Biased residuals of core flow models from satellite-derived 'virtual observatories'

Large satellite vector datasets of the Earth’s magnetic field have become available in recent years. Standard magnetic field models of the internal field are generated by parameterising a small subset of these data through a least-squares spherical harmonic representation. An alternative approach is to create a set of ‘virtual observatories’ (VO) in space, mimicking the operation of fixed ground-based observatories. We derive VO datasets from both CHAMP and Ørsted satellite measurements. We calculate and directly invert the secular variation (SV) from these VO datasets, to infer flow along the core-mantle boundary using an L1 (or Laplacian) norm method (to reduce the effect of outliers). By examining the residuals from the flow models, we find temporally and spatially varying biases and patterns in the vector components.We investigate potential causes for these patterns, for example, by selecting night-side only vector data and applying corrections to the input data, using external and toroidal fields calculated by Comprehensive Model 4 (CM4).We test the effect of a number of data selection and correction criteria and find evidence for influence from fields both internal and external to the satellite, orbital configuration and effects from the method of binning data to produce VO. The use of CM4 to correct the satellite data before calculating the VO SV grid removes a strong bias from external sources but, on average, does not greatly improve the fit of the flow to the data. We conclude that the best fit of the flows to the data is obtained using satellite night-side only data to generate VO. We suggest that, despite best efforts, external fields effects are not completely removed from SV data and hence create unrealistic secular acceleration.