Abstract
A statistical pattern recognition technique called IREW is described. IREW has several strengths, such as fast execution, small storage requirements, and increments learning, that may make the technique useful for many meteorological pattern recognition problems. A weakness of IREW is that it may not recognize complex patterns as well as more sophisticated pattern recognition techniques do.
IREW analyzes data by using a derivative of Bayes' rule to multiplicatively combine the influence of predictors. The weight assigned to each predictor is determined empirically from a training dataset containing the data and verifications for multiple cases. IREW uses several methods to select useful subsets of a large set of predictors.
The objective identification of surface fronts in Nested Grid Model forecasts is described to illustrate how IREW can be applied to a typical pattern recognition problem. The work consisted of identifying factors related to fronts and using some of those factors to make analyses. Given 27 000 predictors, IREW selected many that meteorologists associate with fronts. IREW's analyses were compared to subjective analyses for seven test cases. In this limited test, IREW performed similarly to meteorologists in terms of the number of grid points correctly classified as frontal or non-frontal.
