A Weighted Adaptive Range-Averaging Technique to Improve the Precision Consistency of Polarimetric Variable Fields

Abstract

Abstract The two main metrics for the performance evaluation of radar-variable estimators are the bias and standard deviation (SD) of estimates. Depending on the estimator properties, the bias may increase as the signal-to-noise ratio (SNR) decreases. The standard deviation, however, always rises as the SNR becomes smaller. For instance, if estimates are computed from 16 samples (typically used for WSR-88D surveillance scans) using a rectangular data window and the maximum unambiguous velocity is ∼9 m s−1, the standard deviation of reflectivity estimates increases 1.6 times as the SNR drops from 20 to 2 dB. But for estimates of differential reflectivity, differential phase, and copolar correlation coefficient, SDs increase ∼6.7, ∼6, and ∼54 times, respectively. Hence, this effect impacts the polarimetric variables substantially more than the spectral moments. Additionally, the polarimetric variable SD is also sensitive to the correlation between signals in horizontal and vertical channels leading to reduced data quality in the regions where the correlation coefficient is low. Such increases in the variability of estimates are observable in the fields of dual polarization variables as an increased spatial inhomogeneity (or noisiness) in the areas where radar echoes exhibit low-to-moderate SNRs and/or decreased correlation coefficient. These effects can obscure the visual identification of weather features as well as adversely impact algorithms. Herein, a novel method that applies variable smoothing in the range where the smoothing intensity depends on the SDs of estimates is presented. It applies little or no range averaging in the regions where data SDs are deemed adequate while using more aggressive smoothing in areas where data appear noisy. Significance Statement The noisiness in the fields of polarimetric variables is an issue that has plagued dual-polarization weather radars since their inception. This is because standard deviations of polarimetric variable estimates increase significantly more with the reduction in SNR than those of spectral moment estimates. A typical mitigation approach that indiscriminately averages a fixed number of estimates in the range may lead to unnecessary loss of range resolution in regions where data appearance is satisfactory. Further, such an approach can be inadequate in regions with a high variability of estimates leading to insufficient enhancement of weather feature visibility. In this study, a method that mitigates these issues is proposed.