Magnetic Anomalies Induced by Sources with Unknown Geometry

This research paper addresses the challenges of interpreting magnetic anomalies arising from subsurface sources with unknown or complex geometries, a common issue in geophysical exploration when geological structures deviate from standard, idealized shapes. Traditional inversion methods often rely on geometric assumptions, leading to ambiguities when faced with natural, irregular sources. The study proposes an integrated, geometry-agnostic workflow combining nonparametric equivalent layer modeling, Bayesian Markov Chain Monte Carlo (MCMC) uncertainty quantification, and convolutional neural network (CNN) classification. Synthetic magnetic data generated from amorphous and fractal bodies serve as the basis for validating the method. The equivalent layer inversion reconstructs broad magnetization distributions without the need for explicit geometric constraints, while Bayesian MCMC provides probabilistic estimates and quantifies uncertainty in source parameters such as depth and magnetic moment. This probabilistic approach acknowledges the inherent non-uniqueness of the inverse problem. Additionally, a CNN trained on synthetic datasets can classify magnetic anomalies into source shape categories (bulky, elongated, irregular) with associated uncertainty, enhancing interpretive confidence in complex cases. The study further analyzes sensitivity to noise and magnetization direction variability, demonstrating that these factors critically affect both inversion accuracy and classification performance. Results from synthetic experiments underscore the importance of integrating uncertainty analysis and automated learning in early-stage exploration scenarios, especially when geological information is limited or ambiguous. The proposed framework is shown to enhance the reliability and objectivity of magnetic anomaly interpretation, with future directions involving multi-physics integration and scalable 3D analysis for large regional surveys.