Data Assimilation is commonly used in environmental sciences to improve forecasts, obtained by meteorological, oceanographic or air quality simulation models, with observation data. It aims to solve an evolution equation, describing the dynamics, and an observation equation, measuring the misfit between the state vector and the observations, to get a better knowledge of the actual system's state, named the reference. In this article, we describe how to use this technique to recover missing data and reduce noise on satellite images. The recovering process is based on assumptions on the underlying dynamics displayed by the sequence of images. This is a promising alternative to methods such as space-time interpolation. In order to better evaluate our approach, results are first quantified for an artificial noise applied on the acquisitions and then displayed for real data.