Data Assimilation Meetings at Reading

Date Meeting type Speakers
11 March 2015
Agric 1L14
Invited speaker Melanie Ades (University of Reading)
Estimating the full posterior pdf with particle filters
The predominant reason that Particle filters have been advocated as a Data Assimilation scheme is that theoretically they are able to represent the full posterior pdf. Unfortunately, the issue with the standard implementation of the Particle Filter (the Sequential Importance Resampling (SIR) filter) is that it suffers from filter degeneracy, which prevents it producing reasonable results in high dimensional systems. Recently, several adaptations of the SIR filter have been proposed that reduce or avoid the issue of filter degeneracy. This work looks at the effect these adaptations have on the representation of the posterior pdf and how this is impacted when the number of particles are reduced.

Page navigation