Discovering spatially contiguous clusters in multivariate geostatistical data through spectral clustering

Abstract

Spectral clustering has recently become one of the most popular modern clustering algorithms for traditional data. However, the application of this clustering method on geostatistical data produces spatially scattered clusters, which is undesirable for many geoscience applications. In this work, we develop a spectral clustering method aimed to discover spatially contiguous and meaningful clusters in multivariate geostatistical data, in which spatial dependence plays an important role. The proposed spectral clustering method relies on a similarity measure built from a non-parametric kernel estimator of the multivariate spatial dependence structure of the data, emphasizing the spatial correlation among data locations. The capability of the proposed spectral clustering method to provide spatially contiguous and meaningful clusters is illustrated using the European Geological Surveys Geochemical database.