Application of a Bayesian Classifier of Anomalous Propagation to Single-Polarization Radar Reflectivity Data

Abstract

A naïve Bayes classifier (NBC) was developed to distinguish precipitation echoes from anomalous propagation (anaprop). The NBC is an application of Bayes's theorem, which makes its classification decision based on the class with the maximum a posteriori probability. Several feature fields were input to the Bayes classifier: texture of reflectivity (TDBZ), a measure of the reflectivity fluctuations (SPIN), and vertical profile of reflectivity (VPDBZ). Prior conditional probability distribution functions (PDFs) of the feature fields were constructed from training sets for several meteorological scenarios and for anaprop. A Box-Cox transform was applied to transform these PDFs to approximate Gaussian distributions, which enabled efficient numerical computation as they could be specified completely by their mean and standard deviation. Combinations of the feature fields were tested on the training datasets to evaluate the best combination for discriminating anaprop and precipitation, which was found to be TDBZ and VPDBZ. The NBC was applied to a case of convective rain embedded in anaprop and found to be effective at distinguishing the echoes. Furthermore, despite having been trained with data from a single radar, the NBC was successful at distinguishing precipitation and anaprop from two nearby radars with differing wavelength and beamwidth characteristics. The NBC was extended to implement a strength of classification index that provides a metric to quantify the confidence with which data have been classified as precipitation and, consequently, a method to censor data for assimilation or quantitative precipitation estimation.

Bayesian Echo Classification for Australian Single-Polarization Weather Radar with Application to Assimilation of Radial Velocity Observations

“…Good discrimination relies on these pdfs overlapping as little as possible. The feature fields are also assumed to be independent (the naïve aspect), as useful results have been shown to be obtained with dependent feature fields (Friedman et al 1997;Peter et al 2013). The alternatives are to use a few fields that are known to be independent, or a much more complicated implementation that was not anticipated to yield benefits matching the effort involved.…”

“…The denominator term is constant and can be ignored. The classifier is described in more detail in Peter et al (2013), with the important difference that here P(c) is not assumed equal for all classes. The NBC has been implemented in the BoM's new inhouse radar data handling software (Ancilla).…”

“…The non-Doppler feature fields were explored by Peter et al (2013), from which the present version was developed, though with texture kernels extended to two dimensions.…”

“…Recently, Peter et al (2013) developed a naïve Bayes classifier (NBC) to discriminate anomalous propagation (AP) sea clutter and precipitation, using reflectivity and feature fields based on reflectivity. These feature fields included echo top height, vertical gradients, spin (Steiner and Smith 2002), and texture (Hubbert et al 2009;Kessinger et al 2005).…”

An Objective Prototype-Based Method for Dual-Polarization Radar Clutter Identification

“…Ground clutter detection has used the textures of reflectivity fields [17], and for non-polarimetric radar, vertical profile of reflectivity has also been used to assist clutter detection taking a Bayesian approach [25]. Polarimetric radars offer more opportunities for ground clutter discrimination, e.g., Zrnić et al [26] investigated probability distributions of correlation coefficient ρ hv for ground clutter, and Ryzhkov and Zrnić [27] considered ρ hv < 0.7 and σ φ dp >~10 • -12 • (variation of differential phase) as a robust criterion for rejecting ground clutter using S-band polarimetric radar observations of summer storms.…”