Forecasting Ionospheric Total Electron Content During Geomagnetic Storms

A geomagnetic storm is a temporary disturbance of the Earth’s magnetosphere that caused magnetic field depression. This activity is associated with solar coronal mass ejections, coronal holes, or solar flares which allows its association with ionospheric Total Electron Content (TEC) and its physical changes. Neural Network (NN) is a modeling technique capable in exhibiting nonlinear properties that comprises physical quantities. This study focuses on establishing an ionospheric TEC prediction model during a geomagnetic storm using feed-forward back propagation neural networks. Geomagnetic events on 18 March and 23 June 2015 with (Kp index of 8 and DST index $\lt-$200nT) over Universiti Kebangsaan Malaysia were selected as the case study. Factors influencing TEC were defined and used as input parameters. TEC was modelled as a function of seasonal variation, diurnal variation and solar activity to establish a valid correlation with magnetic field depression. Model output was analyzed by comparing the predicted TEC with measured GPS-TEC and IRIOI-corr TEC model output to verify the accuracy of TEC modeling using RMSE and MAPE values. The RMSE of the NN prediction model against the measured TEC between 2 – 12 TECU in comparison with the IRIOI-corr TEC model between 7-25 TECU. Whereas the MAPE represents 10% – 54% for NN prediction against 23% - 57% for IRIOI-corr TEC model. Hence, NN method showed a higher performance and better estimates of the TEC compared to the IRIOI-corr. This study concludes that both NN and IROI-corr models are unable to predict accurate ionospheric TEC during major geomagnetic storms. However, the prediction accuracy improved in the recovery phase (post-storm).