Equatorial Spread-F Forecasting Model with Local Factors using the Long-Short Term Memory Network

Abstract The predictability of the nighttime equatorial spread-F (ESF) irregularity is essential to the ionospheric disturbance warning system. In this work, we develop the ESF forecasting model using the Long-Short Term Memory (LSTM) network trained with new local input parameters at Chumphon station, Thailand near the magnetic equator, where the ESF onset typically occurs. The LSTM model is trained and tested using the ionogram data from 2008 to 2018 and 2019 during March and September equinoctial months. The input parameters are considered including diurnal variations, seasonal variations, solar activity, magnetic activity, local F-layer height variation, local F-layer drift velocity, and power spectrum of the atmospheric gravity waves. We analyze the ESF forecasting model in terms of monthly probability, daily probability and occurrence, and diurnal predictions. As a result, the proposed LSTM model can achieve 85.4% accuracy when the local F-layer drift velocity and atmospheric grativity waves are utilized. The local gravity wave parameter exhibits an important role on improvements of the LSTM model during post-midnight. When compare to the IRI-2016 model, the proposed LSTM model can provide lower error. Comparison between ANN and LSTM models are exhibited that the LSTM model provides lower error than the ANN in February, March, April, and October.