CEDAR
Coupling, Energetics and Dynamics of Atmospheric Regions Program
x

Deep Learning Models for Ionospheric Electron Density Time Sequence Prediction

Yang Pan, University of Texas at Arlington, TX, USA
Mingwu Jin, University of Texas at Arlington, TX, USA
Shunrong Zhang, Haystack Observatory, Massachusetts Institute of Technology, MA, USA
Yue Deng, University of Texas at Arlington, TX, USA
First Author's Affiliation
University of Texas at Arlington
Abstract text:

In previous work, we developed multi-layer perceptron neural network models for ionospheric electron density (Ne) prediction at a fixed altitude. However, these models did not explicitly account for the temporal connections among Ne during a time period. To address this limitation, we model Ne as time sequences at different altitudes in this study. We explore two different neural network architectures for Ne time sequence prediction: the recurrent neural network (RNN) and the transformer neural network (TNN). The RNN uses a memory mechanism to enable the output of a node to affect the subsequent input of the same node, while the TNN leverages an attention mechanism to focus on different parts of the input sequence simultaneously. We train and test the models using the Incoherent Scatter Radar (ISR) data from Millstone Hill Observatory spanning the period from 1998 to 2018, and compare the prediction performance of RNN and TNN.

Poster PDF
DATA-17-pan-yang-neuralnetwork-ionosphere.pdf (1.58 MB)
Student in poster competition
Poster category
DATA - Data Assimilation, Data Analytics, Methods and Management
↑