Turbulence in a diffusion flame is modulated by thermal expansion, buoyancy effects and lift-off. Aside from simple shear-generated sources density-velocity correlations, ρ'u''(i)̄ represent additional source terms in the Reynolds-stress equations that distort the shear-generated turbulence anisotropy. A modeled transport equation and several zero-order models of ρ'u''(i)̄ are analyzed and their effect on flame predictions with a Favre averaged second-order moment closure for velocity and scalar transport is investigated. The chemical reaction is described by chemical equilibrium and laminar flamelet modeling. The latter is shown to have limitations in application due to differential diffusion. Two attached, vertical H2/N2- air flames with the same Reynolds numbers but different Froude numbers are investigated numerically and experimentally. The desired data base for an overall comparison is provided by comprehensive 3D-LDV, coherent anti-Stokes Raman spectroscopy and spontaneous Raman spectroscopy measurements. The calculations yield correct results in all measured profiles of velocity, temperature and species concentrations. It is shown that only one zero-order model and the transport equation of ρ'u''(i)̄ are adequate. The neglect of those terms will falsify the prediction of decay rates, fluctuations and flame shapes. The magnitude of errors depends on the local Froude number which decreases downstream. The increase in the influence of buoyancy leads to smaller decay rates of axial velocity and to enhanced scalar mixing. Furthermore, turbulence intensities are reduced, and scalar fluctuations and anisotropy are enlarged. The experimentally observed visible flame length shortening with decreasing, density weighted Froude number is reproduced by the presented model.