A Comparative Evaluation of the Effects of Two and Three Dimensional Clustering on Storm Cell Identification Performance

The ability to properly identify storm cells is a significant challenge with applications in the study of storm cell development. While a number of clustering algorithms have been proposed for the identification of storm cells, one of the key characteristics that differentiate these algorithms is the dimensionality of the space in which the clustering is performed. Many of the current methods of identification perform clustering in one dimension at a time. However, among the most recently proposed methods of identification are two different algorithms using density-based spatial clustering with applications in noise (DBSCAN) that perform clustering in higher dimensional spaces. The first of these two algorithms performs clustering in a two-dimensional space (of azimuth and range), at each elevation slice of a radar volume scan, and then vertically associates each two-dimensional cluster to define a storm cell. The other algorithm performs clustering in a three-dimensional space, with dimensions in azimuth angle, range, and elevation angle. This paper presents a comparative evaluation of these two identification methods and reports the resulting conclusions on how two versus three dimensional clustering effects storm cell identification performance.