A Bayesian nonparametric spatial model with covariate-dependent joint weights

This paper presents a spatial process with covariate-dependent random joint distributions. Our construction is based on an extension of the Gaussian copula model using the Beta-regression process. As a generalized form of stick-breaking processes, the proposed model allows the covariance function to be covariate-driven nonstationary. Also, the resulting labeling process provides a covariate-dependent random partitioning. Markov chain Monte Carlo methods are used to make Bayesian inferences. The effectiveness of this model in segmentation of the domain and prediction performance are assessed through a simulation study. Additionally, results from a real dataset demonstrate that the proposed model possesses better spatial prediction performance over other competing models.