Methods

The scheme follows a broad generate-and-test method. Radar reflectivity images are searched in the vicinity of radar-identified circulations for local minima. Each region that forms a local minimum is then tested using fuzzy rules and classified using the fuzzy classifier.

The BWER detection scheme was developed using three data sets, each of which contained at least one BWER. The Lubbock case contained several BWERs, inflow notches and marginal structures, thus necessitating a discrimination capability. Two test cases were chosen - one containing several BWERs and inflow notches (thus necessitating differentiation) and the other a storm sequence with no BWERs (thus necessitating a low false alarm rate). A meteorologist identified all the BWERs in the data, marking them either as definite or as marginal. The output of the scheme was compared with these ``truth'' values. From the total hits, misses and false alarms, the CSI and $CSI_{mod}$ were computed.