Machine learning-based calibration of the GOCE satellite platform magnetometers

Abstract

AbstractAdditional datasets from space-based observations of the Earth’s magnetic field are of high value to space physics and geomagnetism. The use of platform magnetometers from non-dedicated satellites has recently successfully provided additional spatial and temporal coverage of the magnetic field. The Gravity and steady-state Ocean Circulation Explorer (GOCE) mission was launched in March 2009 and ended in November 2013 with the purpose of measuring the Earth’s gravity field. It also carried three platform magnetometers onboard. Careful calibration of the platform magnetometers can remove artificial disturbances caused by other satellite payload systems, improving the quality of the measurements. In this work, a machine learning-based approach is presented that uses neural networks to achieve a calibration that can incorporate a variety of collected information about the satellite system. The evaluation has shown that the approach is able to significantly reduce the calibration residual with a mean absolute residual of about 6.47nT for low- and mid-latitudes. In addition, the calibrated platform magnetometer data can be used for reconstructing the lithospheric field, due to the low altitude of the mission, and also observing other magnetic phenomena such as geomagnetic storms. Furthermore, the inclusion of the calibrated platform magnetometer data also allows improvement of geomagnetic field models. The calibrated dataset is published alongside this work. Graphical Abstract