A Point-cloud Clustering & Tracking Algorithm for Radar Interferometry

Abstract

In data mining, density-based clustering, which entails classifying datapoints according to their distributions in some space, is an essential method to extract information from large datasets. With the advent of software-based radio, ionospheric radars are capable of producing unprecedentedly large datasets of plasma turbulence backscatter observations, and new automatic techniques are needed to sift through them. We present an algorithm to automatically identify and track clusters of radar echoes through time, using \texttt{dbscan}, a celebrated density-based clustering method for noisy point-clouds. We demonstrate our algorithm's efficiency by tracking turbulent structures in the E-region ionosphere, the so-called radar aurora. Through conjugate auroral imagery, as well as \emph{in-situ} satellite observations, we demonstrate that the observed turbulent structures generally track the motion of auroras. What is more, the radar aurora bulk motions exhibit key qualities of auroral electric field enhancements that has previously been observed with various instruments. We present preliminary statistical results using our new method, and briefly discuss the method's limitations and potential future adaptations.