Some basic questions: 0. How are your residuals dropping? 1. Are you monitoring things at specific points to confirm that your solution has reached a steady profile? (60s, LES without precursor RANS , could be short but IDK depends on dynamics of the system) 2. I personally would be hesitant to use highly anisotropic mesh with LES as it would potentially distort resolved eddies (but turbulence modeling experts might comment better on this) 3. What are your schemes? Have you considered doing a precursor RANS simulation with grid-convergence study? That should give you a decent idea of large-scale vortex strutures 4. Is that an average profile you are plotting or a single instantaneous screenshot? LES solution is bound to move around and the PIV of De Silva et al is bound to be some kind of time-average. Pay attention to that and try to recreate a solution with similar 'statistical' averaging (I only skimmed the paper briefly but unfortunately I didnt spot this information. Still, you would need some kind of averaging to report meaningful comparison with experiments: Ideally over a few cycles of the solution while your LES instantaneous solution moves around but becomes steady on average. See also, #3 above for an long-term approximation... kOmegaSST may be a good idea here since you've already spent time optimizing y+)