r2ogs5: Calibration of Numerical Groundwater Flow Models with Bayesian Optimization in R

Area of Science:

• Environmental Science
• Hydrogeology
• Computational Science

Background:

• Climate trends and extreme weather impact regional groundwater balance, necessitating high-resolution models.
• Increasing model complexity makes conventional calibration methods time-consuming.
• Efficient global optimization methods are crucial for calibrating complex groundwater flow models.

Purpose of the Study:

• To introduce r2ogs5, a tool integrating OpenGeoSys with the R programming language for groundwater model calibration.
• To present a sequential model-based optimization approach combining Bayesian optimization (BO) with surrogate modeling.
• To demonstrate the efficiency of this calibration method compared to conventional approaches.

Main Methods:

• Integration of OpenGeoSys version 5 with the R programming language via the r2ogs5 package.
• Application of a sequential model-based optimization approach using Bayesian optimization (BO) and surrogate modeling.
• Calibration of 4 and 12 parameters for two simple groundwater flow models.

Main Results:

• The r2ogs5 method, using Bayesian optimization and surrogate modeling, required fewer groundwater flow model runs than gradient search and Latin hypercube sampling for a 12-parameter model.
• Demonstrated the feasibility of calibrating computationally expensive groundwater flow models.
• Highlighted the utility of r2ogs5 for accessing R's statistical analysis and visualization capabilities.

Conclusions:

• r2ogs5 provides an efficient and intelligent global optimization method for calibrating complex groundwater flow models.
• The tool facilitates the integration of groundwater flow simulation within R-based environmental analysis workflows.
• r2ogs5 is a valuable resource for geoscientists and groundwater modelers utilizing the R language.