Storm-Time Relative Total Electron Content Modelling Using Machine Learning Techniques

Accurately predicting total electron content (TEC) during geomagnetic storms is still a challenging task for ionospheric models. In this work, neural-network (NN)-based model proposed which predicts relative TEC with respect to the preceding 27-day median TEC, storm time European region (with longitudes 30°W–50°E and latitudes 32.5°N–70°N). The (referred as TEC), latitude, longitude, universal time, solar radio flux index F10.7, global SYM-H activity Hp30 are used inputs output of network TEC. can be converted actual knowing calculated at each grid point over considering data from last 27 days before using ionosphere maps (GIMs) international GNSS service (IGS) sources. A event defined when disturbance Dst drops below 50 nanotesla. was trained storm-time period 1998 until 2019 (2015 excluded) contains 365 storms. Unseen 33 events 2015 2020 were test model. UQRG GIMs because their high temporal resolution (15 min) compared other products different analysis centers. NN-based predictions show seasonal behavior including positive negative phases winter summer, respectively, mixture both equinoxes. model’s performance also Neustrelitz (NTCM) quiet-time model, developed German Aerospace Agency (DLR). has root mean squared error (RMSE) 3.38 units (TECU), an improvement by 1.87 TECU NTCM, where RMSE 5.25 found. This corresponds increase 35.6%. outperforms 1.34 TECU, 28.4% 4.72 TECU. Carrington averaged and, therefore, ideal input instead GIM derived median. We found 0.8 17% 3.92 quiet-time-model predicted solely