Tornado detection using a neuro-fuzzy method

The current NWS operational tornado detection algorithm (TDA) searches for strong and localized shear in the velocity field. However, the shear signature becomes difficult to identify when the tornado is located far from the radar. It has been shown that a tornado vortex has a distinct spectral signature which is usually broad, flattened, or bimodal. In this work a neuro-fuzzy algorithm based on both shear and spectral signatures is developed to detect tornado vortices. A neuro-fuzzy system is a fuzzy inference system in which the parameters of the fuzzy sets are automatically adjusted by learning from the past data through a neural network algorithm. The neuro-fuzzy detection algorithm will be presented and discussed. It is further demonstrated using data collected by the KOUN (research WSR-88D) during two tornadic events on May 8th and 10th, 2003. The detection results are also compared with those obtained by the conventional TDA on the KOUN and the Twin Lakes (KTLX) operational WSR-88D, both located approximately 20 miles apart in central Oklahoma.