Optimizing Dike Source Parameters Via Supervised Learning: Insights into Remnant Magnetization and Ore Deposits

Abstract

The study of dikes is crucial in exploration and geodynamics, as it helps understand various complex magmatic and tectonic activities for dating geological events and ore deposit formations. Necessary dikes such as kimberlite and lamproite dikes can indicate the presence of diamond deposits as a marker for studying high crustal deformation over time. The precise interpretation of magnetic data to estimate the dimensional source parameters of the dike and the impact of remnant magnetization in magnetic anomaly remains unresolved. In this study, a Python-based supervised learning process is utilized to train a complex three-dimensional model using the stochastic gradient descent technique to optimize the dike parameters. Two complex synthetic dike models (simple dike and the other dike with sill) were generated using a 3D rectangular lamina of known parameters. A 10% white Gaussian noise was superimposed onto the synthetic data generated from forward modeling simulations to mimic real-world conditions. The primary aim of this research was to accurately assess the characteristics of the dike, which encompassed determining its length, depth, magnitude, and overall magnetization direction. This involved analyzing both the remnant and induced magnetization components. Initially, the methodology was validated on a simple vertical prism model and then applied to a more complex geological structure comprising a dike and sill. The results obtained for both scenarios were satisfactory. The modeling process produced a set of curves to investigate the relationship between the existing dike-sill model and the Earth's paleomagnetic direction. These curves represented parameters such as remanent magnetization intensity (r), remanent magnetization inclination (Ir), remanent magnetization declination (Dr), induced magnetization intensity (f), induced magnetization inclination (If), and induced magnetization declination (Df). This approach was then used to analyze magnetic anomaly data from the Galinge iron ore deposit in northwest China's Qinghai region. The analysis revealed the presence of a highly magnetized dike with remnant properties such as remnant magnetization inclination (Ir) and remnant magnetization declination (Dr) over the Galinge iron ore deposits. The identification of magnetite-rich dikes indicates that the ore deposit might have originated from a combination of magmatic and hydrothermal activities. This discovery sheds light on the intricate geological processes that have contributed to the formation of this region.

Keywords: Total magnetic field, Gradient-descent algorithm, 3D dike, Remnant and induced magnetization direction