Automated classification of rainfall systems using statistical characterization

Mike Baldwin

Univ of Oklahoma/CIMMS


An automated procedure for classifying rainfall systems has been developed. Statistically-based attributes have been used to describe various aspects of a rainfall system. Attributes related to a system's intensity were obtained by analyzing the overall distribution of hourly rainfall amounts. Information regarding the degree of linear organization of a rainfall system was obtained via geostatistical measures. To test the usefulness of these attributes, classification results using hierarchical cluster analysis were validated against a subjective classification. Using the best results from the cluster analysis experiments, a completely automated classification system was developed. Image processing techniques were used to identify rainfall systems within hourly NCEP Stage IV analyses. Validation results from a random sample of 2002 data showed that the automated classification procedure correctly placed systems into stratiform, linear, and cellular classes with 85% accuracy.