Improved characterization of river-aquifer interactions through data assimilation with the Ensemble Kalman Filter

Exchange processes between rivers and groundwater are an important driver for the hydrological, chemical and ecological environment around streams and the cycling of water at the catchment scale. Management decisions for such systems are very often derived on the basis of model predictions and it is therefore essential to properly estimate the relevant model parameters that govern the interaction between river and aquifer. Various eld studies indicate that hydraulic parameters in and around streams are associated with a considerable uncertainty regarding their temporal and spatial distribution. These uncertainties have to be regarded in the estimation of hydraulic parameters and dierent stochastic inversion methods are available for that task. Among these methods, the Ensemble Kalman Filter (EnKF) has been proven to work well for the characterization of subsurface parameters where its advantage over other stochastic inversion techniques is the calculation of a full posterior probability density function without linearization around an optimum, its computational eciency and its ability to be used for real-time predictions. In this work, EnKF was applied to a 3D groundwater model of a well eld within the Limmat aquifer in Zurich (Switzerland) which is strongly in uenced by river-aquifer interactions. The specic aim was to investigate dierent aspects of the spatio-temporal characterization of river bed properties with EnKF and to explore the worth of dierent conditioning data for this site. In a rst study, the model was used in synthetic experiments where reference runs with temporally varying river bed hydraulic conductivities were generated. Then it was tested, to what extend state-parameter updates with EnKF are able to detect these changes in river bed properties based on a limited set of piezometric head measurements from the reference simulations. In a second study, it was investigated how the spatial representation of heterogeneity in uences the updating behavior of EnKF. In this case, synthetic references with spatially heterogeneous elds of river bed permeabilities were generated and piezometric head data from these references were used to update four dierent parameter ensembles that varied in the spatial representation of heterogeneity (i.e., fully heterogeneous versus zonated leakage parameters).