Temporal, spatial and spatiotemporal data are often nonstationary. This project develops computationally efficient simulation-based methods for Bayesian learning and prediction in locally stationary processes using spectral methods.