Isolating internal secular variation in Geomagnetic Virtual Observatory time series using Principle Component Analysis

Geomagnetic Virtual Observatories (GVOs) are a method for processing magnetic satellite data in order to simulate the observed behaviour of the geomagnetic field at a fixed location. As low-Earth orbit satellites move at around 8 km/s and have an infrequent re-visit time to the same location, a trade-off must be made between spatial and temporal coverage, typically averaging over half the local time orbit precession period, within a radius of influence of 700 km. The annual differences (secular variation, SV) of residuals between GVO time series data and an internal field model at a single GVO location will be strongly correlated with its neighbours due to the influence of large-scale external field sources and the effect of local time precession of the satellite orbit. Using Principal Component Analysis we identify and remove signals related to these noise sources to better resolve internal field variations on sub-annual timescales.

We apply our methodology to global grids of monthly GVOs for the Ørsted, CHAMP, CryoSat-2 and Swarm missions, covering the past two decades. We identify common principle components representing orbit precession rate dependent local time biases, and major external field sources, for all satellites. We find that the analysis is enhanced by focussing on regions of geomagnetic latitude where different external field sources dominate, identifying distinct influences in polar, auroral and low-to-mid latitude regions. Annual differences are traditionally used to calculate SV so as to remove annual and semi-annual external field signals, but these signals can be re-introduced if our corrected SV is re-integrated. We find that by representing secular variation with monthly first differences, rather than annual differences, we can identify and remove annual and semi-annual external field variations from the SV, which then improves the use of re-integrated main field GVO time series. By better accounting for contaminating signals from correlated external fields and aliasing, we are able to produce a global grid of GVO time series which better represents internal secular variation at monthly time resolution.