Modeling Earth's Global Electron Density by Using LSTM‐Attention Neural Networks

Abstract The electron number density is an important physical quantity in the Earth's inner magnetosphere. Modeling the long‐term evolution of electron number density is critical to obtain a physical understanding of the inner magnetosphere dynamics. The plasmaspheric plume is an important density structure, but it's difficult to capture the structure and evolution of the sharp and narrow plume during geomagnetically disturbed times. Based on the LSTM‐Attention neural network, we have built a global electron density (GEO‐NE) model to reconstruct the global distribution of electron density with geomagnetic indices as inputs. The GEO‐NE model successfully constructs the evolution of the plasmasphere and plume during various phases of geomagnetic storm activities, especially the sharp and narrow plume structure during geomagnetic storms. To assess the performance of the GEO‐NE model, the comparison between our model output in geomagnetic storm events outside the training sample and the global images of IMAGE EUV has been implemented. In addition, we have also compared the results of our model and plasmapause test particle (PTP) simulation model. The results indicate the attentional mechanism approach exhibits its advantage in modeling electron number density global distribution. The GEO‐NE model provides a new method for plasma density modeling, which may be beneficial to the dynamic modeling of the inner magnetosphere.