Flux Gate Magnetometer Data Predictive Correction Using LSTM Recurrent Neural Networks

Flux gate magnetometers play an important role in ionospheric research, providing valuable data for understanding Earth's magnetic field variations influenced by processes occurring within the ionosphere, such as geomagnetic storms, solar flares, and ionospheric disturbances. However, these instruments are susceptible to errors induced by magnetic interference; presence of ferromagnetic objects in the vicinity of the sensor; sensor tampering, among others. In this study, we present a novel approach utilizing Long Short-Term Memory (LSTM) recurrent neural networks to predict and correct erroneous data points in magnetometer recordings. Our model uses the temporal dependencies inherent in magnetometer data to accurately identify and rectify anomalies caused by external disturbances. By training on historical datasets with known errors, the model learns to predict the correct magnetic field values corresponding to erroneous readings.