An automated technique to categorize storm type from radar and near-storm environment data

An automated approach to storm classification that relies on identifying storms from observed radar data and classifying them based on their shape, radar, and near-storm environmental parameters is described in this paper. Storms are identified and clustered within CONUS radar and environmental data using a combined watershed segmentation and k-means clustering technique. Storms were manually classified into short-lived convective cells, supercells, ordinary cells, or convective cells at two scales, using data from selected severe weather events between May 2008 and July 2009. Objects of composite reflectivity were identified and tracked using a clustering technique at two spatial scales, and attributes for every storm cluster were extracted based on radar and near-storm environment data from model analysis fields. Quinlan decision trees were trained on these individual attributes and implemented to nowcast storm types for both scales. It is shown in this paper that storms can be automatically identified and classified using a decision tree, and that these automatic classifications have different climatological properties, which are potentially useful for short-term forecasting.