Optimization of the matrix of weights in the polarimetric algorithm for classification of radar echoes

Polarimetric radar is uniquely suited for discriminating between different classes of meteorological and nonmeteorological echo. Currently used classification algorithms are based on the principles of fuzzy logic and utilize multiparameter radar measurements with certain weight assigned to each radar variable to account for its classification efficiency. No objective justification has ever been given for the choice of the vector of weights.

In this study, we introduce the matrix of weights which characterizes classification power of each variable with respect to every class of radar echo. A methodology for optimizing such a matrix is also suggested. Instrumental quality of each radar measurement (such as statistical error and bias) is defined by a confidence factor which is used in combination with the matrix of weights. Such a combination provides much greater flexibility to a classification scheme than the existing methodologies.

The data from several storms observed with the polarimetric prototype of the WSR-88D radar in Oklahoma are used to optimize the matrix of weights in the classification algorithm. This algorithm aims at discriminating between 10 classes of radar echo using 6 radar variables.