Introduction:
Current vertical association techniques in the existing NSSL algorithms employ a centroid-relative method to associate vertically adjacent 2D features (vortices, storm components, etc.). In short, the techniques used for vertical association use search radii to identify candidate 2D features with which to vertically associate. It is apparent from realtime and offline testing that the vertical association could be improved upon and that such improvement is expected to result in more accurate trends of storm and vortex diagnostic parameters. Observations suggest that given different environmental conditions, the identification of storms and vortices could vary. Also, the growth and decay of storms in a quiescent environment poses a significant challenge to vertically associate the correct 2D components comprising an actual storm. Both storm and vortex detection algorithms suffer from the problem of choosing the most appropriate 2D feature if there is more than one within a given search radius. Thus, it is proposed to examine the Level II data for a few select data cases representing unique storm types in order to improve the vertical association techniques currently employed in the NSSL suite of severe weather detection/prediction algorithms. The proposed storm types that are to be examined include but are not limited to high shear/fast moving (supercell), squall line, Southwest US monsoon and weak shear/pulse storms (uni/multi-cellular).
Methodology:
The initial phase of this study will include a failure analysis to determine how often and the cause for vertical association failures. It is proposed that 10 volume scans from each case example be examined for vertical association failures. Failures will be determined by human determination of the appropriate 2D features for vertical association. A log will be maintained to determine the cause of these failures which is expected to aid in determining a more appropriate scheme for constructing 3D storms and vortices. It is proposed to begin examination of the vertical association of vortices in both the MDA and TDA. The Radar and Algorithm Display System displays 2D vortices from the Mesocyclone Detection Algorithm (MDA) which facilitates much easier examination 2D vortices and their vertical association. It is possible that some creative source coding could be done in order for the 2D vortices identified by the TDA to be written to the appropriate MDA 2D vortex array for display in RADS. Otherwise, examination of the vertical association in the TDA will be accomplished through the more cumbersome method of examining the intermediate 2D text output in the fort.42 output file. Also, the intermediate 2D text output from the MDA will be employed if needed. The same methodology will be applied to SCIT as well.
Ideas gleaned from other projects (e.g. algorithm recode project) will be kept in mind to determine whether they may improve the vertical association scheme. These ideas include correcting for storm motion and the use of an areal overlap of 2D components/vortices in lieu of centroid-based schemes. In addition, there may be techniques from the scale-separation being employed in SCIT that may be useful.
The following cases have been identified for use in this investigation:
Deliverables:
Design review by 30 September 1999.
Final report describing results due to the OSF by 1 December 1999.
Issues/Concerns:
Since the study lacks many of the display tools that would make for a more expeditious analysis, much of the time could be spent examining the intermediate text output from the algorithms to determine the best vertical association of 2D components/vortices. It is possible that the analysis could consume all the time allotted for this task. The first case that is examined will help to determine whether 10 volume scans for analysis is too much in order to complete an analysis of all cases listed.
Steering Committee: Chris Porter, DeWayne Mitchell, Arthur Witt, Greg Stumpf, John Krause, Mike Eilts, Bob Lee, Mike Fresch