Efficient Calibration of Groundwater Contaminant Transport Models Using Bayesian Optimization.

Area of Science:

• Environmental Science
• Hydrogeology
• Computational Science

Background:

• Numerical modeling is crucial for understanding groundwater contaminant transport dynamics.
• Automatic calibration of complex numerical models is computationally intensive and inefficient.
• Existing methods require numerous model evaluations, limiting calibration efficiency.

Purpose of the Study:

• To present a Bayesian optimization (BO) method for efficient numerical model calibration in groundwater contaminant transport.
• To address the high computing overhead associated with traditional calibration techniques.
• To improve the efficiency and effectiveness of contaminant transport model calibration.

Main Methods:

• Developed a Bayesian optimization framework for model calibration.
• Utilized a probabilistic surrogate model to approximate the objective function.
• Employed an expected improvement acquisition function to guide parameter selection.
• Built a Bayes model to define calibration criteria and the objective function.

Main Results:

• The BO method significantly reduces the number of required numerical model evaluations.
• Achieved effective calibration within approximately 200 evaluations in case studies.
• Demonstrated efficiency in parameter inversion, objective function minimization, and adapting calibration criteria.
• Successfully calibrated a Cr(VI) transport model.

Conclusions:

• Bayesian optimization offers an effective and efficient approach for calibrating groundwater contaminant transport models.
• The BO method substantially reduces the computational budget for model calibration.
• This approach enhances the practical application of numerical modeling in hydrogeology.