Groundwater flow models are critical elements for assessment using numerical simulation. However, it is difficult to create a valid model that can simulate actual observations properly because information about a hydrogeological structure and hydraulic properties are often limited. In this study, we develop a model identification method for groundwater flow simulation based on a sequential data assimilation technique. The ensemble Kalman filter (EnKF) is employed to deal with the nonlinearity of a groundwater flow model and applied to a 3-D saturated-unsaturated flow model.
To assess the identifiability of the groundwater flow model through data assimilation, we perform numerical experiments using synthetic observations of an underground fuel storage facility. Groundwater level and inflow measurements are assimilated into the model to identify the hydraulic parameters of geological layers and fracture zones. The results show that the true hydraulic parameter values can be obtained by the EnKF algorithm with a reasonable ensemble size. The influence of non-linearity, which is mainly derived from unsaturated hydraulic properties of rock, is evaluated from probability density functions that are provided by the EnKF. The nonlinearity does not affect the assimilation result significantly in this case. In addition, the applicability of the method is demonstrated on the basis of real field measurements. The results of data assimilation experiments show that an estimated model can simulate the actual behavior of groundwater flow except in a few observation points. However, we find that accuracies of an assumed hydrogeological structure and assigned boundary conditions are insufficient in particular areas. These results suggest that the developed method has abilities not only to identify and calibrate the groundwater flow model with dynamic data, but also to validate the model.
