Session
Weekend Session V: Automation - Research and Academia
Location
Utah State University, Logan, UT
Abstract
Equatorial plasma bubbles are a space weather phenomenon that occur at low latitudes within the Earth’s ionosphere. These bubbles are regions of low-density plasma that form at the base of the ionosphere and expand upward through the peak and into the topside. They form in the early evening and persist through the nighttime, stretching north and south along magnetic field lines and effecting a sector of longitudes a few degrees wide. These bubbles cause scintillations on radio signals that pass through them, disrupting the performance of systems, such as GPS, throughout the night in regions around the Earth. Due to their potential social impact, there is a desire to know in real time if equatorial plasma bubbles are occurring. This can be achieved using a small satellite in orbit with automated processing to locate bubbles by processing data from in-situ sensors. Inter-satellite communications to a LEO communications constellation allows the possibility of real time detection of equatorial plasma bubbles from such a satellite. Langmuir probes, impedance probes, and ion drift meters are all instruments capable of detecting plasma bubbles. Sensors such as the Scintillation Prediction Observations Research Task (SPORT) Brazil/US CubeSat mission routinely detect equatorial plasma bubbles. This work investigates the application of various machine learning techniques for the detection of plasma bubbles on a CubeSat using time series plasma density data from the SPORT instrument. This paper reviews four machine learning approaches: Auto-regressive Integrated Moving Average (ARIMA) models, Random Forests, Gradient Boosting Machines, and Recurrent Neural Networks (RNN) such as Long Short-Term Memory (LSTM) networks. The models are evaluated using common metrics derived from the confusion matrix (e.g. accuracy, sensitivity, positive predictive value, and F1 score), as well as the computational complexity of the model. Building off of this work, USU will implement a plasma bubble detection algorithm on a low-power edge computing device built for a CubeSat to provide near real-time plasma bubble reporting. USU’s Low-power Array for CubeSat Edge Computing Architecture, Algorithms and Applications (LACE-C3A) project is being funded by NASA’s University SmallSat Technology Partnership Program to develop an FPGA-based compute board that will be capable of performing machine learning algorithms on a CubeSat scale. The near real-time plasma bubble reporting will be one of the applications USU will demonstrate on LACE-C3A.
Applying Machine Learning to Equatorial Plasma Bubble Now Casting on CubeSats
Utah State University, Logan, UT
Equatorial plasma bubbles are a space weather phenomenon that occur at low latitudes within the Earth’s ionosphere. These bubbles are regions of low-density plasma that form at the base of the ionosphere and expand upward through the peak and into the topside. They form in the early evening and persist through the nighttime, stretching north and south along magnetic field lines and effecting a sector of longitudes a few degrees wide. These bubbles cause scintillations on radio signals that pass through them, disrupting the performance of systems, such as GPS, throughout the night in regions around the Earth. Due to their potential social impact, there is a desire to know in real time if equatorial plasma bubbles are occurring. This can be achieved using a small satellite in orbit with automated processing to locate bubbles by processing data from in-situ sensors. Inter-satellite communications to a LEO communications constellation allows the possibility of real time detection of equatorial plasma bubbles from such a satellite. Langmuir probes, impedance probes, and ion drift meters are all instruments capable of detecting plasma bubbles. Sensors such as the Scintillation Prediction Observations Research Task (SPORT) Brazil/US CubeSat mission routinely detect equatorial plasma bubbles. This work investigates the application of various machine learning techniques for the detection of plasma bubbles on a CubeSat using time series plasma density data from the SPORT instrument. This paper reviews four machine learning approaches: Auto-regressive Integrated Moving Average (ARIMA) models, Random Forests, Gradient Boosting Machines, and Recurrent Neural Networks (RNN) such as Long Short-Term Memory (LSTM) networks. The models are evaluated using common metrics derived from the confusion matrix (e.g. accuracy, sensitivity, positive predictive value, and F1 score), as well as the computational complexity of the model. Building off of this work, USU will implement a plasma bubble detection algorithm on a low-power edge computing device built for a CubeSat to provide near real-time plasma bubble reporting. USU’s Low-power Array for CubeSat Edge Computing Architecture, Algorithms and Applications (LACE-C3A) project is being funded by NASA’s University SmallSat Technology Partnership Program to develop an FPGA-based compute board that will be capable of performing machine learning algorithms on a CubeSat scale. The near real-time plasma bubble reporting will be one of the applications USU will demonstrate on LACE-C3A.
