GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.1 GNGTS 2024 from Geophex (Haoping & Won, 2003). FDEM measurements were collected at the following frequencies: 1025, 1525, 2875, 5825, 7775, 12775, 15325, 25525, 36225, 63025, and 80225 Hz. A total of 153,621 soundings were acquired. To model uncertaintes inherent in the petrophysical propertes of the peatland structure, a probabilistc inversion procedure was employed. The procedure utlized an independent extended Metropolis algorithm - a variant of the conventonal extended Metropolis algorithm proposed by Mosegaard and Tarantola (1995). In our specifc implementaton, an infnitely long step-length was selected; this implies that each new model proposal from the prior distributon represented a completely independent realizaton. Based on our expectaton about the target, we created a prior ensemble of 10 5 1D models consistng of 200 layers (each 0.1 m thick),. The synthetc set of 1D models is realized by establishing a prior distributon based on the assumpton of two distnct lithologies: peat and clay. Each geological model is considered made of three lithological units, allowing for varied confguratons without imposing specifc vertcal arrangements, with interface depths and lithology assignments randomly generated. Resistvity values for each layer are drawn from distributons representng expected resistvity ranges for peat and clay, followed by a smoothing process to introduce vertcal spatal coherence. The resistvity ranges associated with each category are consistent with relevant literature examples for peat and clay soils. For each sounding locaton, the response of each prior realizaton was compared
 against the observed data using the conditonal probability (and assuming normally distributed noise in the measurements): , (1) where is a normalizaton factor, takes into account the data covariance, and ofen selected such that , with each component of the vector being the standard deviaton of the corresponding data element. Notably, this inversion scheme was implemented using the SIPPI toolbox (Hansen et al., 2013a; Hansen et al., 2013b). Results Figure 1 shows the horizontal slices at the depths of 3.0 m and 7.0 m, depictng the realizaton of the posterior distributon with maximum likelihood and the probabilites of encountering peat and clay. Peat is predominant in the northern-central area, extending from the surface to depths exceeding 7.0 m, with a shallower deposit in the southern part. The spatal coherence of the results is noteworthy, considering that each sounding was inverted separately without spatal constraints. Verifcaton against boreholes reveals the geophysical reconstructon aligns well with the morphology of the peat unit (Fig. 2). Some mismatches can be potentally atributed to 3D efects (Bai et al. 2021) or actual distance from the inverted and the borehole locatons (for instance, borehole B31* is approximately 4 m away from the inverted secton). Moreover, boreholes were originally drilled aiming at reaching the top of the clay layer but that was not always possible. In this respect, only those actually hitng the top of the clay unit are denoted with an asterisk, such as B31*. The maximum likelihood model exhibits spatal coherence without explicit regularizaton, emphasizing the impact of prior informaton on stability. Comparisons in terms of data ftng show the maximum likelihood model closely matches the observatons, exhibitng a chi-squared value F ( m ) d p ( m ) = k d exp ( − 1 2 ( d − F ( m ) ) T C −1 d ( d − F ( m ) ) ) k d C d diag ( σ d ) −2 = C −1 d σ d