4:00 PM - 4:15 PM
[PEM11-03] A novel Machine learning technique to Parameterize EneRgetic Electron maps (AMPERE)
★Invited Papers
Energetic electron precipitation (EEP) is one of the main processes contributing to the loss of energetic
electrons in the outer radiation belt and has important implications in the coupled atmosphere-ionosphere-magnetosphere system and in space weather (e.g., satellite radiation monitoring, satellite drag, etc.). The lack of global observations of EEP is a major limiting factor in advancing our knowledge on EEP. To circumvent this, we can accurately parameterize EEP by developing global EEP maps through the use of machine learning (ML) techniques. These maps will be based on measurements from the long-lived NOAA’s POES/MetOp satellites and will be produced given a time history of geomagnetic activity. In addition to training the data using global geomagnetic proxies (i.e. Kp), we utilize the entire suite of regional indices provided by the SuperMAG measurements. This provides 1 MLT longitudinal data for our ML model to account for the large MLT dependence on EEP. Our model produces count rates mimicking the E1-E4 POES integral channels at 1 MLT x 1 L resolution. Preliminary validation shows reasonably accurate results when compared with the true count rates. We present results during chorus and EMIC wave activity as well as quiet geomagnetic times to test the model accuracy over a wide range of conditions. We also show the wide range of use cases for such as a data set, which include climate modeling (both forensic and future), validation of other electron detectors, and even some possible nowcasting applications.
electrons in the outer radiation belt and has important implications in the coupled atmosphere-ionosphere-magnetosphere system and in space weather (e.g., satellite radiation monitoring, satellite drag, etc.). The lack of global observations of EEP is a major limiting factor in advancing our knowledge on EEP. To circumvent this, we can accurately parameterize EEP by developing global EEP maps through the use of machine learning (ML) techniques. These maps will be based on measurements from the long-lived NOAA’s POES/MetOp satellites and will be produced given a time history of geomagnetic activity. In addition to training the data using global geomagnetic proxies (i.e. Kp), we utilize the entire suite of regional indices provided by the SuperMAG measurements. This provides 1 MLT longitudinal data for our ML model to account for the large MLT dependence on EEP. Our model produces count rates mimicking the E1-E4 POES integral channels at 1 MLT x 1 L resolution. Preliminary validation shows reasonably accurate results when compared with the true count rates. We present results during chorus and EMIC wave activity as well as quiet geomagnetic times to test the model accuracy over a wide range of conditions. We also show the wide range of use cases for such as a data set, which include climate modeling (both forensic and future), validation of other electron detectors, and even some possible nowcasting applications.