Spatial data, describing surfaces via variables on a 2D plane, is crucial in various domains like ecology and meteorology. Kriging, a prevalent method, models surfaces by interpolating spatial data via a Gaussian process, weighing closer observations more heavily. However, its Gaussianity assumption can be restrictive and hard to verify. Hence, non-parametric and semi-parametric approaches have been explored. This paper proposes a Bayesian nonparametric technique employing Dirichlet processes for inference and prediction in spatio-temporal settings. Dirichlet processes offer flexibility and easy implementation, crucial for their popularity in Bayesian nonparametrics. The method extends finite mixture models to consider infinite mixtures, avoiding the need to select the number of components. The paper discusses the stick-breaking representation of Dirichlet processes, introducing spatially and temporally varying weights to construct evolving clusters for spatio-temporal data.