High-fidelity computational fluid dynamics (CFD) has become an essential tool for the investigation of complex, multi-scale flow phenomena. Its application to tasks such as uncertainty quantification, optimal control and design optimization, however, is severely limited by the high computational cost associated with repeated or near real-time simulations. To overcome this limitation, the present study develops a data-driven reduced-order model (ROM) comprising distinct offline and online stages. In the offline stage, the dominant spatial structures and their associated temporal coefficients are extracted using proper orthogonal decomposition (POD), after which high-order dynamic mode decomposition (HODMD) is employed to represent the temporal dynamics in a Fourier-like form. In the online stage, interpolation in the tangent space of the Stiefel manifold (ITSSM) is used to predict POD modes for previously unseen flow parameters, while mode-realigned pointwise interpolation (MRPWI) is applied to reconstruct the corresponding fundamental frequency and HODMD modes. The flow fields associated with unseen parameters are then reconstructed from the predicted POD modes, fundamental frequency and HODMD modes. The performance of the proposed methodology is assessed using three classical benchmark problems: flow past a circular cylinder and a square cylinder at low Reynolds numbers, and flow over a NACA0012 airfoil at different angles of attack. The predicted velocity fields, drag and lift coefficients show good agreement with high-fidelity numerical simulations. The principal advantage of the present approach lies in its ability to substantially reduce computational cost while preserving the essential flow physics, thereby enabling rapid parameter exploration and optimization across a wide range of applications.