Data assimilation with state alignment using high-level image structures detection

Sequential and variational assimilation methods allow tracking physical states using dynamic prior together with external observation of the studied system. However, when dense image satellite observations are available, such approaches realize a correction of the amplitude of the different state values but do not incorporate the spatial errors of structure positions. In the case of the position of a vortex, for example, when there is misfit between state and observation, the processes can be long to converge and even diverge when high dimensional state spaces are treated with few iterations of the assimilation methods as it is the case in operational algorithms. In this paper, we tackle this issue by proposing an alignment method based on modern object detection methods that uses visual correspondences between the physical state model and the structural information given by a sequence of image observing the phenomena.