Beginnings of two algorithms

A quick glimpse into my work week:

(1) A bunch of us here at the lab are talking to several groups of researchers and forecasters in the Carolinas. They are interested in an automated algorithm to identify and track sea breezes because sea breezes often lead to late-afternoon thunderstorms. So, I pulled together a dataset to see if an automated algorithm would be feasible ... Here's my brief (Powerpoint)

In the picture, the wing-like structures parallel to the coast are the seabreeze extracted from radar reflectivity images. Unfortunately, the technique also picks up the strong gradients at the land-sea boundary right at the coast line and a storm cell north-east of the radar -- these will have to be removed in the final algorithm.



(2) An undergraduate student is working with me this summer. We are trying to evaluate whether some types of tornadic storms are easier or more difficult to forecast than others. In order to do that, we need to automatically determine storm type. I wrote the clustering algorithm and trained the decision tree to do that. Here's one of Eric's talks (also Powerpoint) on the subject.

In the picture, "S" is a supercell and "L" is a convective line. The other two categories we used were "P" for pulse storms and "n" for non-organized.





Both of these ideas are of course in the first stages of a long process before they become real-time routine algorithms (if they ever do).