Calibration of parameter fields consisting of multiple statistical populations

Area of Science:

• Geosciences
• Hydrogeology
• Computational Modeling

Background:

• Parameter fields in groundwater models often comprise multiple statistical populations with uncertain spatial distributions.
• Conventional calibration methods typically assume fixed population locations, ignoring spatial uncertainty.

Purpose of the Study:

• To demonstrate the effectiveness of a proposed multimodal calibration method for parameter fields with multiple statistical populations.
• To assess the impact of spatial population uncertainty on groundwater model calibration and posterior reliability.

Main Methods:

• Application of a recently developed multimodal calibration method to real-world groundwater models.
• Comparison of the multimodal calibration approach with a conventional method that fixes population positions.
• Utilizing synthetic calibration runs to analyze the effects of ignoring spatial uncertainty.

Main Results:

• The multimodal method successfully calibrates both the spatial distribution and parameterization of statistical populations.
• Ignoring spatial population uncertainty leads to inaccurate posterior distributions and poorer observation reproduction.
• The proposed method honors the complete prior geostatistical definition of multimodal parameter fields.

Conclusions:

• Multimodal calibration provides a more comprehensive treatment of uncertainties in groundwater modeling.
• This approach prevents adverse effects associated with fixed population assumptions, yielding more trustworthy posterior models.
• The method enhances the reliability of groundwater model calibration by incorporating spatial distribution uncertainty.