Bayesian nearest-neighbor analysis via record value statistics and nonhomogeneous spatial Poisson processes

This article proposes record value statistics and nonhomogeneous spatial Poisson processes as new nearest-neighbor approaches for detecting spatial patterns. A Markov chain Monte Carlo method with data augmentation was developed to compute the Bayes estimates of posterior quantities of interest. Simulation studies showed that the new approaches yield high detection rates and low false positive rates. We applied the new approaches to detect localized clusters of specific trees and to outline seismic faults in the study space.