Choosing Between Heterogeneity and Anisotropy - What's in the Data and What Do Your Purposes Require?

Fluvial gravel aquifers are characterized by a spatial variability of hydraulic conductivity due to the internal structure of sedimentary deposits. The depositional character of such aquifers introduces strong anisotropic flow conditions on larger scales. The required strength of anisotropy in models depends on the resolution of heterogeneity. Resolving subsurface heterogeneity and the associated anisotropic hydraulic conductivity by means of inverse modeling requires collecting sufficient and conclusive data. Restricted field budget as well as choosing a representative simplified model of the heterogeneous system, however, introduces great uncertainties in parameter estimation and prediction of groundwater management measures. We propose a novel approach for jointly optimizing the measurement and modeling strategies to reduce uncertainty in the prediction of heterogeneity-induced hydraulic anisotropy on larger scales. We numerically simulate radially symmetric steady-state groundwater flow to a partially penetrating well and observe the corresponding drawdown in multi-level observation wells at various radial distances using a large ensemble of highly resolved vertical profiles of hydraulic conductivity. We consider different combinations of a set of observation wells and mimic a best-estimate model calibration for each combination of wells, for each ensemble member, and for each member of a set of simplified models with lower vertical resolution. Our target is to assess a fully upscaled horizontal and vertical hydraulic conductivity, its anisotropy ratio, and the pumping rate needed to dewater a construction pit. Results reveal that the choice of an appropriate measurement strategy improves the prediction of parameters, but depends on the primary targeted property. With increasing radial observation distance, resolving the effectively heterogeneous subsurface with simplified models becomes unfeasible and the prediction uncertainty of hydraulic anisotropy increases. Our study indicates that selecting a model and appointing a measurement strategy are strongly interrelated. We demonstrate that jointly optimizing observation networks and model selection can help to reduce the uncertainty in predicting hydraulic anisotropy and additionally can facilitate resilient measurement and modeling strategies.