Predicting the evolution of geophysical fluids (ocean, atmosphere, continental water) is a ma jor scientific and societal challenge. Achieving this goal requires to consider all the available information: numerical models, observations, error statistics... In order to combine these heterogeneous source of information one uses the data assimilation techniques. During the last two decades, many visible and infrared band sensors have been launched on different satellites. This provide a large amount sequence of images of the earth system. This kind of information is underused in current data assimilation systems. In this paper we will describe how to use optimal control methods for data assimilation and in particular we will emphasise on techniques allowing to assimilate sequences of images.