8.3
An Improved Storm Cell Identification and Tracking (SCIT) Algorithm based on DBSCAN Clustering and JPDA Tracking Methods
Presentation PDF (716.2 kB)
The new cell identification method utilizes a density-based unsupervised clustering algorithm which requires no a priori knowledge of the number of existing cells and is not sensitive to reflectivity “dropouts”. Furthermore, storm cells are identified and stored according to the entire area of the storm cell, which is contrary to the current method of maintaining just a centroid point. This information becomes very useful in applications that require association of storm cells with other meteorological phenomena such as tornadoes and lightning.
In addition to improved storm cell identification, a superior tracking and association algorithm is presented. As previously mentioned, storm cell areas are determined and tracked rather than centroid locations. A scheme of joint probabilistic data association (JPDA) problems are formed to associate storm cells. A traditional combinatorial optimization algorithm, the Hungarian Method, is performed on a particle representation of storm cells. This, in turn, produces a cost matrix which reflects the overall probability of assignment between two storms. Lastly, two iterations of a modified Hungarian Algorithm, capable of making assignments that reflect splitting and merging cells, produce the final storm cell associations. Overall, storm cells are identified and tracked with a much higher degree of fidelity than the currently implemented SCIT algorithm.