Bedload transport governs sediment flux in riverine and engineered systems through turbulence-modulated interactions between near-bed coherent structures and granular particles. This study employs a coupled Computational Fluid Dynamics-Discrete Element Method (CFD-DEM) framework to evaluate turbulence modeling approaches for bedload transport. We conduct a comparison between Large Eddy Simulation (LES) and Reynolds-Averaged Navier-Stokes (RANS) across subcritical to supercritical Shields regimes. Simulation results capture: (1) Turbulence-driven particle entrainment initiating below the critical Shields threshold via intermittent sweeps and ejections; (2) Bedform evolution from small-scale ripples to dunes linked to turbulent kinetic energy anisotropy; (3) Particle saltation dynamics modulated by near-bed turbulence intensity. While RANS offers computational advantages, it exhibits systematic underprediction of near-bed turbulent kinetic energy. These results establish a mechanistic link between microscale turbulence structures and bulk sediment flux, providing a physics-based framework to refine predictive models through explicit incorporation of turbulence-particle phase interactions.