Tornado Detection Using a Neuro–Fuzzy System to Integrate Shear and Spectral Signatures

Abstract

Tornado vortices observed from Doppler radars are often associated with strong azimuthal shear and Doppler spectra that are wide and flattened. The current operational tornado detection algorithm (TDA) primarily searches for shear signatures that are larger than the predefined thresholds. In this work, a tornado detection procedure based on a fuzzy logic system is developed to integrate tornadic signatures in both the velocity and spectral domains. A novel feature of the system is that it is further enhanced by a neural network to refine the membership functions through a feedback training process. The hybrid approach herein, termed the neuro?fuzzy tornado detection algorithm (NFTDA), is initially verified using simulations and is subsequently tested on real data. The results demonstrate that NFTDA can detect tornadoes even when the shear signatures are degraded significantly so that they would create difficulties for typical vortex detection schemes. The performance of the NFTDA is assessed with level I time series data collected by the KOUN radar, a research Weather Surveillance Radar-1988 Doppler (WSR-88D) operated by the National Severe Storms Laboratory (NSSL), during two tornado outbreaks in central Oklahoma on 8 and 10 May 2003. In these cases, NFTDA and TDA provide good detections up to a range of 43 km. Moreover, NFTDA extends the detection range out to approximately 55 km, as the results indicate here, to detect a tornado of F0 magnitude on 10 May 2003.