Assessment of two techniques used to identify ZDR arcs automatically in radar observations

It has been hypothesized that some measurable properties of ZDR arcs in supercells may change in the minutes prior to tornadogenesis and tornadogenesis failure. Automated methods have been developed to objectively identify these features, but thus far there has been no assessment of how accurately they identify these features when compared to a trained human. In this study, we assess two automated methods of identifying ZDR arcs in simulated supercell thunderstorms – the enhanced watershed algorithm (EWA), and a random forest algorithm (RFA) -- against manual identification (“truth”). We hypothesize that (1) the RF algorithm will outperform the EWA at identification of the areal extent of the ZDR arc, and (2) that the EWA will have a higher probability of detecting the ZDR arcs than the RFA. Issues which arose when comparing the objects identified by the different techniques will be discussed.