Abstract: In order to deeply understand the dynamic stall characteristics of elliptic airfoils and capture the main unsteady structures of the flow field, a numerical method suitable for simulating the flow field of elliptic airfoils was established based on unsteady Reynolds-averaged Navier-Stokes (URANS) equations and k-ω SST, γ-Re_θt turbulence models. Subsequently, simulations were carried out for the case of deep dynamic stall of an elliptical airfoil with a relative thickness of 16%, and the unsteady characteristics of the flow field and the variations of aerodynamic coefficient were thoroughly discussed in the time domain. The characteristics of velocity and pressure fields were extracted by dynamic mode decomposition (DMD) technique, and the flow field reconstruction and error assessment were carried out by establishing truncation methods of conjugate modes based on the theory of mode energy proportion. The results show that the appearance of separation bubbles generated by the elliptic airfoil at the leading edge during the upstroke process is an important sign of the dynamic stall vortex formation. The modes of each order can effectively characterize the dynamics of dynamic stall, which are consistent with main flow features in the time domain. The conjugate modes contribute to the unsteady components of the flow field, while the first-order DMD mode reflects the uniform flow field showing characteristics of deep stall of the airfoil. Excluding the first-order mode, the reconstructed flow field using 75%-95% energy proportion of the conjugate modes can capture variations of the aerodynamic coefficients in the time domain to a certain extent, but fails to describe fine unsteady details accurately. Although the reconstructed flow field using 99% energy proportion of the conjugate modes has certain errors when the airfoil undergoes severe flow separation, and the accuracy in capturing fine flow details of near-wall separation bubbles remains limited, it still presents a decent accuracy in aerodynamic coefficient prediction and effectiveness in dimensionality reduction, showing its value in engineering applications.