In river hydraulics, observations are available only in very small quantities. Consequently, in order to take full advantages of data assimilation for river and flooding models, we think to use remote sensing data which can bring extra information on the flow. Here, we present two studies related to variationnal data assimilation methods for river flows when remote sensed data are available: 1) a real flood event (Moselle river) where we identify inflow discharge when one satellite image and pointwise elevation values are partially available; 2) Assimilation of surface flow trajectories (Lagrangian data). Mathematical models are based on St-Venant equations 1.5D and 2D. Numerical results are presented.