SWAT-CUP using SUFI2 algorithm for calibrate and validate Sediment load, at daily time scale (Table 6). The results of the best simulation (based on efficacy measures on daily scale) and its comparison with respect to the observed sediment load along with the observed discharge at weekly scale for 2 years, i.e., 2023–2024 are presented in Figure 9. The statistical performance indicators for the calibration period (2022) revealed that the model achieved an R² value of 0.70 and a Nash-Sutcliffe Efficiency (NSE) of 0.53. For the validation period (2023), the model demonstrated an R² of 0.59 and an NSE of 0.52.

Figure 9, comparing observed and simulated daily sediment yields at site 2 during the 2022 calibration and 2023 validation periods, elucidates both the efficacy and limitations of the SWAT model. In the 2022 calibration period, observed sediment yields exhibit substantial peaks in September and October, with loads reaching up to 1,800 tonnes per day. While the SWAT model captures the overall seasonal pattern, including the pronounced increase during the monsoon and subsequent decline, it underestimates the magnitude of these peaks, particularly during high-flow events. Similarly, during the 2023 validation period, the observed sediment yield peaks in July, exceeding 6,000 tonnes per day, followed by a rapid decline and smaller peaks in subsequent months. The simulated sediment yield generally follows this trend but significantly underestimates the July peak, reaching only approximately 4,000 tonnes per day, and fails to capture some of the acute fluctuations observed in the following months. These results demonstrate the model’s capacity to replicate the general seasonal dynamics of sediment transport but also underscore challenges in accurately simulating extreme events. To enhance the model’s performance, particularly during high-flow periods, further refinement of parameters or the incorporation of additional factors may be necessary.