In this work, a data assimilation method is proposed following an optimise-then-discretise approach, and is applied in the context of computational haemodynamics. The methodology aims to make use of phase-contrast magnetic resonance imaging to perform optimal flow control in computational fluid dynamic simulations. Flow matching between observations and model predictions is performed in luminal regions, excluding near-wall areas, improving the near-wall flow reconstruction to enhance the estimation of related quantities such as wall shear stresses. The proposed approach remarkably improves the flow field at the aortic root and reveals a great potential for predicting clinically relevant haemodynamic phenomenology. This work presents model validation against an analytical solution using the standard 3-D Hagen–Poiseuille flow, and validation with real data involving the flow control problem in a glass replica of a human aorta imaged with a 3T magnetic resonance scanner. In vitro experiments consist of both a numerically generated reference flow solution, which is considered as the ground truth, as well as real flow MRI data obtained from phase-contrast flow acquisitions. The validation against the in vitro flow MRI experiments is performed for different flow regimes and model parameters including different mesh refinements.