Python-Based Model Emulation Workflows with PEST.

Area of Science:

• Environmental modeling
• Computational science
• Geosciences

Background:

• High-fidelity environmental models face computational challenges for uncertainty quantification and optimization.
• A lack of standardized tools hinders the integration of surrogate modeling (emulation) into existing workflows, creating an implementation gap.

Purpose of the Study:

• To bridge the implementation gap by providing a standardized, plug-and-play framework for model emulation.
• To facilitate the integration of emulation techniques into existing environmental modeling workflows.
• To enable direct comparisons between physics-based models and emulators.

Main Methods:

• Developed the Emulator module within the open-source pyEMU package.
• Implemented a plug-and-play architecture for deploying Gaussian Process Regression (GPR) and Data-Space Inversion (DSI).
• Automated workflow components, including non-Gaussian data transformation and PEST interface file generation.

Main Results:

• The Emulator module provides a standardized framework for deploying various model emulation techniques.
• Trained surrogates can serve as direct replacements for physics-based models, simplifying workflow integration.
• Demonstrated utility through a benchmarking optimization problem and a synthetic groundwater model history-matching application.

Conclusions:

• The Emulator module in pyEMU effectively bridges the implementation gap for environmental model emulation.
• Standardized tools like Emulator facilitate the adoption and comparison of emulation techniques.
• This facilitates knowledge building within the community on the effective deployment of surrogate models.