Advanced tornado detection algorithm using super-resolultion and polarimetric data

The neuro-fuzzy tornado detection algorithm (NFTDA) was originally developed to combine shear and spectral signatures observed by weather radar with the goal of providing reliable and early detections. However, some tornado spectral signatures are derived from the level I time series data, which are typically not available from operational weather radar such as WSR-88D. Thus, the NFTDA was modified to include only the Doppler moment of spectrum width and radial velocity and is termed by D-NFTDA. The D-NFTDA was optimized independently for both legacy- and super-resolution data through a training process of neutral network. Compared to legacy-resolution data, the super-resolution data (i.e., 250 m rang resolution and 0.5 degree angular sampling) has the potential to enhance tornadoes' hook and vortex signatures. The performance of D-NFTDA using legacy- and super-resolution was evaluated using 53, and 24 tornado cases, respectively and compared to the results from the tornadic vortex detection algorithm (TDA) in the Warning Decision Support System-Integrated Information (WDSS-II) developed by the National Severe Storms Laboratory (NSSL). Preliminary comparisons have shown that for the legacy resolution the D-NFTDA can provide higher probability of detections (POD) and lower false alarm rate (FAR) than the TDA. Moreover,the D-NFTDA using the super-resolution data shows enhanced detections compared to the D-NFTDA with legacy-resolution.

The negative differential reflectivity and low correlation coefficient have been suggested as the indicators of tornado debris signatures. The upcoming upgrade of the WSR-88D to dual-polarization will allow the measurements of polarimetric data. Therefore, it is of interest to modify the D-NFTDA to include polarimetric variables and it is termed P-NFTDA. The performance of P-NFTDA is investigated using 10 tornado cases collected by a research polarimetric WSR-88D (KOUN) in Norman, Oklahoma during 2003 to 2008. The results have shown that P-NFTDA can effectively suppress the false alarm produced by the D-NFTDA from non-tornadic velocity shear.