Sequential Assimilation of Tracer Test Data by Ensemble Kalman Filter for Fracture Identification

Application of the ensemble Kalman filter (EnKF) to reservoir characterization has the limitation that the observed watercut data cannot be reflected for tuning model variables in some time intervals. The present study assumed a homogeneous isotropic porous medium including a single fracture with adequate width, so that tracer tests could be simulated by the complex variable boundary element method. The resulting tracer effluent behavior was used as observed data, and the EnKF was applied for identifying fracture location to match the data. Using the conventional method, the effluent tracer concentration was measured as the observed variable. Depending on the initial ensemble, the data was not reflected for tuning the fracture location in some time intervals, resulting in a mismatch between the fracture location obtained by EnKF and the referenced location. Furthermore, the poor correspondence of observation data resulted in inaccurate identification even for initial ensembles not including the problematic time interval. Therefore, the observed variable was transformed from the effluent tracer concentration into the time corresponding to the amount of tracer recovery. Consequently, the observed data were adequately assimilated, and the correspondence of the observation data was improved. Accurate localization of the fracture was possible regardless of the initial ensembles.