Open Access

Raphael Badel

• This thesis is concerned with the wide field of feature tracking in time-dependent data. Many solutions already exist for the tracking of various features in scalar fields that are given as discrete time steps. In the field of meteorology, recently published precipitation data of the COSMO-REA2 reanalysis system gave rise to the analysis of precipitation at a convective scale for which a tracking of precipitation cells over time is desired. A previous study indicated that a tracking based on the widely used overlap criterion does not perform well for the reanalysis data. Based on a novel set of requirements, a new approach to the tracking of regions in time-dependent scalar fields is developed in the course of this thesis and applied in a prototypical study to example datasets of the COSMO-REA2 system. Despite the concrete motivation of using the developed method for subsequent studies of precipitation, the tracking approach is designed to be applicable for arbitrary scalar quantities that can conceptually be given on uniform grids of arbitrary dimensions. Based on a detailed description of the utilized methods, this thesis presents a novel tracking solution whose correspondence identification is based on image registration of successive time steps in combination with a hierarchical watershed segmentation by means of which features are extracted. The proposed implementation allows for an efficient generation of tracking results under the premise that the registration-based correspondence information has been precomputed.