Calibration and Reliability in Groundwater Modelling: Credibility of Modelling

(Proceedings of ModelCARE 2007 Conference, held in Denmark, September 2007). IAHS Publ. 320, 2008, 70-75.

Maximum likelihood Bayesian averaging of air flow models in unsaturated fractured tuff

ERIC MORALES-CASIQUE¹, SHLOMO P. NEUMAN¹ &
VELIMIR V. VESSELINOV²

1 Department of Hydrology and Water Resources, The University of Arizona, 1133 E James E. Rogers Way Tucson, Arizona 85721, USA

emorales@hwr.arizona.edu

2 Los Alamos National Laboratory, EES-6, MS T003, Los Alamos, New Mexico 87545, USA

Abstract MLBMA is a maximum likelihood (ML) version of Bayesian model averaging (BMA) that renders it compatible with ML methods of model calibration and thus applicable to cases where prior information about the parameter may be unavailable. We explore the role of prior information in MLBMA by applying it to air flow during a cross-hole pneumatic injection test in unsaturated fractured tuff with and without reliance on packer-test data from six boreholes. We parameterize log air permeability and porosity geostatistically using pilot points and estimate them by calibrating a finite volume pressure simulator (FEHM) against cross-hole pressure data by means of a parallelized version of PEST considering several alternative variogram models. We assess the predictive capabilities of each model based on various model selection criteria and discuss future plans to generate corresponding predictions via MLBMA, cross-validate them against pressure data from the same cross-hole test, and validate them against data from another such test.

Key words Bayesian model averaging; air flow; inverse modelling; maximum likelihood