Bayesian Estimation of Paleointensity: A Criteria Free Approach for Dealing with Data
Abstract
Paleointensity estimates from Thellier type experiments are typically made by rejecting all specimen interpretations that fail an arbitrary set of selection criteria whose purpose is to exclude inaccurate intensity measurements. There are no realistic physical models that describe the "non-ideal" behaviors these criteria are designed to identify and as such selection criteria give inconsistent results. Criteria also frequently discard a majority of the available data, leading to potential biases. Hence, there is no consensus on which set of selection criteria to use. We present a new method for interpreting paleointensity data from Thellier-type experiments. We start from the assumption that the data should follow a straight line relating demagnetization of the original remanence versus that gained when cooled in a laboratory field in a step-wise fashion. We use a Bayesian approach to find a probability distribution of slopes for a given specimen, then normalized to unity. Each distribution is then stacked at the site level to obtain a probability density function for intensity at a site level. The advantages of this method are numerous. Firstly, individual uncertainties for specimens can be estimated allowing specimens that have greater uncertainties (wider probability distributions) to be downweighted when calculating a site level intensity distribution, as opposed to being arbitrarily excluded. Also, using a probability density function to estimate site level intensity means that the significance of a particular estimate can be determined. For example, at sites where multiple intensities were recorded, it is possible to assess the probability that each of these modes contain the true value.
We apply our Bayesian estimation approach (BEST) to a compilation of 33 sites containing 647 specimens (updated from doi:10.1002/2013GC005135 and doi:10.1002/2016GC006307) where the ancient field was known (either from the site's IGRF field or a known lab field). For 93.9% of these sites, the IGRF value falls within the 95% credibility interval and 69.7% fall within the 67% interval, demonstrating the robustness of the technique. The accuracy of this method can be further improved by including other idealized models for Arai plots, such as those for specimens that exhibit multiple components of magnetization.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMGP11A..02C
- Keywords:
-
- 1521 Paleointensity;
- GEOMAGNETISM AND PALEOMAGNETISMDE: 1527 Paleomagnetism applied to geologic processes;
- GEOMAGNETISM AND PALEOMAGNETISMDE: 1540 Rock and mineral magnetism;
- GEOMAGNETISM AND PALEOMAGNETISMDE: 1595 Planetary magnetism: all frequencies and wavelengths;
- GEOMAGNETISM AND PALEOMAGNETISM