GNGTS 2021 - Atti del 39° Convegno Nazionale

389 GNGTS 2021 S essione 3.1 Fig. 2a demonstrates that the NN inversion can infer reasonable 1D models with responses fitting the observation within 5%. When the NN result is compared with the corresponding 1D deterministic inversion in Fig. 2b, the “traditional” deterministic inversion with vertical and lateral smooth constrained is often superior in fitting the data (the “deterministic” data misfit is generally below 2%, as it is visible in Fig. 2b). However, the larger discrepancies between the two solutions occur where the data fitting of the deterministic inversion is larger (e.g., 1850 to 2400 m) and/or in areas characterized by high resistivities; so, the survey’s portions where also the deterministic inversion has difficulties reproducing the observations and characterized by relatively high resistivities are those where the differences with the NN solution are more pronounced. This is in agreement with the fact that, in general, ATEM methods have difficulties in accurately distinguish between different high resistivity values (Bai et al 2020). Discussion The clear advantages of the deterministic inversion come at a price: the NN inversion takes approximately 24 s to invert the entire field example dataset (consisting of around 14,500 soundings with 54 time gates each) by using a standard laptop (equipped with an Intel Core i5- 8250U processor), whereas hours (so, at least, three orders of magnitude more) are necessary to perform the same task by using the 1D deterministic approach and a 64-CPU server. So, the proposed NN approach has at least a few pros: 1. The extreme computational improvements brought by the NN approach potentially paves the road to the optimization of the survey design while the acquisition of the Fig. 1 - (a) The verification dataset. The individual 1D conductivity models of these sections have been used to generate the noise-free synthetic data to be subsequently inverted with the NN discussed in the section “Methodologies”. (b) The inversion results obtained by applying the NN on the data generated by the conductive models in panel (a). The data misfit between the calculated and the original measurements is shown for each individual 1D model location as a red dot (the corresponding axis is on the right in red).