This paper presents an improved Bayesian uncertainty quantification framework for calibrating and comparing wind farm flow models. It quantifies the parameter uncertainty in a joint posterior distribution as well as the model uncertainty. Applied to a large-eddy simulation dataset for wind farm blockage, it compares a standard wake model and an atmospheric perturbation model, showing the latter yields lower model uncertainty. The framework is available in the open-source Python package UMBRA (Uncertainty Modeling toolbox for Bayesian data Re-Analysis).