Classification of SuperDARN Backscatter Observations using Machine Learning Algorithms

The Super Dual Auroral Radar Network (SuperDARN) radars are used to study the ionosphere and its relation to geospace. Data from multiple radars are routinely combined to create ionospheric plasma convection maps. Depending on the ionospheric propagation conditions, SuperDARN radars receive backscatter echoes from different sources, such as the ionosphere, the ground, and meteor trails. It is essential to separate these categories cleanly so that the wrong types of scatter do not contaminate the various data products produced. However, it is challenging to separate these categories because their characteristics in Doppler velocity, spectral width, and other features can overlap, and ground truth is rarely available. The traditional method of distinguishing ionospheric scatter (IS) from ground scatter (GS) classifies data point by point and can misclassify low-velocity IS as GS, particularly in the subauroral region where the ionospheric convection is much weaker. Traditional algorithms ignore correlations between data points that are close in time and location. This study applies unsupervised machine learning methods to cluster backscatter data using a Gaussian Mixture Model (GMM) and several variations of Density-Based Spatial Clustering of Applications with Noise (DBSCAN) using spatial and temporal parameters. We then classify the clusters as IS or GS using the statistical characteristics of the Doppler velocity. We show the newly developed clustering algorithms offer substantial improvements over previous methods by increasing the amount of IS identified by a factor of two on average and identifying undefined scatter and noise categories in SuperDARN data.