The turbulent boundary layer over a two-dimensional Gaussian bump is computed with a hierarchy of turbulence modeling strategies in order to evaluate and improve the near-wall modeling capabilities of hybrid RANS/LES approaches. First, DNS data is used to study the intricate effects of strong pressure gradients and streamline curvature on the boundary layer physics. Partial relaminarization and retransition processes are identified, which lead to complex and atypical turbulence behavior. Second, the bump flow is computed by wall-modeled LES using the hybrid IDDES model. The RANS near-wall model is not predictive of the strong favorable pressure gradient (FPG) effects and results in excessive turbulent diffusion and shear stress. Finally, a physics-based and data-informed correction factor is proposed to address this deficiency and improve the SST k-omega closure. The correction is validated on a number of bump flows, showing significant improvements relative to the baseline model in strong FPG regions.