Ionospheric TEC forecast using universal kriging and recurrent neural network over low-latitude during the X class solar flares occurred in the year 2024

Abstract

In recent years, Statistical and machine learning models have been widely used for ionospheric total electron content (TEC) forecasting. In this research, we constructed a universal kriging (UK) statistical model and a recurrent neural network (RNN) machine learning model to forecast the TEC during six solar flares that happened in February and March 2024. Twelve months (from February 2023 to January 2024) of geomagnetic indices data like Planetary K Index (Kp), Planetary A Index (Ap), Disturbance Storm Time (DST) index and Solar indices data like Radio Flux at 10.7 cm (F10.7), Solar wind (Sw), and Sun Spot Number (SSN) along with GPS TEC values obtained from the IISC station, Bangalore (13.03° N and 77.57° E) were used for training and two months (February and March 2024) of data were used for testing the constructed models to forecast the TEC during the six solar flares (SF) occurred in the year 2024. The forecasted results showed that the UK model obtained root mean square error values (RMSE) of 6.76 during the X 3.38 SF, 5.58 during the X 2.25 SF, 4.85 during the X 1.9 SF, 7.0 during the X 6.3 SF, 12.29 and 6.74 during the X 1.1 SF when compared to the RNN model obtained RMSE values of 12.81, 14.34, 8.01, 9.39, 14.36 and 11.22 respectively. Analysis of TEC variations during February and March 2024 revealed diurnal patterns influenced by solar radiation, with high TEC values during the day and lesser at night. Comparison of UK and RNN predictions during the SF periods highlighted both models' superior ability to capture TEC variations, particularly at peaks and troughs. The linear regression statistical analysis showed high positive correlations (Pearson's r > 0.96) between actual and predicted TEC for both models, with UK demonstrating higher accuracy during intense solar flares (X6.3 and X3.38). Evaluation during the considered dates for the six SF periods indicated that the UK model provided better overall accuracy compared to RNN, though RNN showed competitive performance. The study underscores UK's potential for precise ionospheric TEC forecasting during solar disturbances, which is essential for space weather monitoring and satellite communication systems. However, the RNN model also performed well during high solar activity suggesting its suitability for capturing abrupt ionospheric changes. This study contributes insights into leveraging surrogate and machine learning models for ionospheric studies, demonstrating their effectiveness in predicting TEC variations under varying solar and geomagnetic conditions. The accuracy of prediction depends upon the size of the data set. This research will be useful to mitigate the positional accuracy errors in the navigation systems and also helpful for improved space communication during adverse solar activities.