Evaluating a Storm Tracking Algorithm

Storm tracking algorithms have been evaluated either indirectly based on

fields advected using their motion vectors or by validating each association

decision in a labor-intensive manner. In this paper, we introduce

three bulk statistics that are measures of mismatch error, jumps

and of duration. These statistics allow us to rank commonly proposed

association heuristics (centroid projection, overlap, cost functions, etc.)

and choose a good storm tracking approach.