The development of a storm type climatology using an automated storm classification system

This study investigates the use of a storm classification algorithm to produce a preliminary climatology of storm types throughout the United States. Automatic algorithms could conceivably use probabilities derived from environmental data, radar data, and storm type to nowcast hazards associated with storms. Such products could enhance efficiency of real-time forecasting operations during a busy shift with minimal staffing, such as during a tornado outbreak, when the forecaster cannot supervise all significant attributes of a growing storm.

The automated approach to classifying convective storms is based upon both environmental and radar data by means of a K-Means clustering and watershed segmentation algorithm. Radar and environmental data, which were extracted from the generated clusters, provided the data for the development and training of decision tree models to predict storm types. These decision trees generated a preliminary climatology of storm types for the CONUS. Both seasonal and regional characteristics were investigated for dates between February 2009 through December 2009.