Development of An Automated Approach For Identifying Convective Storm Types Using Reflectivity and Near-storm Environment Data

Abstract

The identification and classification of precipitation systems has been ongoing since the advent of using radar for meteorological applications. Recently such systems have been subjectively classified and compared with model fields (i.e. near-storm modeled soundings) for operational forecasting tools. Other work has attempted automating detection of stratiform versus convective regions of storms, as well as using simplified object classification for storm mode (i.e. cellular vs. linear). The work herein describes a system that combines radar-derived storm attributes with near-storm environmental model data in attempts to classify radar echoes. The work relies heavily on the Warning Decision Support System—Integrated Information (WDSS-II) developed by NSSL. The system readily ingests and processes Level-II NEXRAD data along with RUC-20 data for easy analysis of the storm environment. For a given case, individual storm cells within a radar domain are identified using cell identification techniques found in a Procrustes shape analysis verification tool developed at the University of Missouri-Columbia. For each identified cell, radar-derived attributes (i.e. intensity, aspect ratio, VIL, etc.) are overlaid with selected near-storm environmental variables calculated from a RUC-20 initialization. The data for each cell from multiple cases are fed into a classification and regression tree algorithm and fitted to eight different convective classification regimes. The classification tree is then trained using data from various dates and geographic locations throughout the United States. Output from a test data set is then used and success is judged based on an expert classification of storm types for the new data array.