GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.3 GNGTS 2024 Bayesian imaging method for towed transient electromagnetc data using probabilistc neural networks J. Chen 1 , G. Fiandaca 1 1 The EEM Team for Hydro & eXploraton, Department of Earth Sciences "Ardito Desio", University of Milano, Milano (Italy). Abstract The towed transient electromagnetc (tTEM) can conduct efcient geological surveys on the near- surface of about 100 meters underground. Exploring fast and accurate on-site interpretaton strategies for electromagnetc data is crucial for geoscientsts and engineers to make high-quality decisions and further enhance the applicability of this technology. In this study, we designed a composite probabilistc neural network (cPNN) structure that can simultaneously provide deterministc imaging and Bayesian probabilistc imaging results, providing a comprehensive interpretaton of the observed data and estmatng its uncertainty. We verifed this neural network with nearly 200 km tTEM survey data collected on the Iseo Lake in Italy. The results show that the cPNN network can efectvely characterize the locaton of aquifers and underground clay layers, and the imaging results are consistent with conventonal inversion and sonar bathymetry data. Furthermore, since the cPNN network can obtain the Gaussian distributon of underground resistvity, we can estmate the depth of investgaton (DOI) of the imaging results and extract smooth models from the Gaussian distributon. The cPNN network can obtain approximate Bayesian inversion results for large tTEM dataset in only tens of seconds on a laptop, which has good practcal value. I Introducton The towed transient electromagnetc (tTEM) is a new detecton technology improved on the basis of the ground-based transient electromagnetc (TEM) method in recent years. This technology enables mobile geological surveys with all-terrain vehicles (Auken et al., 2019) or boats (Maurya et al., 2022), similar to the ground-based version of airborne transient electromagnetc (ATEM) technology (e.g. Silvestri et al., 2019). The survey speed of the tTEM system can reach nearly 20 kilometers per hour (Grombacher et al., 2021), and it can conduct efcient and low-cost geological surveys in large survey areas, with a typical depth of investgaton (DOI) ranging from 80 to 100 meters. At present, the tTEM has been successfully applied in various near-surface geological exploraton felds such as agricultural ecological management, groundwater hydrology system mapping