Experimental and numerical approaches are facing strong limitations for the study of complex scenarios. The former, based on particle image velocimetry (PIV), are subject to measurements uncertainties, to limited spatiotemporal resolution, and can be affected by regions of the images where no information on the flow are available. The latter need accurate simulations that may strongly depend on complex unknown boundary conditions, and on a fine and often ad hoc tuning of turbulence model parameters. It follows that complex flows cannot be accurately analysed on the single basis of one of those two kinds of approaches. Here we present our recent research on data assimilation to combine numerical model and experimental observations. Two distinct flow configurations and data assimilation approaches are proposed: (1), a EnWKF assimilation scheme to reconstruct free surface flows form Kinect depth sensor images; (2), a 4DVar assimilation technique, integrating PIV data and DNS, to reconstruct turbulent flows.