Obtaining measurements of atmospheric density variations in the region encompassing roughly 50 to 120 km altitude has been a challenge to the atmospheric science and space community because of its inaccessibility to balloon-borne or space-based instruments.  Our goal is to study the neutral density in this region by observing meteors passing through the atmosphere using High-Power Large-Aperture (HPLA) radar.  Meteoroids decelerate and ablate due to atmospheric collisions, forming plasma known as meteors.  HPLA facilities routinely detect meteor head echoes, which move with the meteoroid through the atmosphere, at altitudes of 70 to 120 km. These observations can be used to monitor atmospheric density over a large subset of the altitudes that are otherwise difficult to access.  The major outcomes of this project include conducting the first simultaneous meteor radar campaign across geographically distinct HPLA facilities, developing new signal processing techniques to interpret these radar observations, developing simulations to understand the population statistics of the observed meteors, and using these meteors to estimate the atmospheric density.

We made simultaneous observations at Jicamarca Radio Observatory (JRO) in Peru, Millstone Hill Observatory (MHO) in Massachusetts, and Resolute Bay (RISR-N) in the Canadian Arctic, on 2019 October 10 and 11, from 09:00 to 13:00 UTC.  This resulted in 2.7 TB of raw radar returns, which continues to provide research output beyond the lifetime of this project.  To identify meteors within the radar data, we developed a machine learning method using a convolutional neural network (CNN) to classify meteor head echoes, and developed a method to simulate head echo observations providing a larger synthetic training set than is possible through manual tagging of real data.  Our CNN architecture accepts raw radar data as input, which enables computationally inexpensive classification of each data segment. The CNNs demonstrate greater than 0.7 overall sensitivity to head echoes at each facility. Sensitivity is significantly higher for stronger head echoes, as one would expect, whereas for the weaker head echoes, it remains above 0.5 across all facilities. The precision drops when cluttering phenomena are present, increasing the occurrence of false positives, but sensitivity remains greater than 0.5.

We applied a signal processing technique to the radar data based on matching the radiofrequency phase of the radar pulses to yield accurate range rates and decelerations for a subset of the meteor populations observed at each facility.  JRO is overall most sensitive to meteor head echoes because of its lower carrier frequency. MHO observes more meteors than RISR-N at the same frequency, indicating that the latitude and beam angle relative to the ecliptic plane is a significant factor in meteor detectability.  From our meteor data, we find that RISR-N does not observe head echoes with range rates faster than 55 km/s, likely due to its high latitude and radar beam pointing away from the ecliptic plane. The RISR-N population also demonstrates a bias toward larger and faster head echoes due to its higher carrier frequency. At JRO near the equator, a larger spread of range rates is observed. A trend of greater decelerations at lower altitudes is observed at RISR-N and JRO.

To map the detected meteors back to their parent meteoroid populations, we developed two simulation tools.  We implemented an ablation model to track the meteoroid’s mass loss through the upper atmosphere prior to radar detection, and an orbital force model to back-propagate the trajectory through the solar system.  Through ablation modeling, we find that small meteoroids (nanogram to microgram) can lose half of their initial mass above 100 km altitude.  This altitude dependence indicates that while nearly all meteoroidal mass deposited at 75 km altitude comes from larger meteoroids, the dominant mass deposition at 100 km and above is from the smaller population.  We developed a new open-source software for determining the heliocentric orbital parameters of meteoroids based on their meteor head echoes.  We use numerical integration, which outperforms analytical approximations and allows us to take perturbative effects into account.  We included solar radiation pressure due to its greater influence on very small particles but showed that it is small compared to drag and third-body perturbations of the Moon and planets.  We compute the uncertainty of the orbital elements using the covariance transform method, which runs faster than Monte Carlo simulations.

Given the detected meteor dataset, we applied the atmospheric density estimation to 1151 meteors (311 at JRO, 410 at MHO, and 430 at RISR-N).  These meteors provide good sampling in the mid-range altitudes of 90-110 km, where most of them are detected, while the estimates outside of this altitude range have high error.  These density profiles show a generally similar trend to MSIS00 but infer some spatial density variation beyond the model output.  The confidence of this meteor-derived density estimation can be increased in future work with continued enhancement in radar sensitivity, signal processing, and modeling techniques.

Last Modified: 11/26/2023
Modified by: Sigrid Elschot