This work is devoted to quantify the predictive uncertainty in RANS simulation of a non-premixed lifted flame due to uncertainty in the model parameters of the scalar dissipation rate transport equation. The uncertainty propagation and the global sensitivity analysis of the effect of such parameters on the quantities of interest (QoIs) is performed employing Polynomial Chaos Expansions as surrogate models of the uncertain response. This approach is applied on a lifted methane-air jet flame in vitiated coflow, already experimentally investigated by Cabra et al [1]. The results show the effectiveness of the approach to provide predictions with estimates of uncertainty. It is shown that the the uncertainty in the mixture fraction and temperature is negligible as long as only pure mixing happens, then it becomes significant in the regions where ignition begins, starting from z/D=30. Of the four parameters considered, i.e., CD1, CD2, CP1 and CP2, main and total effect sensitivity indices show that the largest contribution to the uncertainty in the flame temperature is given by the two dissipation parameters CD1 and CD2, while the production parameter CP2 has almost negligible impact on the predictions. Lastly, the surrogate models are used to determine an optimum set of parameters that minimizes the distance with the experimental measures, leading to improved predictions of the QoIs.