Sensitivity and Uncertainty Analysis of Hydrological Parameters of Rafsanjan Plain Using the SWAT Model

Extended Abstract
Introduction
Rafsanjan Plain, an arid region of Iran, has experienced groundwater depletion and significant hydrological changes in recent decades. This study applied the semi-distributed, process-based SWAT model to simulate hydrological processes and estimate the water balance. Spatial inputs (DEM, land use, soil type, slope) and daily climate data (precipitation, temperature, humidity, wind, solar radiation) for 2002–2024 were used. The watershed was divided into sub-basins and HRUs, and the model was calibrated and validated using SUFI-2 in SWAT-CUP, with sensitivity and uncertainty analyses conducted. Results indicated that curve number, saturated hydraulic conductivity, and baseflow parameters most influenced streamflow simulation. Model performance was acceptable (P-factor 0.42, R-factor 2.62 in calibration; 0.42 and 0.39 in validation). Over 60% of annual precipitation was lost via evapotranspiration, with surface runoff contributing less than 0.1%. These findings demonstrate SWAT’s effectiveness for water resources management and climate impact assessment in arid regions.
Materials and Methods
The Rafsanjan Plain, encompassing 12,513 km² within the Kavir-e Dranjir-Saghand basin, was delineated into 12 sub-basins using a 30-meter Digital Elevation Model. The model configuration incorporated land use/land cover maps (derived from Landsat imagery), soil classification data from the Iranian Soil and Water Research Institute, and slope categories to generate Hydrological Response Units (HRUs) through an overlay process. Daily climatic inputs spanning 2002-2024 included precipitation, maximum/minimum temperature, and relative humidity from the Rafsanjan synoptic station; wind speed and solar radiation were obtained from NASA's POWER database to fill data gaps. Monthly discharge observations from two hydrometric stations facilitated model calibration (2002-2017) and independent validation (2018-2024). The SUFI-2 algorithm in SWAT-CUP performed automated calibration through iterative Latin Hypercube sampling, accounting for parameter uncertainty by bracketing observations within 95% prediction uncertainty bounds. Sensitivity analysis employed the t-test method to rank 27 parameters related to runoff generation, soil water movement, and groundwater flow. Performance evaluation utilized the P-factor (percentage of observations within uncertainty band) and R-factor (average width of uncertainty band normalized by standard deviation), supplemented by coefficient of determination (R²), Nash-Sutcliffe efficiency (NSE), and root mean square error (RMSE). The water balance equation in SWAT quantified precipitation partitioning into evapotranspiration, surface runoff, lateral flow, baseflow, and deep aquifer recharge components.
Results and Discussion
Global sensitivity analysis identified 27 parameters significantly influencing streamflow simulation, with the curve number (CN2), saturated hydraulic conductivity (SOL_K), and baseflow recession constant exhibiting the highest sensitivity based on t-statistics and p-values (<0.05). Calibration achieved P-factor=0.42 and R-factor=2.62, while validation yielded P-factor=0.42 and R-factor=0.39, indicating acceptable model performance according to ASABE guidelines. Statistical metrics demonstrated strong agreement (R²≈0.91, NSE=0.84) between simulated and observed monthly discharge, though the model overestimated peak flows in extreme years (2006, 2014) due to limited availability of sub-daily precipitation data and simplified representation of runoff generation during high-intensity events. Water balance analysis revealed that 63.2% of mean annual precipitation (142 mm) was lost through actual evapotranspiration, 28.4% contributed to deep aquifer recharge, 7.8% generated surface runoff, and return flow constituted merely 0.08%, characterizing typical hyper-arid hydrology. Baseflow dominated river discharge during dry months (June-September), comprising 85% of total flow, while snowmelt contributed significantly to spring peaks. Uncertainty analysis demonstrated that parameters controlling runoff partitioning and soil water retention (CN2, SOL_K, soil available water capacity) contributed 68% of total prediction uncertainty. Seasonal patterns showed that precipitation and runoff peaked in March-April, while potential evapotranspiration reached maximum values during June-August. The model's performance in simulating baseflow recession was superior to its representation of quickflow response, reflecting its conceptual structure and parameterization limitations in capturing rapid runoff processes.
Conclusion
This study successfully calibrated and validated the SWAT model for the Rafsanjan Plain, demonstrating its capability to simulate hydrological processes in data-scarce arid environments with acceptable uncertainty levels. The identification of 27 sensitive parameters, particularly CN2 and SOL_K, highlights that accurate characterization of soil hydraulic properties is critical for reducing simulation uncertainty and improving prediction reliability. Quantification of water balance components revealed severe water loss through evapotranspiration (>60%) and minimal groundwater recharge, emphasizing the unsustainable nature of current water use practices and the urgent need for demand management strategies. While the model effectively reproduced seasonal flow patterns and baseflow dynamics (R²≈0.91), overestimation of peak flows indicates limitations in representing extreme rainfall-runoff events, attributable to coarse temporal resolution of precipitation data and simplified infiltration processes. These findings provide a robust scientific foundation for evaluating climate change scenarios, land use change impacts, and water management interventions such as deficit irrigation and artificial recharge. Future research should integrate SWAT with groundwater quality modules to address salinization, incorporate higher-resolution meteorological forcing data, and couple with optimization algorithms to support multi-objective water allocation decisions. The established parameter ranges and methodological framework offer transferable guidance for hydrological modeling in similar arid watersheds, ultimately supporting evidence-based policies for sustainable water resource management.