Dense Estimation of Layer Motions in the Atmosphere

In this paper, we address the problem of estimating dense motion fields related to a stratified atmosphere which is observed through satellite imagery. Estimating the evolving vertical distribution of horizontal wind fields from satellite image time series is of great importance for the study of atmospheric dynamics. Because of the sparse 3D nature of observations, classical correlation-based techniques are not suited for the dense estimation of layer motion. Moreover, such methods are not necessarily temporally consistent. This paper proposes a sound energy-based estimator producing dense wind fields estimates for each partially observed atmospheric layer. The energy function to be minimized is composed of a data term based on the continuity equation which cancels out the influence of undesirable layers and of a specific div curl regularization term. To preserve the temporal consistency of the estimates, the variational method is initialized by propagation of the previous estimated field according to a velocity-vorticity formulation of Navier-Stokes equations. The relevance of our estimator is demonstrated on Meteosat image sequences