Abstract

Real-time identification of stationary clutter was introduced into the NEXRAD fleet with Clutter Mitigation Decision (CMD), with Build 11 in the summer of 2009. CMD produces a clutter map which dictates where the clutter filter is to be applied. The single-polarimetric version of CMD identifies c... Show moreReal-time identification of stationary clutter was introduced into the NEXRAD fleet with Clutter Mitigation Decision (CMD), with Build 11 in the summer of 2009. CMD produces a clutter map which dictates where the clutter filter is to be applied. The single-polarimetric version of CMD identifies clutter using a fuzzy logic design, with feature fields derived from radar reflectivity and a relatively new base data field called Clutter Phase Alignment (CPA). The dual-polarimetric version of CMD, introduced with build 13, additionally uses feature fields derived from PHIDP and ZDR. However, there is interest to push the detection capabilities of CMD down to lower clutter-to-signal-ratios (CSRs), around -13 dB, since some of the base dual-polarimetric fields are sensitive to clutter down to that level. In this paper, we present a neural network approach which is used to study the relative merits of the input feature fields, to evaluate the effect of different algorithmic decisions (for example, to include or not include a particular feature field), and to act as an alternative algorithm to compare against. Show less