GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.1 GNGTS 2024 Results The feld test produced the two EMI inversions shown in Figure 2, showcasing: 1) the unconstrained outcome (Figure 2a-f, middle column), in which the solutons are obtained by simply minimizing Eq. 1 within the ensemble , and 2) the constrained inversions (Figure 2a-f, rightmost column) that incorporates informaton from both the ensemble and the 2D ERT secton. The unconstrained result tends to be more resistve than the constrained inversions, illustratng the impact of ancillary informaton from the ERT secton on spatal consistency and on the reducton of erratc variatons between adjacent 1D models. Figure 2 ofers a comparatve analysis at diferent depths, in which horizontal slices of GPR refectons' envelope can be checked against the corresponding slices of EMI volumes obtained with and without the ERT constraint. This comparison highlights the efectveness of schemes incorporatng ancillary informaton, as seen in the detecton of a conductve anomaly related to a pipeline, partcularly visible in Figure 2c-d at the depth of 2.1 m to 2.6 m. Diferently from existng spatally constrained schemes, the present approach does not rely (vertcally) on smooth/sharp regularizatons (Klose et al. 2022; Klose et al.2023); stll vertcal consistency is enforced by selectng the model in the ensemble that, by constructon, is generated in accordance with our expectatons (Zaru et al. 2023). In summary, the proposed inversion scheme, utlizing an arbitrarily complex prior distributon and ancillary informaton, generates a unique EMI inversion result. The ability to formalize informaton through prior distributon samples and reference models proves crucial for achieving spatally consistent and reliable reconstructons. In additon, the method efectvely integrates ancillary data, presentng a robust soluton that aligns with geological descriptons, GPR refectons, and borehole logs. Conclusion In conclusion, this research presents some of the results discussed in Zaru et al. 2023 which introduces an inversion scheme for EMI data, employing a realistc prior distributon for vertcal conditoning and incorporatng ancillary informaton from an ERT cross-sectons for lateral consistency in the fnal (pseudo-)3D resistvity volume. Our results highlight the severe non- uniqueness of EMI data inversion, emphasizing the importance of formalizing informaton through prior distributon samples and reference models. The proposed constrained scheme, integratng ancillary observatons ab inito, demonstrate their efectveness in addressing the ill-posedness of the original problem. Comparisons with the GPR cross-sectons validate the proposed approach's capability to detect lithological interfaces and resistvity contrasts with decent resoluton, considering the amount of informaton provided by the EMI survey is contained in just 3 frequencies. The computatonal efciency of the strategy, coupled with its ability to provide a single reliable model, opens avenues for future exploraton, including the potental implementaton of a probabilistc soluton to assess feature probabilites across diferent realizatons of the posterior distributon. Overall, the proposed inversion strategy enhances the reconstructon of 3D resistvity volumes from EMI measurements, ofering improved agreement with independent geophysical observatons. P P P