Abstract: Although the research on flow around bluff body sections can obtain the characteristics of the velocity field through particle image velocimetry (PIV) and computational fluid dynamics (CFD) methods, it is limited by the Reynolds number effect, onsite test conditions, and the accuracy of numerical simulation. However, the full-scale bridge surface pressure measurement technology is more engineering practical. Building upon the inherent coupling between the surface pressure field and velocity field of bridge sections, this paper proposes a reduced-order correlation and prediction model for the velocity field of a 1:5 rectangular section using surface pressure distribution. The developed model‌ integrates dynamic mode decomposition (DMD) and a BP neural network to: (1) extract pressure and velocity field modes across Reynolds numbers (1000–20000); (2) establish their mapping relationship through an implicit neural network; and (3) achieve velocity field prediction from pressure data. ‌Validation results‌ at Re=6000 show prediction errors of merely 0.06 m/s (lateral) and 0.02 m/s (vertical) at the reference point 1.5, 0, demonstrating the model's effectiveness. This research provides valuable insights for wake flow field reconstruction and aerodynamic measure evaluation in bridge sections.