IN039-09
Enabling Robust Real-time Information from Ground-Based Riometers: A Machine Learning Approach
Enabling Robust Real-time Information from Ground-Based Riometers: A Machine Learning Approach
Tuesday, 15 December 2020: 17:54
Virtual
Abstract:
Ground-based riometers can provide quantitative information about processes related to high-energy electron precipitation as well as the HF radio environment local to the instrument. Depending on the nature of the instrument (multi-frequency, imaging, etc.) a riometer has a variety of potential space weather applications. The largest barrier to using riometer data in real-time has been the need to process the data into a usable form by removing the cosmic background signal and creating an absorption data stream.
These efforts are often stifled by the nature of the riometer instrument itself. As radio receivers, riometers are often subject to numerous noise sources (both local and extra-terrestrial) as well as other instrumentation effects. The need to place the instruments in radio quiet regions leaves them susceptible to wildlife and other local impacts that would not be experienced with the more reliable space-based assets. In this paper, we present the preliminary results of a machine learning study to enable the robust usage of riometer data, in real-time, from the Geospace Observatory Canada Widebeam Riometer Network. We will present a description of the key obstacles for utilizing these instruments in real-time, as well as the feature identification used within a neural network-based model of riometer “usability” classification. We also present an optimization pathway for future utilization and implementation of this model at scale. This study is the foundational work that will enable true real-time data feeds from the GO-Canada continent-scale array of riometers.
These efforts are often stifled by the nature of the riometer instrument itself. As radio receivers, riometers are often subject to numerous noise sources (both local and extra-terrestrial) as well as other instrumentation effects. The need to place the instruments in radio quiet regions leaves them susceptible to wildlife and other local impacts that would not be experienced with the more reliable space-based assets. In this paper, we present the preliminary results of a machine learning study to enable the robust usage of riometer data, in real-time, from the Geospace Observatory Canada Widebeam Riometer Network. We will present a description of the key obstacles for utilizing these instruments in real-time, as well as the feature identification used within a neural network-based model of riometer “usability” classification. We also present an optimization pathway for future utilization and implementation of this model at scale. This study is the foundational work that will enable true real-time data feeds from the GO-Canada continent-scale array of riometers.