Ensemble Kalman filtering and conditional normalizing flows for seismic monitoring via data assimilation

Ensemble Kalman filtering and conditional normalizing flows for seismic monitoring via data assimilation (2023 - 2025)

• Applied the ensemble Kalman filter (EnKF) to estimate subsurface CO₂ concentrations using surface seismic observations in a realistic setting.
• Compared EnKF to a class of scalable conditional normalizing flow filters based on neural networks with coupling layers.
• Redesigned the parametrization of the coupling layer to be more robust to weight initialization and data noise.

Code here