16-17 sept. 2021 Fontainebleau (France)
Comparison of simulation-based algorithms for parameter estimation and state reconstruction in nonlinear state-space models
Valérie Monbet  1@  , Pierre Ailliot  2@  , Thi Tuyet Trang Chau  3@  , Pierre Tandeo  4, 5@  
1 : IRMAR
Universite de Rennes 1
2 : Département de Mathématiques [Brest]  -  Site web
Université de Bretagne Occidentale (UBO)
6 avenue Le Gorgeu CS 93837 29238 BREST cedex 3 -  France
3 : Laboratoire des Sciences du Climat et de lÉnvironnement [Gif-sur-Yvette]  -  Site web
Université de Versailles Saint-Quentin-en-Yvelines : UMR8212, Commissariat à l'énergie atomique et aux énergies alternatives : DRF/LSCE, Université Paris-Saclay, Institut National des Sciences de l'Univers : UMR8212, Centre National de la Recherche Scientifique : UMR8212
Bât. 12,avenue de la Terrasse, F-91198 GIF-SUR-YVETTE CEDEX -  France
4 : Laboratoire des sciences et techniques de l'information, de la communication et de la connaissance  (Lab-STICC)  -  Site web
Télécom Bretagne, Institut Mines-Télécom, CNRS : UMR6285
Technopole Brest Iroise BP 832 29285 BREST CEDEX -  France
5 : Département Signal et Communications  (SC)
Télécom Bretagne, Institut Mines-Télécom
Technopôle Brest-Iroise CS 83818 29238 BREST CEDEX 3 -  France

In geosciences, data assimilation (DA) methods aim at reconstructing the state of a system by combining observations with a physical model. The most popular algorithms in the DA community are based on the Ensemble Kalman Filter and Smoother (EnKF/EnKS) and its extensions.

DA is usually based on a state space model. The parameters of the model strongly impact the results of DA algorithms and they are usually unknown and may be difficult to specify. There is hence a need for an efficient method to estimate them. Expectation-Maximization (EM) is the most classical algorithm in the statistical literature to estimate the parameters in models with state variables. It consists in updating sequentially the parameters by maximizing a likelihood function where the state is approximated using a smoothing algorithm. In this paper, we propose to combine a Stochastic EM (SEM) algorithm with various smoothing algorithms and compare the results.


Personnes connectées : 24 Vie privée
Chargement...