Astronomy and Astrophysics – Astronomy
Scientific paper
Dec 2000
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2000geoji.143..799w&link_type=abstract
Geophysical Journal International, Volume 143, Issue 3, pp. 799-808.
Astronomy and Astrophysics
Astronomy
12
Gaussian, Geomagnetic Field Model, Laplace, Robust Modelling
Scientific paper
We consider the consequences of a priori assumptions made about the probability density function used for historical geomagnetic data modelling. We describe a method, based on penalized maximum likelihood estimation, that can model data under any (for p>=1) p-norm measure of misfit, although only p=1 and p=2 are considered here. For p=2 it is implicitly assumed that the errors in the data originate from a Gaussian distribution, whereas p=1 assumes that the errors arise from a double exponential (or Laplace) distribution. We show that the geomagnetic main field models are consistent (in the sense that their residuals give excellent agreement with the assumed error distribution) only when modelled under the 1-norm measure of misfit. Least-squares (2-norm) methods that reject data at a pre-assigned residual level depend critically on the correct assignment of a priori error estimates, which are almost never known accurately. We demonstrate that main field models of the Earth's magnetic field generated by 1-norm modelling are robust, and therefore the ad hoc data rejection procedure, required by 2-norm modelling, should not be implemented.
Jackson Andrew
Walker Matthew R.
No associations
LandOfFree
Robust modelling of the Earth's magnetic field does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Robust modelling of the Earth's magnetic field, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Robust modelling of the Earth's magnetic field will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-1043193