Nowcasting of Ionospheric Total Electron Content Using the Random Forest and Deep Neural Network Machine Learning Methods
Abstract
The total electron content (TEC) is the measure of the amount of electrons in the upper atmosphere formed due to ionization, mainly from the Sun's extreme ultraviolet radiation. The presence of large TEC and perturbations in the electron density distribution in the ionosphere can significantly affect satellite and other communication signals passing through and can impede accurate communications and navigation systems. Consequently, TEC perturbations and abnormalities can be seen as an indicator of space weather impacts that cause significant damage to satellites, power grids and transportation systems. Hence, estimating the present and future TEC will contribute in mitigating adverse space weather events. In this research we present mainly data-driven nowcasting of TEC using the random forest and deep neural network machine learning methods. The data sources are satellite and ground based measurements consisting of a suite of parameters characterizing the solar and solar-terrestrial environments. The set of training parameters include the solar flux, solar wind density and speed, the three components of interplanetary magnetic field, extreme ultraviolet flux, solar zenith angle, Lyman-alpha, the KP, Dst and Polar Cap (CP) indices, electrojet produced magnetic field, etc - a total of 44 parameters. Most of the training data is from the NASA's Space Physics Data Facility (SPDF). The TEC is calculated at 30 minute interval from 2008 - 2013. The TEC data from 2008-2012 is used for training the machine learning nowcasting models and the 2013 data is used for independent testing. We also present the parameter importance estimates and ranking showing their influence for the machine learning models. A few top parameters were then used to forecast the near future TEC using the long short-term memory (LSTM) deep neural network method. One of the advantages of machine learning over traditional methods is that it can approximate the functional relationship between the several input parameters driving the TEC production and the output (TEC). The other advantage is its capability to estimate the TEC at high accuracy using ensemble method as in the random forest bagging.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMNG31A0849Z
- Keywords:
-
- 1914 Data mining;
- INFORMATICS;
- 1942 Machine learning;
- INFORMATICS;
- 7599 General or miscellaneous;
- SOLAR PHYSICS;
- ASTROPHYSICS;
- AND ASTRONOMY;
- 7999 General or miscellaneous;
- SPACE WEATHER