[S-EM14] Electric, magnetic and electromagnetic survey technologies and scientific achievements

[SEM14-17] Simultaneous resistivity imaging of airborne electromagnetic data based on deep learning

*Sihong Wu¹, Qinghua Huang¹, Li Zhao¹ (1.Department of Geophysics, School of Earth and Space Sciences, Peking University, Beijing, China)

Keywords:Airborne electromagnetics, resistivity imaging, subsurface structure, deep learning, electrical property

The airborne electromagnetic (AEM) method is an effective means for subsurface electrical structure imaging with a depth of investigation of several hundred meters. With high acquisition efficiency and terrain adaptability, the AEM technique has seen extensive applications including mineral exploration, groundwater investigation, environmental monitoring and geological mapping. However, due to the vast area of coverage and dense spatial sampling, huge volumes of AEM data present a severe computational challenge for rapid resistivity imaging. Inspired by Google’s neural machine translation system, we develop a deep learning-based inversion system to directly translate AEM data into the resistivity model with the consideration of transmitter altitude. Synthetic tests demonstrate that our proposed inversion system yields more robust results than the conventional Gauss-Newton algorithm in both noise-free and noise-added conditions, while the inversion efficiency is significantly improved. Applied to the field dataset acquired by the U.S. Geological Survey in the Leach Lake Basin in California, our system delivers inversion results of more than 740,000 AEM soundings in a few seconds on a common PC with strong noise robustness. The inverted resistivity images agree well with previous knowledge of local geological environment, delineating the geometries of faults, surrounding mountains and the lake. Our inversion system can support real-time resistivity imaging in AEM surveys and facilitate resource exploration and tectonic studies.