Gharnouki et al. (2026) Assessing Uncertainty in Multi-Source Precipitation for a Semi-Arid Mediterranean Catchment Using SWAT

Identification

• Journal: Earth Systems and Environment
• Year: 2026
• Date: 2026-04-06
• Authors: Ines Gharnouki, Sihem Benabdallah, Jalel Aouissi, Sudoy Kumer Ghosh, Anjon Kumar Das
• DOI: 10.1007/s41748-026-01153-z

Research Groups

• LR17AGR01, InteGRatEd Management of Natural Resources: remoTE Sensing, Spatial Analysis and Modeling (GREEN-TEAM), National Agronomic Institute of Tunisia (INAT), Carthage University, Tunis, Tunisia
• Center for Water Research and Technologies (CERTE), Soliman, Tunisia
• Department of Electrical and Electronics Engineering, Dhaanish Ahmed College of Engineering, Affiliated to Anna University, Chennai, India
• Department of Civil Engineering, Southern Illinois University, Edwardsville, IL, USA

Short Summary

This study evaluates the performance of five multi-source precipitation datasets (observed, CHIRPS, PERSIANN, GPM-IMERG, ERA5) in driving the SWAT hydrological model for a semi-arid Mediterranean catchment in central Tunisia over a 17-year period. It found that CHIRPS and ERA5 provided satisfactory hydrological simulations after recalibration, with CHIRPS demonstrating the best overall performance and highlighting precipitation as the primary source of uncertainty.

Objective

• To evaluate the performance of the SWAT model in simulating streamflow under data-scarce conditions in the semi-arid Haffouz catchment using multi-source precipitation datasets.
• To assess and compare the hydrological performance of CHIRPS, ERA5, GPM-IMERG, and PERSIANN precipitation products in driving SWAT simulations.
• To evaluate the effects of parameter transfer from gauge-calibrated setups versus dataset-specific recalibration on model performance.
• To perform a sensitivity and uncertainty analysis of hydrological model parameters using multi-source precipitation datasets.
• To interpret performance differences based on retrieval algorithms, gauge blending, and regional hydroclimatic characteristics.

Study Configuration

• Spatial Scale: Haffouz catchment, a sub-basin of the Merguellil upstream catchment, central Tunisia. The catchment covers an area of 625 km² with altitudes ranging from 200 m to 1200 m.
• Temporal Scale: 17-year period from January 2001 to December 2017. A one-year warm-up period (2001) was used, followed by a calibration period (January 2002 to December 2011) and a validation period (January 2012 to December 2017). Simulations were performed at a daily time step, with performance evaluated monthly.

Methodology and Data

• Models used:
  • Soil and Water Assessment Tool (SWAT) (version SWAT2012 within ArcSWAT)
  • SWAT Calibration and Uncertainty Procedures (SWAT-CUP) utilizing the SUFI-2 algorithm for calibration and uncertainty analysis.
• Data sources:
  • Precipitation:
    • Observed rainfall: Daily data from eight rain gauges (2001-2017) provided by the Regional Commissariat for Agricultural Development (CRDA) of Kairouan and Siliana governorates.
    • CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data): 5 km spatial resolution, daily.
    • PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks): 0.25° x 0.25° spatial resolution, hourly, daily, monthly, and yearly.
    • GPM-IMERG (Global Precipitation Measurement Integrated Multi-Satellite Retrievals): 0.1° x 0.1° spatial resolution, half-hourly.
    • ERA5 (European Center for Medium-Range Weather Forecasts Reanalysis): Approximately 30 km spatial resolution, hourly.
  • Discharge: Monthly discharge data from the Haffouz hydrometric station (2001-2017) obtained from CRDA.
  • Temperature: Daily minimum and maximum temperatures from the World Weather for Water Data Service (W3S).
  • Digital Elevation Model (DEM): 30 m x 30 m spatial resolution from the United States Geological Survey’s Earth Explorer platform.
  • Land Use: Classified Sentinel-2 images (September 2020 to August 2021) into seven classes (water bodies, olives, cactus, orchards, winter wheat, forest land, pasture) using the Random Forest algorithm.
  • Soil Data: Digitized Tunisian soil map (scale 1:50,000) supplemented by a soil sample database including ten soil properties.

