Assimilation of satellite data into numerical models by optimal control methods

• Main Author: van den Berghe, François
• Other Authors: Centre Énergétique et Procédés (CEP), Mines Paris - PSL (École nationale supérieure des mines de Paris), Université Paris Sciences et Lettres (PSL)-Université Paris Sciences et Lettres (PSL), École Nationale Supérieure des Mines de Paris, Lucien Wald([e-mail])
• Format: Doctoral or Postdoctoral Thesis
• Language: French
• Published: HAL CCSD 1992
• Subjects: télédétection; assimilation de données; équations adjointes; filtre de Kalman; modèle numérique; mesures diffusiomètriques; remote sensing; data assimilation; adjoint equations; Kalman filter; numerical model; scatterometer observations; [SPI.SIGNAL]Engineering Sciences [physics]/Signal and Image processing; [INFO.INFO-TS]Computer Science [cs]/Signal and Image Processing; North Atlantic
• Online Access: https://pastel.hal.science/pastel-00954478; https://pastel.hal.science/pastel-00954478/document; https://pastel.hal.science/pastel-00954478/file/VANDENBERGHE_FranA_ois-rectifiA_.pdf

Satellite data assimilation methods are investigated. There are mainly two kinds of technics: the sequential methods, derived from the Kalman filter and the variational methods, based on the adjoint equations of the optimal control theory. Variational methods are more recent. This work attempts to assess their potentialities in remote sensing through two examples. The first study, carried out with the advection-diffusion equation as a numerical model, demonstrates the feasibility of the variational methods. It is possible to minimize a cost function measuring the distance between the model trajectory and observations distributed in time and space. The second application is performed with actual observations of the Seasat satellite borne-wind scatterometer. Its aim is the mapping of the wind fields at the sea surface. Within a twelve hours time period, data are assimilated in a non linear numerical model. This model discretizes the vorticity equation on a rectangular domain covering approximatively the North Atlantic Ocean. A comparison between the wind fields obtained with the variational assimilation and those produced by a sequential assimilation highlights the adjoint equations ability to propagate information backward in time. The resulting analyzes are better, particulary at the beginning of the assimilation time period, and the analyzed fieds are consistent with the model dynamics. Nous nous intéressons aux méthodes permettant d'assimiler les données de télédétection. Il y a principalement deux types de technique : les méthodes séquentielles, issues du filtrage de Kalman et les méthodes variationnelles, basées sur les équations adjointes du contrôle optimal. Les méthodes variationnelles sont récentes et nous essayons de présenter leur intérêt pour la télédétection à l'aide de deux applications. Une première étude est menée sur un modèle numérique d'advection-diffusion. Elle montre la faisabilité numérique de la méthode : il est possible de minimiser une fonction de coût mesurant la distance entre la ...