AI-Based Sensor Fusion Unveils Auroral Ionosphere Dynamics
ECE Associate Professor Pau Closas, in collaboration with Joshua Semeter from Boston University (PI), was awarded a $1,153,603 NSF grant for “Unveiling Hidden Dynamics of the Auroral Ionosphere Using AI-based Sensor Fusion.”
Abstract Source: NSF
Earth’s ionized outer atmosphere has a significant impact on radio-frequency (RF) signals used for communication, radar, and Global Navigation Satellite Systems (GNSS) such as GPS. These effects intensify and become less predictable during elevated solar activity, particularly at higher latitudes where auroral phenomena produce complex and transient ionospheric structures that cause GNSS signal scintillations and tracking failures. At high magnetic latitudes, Earth’s convergent magnetic field acts as a lens, channeling electromagnetic energy derived from the solar wind into a narrow latitudinal region. Most of this energy is dissipated in the lower ionosphere (<200km) through a complex interplay of particle precipitation, plasma heating, and turbulent transport. This multi-scale dynamic is poorly understood and challenging to observe. This project develops innovative AI-driven methodologies to enhance our understanding of these structuring processes and their implications for technologies we rely on for convenience, safety, and national security. The research will facilitate new approaches for monitoring and mitigating ionospheric effects on RF signals and will guide the creation of next-generation “smart sensors” that incorporate a hybrid suite of sensors alongside on-sensor generation of ionospheric models. This project will support graduate and undergraduate student training.
This project addresses this challenge through a methodology that leverages the complementary nature of GNSS and optical data. Wide-field imaging of select emissions in the aurora and airglow spectrum provide quantitative information about plasma production and loss rates, while dual-frequency GNSS receivers offer precise measurements of path-integrated plasma density, referred to as Total Electron Content (TEC). These measurements are connected through established physics-based models. The objective is to identify ionospheric parameters, such as density, temperature, and ion composition, that are consistent with both observational data and established physical principles. Science questions to be addressed are (1) What are the important multi-scale plasma density patterns that define the electrical load seen by the magnetosphere? and (2) How do variations in latent parameters impact the ionospheric response? This work is ideally suited to a specialized form of AI known as Physics Informed Machine Learning (PIML), which incorporates known physics to uncover hidden states of the ionosphere, providing both detailed ionospheric representations and insights into unobserved parameters. The accuracy of PIML models improves with the amount of data applied. Thus, citizen scientists can play a significant role in this research, as GNSS and optical sensors embedded in consumer smartphones continue to improve.