Deconvolution of long‐core paleomagnetic data of Ocean Drilling Program by Akaike's Bayesian Information Criterion minimization

Abstract

Deconvolution of long‐core paleomagnetic data of the Ocean Drilling Program (ODP) was developed based on Bayesian statistics. Samples of deep‐sea sediments obtained by drilling with an advanced piston corer are subjected to pass‐through measurements by a cryogenic magnetometer aboard the ODP's research vessel (Joides Resolution) which provide continuous paleomagnetic records. Deconvolution of the magnetometer output was developed in order to obtain more detailed variations of the magnetization than permitted by the spatial resolution of the pickup coil (approximately 11 cm). The magnetization vector was modeled as a smoothly changing sequence parametrized by L2 norm of second‐order difference and then deconvolved using a matrix calculation as a smoothness constrained least squares method. The optimum smoothness was obtained by minimizing Akaike's Bayesian Information Criterion which is a measure of the logarithm of the likelihood. This deconvolution scheme was applied to pass‐through data artificially produced by convolving synthetic reversal record and adding a Gaussian noise. The amplitude of the noise was estimated and the variations shorter than the spatial resolution of the sensor were obtained. The deconvolution also was applied to the real pass‐through data from Holes 769A and 769B measured at intervals of 5 mm. The deconvolution revealed variations of the magnetizations with a maximum spatial resolution of about 2 cm. The magnetizations after deconvolution were in good agreement with magnetization values measured separately on cube samples taken at intervals of 5 or 10 cm.