Tracking the Evolution of Water Flow Patterns Based on Spatio-Temporal Particle Flow Clusters

Marine scientists investigate the movement of oceanic water particles with floating measurement devices released in the real ocean, as well as with virtual particles released in numerical model simulations. The detection, visualization, and evolution of clustered particles is key for gaining a comprehensive understanding of the underlying processes in the oceans. Thereby, vast amounts of mobility data (3D coordinates of these particles over time) need to be analyzed using mobility data science methods. In this paper, we describe the application of data science techniques to detect particle clusters and, more importantly, to track the evolution of these clusters over time in order to support the analysis of oceanic flows. In particular, we apply a well-known concept for tracking the cluster evolution from the data mining community that relies on pair-counting and, thus, is rather inefficient. In order to be applicable to large amounts of particles, we further elaborate two heuristic solutions to compute the cluster transitions based on spatial approximations. Experiments on real world data show a considerable speed-up while sacrificing marginal accuracy drops. Our prototype is used by domain experts for the analysis of the large-scale ocean by virtual particle release experiments in ocean simulations.