Probabilistic Modeling of Tornado Path Length Intensity Variation Using F/EF-Scale Damage Data

The intensity of a tornado typically varies over its life cycle and generally includes a formation stage, a mature stage and a dissipation stage. Tornadoes do not typically exhibit their maximum intensity over their entire recorded path length.  As a result, modeling and analysis of tornado path length intensity variation (PLIV) is an important part of tornado windspeed risk analysis. While radar observations provide direct estimates of tornado windspeeds, sufficient radar data does not exist to probabilistically analyze intensity variation along complete tornado path lengths. Radar data shows that intensity can change rapidly as well as persist along the tornado path.

In support of tornado windspeed hazard risk assessment, a probabilistic model of a tornado path length intensity variation (PLIV) was developed from tornado damage data. The data sources used for PLIV modeling include both non-geo-referenced and geo-referenced damage maps, which include intensity ratings along the complete tornado path.  For the non-geo-referenced map analysis, 184 historically mapped tornadoes were analyzed to compute mean fraction tables of the conditional probability of the local path length intensity, given the maximum rated intensity of the tornado. The average spacing of the intensity ratings in this data set varied from about 1 mile to 4 miles.  Catalogs of F/EF scale ratings in the PLIV sequence that they occur (along with each rating’s appropriate length portion) were developed from the tornado damage maps.  PLIV catalogs were created manually for the184 non-geo-referenced tornado maps, assuming equal segment lengths for each rating.

The geo-referenced PLIV analysis was based on the NWS Damage Assessment Toolkit (DAT) data for the years 2010-2014 (1,268 events).  Comparison of the DAT events with the SPC database resulted in 734 events validated by location and date. Least squares line fits to the DAT data were performed for each tornado to provide the PLIV sequence with respect to positive tornado path direction. The ratings were determined by discretizing the tornado paths length-wise and using the DIs within each discretized segment to determine that segment’s rating. Different discretization lengths (from 0.25 to 4 miles) and methods of determining the segment rating from the DI ratings were investigated. Mean fraction conditional probability tables as well as individual event PLIV catalogues were developed for this dataset as well.

The paper presents the PLIV analysis results and discusses needs for improved data collection to support improved PLIV modeling.  The paper concludes with PLIV data converted to windspeeds and illustrates spline fits of example PLIV catalogues in which the EF scale intensity ratings have been converted to windspeeds.  An advantage of this modeling process is that it separates the analysis of intensity variation along a tornado path length from the estimation of windspeeds given a damage rating.