A High-Precision and Lightweight Prediction Model for Global Total Electron Content

Abstract

Precise prediction of the global spatial–temporal distribution of total electron content (TEC) is a challenge in space weather. Existing models are generally able to provide rather good prediction results at the cost of a large amount of computing resources. This limits the application of the method. A lightweight and highly accurate global TEC prediction model was developed in this study. Our model is capable of forecasting the global TEC map up to 12 h in advance with a step of one hour. The predicted results during geomagnetic quiet periods were consistent with measurements, with a maximum and average mean error (ME) of 1.5 TECU and −0.04 TECU under conditions of high solar activity, respectively. Our model also performed well during geomagnetic disturbed periods, with a maximum ME of 4.5 TECU and 2.5 TECU under conditions of high and low solar activities, respectively. Our model significantly reduces the training time (47%) and basic requirement of memory (60%) relative to the model of Liu et al. (2022) with no remarkable loss of model accuracy.

Keywords

• TEC prediction
• machine learning
• ionosphere