Abstract: Spatial processes exhibit nonstationarity in many climate and environmental applications. Convolution-based approaches are often used to construct nonstationary covariance functions in Gaussian processes. Although convolution-based models are flexible, their computation is extremely expensive when the data set is large. Most existing methods rely on fitting an anisotropic, but stationary model locally, and then reconstructing the spatially varying parameters. In this study, we propose a new estimation procedure to approximate a class of nonstationary Matérn covariance functions by local-polynomial fitting the covariance parameters. The proposed method allows for efficient estimation of a richer class of nonstationary covariance functions, with the local stationary model as a special case. We also develop an approach for a fast high-resolution simulation with nonstationary features on a small scale and apply it to precipitation data in climate model outputs.

Key words and phrases: Climate model runs, conditional simulation, large datasets, local likelihood estimation, nonstationary Matérn covariance function, polynomial approximation.