Comparing techniques and reanalysis datasets when diagnosing the relative severity of convective outbreaks

Several recent studies have investigated the accuracy and skill with which fields of severe weather parameters can diagnose the type of severe weather outbreak (e.g., tornado and primarily nontornadic outbreaks) and relative severity of any type of convective outbreak. A number of techniques have been introduced to ascertain the utility of these fields. This study compares two methods. The first method, the principal component analysis technique, requires converting a data matrix composed of gridded meteorological fields centered at the median latitude and longitude of the severe weather reports for a 12-h period surrounding the valid time of the outbreak into a principal component scores matrix. This matrix is then used as input for several statistical algorithms to determine the ability of training models to diagnose the severity of outbreaks on an independent testing set.

The second method, the areal coverage technique, determines the number of grid points in a domain (fixed for each outbreak) that exceed a predetermined threshold for several severe weather parameters. Statistical algorithms use these values as input for training, and again are tested on independent cases. Additionally, the two techniques were implemented using two reanalysis datasets, the NCAR/NCEP reanalysis and the North American Regional Reanalysis datasets, to investigate the differences in performance using substantially different grid spacings. Results suggest skillful and comparable ability in diagnosing the severity of outbreaks with the two techniques and the two reanalysis datasets. Implications of this work will be discussed.