Toward a combined Bayesian frameworks to quantify parameter uncertainty in a large mountainous catchment with high...

Area of Science:

• Hydrology and Water Resources Management
• Environmental Modeling
• Computational Science

Background:

• Hydrological models are crucial for water resource management, but quantifying uncertainties remains a significant challenge.
• Accurate uncertainty quantification is essential for reliable predictions and effective decision-making in water resource management.

Purpose of the Study:

• To investigate and compare the performance of two parameter uncertainty quantification methods: Sequential Uncertainty Fitting version 2 (SUFI-2) and DiffeRential Evolution Adaptive Metropolis (DREAM-ZS).
• To evaluate these methods in predicting runoff using the Soil and Water Assessment Tool (SWAT) model at a multisite flow gauging station.

Main Methods:

• Employed SUFI-2 and DREAM-ZS algorithms with S1 and S2 strategies to explore output uncertainty in the SWAT model.
• Defined S1 strategy using SWAT manual prior ranges and S2 strategy using a compromise between prior and posterior ranges.
• Assessed performance using P-factor, d-factor, Nash-Sutcliffe coefficient (NS), and dimensionless variants of average relative deviation amplitude (S and T).

Main Results:

• The S2 strategy generally outperformed the S1 strategy in reducing prediction uncertainties, showing improved NS, S, and T values.
• DREAM-ZS algorithm enhanced model calibration efficiency and yielded more realistic parameter values for SWAT runoff simulations.
• Parameter uncertainty analysis indicated specific ranges for S and T indices under different strategies.

Conclusions:

• The DREAM-ZS algorithm and the S2 strategy demonstrate superior capabilities in improving hydrological model calibration and reducing prediction uncertainties for runoff simulation.
• The S2 strategy, integrating formal and informal Bayesian approaches, is effective in minimizing prediction uncertainties.
• This research provides valuable insights into selecting appropriate uncertainty quantification methods for hydrological modeling.