Philip Browne, University of Reading
A comparison of the LETKF and the equivalent weights particle filter on the barotropic vorticity model
Abstract:
Data assimilation methods that work in high dimensional systems are crucial to many areas of the geosciences: meteorology, oceanography, climate science etc. Particle filters have recently been shown to scale to problems that are of use to these communities. I will present a systematic comparison of the equivalent weights particle filter with the established and widely used local ensemble transform Kalman filter method of data assimilation. Both methods are applied to the barotropic vorticity equation for different networks of observations. I will show that, in all cases, the local ensemble transform Kalman filter produced lower root mean squared errors than the equivalent weights particle filter. The performance of the equivalent weights particle filter will be shown to depend strongly on the form of nudging used, and a nudging term based on the local ensemble transform Kalman smoother improves the performance of the filter.
I will highlight the reasons why such a comparison may not be, in a sense, fair. For example, the mean of the equivalent weights particle filter analysis may not be the best measure of error in the case when the posterior is non-Gaussian.