Principal Component Analysis of Geomagnetic Observatory Data for the Past 50 Years

Smirnov and Egbert (2012) have developed methods for robust principal components analysis (PCA) for geomagnetic arrays with large data gaps and only partial overlap of site occupations. Here we discuss application of these methods to analyze hourly means from the global array of geomagnetic observatories, using all available data from the past 50 years. The frequency domain PCA approximates the global geomagnetic variation signal in the period bands from 6 hours to 100 days as a sum over a relatively small number of spatio-temporal modes. Our principal motivation for this analysis is to estimate the dominant coherent spatial modes, for use as input data for long period EM induction studies, but our approach may have other important applications for modeling of external geomagnetic variations, in both quite and active conditions. For example, at daily variation periods, approximately 10 modes are required to adequately model spatial structure at all latitudes. In contrast, climatological models (such as CM-4) have assumed spatial structure estimated from quite days, allowing for seasonal modulations--essentially two spatial modes. Our results thus suggest extensions of empirical models to include a broader range of latitudes and geomagnetic conditions. The PCA analysis also may be useful for studies of temporal variations, as we obtain essentially continuous temporal variations for each spatial mode over the past 50 years.