Classification of High Density Regions in Global Ionospheric Maps With Neural Networks

Abstract

The database of Global Ionospheric Maps (GIMs) produced at Jet Propulsion Laboratory is analyzed. We define high density Total Electron Content (TEC) regions (HDRs) in a map, following certain selection criteria. For the first time, we trained four convolutional neural networks (CNNs) corresponding to four phases of a solar cycle to classify the GIMs by the number of HDRs in each map with ∼76% accuracy on average. We compared HDR counts for GIMs across ten years to draw conclusions on how the number of HDRs in the GIMs changes throughout the solar cycle. Occurrence of HDRs during different geomagnetic activity conditions is discussed. Catalogue of selected HDRs for ten years and four CNN-based models that can be used to extend classification to other years are provided for the community to use.

Statistical Characteristics of Total Electron Content Intensifications on Global Ionospheric Maps

“…First, the study relies on the TEC intensification data set generated with the TEC feature extraction software. The robustness of the software in pattern recognition has been demonstrated via visually inspecting the identified TEC intensification regions for ten thousands of TEC maps randomly selected (Verkhoglyadova et al., 2021). The software can successfully recognize TEC intensification regions of various shapes over a vast majority of the TEC maps.…”

“…The TEC feature extraction software is built upon the labeling script reported in Verkhoglyadova et al. (2021). The previous work utilized several methods from OpenCV's image processing library (https://opencv.org/) to recognize boundaries of high density TEC regions (HDRs), that is, TEC enhancement regions, that meet the defined criteria.…”

“…The script is used to determine the number of HDRs for approximately 10,000 TEC maps selected from 10 years of the JPL GIM data. The occurrence of different HDR counts and its variation with solar cycle phase, season, and geomagnetic activity are examined (Verkhoglyadova et al, 2021(Verkhoglyadova et al, , 2022. For this work, we have modified and improved the labeling script to construct the feature extraction software.…”

“…Different types of global‐scale longitudinal‐latitudinal maps have been utilized to investigate EIAs or EIA‐like features. Examples include climatological maps of the F2‐layer peak electron density obtained from a global collection of ionospheric radio occultation measurements (e.g., Tulasi Ram et al., 2009; Yue et al., 2015), global TEC maps constructed from Global Navigation Satellite System (GNSS) ground‐based measurements (e.g., Verkhoglyadova et al., 2021), and more recently, OI 135.6 nm emission maps for the nighttime ionosphere from Global‐scale Observations of the Limb and Disk (GOLD) imager (Cai et al., 2023; Eastes et al., 2019). These studies offer valuable global‐scale characterization of EIAs and EIA‐like features.…”

Understanding Large-Scale Structure in Global Ionospheric Maps With Visual and Statistical Analyses

“…Alternatively, the image processing library OpenCV for Python together with edge-enhancing technique was applied to identify HDRs in a selected GIM dataset with visual inspection. This is an improvement upon our image approach (Verkhoglyadova et al, 2021). First, for each TEC map, represented by gridded TEC values, we round the float TEC values to integer numbers and linearly scale the TEC values to numbers between 0 and 255.…”