Image assimilation with the weighted ensemble Kalman filter

Abstract : We present a sequential data assimilation method based on the combination of two assimilation techniques: the ensemble Kalman filter (EnKF) and the particle filter. Both techniques are based on Monte Carlo sampling allowing approximate solving of non linear stochastic filtering problems. However, while the EnKF is still based on a Gaussian assumption, the particle filter does not rely on such an assumption but is known to be less efficient when the number of available samples is small. In practice, both techniques are combined in the sense that the sampling step of the particle filter is based on the EnKF technique, followed by a weighting of samples using observations. The association of these two approaches is a step toward an efficient application of ensemble techniques to high-dimensional and non linear / non Gaussian problems, such as those encountered in meteorology or oceanography. We show the performances of this new approach on high-dimensional problems where the goal is to filter turbulent velocity fields from image observations. The assimilation technique associates a non linear stochastic dynamical model to linear observations extracted from the image sequences, or directly to the image data through a non linear observation operator. The method has been validated on a synthetic sequence, and applied to real oceanographic satellite image sequences of SST (sea surface temperature). This work corresponds to an extension of the preliminary study published in Tellus (N. Papadakis, E. Memin, A. Cuzol, N. Gengembre. Data assimilation with the Weighted Ensemble Kalman Filter. Tellus A, vol.62(5), p. 673-697, 2010). In a second part, we present a way to improve the assimilation when the time step between observations (images) is very long. We make use of a conditional simulation technique in order to reduce dynamical discontinuities produced in that case by the sequential techniques.
Type de document :
Communication dans un congrès
International Workshop on Adjoint Model Applications in Dynamic Meteorology, Oct 2011, Cefalu, Sicilia,, Italy. 2011
Liste complète des métadonnées
Contributeur : Etienne Memin <>
Soumis le : dimanche 6 mai 2012 - 17:21:37
Dernière modification le : mercredi 11 avril 2018 - 02:00:59


  • HAL Id : hal-00694782, version 1



Cuzol Anne, Etienne Memin. Image assimilation with the weighted ensemble Kalman filter. International Workshop on Adjoint Model Applications in Dynamic Meteorology, Oct 2011, Cefalu, Sicilia,, Italy. 2011. 〈hal-00694782〉



Consultations de la notice