Abstract: To improve the prediction performance of the RANS model for flow separation, the model constants of the SST turbulence model were recalibrated using the unsteady ensemble Kalman filter (EnKF) data assimilation (DA) combined with the particle image velocimetry (PIV) data of the flow around a NACA0012 aerofoil. The differences in prediction between steady DA and unsteady DA with different model constant perturbations and ensemble sizes were compared and analyzed. The results show that the unsteady simulation can enhance the robustness of the numerical simulation and improve the initial prediction distribution of the RANS model compared to the steady simulation. The steady DA has obvious defects for large model constant perturbation or small ensemble size. The unsteady DA is more robust and can obtain the optimal turbulence model constants with larger perturbation and smaller ensemble size, resulting in more accurate prediction of the flow fields.