Sßnchez, Jose Luis (Universidad de Leon)
Fraile, Roberto (same affiliation)
Fuente, Maria Teresa de la (same affiliation)
Marcos, Jose Luis (same affiliation)
Since 1986 the Laboratory for Atmospheric Physics at the University of Leon has been developing a Project to analyze the characteristics of the severe storms which affect the province of Leon, utilizing a wide variety of meteorological data: METEOSAT images, meteorological radar, tropospheric sondes, meteorological observation networks at ground level, and so forth. The study zone has an area of close to 10,000 km2, and by means of these various information systems, detailed information about storm activity in the summer months can be obtained.
One of the aims of the Project is to evolve techniques for short range forecasting of severe storms, and some forecasting models have been developed to try to achieve this.
In the present study, reference will be made to the results achieved by applying the mathematical model known as discriminant analysis. This is a technique which classifies individuals or objects into mutually exclusive groups or classes depending on the values of a set of variables measured upon the objects, and which thus minimizes the probability of an erroneous discrimination.
For the purposes of this study, the preliminary objective was to classify the meteorological situations into two groups: Risk/No Risk of Storms, and Risk/No Risk of Hail. To do this, values for the following variables were used: temperature, dewpoint temperature at the time of launching a radiosonde, and atmospheric pressure at ground level; minimum temperatures recorded the previous night, convective temperature, altitude of the convective condensation level, altitude of the 0░C level, altitude of the 0░C dewpoint temperature level, altitude of the 500 hPa level, altitude of the 300 hPa level and, finally, five indices of atmospheric stability (Showalter, Galway, Total Totals, K and H).
The application of discriminant analysis to the values provided by the fifteen variables permitted, in a single operation, the risk of a thunderstorm (or hailstorm) to be forecast for a determined summer day.
In order to define the discriminant function, geometric criteria were used. Once the four subpopulations (namely, Storm Days/No Storm Days, Hail Days/No hail Days) had been characterized by means of each variable, a determined day would be allocated to the subpopulation it most closely resembles. To differentiate populations by means of random variables, it was necessary to define a distance. The covariances matrix of the fifteen variables measured for the subpopulations Storm Days/No Storm Days, was shown through analysis not to be the same, so the use of linear discriminators could be ruled out. The same result has been shown for the subpopulations Hail Days/No Hail Days. Instead, the maximum likelihood criterion was applied to define the discriminator as a quadratic function of the fifteen established variables.
The above model was applied to a sample of 78 Storm Days, 151 No storm Days, 43 Hail Days and 186 No Hail Days. The results demonstrated for the case of Storm/No Storm days a value for the True Skill Score of 0.7513 and 0.8423 for the Hail/No Hail case, which made it clear that the discrimination is more satisfactory when the meteorological conditions favor the formation of severe storms.