Relativistic electrons with energies exceeding 2 MeV in geosynchronous orbit pose significant risks to spacecraft, particularly during periods of intense geomagnetic activity when their flux increases drastically. To predict their evolution, we constructed an artificial neural network (ANN) based on observations from both Geostationary Operational Environmental Satellites-15, incorporating the Fengyun-2G spacecraft. The last orbital measurements in each of all 24 different MLTs (magnetic local times) and their two MLT-adjacent, as well as the historical geomagnetic and solar wind indices, are adopted into the model inputs. The current model achieves a better prediction performance than any previous models, with overall root mean square errors of 0.233, prediction efficiencies of 0.886, and Pearson-correlation-coefficients of 0.943. Furthermore, 99.6% of the samples display an observation-prediction discrepancy of less than one order of magnitude, while more than 96.0% reveal a difference of less than 0.5 orders. Subsequent investigation demonstrated its capability to accurately monitor global changes of the relativistic electron flux at Ek 2 MeV. By incorporating multi-satellite observations, the current model significantly enhances the predictive ability of the ANN model, which is valuable for studying radiation belt dynamics and developing a warning system for space weather disasters.
Zou et al. (Sun,) studied this question.