Numerical error or model error?

If a CFD result deviates from measurement there are two candidates: you may be solving the numbers wrongly, or you may be solving the wrong equations. They are different things and call for different responses.

How to tell them apart: change the mesh, keep the model

Numerical error decreases systematically as the mesh is refined and converges to zero. Model error does not — it stays the same however fine the mesh becomes, because its source is not the discretisation but the closure assumption.

The practical order:

1. Demonstrate grid independence (and time-step independence if transient). The remaining difference is then not numerical.
2. At that point change the model — turbulence model, wall treatment, radiation model. If the result changes, that difference is the model uncertainty.

Reversing this order is the most common mistake: changing turbulence models without having shown grid independence and declaring "this model agreed better" conflates model error with mesh error. See the verification and validation note.

The dangerous scenario: errors masking each other

If mesh error pushes the result up and model error pulls it down, the total lands very close to the measurement. The study looks successful, and then the agreement falls apart when the mesh is refined — because it was a coincidence.

For this reason "agrees with measurement to within 2%" is not evidence of quality unless grid independence has been shown.

A third candidate: input error

The two candidates are really three. The source of the deviation may also be a boundary condition, a material property or a geometry difference — and in industrial problems that third item is often the largest. Before blaming the model, verify that the inputs match the measured condition. See the boundary condition note and the significant digits note.

References

1. Celik, I. B., Ghia, U., Roache, P. J., Freitas, C. J., Coleman, H. & Raad, P. E. (2008). Procedure for estimation and reporting of uncertainty due to discretization in CFD applications. Journal of Fluids Engineering, 130(7), 078001. DOI: 10.1115/1.2960953
2. ASME (2009, R2021). V&V 20 — Standard for Verification and Validation in Computational Fluid Dynamics and Heat Transfer. American Society of Mechanical Engineers, New York.
3. Roache, P. J. (1994). Perspective: A method for uniform reporting of grid refinement studies. Journal of Fluids Engineering, 116(3), 405–413. DOI: 10.1115/1.2910291