A hierarchical Bayesian mixture approach for modeling reflectivity fields with application to Nowcasting

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] We study a hierarchical Bayesian framework for finite mixtures of distributions. We first consider a Dirichlet mixture of normal components and utilize it to model spatial fields that arise as pixelated images of intensities. We demonstrate our models using results from simulated data as well as using "real-world" weather radar reflectivity fields. We propose model adequacy and verification tests to further illustrate the effectiveness of the model. We then consider and define spatio-temporal processes using a hierarchical Bayesian mixture model to help us predict the evolution of these processes based on several radar reflectivity fields observed over a short-term time period. We illustrate the methodology with simulated data and apply verification methods to demonstrate the ability of the methods to model such data. We implement these models in nowcasting the evolution of storm systems observed around the area of Kansas City, Missouri, on June 7, 2007.