In this study, we would like to evaluate and improve the performance of Wall-Modeled Large-Eddy Simulation (WMLES) on the modeling of a pipe flow for which Direct Numerical Simulation (DNS) data is available [1] and considered as a reference for further comparisons. Models used in WMLES may raise problems of accuracy which come from the uncertain values of model parameters and model simplifications. In this study, we focus firstly on the impact of the model parameter uncertainties on the simulation results, and then on the reduction of these uncertainties via data calibration. These studies using sampling-based approaches can be unaffordable when coupled with a high-fidelity simulation that requires several CPU hours for a single execution. To reduce the computational cost while maintaining a target accuracy, we propose to build surrogate models based on Gaussian Processes for simulations outputs, and replace the simulator for evaluating the large size sampled sets. For this study, a CFD-UQ methodology is developed which couples our internal UQ tool and a CFD solver. It has been applied on a turbulent pipe flow case that allows us to validate its implementation.