PEST++IES How Many Iterations and Realizations, Finding the Point of Diminishing Returns

Area of Science:

• Groundwater hydrology
• Computational modeling
• Geostatistics

Background:

• PEST++IES is a popular tool for groundwater model calibration and uncertainty analysis.
• Its ensemble smoother approach is efficient for highly parameterized models.
• Determining the optimal number of ensemble realizations and iterations is crucial for efficiency.

Purpose of the Study:

• To investigate the optimal number of ensemble realizations and iterations for PEST++IES.
• To evaluate the impact of ensemble size on model performance in groundwater modeling.
• To assess the trade-off between computational cost and accuracy in history matching.

Main Methods:

• A modified Freyberg model was used for simulations.
• Four iterations were performed with ensemble sizes ranging from 10 to 2000.
• Hydraulic conductivity, recharge, river conductance, and well flow rates were adjusted.
• Results were compared against a "truth" model using risk-based well capture zones and hydraulic conductivity fields.

Main Results:

• Ensemble sizes of 100 to 250 realizations generally yielded good results.
• Two PEST++IES iterations were found to be sufficient for most scenarios.
• Smaller ensemble sizes (e.g., 10-50) showed diminished performance.
• Larger ensemble sizes (e.g., >500) offered minimal additional improvement.

Conclusions:

• An ensemble size of 100-250 realizations and two iterations represents an efficient and effective configuration for PEST++IES.
• This finding helps optimize computational resources in groundwater modeling.
• The study provides practical guidance for users of PEST++IES for history matching and uncertainty analysis.