Fast and practical inversion for semi-airborne transient electromagnetic data based on supervised descent learning technique

• Published in: Journal of applied geophysics Vol. 241; p. 105806
• Main Authors: Rao, Liting, Wu, Xin, Dang, Bo, Gao, Jianshen
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.10.2025
• Subjects: Gauss Newton method; Inversion; Machine learning; Semi-airborne transient electromagnetic; Supervised descent method; Semi-airborne transient electromagnetic; Supervised descent method; Inversion; Machine learning; Gauss Newton method
• ISSN: 0926-9851
• DOI: 10.1016/j.jappgeo.2025.105806

Semi-airborne transient electromagnetic method (SATEM) is an innovative geophysical exploration technique with high efficiency, low cost and strong adaptability to complex terrains, particularly suitable for China's diverse geological conditions. Due to the complex electromagnetic detection theory and large volume of observational data associated with SATEM, there is a critical demand for reliable inversion method with higher efficiency to obtain subsurface geoelectric structure. Therefore, this study introduces supervised descent learning technique in machine learning, integrating it with Gauss Newton method to form a fast and practical inversion scheme for SATEM data. The proposed inversion framework consists of three stages: offline training, online prediction and model modification. During the offline training, the average descending direction of implicit model characteristics is obtained by integrating prior information into training set. In the online prediction, the parameters of geoelectric model are reconstructed rapidly by using physical modeling function and descending direction obtained by training. Subsequently, the predicted models are modified by Gauss Newton method to minimize data residuals further. Synthetic and field data examples demonstrate that the hybrid inversion method effectively combines the strengths of both techniques. Specifically, the inversion efficiency significantly surpasses that achieved by employing the Gauss Newton method alone, while the accuracy of the inversion results exceeds those obtained solely through the supervised descent method. Moreover, the model modification process ensures the reliability of the inversion scheme by reprocessing any failed predicted models, thereby enhancing the overall robustness and applicability of the method. •SATEM requires more efficient and reliable inversion method due to complex electromagnetic theory and large data volumes.•A hybrid inversion scheme combines supervised descent method and Gauss Newton method for fast and reliable SATEM imaging.•The scheme includes offline training, online prediction, and model modification to boost efficiency and reduce residuals.•Tests show the hybrid method improves speed over Gauss Newton method and accuracy over supervised descent method alone.