One‐Day Forecasting of Global TEC Using a Novel Deep Learning Model

• Published in: Space Weather Vol. 19; no. 1
• Main Authors: Lee, Sujin, Ji, Eun‐Young, Moon, Yong‐Jae, Park, Eunsu
• Format: Journal Article
• Language: English
• Published: Washington John Wiley & Sons, Inc 01.01.2021
• Subjects: Deep learning; Equatorial regions; Forecasting; Geomagnetic activity; Geomagnetism; Ionosphere; Ionospheric disturbances; KP index; Mathematical models; Model testing; Orbit determination; Prediction models; Solar activity; Solar cycle; Solar maximum; Solar minimum; Teaching methods; TEC forecasting; Total Electron Content; Training
• ISSN: 1542-7390; 1539-4964; 1542-7390
• DOI: 10.1029/2020SW002600

In this study, we make a global total electron content (TEC) forecasting using a novel deep learning method, which is based on conditional generative adversarial networks. For training, we use the International GNSS Service (IGS) TEC maps from 2003 to 2012 with 2‐h time cadence. Our model has two input images (IGS TEC map and 1‐day difference map between the present day and the previous day) and one output image (1‐day future map). The model is tested with two data sets: solar maximum period (2013–2014) and solar minimum period (2017–2018). Then, we compare the results of our model with those of 1‐day Center for Orbit Determination in Europe (CODE) prediction model. Our major results can be summarized as follows. First, we successfully apply our model to the forecast of global TEC maps. Second, our model well predicts daily TEC maps with 1 day in advance using only previous TEC maps. The averaged root mean square error, bias, and standard deviation between AI‐generated and IGS TEC maps are 2.74 TECU, −0.32 TECU, and 2.59 TECU, respectively. Third, our model generates some peak structures around equatorial regions. Fourth, our model shows better performance than 1‐day CODE prediction model during both solar maximum and minimum periods. Fifth, another model with additional input data Kp index gives a slight improvement of the results. Our study shows that our deep learning model based on an image translation method will be effective for forecasting of future images using previous data. Plain Language Summary The main causes of ionospheric disturbance are solar activity and geomagnetic activity. The total electron content (TEC) is one of the important parameters of ionosphere and it can be used to investigate ionospheric disturbances. We develop a global TEC 1‐day forecasting model using a novel deep learning method. For training, we use the International GNSS Service TEC maps from 2003 to 2012 which include both solar maximum and minimum periods. Our model successfully forecasts the global TEC maps 1 day in advance. Key Points We make a global total electron content (TEC) 1‐day forecasting using a deep learning model based on conditional generative adversarial networks Our model shows better performance than 1‐day Center for Orbit Determination in Europe prediction model during the solar maximum and solar minimum periods We successfully apply our model to the forecast of global TEC maps using only previous TEC data