How to identify and understand large-scale structuring in global ionospheric maps?

Global ionospheric state is frequently characterized by the total electron content (TEC) that is vertically integrated electron density. Global TEC distribution features prominent daytime equatorial ionization anomalies (EIAs) and other regions with elevated TEC, i.e., high density regions (HDRs). Global ionospheric map (GIM) is a gridded 2D data product for TEC that is commonly used to visualize global ionospheric state. How many anomalies and how many high density regions (HDRs) are present in a GIM? How does the number of the HDRs and their TEC magnitudes depend on solar and geomagnetic activity? To address these questions, we apply computer vision and statistical analysis techniques to the GIM dataset (binned 1 degree by 1 degree every 15 minutes) produced by Jet Propulsion Laboratory, California Institute of Technology. We utilize an unsupervised Gaussian Mixture method extended for periodic boundary conditions to identify unique TEC clusters defined as HDRs. Alternatively, we apply the image processing library OpenCV together with edge-enhancing technique to identify HDRs in a training and testing datasets with manual inspection. Using the labeled dataset, four convolutional neural networks are trained (one for each phase of the solar cycle) to classify the number of HDRs over two solar cycles. We found that three and multiple HDRs are common features of GIMs. Our analysis does not show a preference for multiple HDRs to occur during periods of elevated geomagnetic activity. We believe these different approaches to structure identification are complimentary and aid our understanding of the properties of EIAs and ionospheric HDRs and how they can be identified.