Ground motion prediction maps using seismic-microzonation data and machine learning

Abstract. Past seismic events worldwide demonstrated that damage and death toll depend on both the strong ground motion (i.e., source effects) local site effects. The variability of earthquake distribution is caused by stratigraphic and/or topographic setting buried morphologies (e.g., irregular sub-interface between soft stiff soils) can give rise to amplification resonances with respect expected at reference site. Therefore, conditions affect an area related full collapse or loss in functionality facilities, roads, pipelines, other lifelines. To this concern, near-real-time prediction variation over large areas a crucial issue support rescue operational interventions. A machine learning approach was adopted produce maps considering morphological conditions. set about 16 000 accelerometric data points 46 geological geophysical retrieved from Italian European databases. intensity measures interest were estimated based nine input proxies. regression model Gaussian process regression) allows for improving precision accuracy estimation available methods equation ShakeMaps). In addition, 50 m × resolution generated, providing agreement results advanced numerical simulations detailed subsoil models.