On Oct. 4th, 1999, a meeting was held to discuss the approach for the NSSL/OSF MOU Task 5.2. In attendance were, DeWayne Mitchell, Chris Porter, Mike Eilts, Greg Stumpf, Arthur Witt, John Krause, Bob Lee, and Mark Fresch. Here's the task description.

"Improve vertical association. The NSSL will investigate ways to improve the 3D vertical association routines of its algorithms. It is recommended that the NSSL analyze failures of the current algorithms as a basis for trials to improve the algorithms. Evaluation should be against human truth using algorithm-detected 2D features. Corrections should be made for high-shear environments and storm-motion. The investigation should include cases with the following spectrum of storm types: high shear/fast moving storms, squall-line, Southwestern U.S. monsoon, and weak shear /pulse storms. It is anticipated that improvements to vertical association will lead to performance improvements in algorithm tracking and trends, and inter-relationships between other "down stream" algorithms (e.g., SCIT improvements will lead to HDA improvements)."

To start the meeting, DeWayne Mitchell identified the slight differences between vertical association methods used by MDA, TDA, and SCIT and identified level II cases of the recommended storm types. The group agreed studying every storm and circulation in all the cases could take too long for the 4 month task and discussed ways to increase productivity. The group agreed with John Krause's suggestions to use 1) candidate tables to track which 2D features are being selected and rejected in the vertical association and 2) a display tool, developed by Open Systems folks, to graphically depict 2D features. The group suggested a failure analysis be done and to take advantage of prior failure analysis by Pam MacKeen, Travis Smith, and Arthur Witt. Some concern was voiced that improvements may be found to fix the failure which hurt overall vertical association performance. Mike Eilts said they expect to have an improved vertical association technique coded by the end of the task.

Mike Eilts suggested NSSL develop a completely new method of vertical association, perhaps using a neural network. Although the group agreed the idea had merit, the some felt developing and testing a completely new method is beyond the scope of the task and would require more time than allotted with no guarantee of improvement.

At meeting's end, the group agreed on the following initial approach:

1) Investigate and categorize failure analysis
2) Put together an list of ideas for improvement
3) Meet again in 1/2 - 1 month