Uncertainty quantification using computationally intensive engineering models is often prohibitive, and surrogate models are needed. One possible surrogate for random field quantities of interest (e.g., quantities that are spatially or time dependent) is the Karhunen-Loeve expansion (KLE). In particular, truncated KLEs are sought for efficient low-rank approximations, and they are constructed numerically from simulations of the original model at a number of parameter sampling points. However, when each simulation is high-fidelity and expensive, the number of model runs needed to achieve reasonable KLE quality can still be very expensive. We thus seek a multifidelity KLE approach to take advantage of available lower fidelity models (e.g., with coarser grids or simplifying physics) that are computationally faster. Our approach is to combine the KLEs corresponding to the different models by associating these KLEs with their common uncertain parameters, enabled by the use of polynomial chaos expansions (PCEs) inside the KLEs. We demonstrate our method on benchmark examples and an application of turbulent round jets for mitigating aircraft jet noise.