Exploration for minerals in brownfields requires mining near large industrial infrastructure. High-voltage transmission lines, pipelines, and metallic fencing lead to electromagnetic interference, which masks weak subsurface signals (Beauvais, 2026; Rodriguez et al., 2006). Simulation models have found that the current from power lines causes high-amplitude signals that simulate geological fault zones or conductive minerals (Beauvais, 2026). Thus, geologists have to establish an organized approach to separating real subsurface signals from anthropogenic noise. The following is a summary of three main approaches: (a) physical signal analysis while collecting data, (b) filtering through computational means using artificial intelligence, and (c) geological confirmation through multiphysics approach.
Primary fields generate secondary electromagnetic fields in both geological materials and man-made structures. Cultural noise features sharp spatial damping, fixed power-grid frequency at 50 or 60 Hz, and brief high-amplitude transients (Rodriguez et al., 2006; Tietze et al., 2026). True sub-surface secondary fields display smooth time attenuation in time-domain EM systems and consistent phase tensor spatial relationships in magnetotelluric surveys (Tietze et al., 2026; Wang et al., 2024). Surveying crews reduce surface noise by positioning measuring stations away from power lines and maintaining strict noise levels (Beauvais, 2026). Hybrid surveying schemes, like a combination of full magnetotelluric stations and dense electric-field measurements, increase spatial sampling at limited-access locations (Tietze et al., 2026). Shallow screening tools, such as ground conductivity meters and metal detectors, are used to find shallow infrastructure before doing deeper inversions (Pyramid Geophysical Services, 2021).
Special signal processing is done on raw time-series electromagnetic data collected close to development to reduce noise. Digital notch filters are used to filter out grid hum and fundamental harmonics while keeping the rest of the bandwidths of the signals (Zhang et al., 2016). Remote referencing reduces uncorrelated local cultural noise by cross correlating local electric and magnetic signals with magnetic signals from remote locations (Tietze et al., 2026). Multi-transient electromagnetic surveys use cross correlation and selective stacking to extract subsurface impulse responses in disturbed environments (Zhang et al., 2016). Modern methods based on deep learning provide solutions for noise removal in complicated environments. DREMnet utilizes decoupled representation learning to divide the input signals into content factors that are the underground signals and context factors that are the environmental noise (Wang et al., 2024). Neural networks generate smooth attenuation curves across all receiver channels, recovering weak geological responses beneath motion and industrial interference (Wang et al., 2024).
The numerical inversion transforms the cleaned electromagnetic data into resistivity models in 2D and 3D. The conductive anomalies observed must represent realistic geometry in the subsurface such as dipping fault horizons or strata bound bodies and not something restricted by the surface (Rodríguez et al., 2006). The resistivity models from inversion are verified using the borehole induction logs and petrophysical results from laboratory core analysis (Rodríguez et al., 2006; Tietze et al., 2026). The petrophysical analysis is done to verify if the conductive model represents volcanic altered rocks, clay-bearing rocks, or mineralized shales (Rodríguez et al., 2006; Tietze et al., 2026). Reprocessing the seismic reflections over the electromagnetic anomalies reveals their correlation with fluid-filled fault structures and conduits (Tietze et al., 2026).
The consistent determination of the true subsurface target among cultural noise calls for a well-planned methodology. The process of effective site selection, reference collection, efficient filtering, use of artificial intelligence technology to disentangle, and multiphysics validation will generate a valid interpretation. Modern denoising techniques such as DREMnet have facilitated the generation of accurate geological models from the noisy industrial surveys (Wang et al., 2024). This process helps geologists to evaluate deep mineral, groundwater, and environmental targets in a highly developed brownfield area.
References
Beauvais, V. (2026). Anthropogenic Electromagnetic Noise from Transmission Lines in VLF-EM Surveys. Honors Theses, 4016, Western Michigan University.
Pyramid Geophysical Services. (2021). Geophysical Investigation to Identify Buried Infrastructure and Waste. Technical Case Study Report.
Rodriguez, B. D., Deszcz-Pan, M., & Sawyer, D. A. (2006). Electromagnetic Studies and Subsurface Mapping of Electrical Resistivity in the La Bajada Constriction Area, New Mexico. U.S. Geological Survey Professional Paper 1720-F, 121–163.
Tietze, K., Platz, A., Ritter, O., Weckmann, U., Holdstock, M., Rieger, P., Susin, V., Melo, A., Torremans, K., & Kiyan, D. (2026). Magnetotelluric Exploration of Zn Pb Mineralization in Stonepark, Irish Midlands. Geophysical Prospecting, 1–25.
Wang, S., Guo, M., Wang, X., Deng, F., Mao, L., Wang, B., & Gao, W. (2024). DREMnet: An Interpretable Denoising Framework for Semi-Airborne Transient Electromagnetic Signal. arXiv preprint arXiv:2401.00000, 1–12.
Zhang, W., Di, Q., & Lei, D. (2016). Estimation of the Earth Impulse Response of MTEM in Very Noisy Environment. Atlantis Press, 133–147.
