Section: New Results
Image data assimilation for oceanography
The objective of this study is to compute the circulation velocity from a sequence of satellite images of Sea Surface Temperature (SST), in order to assimilate it into an oceanographic circulation model.
It is possible to assess velocities from several images by solving a PDE system (conservation equation, regularity) defined using derivatives of the image computed at every pixels. This is not satisfying in the case of oceanographic observations, since a significant part of the observations of the sea surface is occulted by clouds.
We therefore propose a velocity estimation method with a similar principle as data assimilation: the available image data (not occulted by clouds) constitute observations of SST, to be assimilated within a dedicated Image Model , used to describe the evolution of image information. The Image Model implements a simplified version of the transport of temperature by advection and diffusion: the transport is expressed in the 2D image space, source and sink terms being ignored. The state space is defined by the SST and the 2D components of the surface velocity. The latter are assumed constant during the assimilation time window.
Observations of sea surface temperature are assimilated within the Image Model, by minimizing a functional expressing the discrepancy between the modelled and the observed SST, plus an additional second-order div-curl regularization term applied to the retrieved velocity field.
This study is a crucial step towards the formalisation of assimilation techniques for image data. Current research works address the application of this technique to the assimilation of structured image data such as fronts, instead of considering an image as a collection of independent measurements.