SA51A-2390
Prediction of the Nighttime VLF Subionospheric Signal Amplitude by Using Nonlinear Autoregressive with Exogenous Input Neural Network Model

Friday, 18 December 2015
Poster Hall (Moscone South)
Hendy Santosa, Bengkulu University, Electrical Engineering, Bengkulu, Indonesia; University of Electro-Communications, Tokyo, Japan, Yasuhide Hobara, Earth Environment Research Station, The University of Electro-Communications, Chofu, Tokyo, Japan; The Univ. of Electro-Comms, Chofu-City, Tokyo, Japan and Michael A Balikhin, Department of Automatic Control and Systems Engineering, The University of Sheffield, Sheffield, United Kingdom
Abstract:
Very Low Frequency (VLF) waves have been proposed as an approach to study and monitor the lower ionospheric conditions. The ionospheric perturbations are identified in relation with thunderstorm activity, geomagnetic storm and other factors. The temporal dependence of VLF amplitude has a complicated and large daily variabilities in general due to combinations of both effects from above (space weather effect) and below (atmospheric and crustal processes) of the ionosphere. Quantitative contributions from different external sources are not known well yet. Thus the modelling and prediction of VLF wave amplitude are important issues to study the lower ionospheric responses from various external parameters and to also detect the anomalies of the ionosphere.

The purpose of the study is to model and predict nighttime average amplitude of VLF wave propagation from the VLF transmitter in Hawaii (NPM) to receiver in Chofu (CHO) Tokyo, Japan path using NARX neural network. The constructed model was trained for the target parameter of nighttime average amplitude of NPM-CHO path. The NARX model, which was built based on daily input variables of various physical parameters such as stratosphere temperature, cosmic rays and total column ozone, possessed good accuracies.

As a result, the constructed models are capable of performing accurate multistep ahead predictions, while maintaining acceptable one step ahead prediction accuracy. The results of the predicted daily VLF amplitude are in good agreement with observed (true) value for one step ahead prediction (r = 0.92, RMSE = 1.99), multi-step ahead 5 days prediction (r = 0.91, RMSE = 1.14) and multi-step ahead 10 days prediction (r = 0.75, RMSE = 1.74). The developed model indicates the feasibility and reliability of predicting lower ionospheric properties by the NARX neural network approach, and provides physical insights on the responses of lower ionosphere due to various external forcing.

<< Previous Abstract | Next Abstract >>