A novel approach of tornado detection using a machine intelligence system based on shear and spectral signatures

Current tornado detection algorithm (TDA) searches strong and localized azimuthal shears in the field of mean Doppler velocities. Recent studies have shown that wide and flat Doppler spectra can often be observed in the region of tornadoes using the research Weather Surveillance Radar- 1998 Doppler (WSR-88D). Unlike shear signatures, tornado spectral signatures (TSS) are not sensitive to the radar smoothing effect which is introduced by the increasing size of radar resolution volume with increasing range. Therefore, TSS has the potential to facilitate and possibly to improve tornado detection, especially at far ranges. In this work, three parameters are developed to characterize TSS and the dependence of these three parameters on various factors such as the tornado size, reflectivity structure, and range are studied. Moreover, a fuzzy logic system is developed to integrate multiple parameters including both spectral and shear signatures to subjectively detect the existence of tornadoes. This system is further refined by the addition of a neural network to provide optimal decision criteria through a self-learning process. This approach is termed Neuro-Fuzzy Tornado Detection Algorithm (NFTDA). The performance of NFTDA is studied statistically for various conditions using simulations. In addition, the NFTDA is demonstrated for two tornado cases with data collected by the research WSR-88D (KOUN) in Norman in 2003. The detection results from NFTDA are compared with those from conventional shear-based detection algorithms. Preliminary results have shown that NFTDA can provide accurate tornado detection at further ranges while conventional TDA has limited performance.