Volumetric Storm Cell Classification with the Use of Rough Set Methods

A radar data processing system gathers meteorological volumetric radar data by conducting a volume scan. Meteorologists use these radar data to detect thunderstorms. Radar subsystem exists to allow operational meteorologists to focus their attention on the regions of interest within the volumetric radar scan known as storm cells. When a storm is found, a number of parameters are computed. There are 22 derived features and 1 decision used in the analyses such as: height offset, extent, core volume, core height, supercell severity, wind gust severity, hail occurrence, core tilt angle, supercell flag, joint count, split count, core tilt vector, velocity set flag, velocity, core size, orientation, cell type – as a decision. But it is difficult to classify detected storm cells into a specific type of storm event due to a number of confounding factors such as incomplete data, complex evolution of storm cells and high dimensionality of the data. Our objective is to identify patterns in the data that indicate, with a high degree of accuracy, the onset of a severe weather event using either the derived features of matched–cell files from the Radar Decision Support System (RDSS) database of Environment Canada [9], or the raw data of the volume scans. The classification of storm cells is a difficult problem. In this paper, the cross-validation method based on the rough set approach is used to classify storm events. We assume that the reader is acquainted with the basic notions of rough set theory [3] so all fundamental definitions are passed over. The analysis results obtained for volumetric storm cell data by using the cross-validation method are promising. In the next paper, this method will be compared with other methods, with respect to the accuracy coefficient in the classification over testing data.