Tornado identification using a neuro-fuzzy approach to integrate shear and spectral signatures

Traditional tornado identification is primarily dependent on reflectivity and/or shear signatures. The detection of potential tornadic storms by identifying hook echoes was documented by Stout and Huff (1953), and was realized as a tornado precursory signature after the Illinois tornado (Fujita 1958). The NSSL Tornado Detection Algorithm (NSSL TDA) (Mitchell et al. 1998) searches for strong and localized azimuthal shears. However, if a tornado is located at far ranges or if the tornado is small compared to the radar resolution volume, the shear signature becomes difficult to identify (Brown and Lemon 1976). Recently, a half-degree angular sampling was proposed to improve the shear signatures (Brown et al. 2002). However, the statistical error of the spectral moment estimates increases due to that fewer samples are used. Because tornado debris consists of an assemble of particles of various sizes and irregular shapes, some of them have distinct polarimetric signatures that are different from hydrometers, and can be used to improve tornado detection (Ryzhkov et al. 2005). The Doppler spectra from a tornadic region are different from Gaussian-like spectra from by other regions of the storm. Broad and flat tornado spectral signatures were found in simulated data, and bimodal spectra were observed by pulse Doppler radar (Zrnić and Doviak 1975). These distinct spectral signatures can be used to improve the detection capability. In this work, a neuro-fuzzy method is developed based on spectral analysis, neural network and fuzzy logic, in which tornado signatures in both velocity and spectrum domains are integrated to improve the tornado detection.