Résumé : Since several decades many satellites have been launched for the observation of the Earth for a better knowledge of the atmosphere and of the ocean. The sequences of images that such satellites provide show the evolution of some large scale structures such as vortices and fronts. It is obvious that the dynamic of these structures may have a strong predictive potential. Extracting these structures and tracking their evolution automatically is then essential for future forecast systems. In this paper we consider extraction of spatio-temporal geometric structures from image sequences of geophysical fluid flow using three-dimensional (3D) curvelet transform and total variation minimization. Numerical experiments on simulated geophysical fluids and real video data by remote sensing show good performances of the proposed method in terms of denoising and edge structural extraction. This work is partially motivated by a sequent application to image sequence assimilation of geophysical fluids.