Emulating functional outputs from high-fidelity simulators often involves constructing a metamodel combining dimension reduction techniques, which describe high-dimensional data by a significantly smaller number of variables, and regression on these variables. In scenarios with limited high-fidelity data, simulators of lower fidelities that are computationally cheaper to run can be used to improve the prediction accuracy, resulting in a multi-fidelity metamodel. Such metamodels are essential tools for performing uncertainty quantification (UQ), which requires the execution of numerous simulations. Most of the multi-fidelity metamodels with functional outputs proposed in the literature [1] lack an estimator of the prediction error, which could be instrumental in an active learning process. Building on an existing multi-fidelity metamodel [1], a model for estimating the prediction error is developed in this work. This model accounts for errors originating from both dimension reduction and regression. For dimension reduction, we explore both a linear approach, specifically a cross-validation-based proper orthogonal decomposition with an inherent error model [2], and a nonlinear manifold learning technique. Regression is conducted using Gaussian-process interpolation, which also comes with an error model. The dimension reduction and regression errors are then combined to provide a prediction error on the functional output of the metamodel. We employ this novel error model in a UQ framework to assess the prediction error of a quantile field across various analytical test cases of escalating complexity, including an aerospace application. Additionally, this model is benchmarked against another prediction error model from the literature [2]. [1] B. Malouin, J.-Y. Tr´epanier and M. Gariepy, Interpolation of transonic flows using a proper orthogonal decomposition method. Int. J. Aerosp. Eng., Vol. 2013, 2013. [2] B. Kerleguer, Multifidelity surrogate modeling for time-series outputs. SIAM/ASA J. Uncertain., Vol. 11, pp. 514–539, 2023. This work is co-funded by ONERA and the Agence de l’innovation de défense (AID).