GNGTS 2015 - Atti del 34° Convegno Nazionale

Results. The results of the inversions were compared between themselves and with those obtained from a test on field data carried out in Piana degli Albanesi in an area characterized by high heterogeneity. For the inversion of both simulated and experimental data were used the same algorithms and parameters. Results of model #2 are shown as example (Fig. 1), which highlight how the resolution strongly depends on the choice of the data set and how the information recovery decreases with the target depth. Results obtained by noisy data (Fig. 1b) are poorer of information than noise-free data (Fig. 1a) but the gap between results decreases considering more complex and time requiring data sets. In these examples the comparison is made between data set 1 ( a max =1; n max =35) and data set 5 ( a max =5; n max =6) for DD, PD and WS, and between data set 1 ( e=1-8; c = 1) and data set 6 ( e=1-8; c = 6) for MG. Generally, as the data set number increases, the model misfit (Fig. 1 d) decreases in correspondence of the targets and overall the shapes are better resolved and artifacts are more limited. This can be explained by the comparison of the correspondent images of relative model sensitivity that show a more uniform distribution and higher values at greater depths (Fig. 1e). The comparison of the data misfit of inversions for each data set of DD, PD, WS and MG arrays obviously shows sensibly higher values for noisy and field data than for noise-free data (Fig. 2a). Moreover, there is a similar trend for both field and noisy data. As the complexity and time required of data sets increases (data set number increasing from 1 to 8) the RMS% decreases. For DD it starts from high values and decreases rapidly, for WS and PD it starts from lower values and decreases more slowly, instead practically it does not vary for MG. The average model misfit (Fig. 2b) is highly depending on resistivity values of the subsoil. Generally, the trends show a decreasing parameter as data set number increases (very strong for the MG) that probably is related to the increase of number of current dipoles. These trends generally reverse after the data set n. 5 or 6. The average sensitivity (Fig. 2c) is few influenced by the presence of noise and its trend is very similar between noisy and field data because probably the only difference is due to the different resistivity trend in the subsoil. WS data sets show the highest values and DD ones the lowest. In every array, sensitivity increases with data set number, very quickly for MG. For models #1 and #2 (resistive and conductive targets) we also estimated the trends of the normalized values of the average sensitivity, on windows of the same size of the target, as a function of the depth of the target. Results for model #2 are shown in Fig. 3. Colored zones show the areas of variation of the parameter for each array, from the lower data set number (dotted line) to the highest one (solid line). Trends show the exponential decrease of the sensitivity values as the target depth increases. We note that the slope of the curves decreases considering more complex data sets. This means that not only these latter have a trend for higher sensitivity in every area of the section, but also that they resolve better the deeper cells of the models. Ultimately, the best sensitivity is obtained for WS array, and the worst for DD array. Fig. 2 – Parameters that express the quality of inversions depending on the data set number: a) data misfit (RMS error %); b) average model misfit; c) average relative model sensitivity. 104 GNGTS 2015 S essione 3.2