[M-GI29] Data-driven geosciences

[MGI29-P06] Conversion of grid search result to probability distribution – application for source estimation problems of geomagnetic anomaly

*Hiroshi Ichihara^1,2, Tatsu Kuwatani¹ (1.Research Institute for Marine Geodynamics, Japan Agency for Marine-Earth Science and Technology, 2.Graduate school of environmental studies, Nagoya University)

Keywords:grid search, probability distribution, magnetic anomaly, marginal distribution

Statistical evaluation of estimated parameters is essential for geophysical problems. Grid search is widely used in these problem to determine model parameters explaining observed data. Residual between observed data and model predicted response, such as root mean squired misfit (RMSm), is usually used as a criteria of model goodness. However, RMSm is not a direct criterion to evaluate the model parameters. In this study, we proposed a method to convert distribution of RMSm obtained from the grid search into possibility distribution to evaluate the result statically. We applied the proposed method for synthetic and real geomagnetic anomaly data sets to search source position and magnitude of magnetic moment. The test for synthetic data sets indicates availability of the proposed method including ill-posed problems. The real data sets consist of time variations of magnetic field around Nishinoshima volcano. The analyses of the real data constrained the source of magnetic change in narrowed area. We also emphasize that marginal probability distribution constrained confidential limit of each parameter, that the RMSm hardly evaluate. The method is convenient and thus can directly be applied for the previous results of grid search. The proposed method should be especially useful for the problems to search source location of earthquake, surface geodetic deformation, magnetic change and their joint analyses in active volcanic areas.