Covariance Localization and Data Transformation within Deterministic Ensemble Kalman Filter for Reservoir History Matching

During past decade, Ensemble Kalman filter, EnKF, has emerged as a promising method for reservoir history matching. Although EnKF overcomes KF inability for large scale model, butEnKF, also, requires considerations: Using small ensemble for practical implementation leads toappearance of spurious correlation in conariance structure. On the other hand, non-Gaussian variables in EnKF state vector violate KF assumption resulting suboptimal results. In this study weimplemented an exponential distance base covariance localization and non-parametric Gaussiananamorphosis transformation to mitigate the former and the later issue. The results indicate that localization is a crucial mitigation step which is a must meanwhile data transformation along with localization preserves system total variance the most.

Keywords:

Reservoir History Matching , Ensemble Kalman Filter , Covariance Localization , Gaussian Transformation