Ensemble-Based Uncertainty Estimation in Full Waveform Inversion
Résumé
Uncertainty Quantification is a major issue for most tomography problems. In this work, We propose an original application of ensemble-based methods in Full Waveform Inversion (FWI). The method relies on classical FWI-schemes coupled with a deterministic version of an Ensembl Kalman Filter (EnKF). This approach allows accessing a low-rank estimation of the posterior covariance matrix, giving us access to quantitative uncertainty measurements as long as we are close enough to the global minimum. Applications to 2D FWI in the frequency domain have proven to be successful, and the variance maps obtained from joint EnKF-FWI are encouraging premise for this method.