Improved seismic methodologies in lithospheric imaging
About the Event:
The dispersion of surface waves implies that their frequencies are sensitive to different depth ranges within the Earth; conveniently, this allows us to infer the sub-surface structure. Velocity and attenuation of surface waves provide complementary information. In principle, they can be used to constrain the Earth’s parameters (e.g., temperature, composition, and viscosity) and reconstruct the geodynamic processes that slowly shape our planet. This presentation reviews some of the most widespread methodologies employed in the imaging of the Earth’s lithosphere, including ambient-noise interferometry and the (earthquake-based) two-station method. Particular attention is paid to the development and validation of such methodologies, to what can be achieved by their application, and to their intrinsic limits. The latter considerations are made dynamically, while discussing the results of the tomographic studies that I carried out on (i) Sardinia island, (ii) central-western Mediterranean, and (iii) North America. The last part of the presentation focuses on the application of a supervised machine- learning algorithm to detect small, local earthquakes recorded at single receivers. Based on more than one million seismograms that I collected worldwide and labeled as “earthquake” or “noise”, a simple Convolutional Neural Network (CNN) achieves an accuracy of 96.7, 95.3, and 93.2% on training, validation, and test set, respectively. Considering the large variety of different geologic and tectonic settings included in the set of labeled seismograms, this proves the generalization capability of the algorithm and suggests its application to real-time detection of local events.
About the Speaker:
Fabrio Magrini is a PhD student studying geophysics at Roma Tre University in Italy.