Development and Assessment of Nuero-fuzzy gust front detection algorithm

The strong wind and shear associated with gust front can be hazardous and endanger air terminal operations especially during landing and takeoffs. In the past, the gust front detection algorithm with radars searches for thin-line features in Doppler measurements. Recently, the national network of S-band Weather Surveillance Radar - 1988 Doppler (WSR-88D) has been upgraded to be dual-polarization and the added polarimetric variables have the potential to improve gust front detection. In this work, a novel Neuro-fuzzy gust front detection algorithm (NFGFDA) was developed to detect gust front, which capitalizes on both thin-line feature and polarimetric signatures.) The inputs to the detection include reflectivity, differential reflectivity, co-polar correlation coefficient, standard deviation of radial velocities and differential phases, and the thin-line feature from both reflectivity factor and the motion of reflectivity. NFGFDA is optimized by a training procedure with neural network.

In this work, the characteristics of gust fronts were first analyzed and discussed. Moreover, the performance of NFGFDA was assessed using nine cases observed by the polarimetric WSR-88D, in which six cases contain gust front and three null (non-gust front) cases contain a tornado outbreak, a squall line and a clear air. The total number of volume scans is 102. Preliminary results have shown that gust fronts with different orientations and intensity can be detected reliably using NFGFDA. Specifically, the probability of detection and the probably of false alarm resulted from these cases are 100 % and 0 %, respectively. Moreover, two other performance measures that were used in the literatures were also implemented. It has shown that the probably of length detected is 82 % and the probably of false detection length is 1 %.