Publications

Continuous geomagnetic field models spanning several millennia have recently been developed using various selections of archeo- and paleomagnetic data and their inferred ages. In each case the geographic and temporal distribution of available data is far from uniform and both the magnetic data and ages have large uncertainties. We estimate error bars for both the models and their predictions using two statistical resampling techniques and a combination thereof. First, we used what we call the spatial and temporal (ST) bootstrap yielding different spatial and temporal distributions taken randomly from the original dataset. Second, we kept the original (temporal and spatial) distribution of data, but varied each datum randomly within the expected distributions of uncertainty in both the magnetic observation and assigned ages. We call this the magnetic/age (MA) Bootstrap. We produced a large number of models based on resampled data using each of the ST and MA bootstrap methods and then obtain standard deviations for both global model coefficients and predictions of field components. The ST and MA methods yield model uncertainties of the same order of magnitude. A sequential combination of MA and ST resampling takes into account the influence of uncertainties in both magnetic elements and ages as well as the unsatisfactory data distribution. We present global and regional results from this analysis and compare the uncertainties obtained from model predictions to the assigned data errors. The uncertainties obtained for magnetic field elements vary depending on whether they are obtained by error propagation from uncertainties in the model coefficients or by computing the standard error in the individual element predictions for all resampled models. The propagated uncertainties do not currently allow for covariance among the coefficients. Hence, they can be too large in some geographic regions and time intervals with good data coverage. Individual element uncertainty predictions incorporate any such covariance automatically, and can in principle better accommodate regional variations in model accuracy. DE: 1503 MN: 2008 Fall Meeting

---

Korte, M.; Constable, C. G.; Donadini, F. (2008): Uncertainty Estimates for Millennial Scale Geomagnetic Field Models. AGU 2008 Fall Meeting (San Francisco, USA 2008).