Conceptualization and Calibration of Anisotropic Alluvial Systems: Pitfalls and Biases |
| |
| Authors: | Guillaume Gianni John Doherty Philip Brunner |
| |
| Institution: | 1. Watermark Numerical Computing, Brisbane, Australia;2. Center for Hydrogeology and Geothermics, University of Neuchâtel, Neuchâtel, Switzerland |
| |
| Abstract: | Physical properties of alluvial environments typically feature a high degree of anisotropy and are characterized by dynamic interactions between the surface and the subsurface. Hydrogeological models are often calibrated under the assumptions of isotropic hydraulic conductivity fields and steady-state conditions. We aim at understanding how these simplifications affect predictions of the water table using physically based models and advanced calibration and uncertainty analysis approaches based on singular value decomposition and Bayesian analysis. Specifically, we present an analysis of the information content provided by steady-state hydraulic data compared to transient data with respect to the estimation of aquifer and riverbed hydraulic properties. We show that assuming isotropy or fixed anisotropy may generate biases both in the estimation of aquifer and riverbed parameters as well as in the predictive uncertainty of the water table. We further demonstrate that the information content provided by steady-state hydraulic heads is insufficient to jointly estimate the aquifer anisotropy together with the aquifer and riverbed hydraulic conductivities and that transient data can help to reduce the predictive uncertainty to a greater extent. The outcomes of the synthetic analysis are applied to the calibration of a dynamic and anisotropic alluvial aquifer in Switzerland (The Rhône River). The results of the synthetic and real world modeling and calibration exercises documented herein provide insight on future data acquisition as well as modeling and calibration strategies for these environments. They also provide an incentive for evaluation and estimation of commonly made simplifying assumptions in order to prevent underestimation of the predictive uncertainty. |
| |
| Keywords: | |
|
|
