Neuro-Fuzzy Gust Front Detection Algorithm With S-Band Polarimetric Radar

Abstract

A gust front (GF) is the leading edge of the cold outflow from a thunderstorm. The upgrade of the S-band Weather Surveillance Radar-1988 Doppler (WSR-88D) to dual-polarization has been completed recently in the U.S. Therefore, it is timely to exploit the added benefits of polarimetric variables to identify GFs. In this paper, six signatures derived from polarimetric WSR-88D data are developed to characterize GFs, including medium reflectivity, apparent thin line feature in reflectivity, and the motion of reflectivity quantified by a line feature parameter, high differential reflectivity, low copolar cross-correlation coefficient, apparent convergence manifested by the large radial shear, and large standard deviation of differential phase. These signatures are fuzzy in nature, and therefore, a novel neuro-fuzzy GF detection algorithm (NFGDA) is developed using a fuzzy logic inference system, which is optimized by a training process using a neural network. WSR-88D data from 11 cases (totaling 121 volume scans) are used to evaluate the performance of NFGDA and compared to the operational machine intelligent GF algorithm (MIGFA) with single polarization data. The results show that NFGDA can provide improved performance with a higher probability of detection of 92% (versus 78% with MIGFA), lower false alarm ratio of 0% (versus 9%), and higher percentage correct of 93% (versus 74%). Additional length-based scoring schemes show that NFGDA can correctly detect 62% (41% with MIGFA) of the total length of GFs, and minimize falsely detected length to 7% (61%).