Main Results

• Calibration with observed precipitation: The SWAT model achieved satisfactory performance with a Nash–Sutcliffe Efficiency (NSE) of 0.63, a Kling–Gupta Efficiency (KGE) of 0.72, and a Percent Bias (Pbias) of 18%. The 95% prediction uncertainty (95PPU) encompassed 64% of observed data (P-factor = 0.64) with a low uncertainty band thickness (R-factor = 0.93).
• Validation with observed precipitation: Model performance was unsatisfactory (NSE = 0.33, KGE = 0.35, Pbias = 10%), primarily due to the prevalence of low-flow conditions, high interannual variability, and the semi-arid climate.
• Impact of parameter transfer (using gauge-calibrated parameters with alternative precipitation products):
  • CHIRPS showed acceptable performance (KGE = 0.32) even without recalibration, but its performance declined with transferred parameters due to increased overestimation (Pbias increased from 21% to 61%).
  • ERA5 and PERSIANN showed significant, but still unsatisfactory, improvements in performance metrics (e.g., ERA5 NSE increased from -1.2 to 0.25, KGE from -0.12 to 0.19; PERSIANN NSE increased from -11 to -0.04, KGE from -3.0 to 0.29).
  • GPM-IMERG showed minimal sensitivity to calibration parameters, with largely unchanged performance.
• Recalibration with CHIRPS and ERA5: Both datasets achieved satisfactory performance after recalibration, with CHIRPS yielding NSE = 0.51 and KGE = 0.58 (Pbias = -15% underestimation), and ERA5 yielding NSE = 0.50 and KGE = 0.61 (Pbias = 7.6% overestimation). Both products tended to underestimate high streamflow events.
• Parameter Sensitivity: The SCS runoff curve number (CN2), Effective hydraulic conductivity in main channel alluvium (CH_K2), and Plant uptake compensation factor (EPCO) were identified as the most sensitive parameters (p < 0.05).
• Uncertainty Analysis (P-factor, R-factor) after recalibration: CHIRPS achieved a P-factor of 0.5 and R-factor of 1.05, while ERA5 achieved a P-factor of 0.45 and R-factor of 0.9. These values indicate satisfactory prediction uncertainty for both products in the semi-arid context.
• Different precipitation inputs resulted in distinct optimal parameter values and ranges, confirming the sensitivity of the SWAT model to precipitation data and their spatial distribution.

Contributions

• Provides a systematic evaluation of multiple satellite and reanalysis precipitation products (CHIRPS, ERA5, GPM-IMERG, PERSIANN) for hydrological modeling in a semi-arid Mediterranean catchment using the SWAT model.
• Assesses the impact of both parameter transfer from gauge-calibrated setups and dataset-specific recalibration on model performance, offering insights into model robustness under different forcing data.
• Offers a process-based interpretation of performance differences, linking them to precipitation retrieval algorithms, gauge blending strategies, and regional hydroclimatic characteristics.
• Confirms that precipitation is the principal source of uncertainty in hydrological modeling and that each precipitation dataset imposes specific ranges on calibrated parameters.
• Identifies CHIRPS as the best-performing satellite-based product for the Haffouz catchment, demonstrating its value as a reliable alternative in data-scarce regions for improving hydrological simulations.

Funding

The project ALTOS (Managing water resources within Mediterranean agro-systems by accounting for spatial structures and connectivity) within the framework of the PRIMA-Section II, funded by the Ministry of Higher Education and Scientific Research of Tunisia.

Citation

@article{Gharnouki2026Assessing,
  author = {Gharnouki, Ines and Benabdallah, Sihem and Aouissi, Jalel and Ghosh, Sudoy Kumer and Das, Anjon Kumar},
  title = {Assessing Uncertainty in Multi-Source Precipitation for a Semi-Arid Mediterranean Catchment Using SWAT},
  journal = {Earth Systems and Environment},
  year = {2026},
  doi = {10.1007/s41748-026-01153-z},
  url = {https://doi.org/10.1007/s41748-026-01153-z}
}