This study performs a comprehensive analysis of the effect of flow separation and reattachment, convective conditions and Pr to understand their effect on heat transfer characteristics and the predictive capability of low- and hig-fidelity turbulence models are assessed. To achieve the objective DNS is performed for plane channel flow at Reτ = 640, Pr = 0.71 and 0.025 involving mixed forced and natural convection condition, and RANS, hybrid RANS/LES, and LES calculations are performed for backward backing step with expansion ratio 1.5, Pr = 0.71 and 0.0088 and Ri = 0 and 0.338. Channel flow simulations reveal that the convective conditions affect the near-wall turbulent structures and thermal diffusion more significantly in high-Re flows that in low-Re flows. Thus, the generated DNS database provides a challenging test case for turbulence model validation. For backward facing step case, all the turbulence models predict the overall flow characteristics, and Ri = 0 case is a more challenging validation test case than Ri = 0.338, as the former involves complex turbulent diffusion, whereas the latter is dominated by large scale buoyancy driven convection. Results show that well resolved PANS and LES predictions can help in improve understanding of turbulent diffusion under complex convection and flow separation/ reattachment regimes. RANS results are also quite encouraging and indicates that they may represent a reasonable compromise between computational expense and accuracy for cases in which high resolution simulations are not feasible.