A nested sampling particle filter for nonlinear data assimilation
Type
ArticleKAUST Department
Earth Fluid Modeling and Prediction GroupEarth Science and Engineering Program
Environmental Science and Engineering Program
Physical Science and Engineering (PSE) Division
Date
2014-04-15Online Publication Date
2014-04-15Print Publication Date
2014-07Permanent link to this record
http://hdl.handle.net/10754/563498
Metadata
Show full item recordAbstract
We present an efficient nonlinear data assimilation filter that combines particle filtering with the nested sampling algorithm. Particle filters (PF) utilize a set of weighted particles as a discrete representation of probability distribution functions (PDF). These particles are propagated through the system dynamics and their weights are sequentially updated based on the likelihood of the observed data. Nested sampling (NS) is an efficient sampling algorithm that iteratively builds a discrete representation of the posterior distributions by focusing a set of particles to high-likelihood regions. This would allow the representation of the posterior PDF with a smaller number of particles and reduce the effects of the curse of dimensionality. The proposed nested sampling particle filter (NSPF) iteratively builds the posterior distribution by applying a constrained sampling from the prior distribution to obtain particles in high-likelihood regions of the search space, resulting in a reduction of the number of particles required for an efficient behaviour of particle filters. Numerical experiments with the 3-dimensional Lorenz63 and the 40-dimensional Lorenz96 models show that NSPF outperforms PF in accuracy with a relatively smaller number of particles. © 2013 Royal Meteorological Society.Citation
Elsheikh, A. H., Hoteit, I., & Wheeler, M. F. (2014). A nested sampling particle filter for nonlinear data assimilation. Quarterly Journal of the Royal Meteorological Society, 140(682), 1640–1653. doi:10.1002/qj.2245Publisher
WileyDOI
10.1002/qj.2245ae974a485f413a2113503eed53cd6c53
10.1002/qj.2245