Abstract
This paper discusses the problems which arise in producing a numerical (spherical harmonic) model of the Earth's main magnetic field for a particular epoch when using methods based on least squares. The input data are contaminated by ionospheric, magnetospheric, and crustal fields, and are not simultaneous in time. Some selection of the data is usually needed, and relative weights have to be assigned. Standard least-squares fitting, the problem of truncation, and the modifications used to avoid this and some other problems are outlined. The problem of accurately and unambiguously estimating the uncertainty in the resultant model is emphasised.
