Uncertainty Quantification for 3D FWI at an Operational Level: an Ensemble Transform Kalman Filter approach
Résumé
Gaining knowledge on the variability of the FWI solutions is crucial. It shall help control their quality and guide their geological interpretation. The main obstacle for designing efficient uncertainty quantification strategy remains the computational cost. We propose in this study an adaptation to FWI of a low-rank approach developed in the data assimilation community named Ensemble Transform Kalman Filter. The method works in a predictor-corrector manner for an ensemble of models instead of considering a single model. In our strategy, the prediction step implies solving the FWI problem for each model of the ensemble. The correction step involves only low-rank linear algebra operations and is thus computationally feasible. We present the application of this strategy to a 3D OBC dataset from the North Sea. We show how we gain insight on the variability of the solution through the estimated variance, and on the local resolution through the estimated covariance. We also test the sensitivity of the method with respect to the ensemble size and find that a number of 50 models makes it possible to capture most of the information.
Origine | Fichiers produits par l'(les) auteur(s) |
---|