Item

Abstract

Recent applications of ANN, in geophysics include, e.g., the prediction of site amplification response (Boudghene Stambouli et al. 2017; Salameh et al. 2017; Bergamo et al. 2021) and seismic site classification (Zhao et al. 2006; Ghasemi et al. 2009; Ji et al. 2021; Díaz et al. 2022). The identification of significant peaks on microtremor H/V curves is traditionally assessed by means of the popular SESAME (2004) criteria, which, however, cannot discriminate between the stratigraphic and non-stratigraphic nature of them. Non-stratigraphic peaks are typically caused by machineries acting in the proximity of the recording station or result as an effect of vibrations induced from large nearby structures. Low frequency artefacts perturb the noise wavefield up to large distances from the source. The frequency of anthropogenic peak can also coincide or be very close to the frequency of a peak of stratigraphic origin. H/V curves are easily affected by artefacts that it is mandatory to discriminate them, before attempting any interpretation of the H/V curve itself. Given the worldwide extended use of H/V methods in seismic microzonation, being altered about the possible presence of non-stratigraphic peaks is important to avoid misclassifications of soil resonance frequencies or wrong stratigraphic interpretations when the H/V method is combined with, e.g., surface wave multichannel methods to achieve Vs profiles. Given the limited capability of standard techniques in this type of signal classification, we decided to explore the performance of different NNs in this sorting problem consisting in discriminating between stratigraphic and non-stratigraphic features in microtremor signals.