Uncertainty Analysis for Computationally Expensive Models with Multiple Outputs.

Area of Science:

• Environmental Modeling
• Computational Statistics

Background:

• Bayesian MCMC calibration is crucial for complex environmental models.
• Computationally expensive models pose significant challenges for calibration and uncertainty analysis.

Purpose of the Study:

• To implement the SOARS methodology for efficient Bayesian MCMC calibration.
• To enhance the efficiency of the GRIMA algorithm for model evaluation.

Main Methods:

• Utilized SOARS (Statistical and Optimization Analysis using Response Surfaces) methodology.
• Employed a radial basis function interpolator as a surrogate model (emulator).
• Located high posterior density regions (HPDR) using global optimization and GRIMA algorithm for efficient model evaluation.

Main Results:

• Successfully applied SOARS to an eight-parameter SWAT2005 model for the Town Brook watershed.
• Modeled daily stream flows and phosphorus concentrations, demonstrating the methodology's effectiveness.

Conclusions:

• SOARS provides an efficient approach for Bayesian MCMC calibration of computationally expensive models.
• The enhanced GRIMA algorithm improves the efficiency of model evaluation within HPDRs.