Utilizing XGBoosts to correct arcjet contamination in magnetic field measurements from GOES missions

The magnetometers onboard the Geostationary Operational Environmental Satellites (GOES) provide crucial measurements for space weather monitoring and scientific research. However, periodic arcjet thruster firings introduce contamination in the measured magnetic field, affecting data accuracy. The currently used correction matrix approach mitigates these effects but struggles with transient variations and residual errors. In this study, we present an alternative correction method using XGBoost, a machine learning algorithm, to correct arcjet-induced contamination in the GOES-17 magnetometer data using GOES-18 as ground truth. Using cross-satellite comparisons and supervised learning techniques, our model is effective in reducing artificial disturbances, especially non-linear variations. We found that the XGBoost method works better than the existing correction matrix approach for E and P components, while the correction matrix performs better for the N component. Although some limitations remain due to training data constraints, our results highlight the importance of machine learning to improve magnetometer data quality by recognizing and correcting complex satellite-driven artifacts. The collocation of GOES-17 and GOES-18 provided a unique opportunity for cross-satellite calibration and validation, and with a longer collocation period, the XGBoost method shows significant promise for better correction of operational data, emphasizing the need for such configurations in future satellite missions.