Exploring Cyclone Evolution with Hierarchical Features

Towards Benchmark Data Generation for Feature Tracking in Scalar Fields

“…This corresponds to a set of extrema falling in a level set adding (or respectively subtracting) a local offset threshold from its dominant extremum. This feature definition has been used for cyclone definition [7,20] and also in the medical context of building a patient-specific brain atlas [24]. There are other more application-specific feature definitions using critical points.…”

“…Another approach is to extract features per time step and then relate these features to each other in a second tracking step. To generate the final tracking graphs, often in an additional step, semantic filtering is performed, e.g., by excluding unrealistic large distances in the cyclone tracking work by Nilsson et al [20], or by performing global match optimization [28].…”

“…For tracking in the visualization, this method was used as Combinatorial Feature Flow Fields as introduced by Reininghaus et al [25]. Since then, it has been modified and used in several applications [7,11,20]. For a more detailed summary of topology-based tracking, see the review article by Yan et al [38].…”

“…Feature tracking in scalar fields has become a standard application for topological data analysis [38]. Application examples span over many different domains, e.g., tracking of cyclones [20], vortices [25,26], or burning structures in combustion simulations [1].…”

“…Now, split events can also be captured and we have some notion of strength of connection between minima in consecutive steps, with an indication of "strong" connection between m i of m j if the they are mapped to each other via the forward and backward maps, and "weak" connection if only one of the forward or backward map establishes a connection between the pair of minima. These gradient-based tracking approaches either explicitly compute the integral lines of point features [20], or retrieve that information from precomputed ascending/descending manifolds [7]. Our approach follows the latter, where the manifolds can efficiently be computed via path compression [15].…”