A new classification approach for detecting severe weather patterns

Abstract

Early detection of possible occurrences of severe convective events would be useful in order to avoid, or at least mitigate, the environmental and socio-economic damages caused by such events. However, the enormous volume of meteorological data currently available makes difficult, if not impossible, its analysis by meteorologists. In addition, severe convective events may occur in very different spatial and temporal scales, precluding their early and accurate prediction. In this work, we propose an innovative approach for the classification of meteorological data based on the frequency of occurrence of the values of different variables provided by a weather forecast model. It is possible to identify patterns that may be associated to severe convective activity. In the considered classification problem, the information attributes are variables outputted by the weather forecast model Eta, while the decision attribute is given by the density of occurrence of cloud-to-ground atmospheric electrical discharges, assumed as correlated to the level of convective activity. Results show good classification performance for some selected mini-regions of Brazil during the summer of 2007. We expect that the screening of the outputs of the meteorological model Eta by the proposed classifier could serve as a support tool for meteorologists in order to identify in advance patterns associated to severe convective events.