PEST

PEST implements the pestpp-ies iterative ensemble smoother (IES) for parameter estimation, history matching, and uncertainty quantification in large-scale environmental models via the PEST model coupling protocols.

Key Features:

• Iterative Ensemble Smoother (IES): Implements an ensemble-based variant of the Gauss–Levenberg–Marquardt algorithm to update parameters using ensembles of model realizations.
• Ensemble uncertainty propagation: Propagates parameter uncertainty through ensembles and updates parameters iteratively using linearized covariance relationships.
• Model coupling: Interfaces non-intrusively with existing simulators via the standard PEST model coupling protocols to enable model-independent application.
• Parallelization: Provides automated parallelization across ensemble members for scalable computation.
• Lambda regularization testing: Includes automated lambda regularization testing for stabilized parameter updates.
• Fault tolerance: Maintains operation in the presence of failed model evaluations through fault-tolerant ensemble handling.
• High-dimensional conditioning: Conditions thousands to tens of thousands of parameters and delivers posterior predictive uncertainty estimates for forecasts.
• Computational efficiency: Uses ensemble-based calibration and scalable uncertainty propagation to reduce computational burden relative to traditional gradient-based or Markov chain methods.

Scientific Applications:

• Parameter estimation: Calibration of model parameters in large-scale environmental models.
• History matching: Matching model outputs to observational records for model conditioning.
• Uncertainty quantification: Quantifying parameter and predictive uncertainty, including posterior predictive uncertainty for forecasts dependent on large parameter sets.
• Groundwater and solute transport modeling: Application to groundwater flow and solute transport models and other environmental domains via model coupling.
• High-dimensional model calibration: Conditioning and calibration of models with thousands to tens of thousands of parameters.

Methodology:

Uses the pestpp-ies ensemble-based Gauss–Levenberg–Marquardt algorithm, propagates uncertainty through ensembles, updates parameters using linearized covariance relationships, performs automated parallelization across ensemble members, applies lambda regularization testing, and implements fault tolerance for failed model evaluations while coupling to simulators via the PEST model protocols.