Seismic-ionospheric Precursor Prediction Using Deep Learning

Numerous studies have indicated the potential involvement of seismic-ionospheric precursors (SIPs) in catastrophic earthquakes. These SIPs have been extensively analyzed in terms of characteristics such as the degree of decrease or increase, timing of occurrence, and duration, with a specific focus on electron density or total electron content (TEC). While deep neural networks have demonstrated remarkable accuracy in various applications, their capability to estimate SIPs has been severely limited. In this study, we developed a novel AI prediction model for forecasting SIPs based on a six-channel data sequence: TEC, median TEC, lower bound TEC, upper bound TEC, positive polarity, and negative polarity. To leverage the power of deep learning image classification methods, we treated this data as a six-dimensional image. This approach yielded a maximum accuracy of 67.1 percent.