Short-time forecast of ionospheric irregularities using long short-term memory networks over equatorial and low-latitudes regions

Predicting ionospheric conditions is becoming increasingly important towards the operational efficiency of both ground-based and space-borne radio communication systems with a view to compensate for the effects of space weather. This study focuses on predicting ionospheric irregularities in the complex and variable equatorial ionosphere which is deemed critical for optimal space-based application. We utilized the Long-Short-Term-Memory (LSTM) deep learning algorithm to develop a predictive model for forecasting disturbances in the equatorial ionization anomaly (EIA) region using Global Navigation Satellite Systems (GNSS) data. We utilized fifteen-year worth of data (2005–2020) to train, validate and test the performance of the model and assessed the results against a baseline model relying on daily and hourly Rate of Change of TEC Index (ROTI) values and utilized evaluation metrics such as correlation (R), determination coefficient ($R^2$), and mean squared error (MSE). Remarkably, the LSTM Predictive Model consistently outperformed the Baseline Model across various stations, demonstrating higher R and R² values and significantly lower MSE. These results indicate the LSTM model's superior accuracy in forecasting ionospheric disturbances, essential for space-based applications. The distribution analysis of residual errors highlighted the LSTM model's ability to better capture underlying patterns and variability in the target variable. This study contributes to enhancing ionospheric forecasting models for space applications, ensuring the dependability of space-based systems.