Time-consistent estimators of 2D/3D motion of atmospheric layers from pressure images

In this paper, we face the challenging problem of estimation of time-consistent layer motion fields at various atmospheric depths. Based on a vertical decomposition of the atmosphere, we propose three different dense motion estimator relying on multi-layer dynamical models. In the first method, we propose a mass conservation model which constitutes the physical background of a multi-layer dense estimator. In the perspective of adapting motion analysis to atmospheric motion, we propose in this method a two-stage decomposition estimation scheme. The second method proposed in this paper relying on a 3D physical model for a stack of interacting layers allows us to recover a vertical motion information. In the last method, we use the exact shallow-water formulation of the Navier-Stokes equations to control the motion evolution across the sequence. This is done through a variational approach derived from data assimilation principle which combines the dynamical model and the pressure difference observations obtained from satellite images. The three methods use sparse pressure difference image observations derived from top of cloud images and classification maps. The proposed approaches are validated on synthetic example and applied to real world meteorological satellite image sequences