Riverbed hydraulic conductivity (K) is a critical parameter for the prediction of exchange fluxes between a river and an aquifer. In this study, the role of heterogeneity patterns was explored using the fully integrated hydrological model HydroGeoSphere simulating complex, variably saturated subsurface flow. A synthetic 3-D river-aquifer reference model was constructed with a heterogeneous riverbed using non-multi-Gaussian patterns in the form of meandering channels. Data assimilation was used to test the ability of different riverbed K patterns to reproduce hydraulic heads, riverbed K and river-aquifer exchange fluxes. Both fully saturated as well as variably saturated conditions underneath the riverbed were tested. The data assimilation experiments with the ensemble Kalman filter (EnKF) were carried out for four types of geostatistical models of riverbed K fields: (i) spatially homogeneous, (ii) heterogeneous with multi-Gaussian distribution, (iii) heterogeneous with non-multi-Gaussian distribution (channelized structures) and (iv) heterogeneous with non-multi-Gaussian distribution (elliptic structures). For all data assimilation experiments, state variables and riverbed K were updated by assimilating hydraulic heads. For saturated conditions, heterogeneous geostatistical models allowed a better characterization of net exchange fluxes than a homogeneous approximation. Among the three heterogeneous models, the performance of non-multi-Gaussian models was superior to the performance of the multi-Gaussian model, but the two tested non-multi-Gaussian models showed only small differences in performance from one another. For the variably saturated conditions both the multi-Gaussian model and the homogeneous model performed clearly worse than the two non-multi-Gaussian models. The two non-multi-Gaussian models did not show much difference in performance. This clearly shows that characterizing heterogeneity of riverbed K is important. Moreover, particularly under variably saturated flow conditions the mean and the variance of riverbed K do not provide enough information for exchange flux characterization and additional histogram information of riverbed K provides crucial information for the reproduction of exchange fluxes.
- Data assimilation
- Ensemble Kalman filter
- Fully-coupled physically-based flow modelling
- Riverbed characterization