Influence of method selection on clustering analyses of point-like features: Examples from three zones of distributed volcanism

Abstract

Most regions that have experienced spatially distributed volcanism also present a clustered structure of eruptive vents. Attempts to identify patterns from the spatial distribution of vents have resorted to various methods. The influence that specific methods might have on the identified patterns, however, has remained largely unexplored. In this work I assess the ability of common “out-of-the-box” algorithms to yield consistent results. The geological significance of those groupings is also tested. The results of those tests indicate that method selection exerts a strong influence, and therefore introduces undesired biases on the identification of groups on zones of distributed volcanism. The results reported below indicate that it is almost impossible to capture the fine structure of volcanic clusters by focusing on a single value that “best” describes spatial distribution. Nevertheless, an alternative approach that could reduce the negative influence of such method-dependence is outlined. This method searches for core-groups as the membership identified on each group by more than one algorithm. Despite the limited ability of this method to reproduce pre-existing geologic or tectonic partitions, it provides grounds for a more robust interpretation of results than the use of only one out-of-the box method in isolation.