312 A 3-Dimensional DBSCAN Storm Tracking and Identification Algorithm – Description and Implementation

Monday, 24 January 2011
Washington State Convention Center
Jenny L. Matthews, Georgia Tech Research Institute, Severe Storms Research Center, Atlanta, GA; and J. Trostel

A newly developed storm cell identification and tracking algorithm is presented. The new cell identification method utilizes a three-dimensional density-based unsupervised clustering algorithm, DBSCAN, which requires no a priori knowledge of the number of existing cells. It is an extension of a previously developed algorithm that employed the DBSCAN algorithm in two dimensions. Similar to other past algorithms, the previously developed algorithm identified two-dimensional clusters in each elevation slice or the radar's volume scan and attempted to vertically associate these clusters to identify storm cells in three dimensions. While this method has the advantage of reduced dimensionality and increased computational efficiency, it is ill-fitted for identifying the complex shapes that the storm cells form. In contrast, the newly developed algorithm identifies storm cells with a single three-dimensional algorithm making it well-suited for the task at hand.

Using this new algorithm, several properties of the identified storm cells are calculated and stored. Furthermore, the exact area over which the storm cells span in addition to the area of the storm cells' cores are stored and used for tracking and spatial association of other meteorological phenomena. In addition to improved storm cell identification, a joint probabilistic data association (JPDA) tracking and association algorithm is applied to the identified storms. An initial comparison of the performance of the improved 3D DBSCAN algorithm to the earlier 2D algorithm will be presented.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner