Hybrid Luus–Jaakola and Levenberg–Marquardt inversion for magnetic anomalies over 2D fault-like structures

Abstract

This study presents a shortened summary focusing on the hybrid Luus–Jaakola–Levenberg–Marquardt inversion and its superior performance on both synthetic and Perth Basin data.

In this study, a hybrid approach is introduced and evaluated for estimating the geometric and physical parameters of 2D fault structures from magnetic data. This method integrates the global search capability of the Luus–Jaakola (LJ) algorithm with the rapid convergence of the Levenberg–Marquardt (LM) algorithm, enhanced by two components: a multi-start strategy and a controlled perturbation scheme. This combination overcomes the limitations of initial value dependency and the instability of certain parameters, particularly the effective inclination angle (). The performance of the proposed method was assessed using both synthetic and real-world data. The results demonstrate that the hybrid approach provides higher accuracy and stability in parameter estimation compared to each individual algorithm, especially under noisy conditions. Perturbation experiments and sensitivity analysis revealed a clear parameter hierarchy: depth parameters exhibit the strongest amplitude influence (approximately 200–267% variation), followed by the amplitude coefficient (approximately 85%) and the effective magnetization angle (approximately 63%), while the fault dip angle shows minimal amplitude sensitivity (approximately 5%). The analysis of the effective magnetization angle demonstrated its critical role in identifying optimal starting regions for stable estimations, with the location of these regions contingent on the state of other parameters. In the real-data application, the hybrid method yielded parameters consistent with plausible geological models of the area. The findings of this study offer a practical framework for improving magnetic data inversion, particularly in noise-contaminated environments.

In this study, a hybrid approach combining the Luus–Jaakola (LJ) and Levenberg–Marquardt (LM) algorithms is introduced for estimating the geometric and physical parameters of 2D fault structures from magnetic data. The method leverages the global search capability of LJ and the rapid convergence of LM, while incorporating a multi-start strategy and controlled perturbation to overcome initial value dependency and instability in critical parameters such as the effective inclination angle (α). The performance was assessed using both synthetic and real-world data. For synthetic data with 15% noise, the hybrid method achieved an RMS of 4.45 nT, significantly outperforming standalone LJ (17.44 nT) and LM (4.60 nT), while reducing the estimation error of α from 25.40% to 9.39% through controlled perturbation. When applied to real magnetic data from the Perth Basin, Australia, the hybrid approach yielded an RMS of 1.33 nT compared to 2.16 nT (LJ) and 10.61 nT (LM), producing parameters consistent with the plausible geological model of the area. These findings demonstrate that the hybrid framework offers superior accuracy and stability for magnetic data inversion, particularly under noisy conditions.