Full text loading...
Single station ambient noise measurements are widely used in both seismic risk assessments and stratigraphic reconstructions of the subsurface. In mining research environments, it is common to come across surveys of several hundreds of measurements aiming at preliminary bedrock mapping. With these premises, having simple tools to use for immediate evaluation of the quality of the collected measurements and providing guidance on their interpretation is highly valuable. To address this need, we trained a supervised neural network that, by analyzing different regions of microtremor spectra, automatically provides a general quality assessment of the measurements. It also informs the user about the presence of stratigraphic (1D or 2D) resonances, velocity inversions, artifacts, malfunctioning sensors or disturbances that need to be removed to improve the measurement interpretation. We will discuss the advantages of this neural network approach over existing microtremor H/V peak classification methods and highlight its limitations.