A deep-learning based approach for predicting high latitude ionospheric scintillations using geospace data and auroral imagery
Abstract
Small scale fluctuations and irregularities in ionospheric electron density impact the propagation of transionospheric high frequency radio waves such as those used by the Global Navigation Satellite System (GNSS), degrading signal quality. These disruptions in radio signals are termed ionospheric scintillations. Ionospheric scintillations at high latitudes are highly driven by space weather and therefore are extremely dynamic in nature. No existing physics-based model is capable of predicting these ionospheric irregularities with good spatial and temporal resolution. Recent work has demonstrated that a data-driven machine learning model based on time-series data from ground-based GNSS receivers, solar wind, and geospace parameters has some skill in predicting ionospheric scintillations 1 and 3 hours in advance [McGranaghan et al. 2018]. Ionospheric irregularities associated with the occurrence of scintillations have also been correlated with the presence of structures in visible aurora, although it is not straightforward to directly incorporate these observations into predictive models. Deep learning, which has led to significant advances in the field of natural image processing, offers a promising approach for image feature extraction that may be useful for problems in the physical sciences. The success of deep learning in image processing is one particular instance of its effectiveness in extracting task-specific information from input data given little explicit information about the structure of the data. In the context of the physical sciences, this property may help to make use of data where explicit physics-based models cannot practically be applied. Here we focus on developing a deep-learning based predictive model for ionospheric phase scintillations incorporating both time series data from ground-based magnetometer and GNSS receivers, solar and magnetospheric parameters, and auroral imagery from all-sky imagers in the Time History of Events and Macroscale Interactions during Substorms (THEMIS) network in Northern Canada. This approach improves upon state-of-the-art predictive models based on time series data alone by incorporating the auroral imagery.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMNG21A..08M
- Keywords:
-
- 1914 Data mining;
- INFORMATICS;
- 1942 Machine learning;
- INFORMATICS;
- 7599 General or miscellaneous;
- SOLAR PHYSICS;
- ASTROPHYSICS;
- AND ASTRONOMY;
- 7999 General or miscellaneous;
- SPACE WEATHER