Separation of magnetic-field data using the differential Markov random-field (DMRF) approach

Abstract

In this paper, the differential Markov random-field (DMRF) method is introduced and applied to the magnetic anomaly separation problem, in which residual anomalies are separated from a regional field. The DMRF method is an unsupervised statistical model-based learning approach that does not require prior knowledge. A data-adaptive program, based on the evaluation of noise and superimposed effects of various geologic structures, is presented by considering a statistical maximum a posteriori (MAP) criterion. The aim of our method is to capture the intrinsic properties of geologic structures and then to identify and hence understand the behavior of the observed magnetic-anomaly map. The magnetic-anomaly map is modeled using a 2D matrix. In the DMRF approach, each pixel of the matrix is evaluated considering neighboring pixels. In synthetic models, anomalies of magnetic dipoles are tested for different depths, orientation angles, and lengths. The DMRF method also is applied to the vertical magnetic-anomaly map of the Sivas-Divrigi region in Turkey, which contains the Dumluca iron ore reserves. Shallow reserves are detected clearly by the DMRF method, proving greater accuracy than classical filtering techniques. The results are confirmed by Technical Ore Research of Turkey (MTA) drilling reports.