David Livings, University of Reading

The Probability of Weight Collapse in the Weighted Ensemble Kalman Filter


The weighted ensemble Kalman filter (WEnKF) is a particle filter that uses an ensemble Kalman filter (EnKF) to provide a proposal distribution. It offers the possibility of converting an existing EnKF into a filter that will solve the full, nonlinear, non-Gaussian, filtering problem. Unfortunately, experiments with a simple dynamical system have shown that the WEnKF is prone to catastrophic weight collapse in some circumstances. This presentation will show how to make rough calculations of the probability of such a collapse happening.

File available here