Operational forecasting of tornado locations: a verification study of "pathcasts"

During the past decade, attempts to forecast the location of tornadoes have become more frequent in Tornado Warnings issued by NOAA/National Weather Service (NWS). These "pathcasts" are created by linear extrapolation of radar mesocyclone signatures by forecasters, in some cases enhanced by real-time spotter reports. Beginning in 2008, the initial time, latitude, longitude and motion vector associated with each pathcast have been included with the text warning. This allows for the first time a quantification of errors of forecast tornado locations associated with linear extrapolation.

All tornadoes rated EF2 or higher in 2008 were examined to determine whether any pathcasts were included in the associated NWS Tornado Warnings or Severe Weather Statements. If pathcasts were found, the extrapolated motions were compared with the corresponding tornado tracks to determine forecast location errors as a function of lead time. Further, the tornado tracks were compared to the corresponding motion estimates derived from environmental data (right-moving supercell motion using the Internal Dynamics method) and from an advanced radar storm motion algorithm (via K-means clustering from the NSSL Warning Decision Support System - Integrated Information). Along with mean errors and standard deviation, these data were examined to determine whether the forecasts have left/right or slow/fast biases.

Examination of these data reinforce previous studies that reveal the challenges of making precise tornado path forecasts and communicating them effectively. However, these results will allow forecasters to improve their operational motion estimates and determine how much to "fan out" their warning polygons downstream. In addition, these results will provide information on how automated motion guidance might be used to enhance the warning-making process. Finally, these results may be used as a baseline against which to evaluate future explicit tornado path forecasts from "warn-on-forecast" numerical models.