Abstract Equatorial plasma bubbles (EPBs) are plasma density depletion structures that produce magnetic field perturbations through the diamagnetic effect and therefore need to be identified and excluded in high‐precision geomagnetic field modeling. Traditional EPB detection methods typically rely on in situ electron density or total electron content data, which are unavailable on the Macau Science Satellite 1 (MSS‐1) mission, due to the lack of relevant payloads. To detect the possible diamagnetic signatures caused by EPBs, we propose a machine learning (ML) approach, BMML (Bubble detected from Magnetic data of MSS‐1 using Long Short‐Term Memory network), to automatically identify EPBs using only magnetic field measurements of MSS‐1. The BMML model is trained on time‐frequency characteristics of Swarm magnetic data labeled by its Level‐2 Ionospheric Bubble Index (IBI) product. Statistical analysis shows that our BMML achieves an accuracy of 0.969 on the test data set of Swarm. In addition, we find that BMML model sometimes is even better than the IBI product in detecting EPBs from Swarm data, for example in identifying small‐scale EPBs and distinguishing closely spaced EPBs. When applied to the independent MSS‐1 data, BMML shows well generalization, successfully identifying EPB‐related diamagnetic signatures. Further statistical analysis of the identified EPBs from MSS‐1 shows consistent occurrence distributions with previous studies, which confirms the reliability of our model. The BMML has its advantage to reduce the subjectivity and limitations associated with thresholds setting in traditional approaches for identifying EPBs. In addition, the BMML model can enhance the reliability and accuracy of geomagnetic models of MSS‐1 mission, by detecting and excluding EPB‐contaminated magnetic data.
Rang et al. (Tue,) studied this question.