Classification of Non-meteorological Echoes from Dual Polarization Weather Radar Observations using Deep Learning Models

Dual polarization Doppler weather radar is a powerful tool for observing the wide spatial variability of precipitation instantly. Polarimetric signals scattered by scatterers of various types have different characteristics. These different characteristics of dual polarization observations have been widely studied to understand meteorological echoes [1]. Identifying non-meteorological echoes, coexisting with meteorological ones is crucial for quality control of radar observations. Quality control of radar observations is a challenging task and requires advanced knowledge of weather radar signal characteristics. This paper aims to enhance the understanding of the radar observation characteristics of non-meteorological echoes such as ground clutter and biological particles and presents a method for classifying non-meteorological echoes at each range sample using deep learning techniques. Dual polarization products such as reflectivity, differential reflectivity, correlation coefficient, and texture information of differential propagation phase shift are used to classify as meteorological/non-meteorological echoes. Differential reflectivity provides the shape of the echoes and correlation coefficient indicates the similarity between horizontal and vertical scattering of the echoes [1]. Observations from the Colorado State University CHILL (CSU-CHILL) S-band dual polarization radar were used to train the deep learning models. This paper analyzes classifications using deep learning models, including logistic regression, random forest, and neural networks. The accuracy of deep learning based classification was evaluated, and the preliminary results show high concordance with the classification derived from various dual polarimetric products analysis available in the literature.

References

[1] Chandrasekar V, Beauchamp RM, Bechini R. Introduction to Dual Polarization Weather Radar: Fundamentals, Applications, and Networks. Cambridge University Press; 2023