H33C-0835:
Large-scale hydrological model prediction uncertainties estimated from an ensemble of hydrostratigraphic models based on resistivity and borehole data

Wednesday, 17 December 2014
Pernille Aabye Marker1, Ty P.A. Ferré2, Nikolaj Foged3, Anders Vest Christiansen3, Esben Auken3, Klaus Mosegaard4 and Peter Bauer-Gottwein1, (1)Technical University of Denmark, Department of Environmental Engineering, Lyngby, Denmark, (2)University of Arizona, Tucson, AZ, United States, (3)Aarhus University, Department of Geoscience, Aarhus, Denmark, (4)Niels Bohr Institute - University of Copenhagen, Copenhagen, Denmark
Abstract:
Large-scale hydrological models are important tools for water resources management. Model predictions are used in agricultural, contamination, water scarcity, and groundwater depletion applications and for well-field management. The predictions, used for management and practical decision-making, are sensitive to variations in hydrostratigraphy, thus uncertainty can be addressed by sampling the structural model space. Hydrostratigraphic input to large-scale hydrologic models is commonly based on one-truth geologic models.

High resolution airborne electromagnetic (AEM) data with extensive spatial coverage are valuable for use in hydrostratigraphic modeling. In particular, geological structures and within-unit heterogeneity, which are poorly identified with spatially scarce borehole lithology data, are well resolved by AEM data. The challenge is to combine geophysical and hydrological information in a common parameter space. We propose to estimate hydrological model prediction uncertainties using an ensemble of resistivity and borehole based hydrostratigraphic models.

Single hydrostratigraphic models are created using a semi-automatic sequential hydrogeophysical inversion method, which integrates AEM and borehole data. A spatially variable translator function converts electrical resistivities obtained from geophysical inversion into clay fractions through correlation with borehole lithological observations. The subsurface domain is divided into zones by k-means clustering on the inferred clay fractions and electrical resistivities. Hydraulic conductivities of the zones are estimated through hydrological model calibration using head and discharge observations. An ensemble of behavioral hydrostratigraphic models is sampled based on a threshold value of the objective function of the clustering algorithm, and goodness of fit to hydrological data. Hydrological predictions are capture zones and drawdown responses to pumping in areas of interest for management. Results will be shown for a Danish case study.