Premium
Applying the Burr Type XII Distribution to Decompose Remanent Magnetization Curves
Author(s) -
Zhao Xiangyu,
Fujii Masakazu,
Suganuma Yusuke,
Zhao Xiang,
Jiang Zhaoxia
Publication year - 2018
Publication title -
journal of geophysical research: solid earth
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.983
H-Index - 232
eISSN - 2169-9356
pISSN - 2169-9313
DOI - 10.1029/2018jb016082
Subject(s) - log normal distribution , spurious relationship , mathematics , gaussian , normal distribution , statistical physics , skew , remanence , distribution (mathematics) , statistics , magnetization , mathematical analysis , computer science , physics , telecommunications , quantum mechanics , magnetic field
Discriminating magnetic minerals of different origins in natural samples is useful to reveal their associated geological and environmental processes, which can be achieved by the analysis of remanent magnetization curves. The analysis relies on the choice of the model distribution to unmix magnetic components. Three model distributions were proposed in past studies, namely, the lognormal, skew normal, and skewed generalized Gaussian distributions, which are related to the normal distribution. In this study, the Burr type XII distribution is tested and compared with existing model distributions. An automated protocol is proposed to assign parameters necessary to initiate the component analysis, which improves the efficiency and objectivity. Results show that the new model distribution exhibits similar flexibility to the skew normal and skewed generalized Gaussian distributions in approximating skewed coercivity distributions and can fit end‐member components better than the commonly used lognormal distribution. We demonstrate that the component analysis is sensitive to model distribution as well as measurement noise. As a consequence, the decomposition is subject to bias that is hard to identify due to the lack of ground‐truth data. It is therefore recommended to compare results derived from various model distributions to identify spurious components.