Machine Learning for Quantifying Effects of Seasonal Variation and Natural Phenomenon on the D-region Ionosphere

Abstract:

It is difficult to measure properties of the D-region ionosphere, such as electron density, due to the altitude range (60-90 km). Sounding rockets can provide direct measurements, but only at a single location and time. Rockets are also prohibitively expensive for large numbers of measurements. It is possible to infer various properties of the D-region using very low and low frequency (VLF/LF, 30-300 kHz) radio waves, however this technique is often ambiguous, meaning multiple possible ionospheric solutions lead to the same measurements. Recently, Gross and Cohen [2020] used an artificial neural network (ANN) to determine D-region waveguide conditions across a network of transmitter-receiver paths within and near the continental US. The ANN was trained to predict a single day’s Wait and Spies parameters (h’ and beta) for electron density. The model required clean and uninterrupted data for the entire network of receivers and transmitters. As such, days with clean data were manually selected which hindered the ability for the model to be used on a larger scale. For example, each transmitter has a weekly maintenance day, so many days were unusable.

An automatic quality assessment tool was used to select all the useful receiver-transmitter paths for each day. We then trained the ANN for a wide variety of network configurations after speeding up and parallelizing the computation and training time. By removing the need for the full network of receivers and transmitters, we are currently able to apply the ANN approach to hundreds of days, tracking h’ and beta on up to 21 paths for the daytime hours. With this expanded dataset, we examine the effects of seasonal variation and natural phenomenon on the ionosphere’s D-region. The modelling technique is now able to quantify the ambient daytime ionosphere throughout the year. We explore effects of several types of ionospheric disturbances, including the 2017 solar eclipse, solar flares, and hurricane-induced gravity waves. We are also working to expand the ionospheric estimation into the nighttime hours, whose ambient conditions flutter much more, and which is subject to a much broader range of geophysical disturbances