GNGTS 2018 - 37° Convegno Nazionale

738 GNGTS 2018 S essione 3.3 resistivity (or phase) value estimated for the previous period. In other words, for each frequency k , the “best” resistivity (or phase) value, k , is the one that satisfies the following relation: min | Log ( ρ i,k −1 ) − Log ( ρ i, k ) | , (1) where i is the index of the ith resistivity obtained for the specific frequency k. As an example, Fig. 2 shows the results of the application of the selection criterion to synthetic MT data affected by Gaussian white random noise (SNR=0.83) in a window of length equals to one-fifth of the entire time series length. As it can be seen, most of the resistivities ρ xy and ρ yx and corresponding phases, retrieved after the application of the filtering procedure (diamond symbols), fit well with the synthetic noise-free curves (blue straight lines). Similar results have been found by applying noise with a chi−square distribution as well as by varying the noise window length affecting the original synthetic dataset. Finally, the proposed filtering procedure has been applied to real MT data acquired in the Yellowstone caldera (Yellowstone National Park, Wyoming, United States). In Fig. 3, a comparison between apparent resistivities and phases retrieved with and without the application of the SOM filtering procedure is shown. As it can be seen in Fig. 3a, the curves estimated after the filtering have a generally smooth shape. In particular, the filter clearly improves the quality of the apparent resistivity estimates in the range of periods between 3 s and 20 s. Indeed, in this range, the apparent resistivities obtained without the clustering filter are very scattered and definitely unreliable, because, as mentioned above, the apparent resistivity variations generally show a smooth pattern. The filter application improves also the estimates of the yx component (Fig. 3b), particularly in the range of periods between 0.05 s and 0.3 s, where the resistivity values obtained without the SOM filter tend to decrease faster than those obtained after filtering. This behaviour is likely due to the presence of noise, as it is suggested by the occurrence of the two large clusters below the red crosses, which indicate the ρ yx values obtained without the SOM filter. Indeed, these clusters refer to portions of signal that generate apparent resistivities lower than those obtained in the remaining part of the signal. Fig. 3 - Comparison between the apparent resistivity and phases estimates retrieved with (circles) and without (cross) the application of the clustering procedure for both xy (a) and yx (a) components.