Volumetric Imaging of Advanced Modular Incoherent Scatter Radars (AMISR) Data Using Neural Network
Year: 2026
Type: article
Abstract: Ionospheric electron density is highly dynamic and structured, which can cause major disturbances with satellite communication. Advanced Modular Incoherent Scatter Radars (AMISR) can measure three-dimensional electron density movement over time. Visual representations of these interactions exist in two-dimensional figures or plots, along with the development of three-dimensional images of electron density structures across the science community. This paper presents our Artificial Neural Network (ANN) machine learning algorithm, which can reconstruct three-dimensional images of electron density structures. Utilizing data collected from the two AMISR facilities in North America, three-dimensional images of polar cap interactions (Resolute Bay, Canada) and auroral activities (Poker Flat, Alaska) can be depicted. AMISR data files are spatially sparse and contain real uncertainties and artifacts, causing a multitude of issues for three-dimensional reconstruction. Our machine learning algorithm is trained to predict electron density values between discrete measurements for a complete three-dimensional image of the ionosphere. This paper investigates four hyperparameters dropout layers, early stopping conditions, and parameter weights that guided the prediction process. The impact of each function is individually analyzed for various electron density structures. Common electron density configurations that were replicated are Chapman layers, circular clusters, large gradients, wave patterns, and combinations of these designs. Complex structures were prone to having greater prediction errors, while structures such as Chapman layers and the combined Chapman and circular structures consistently had lower errors. The ANN hyperparameters are presented with structure-specific function weight values, which produce three-dimensional representations of ionospheric electron-density structures that can support visualization and quantitative analysis of spatially varying ionospheric features.
Source: Machine Learning Earth
Language: English
Subfield: Space and Planetary Science
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,125
Funder: National Science Foundation
Awards: CAREER: Characterization of Sources of Ionospheric Scintillation and Space Weather Prediction through Analytics and Machine Learning, REU Site: Research Experiences for Undergraduates in Atomic, Molecular, and Biological Physics, Collaborative Research: Characterizing High-latitude Ionospheric Fluid Turbulence and Radio Scintillation with New Observations and Data-Driven Modeling
OpenAlex ID: W7235281718
