Spatiotemporal clustering using Gaussian processes embedded in a mixture model

The categorization of multidimensional data into clusters is a common task in statistics. Many applications clustering, including the majority tasks ecology, use that inherently spatial and often also temporal. However, spatiotemporal dependence typically ignored when clustering multivariate data. We present finite mixture model for incorporates autocorrelation by appropriate Gaussian processes (GP) mixing proportions. allow flexible semiparametric on environmental covariates, once again using GPs. propose to Bayesian inference through three tiers approximate methods: Laplace approximation allows efficient analysis large datasets, both partial full Markov chain Monte Carlo (MCMC) approaches improve accuracy at cost increased computational time. Comparison methods shows useful alternative MCMC methods. A decadal 253 species teleost fish from 854 samples collected along biodiverse northwestern continental shelf Australia between 1986 1997 added clarity provided accounting autocorrelation. For these data, temporal comparatively small, which an important finding given changing human pressures over this