Research Projects

Machine Learning (ML) tools are increasingly being used in geo-scientific research to develop robust seismic velocity models, which can produce synthetic datasets similar to observed ones above a specific seismic site condition. Seismic site characterization is crucial for analyzing structures and providing parameters for smart city planning. The proposed research will utilize the concept of Horizontal-to-Vertical Spectral Ratio (HVSR), first proposed by Nogoshi & Igarshi (1971) and widely spread by Yatuka Nakamura (1989). HVSR curves are generated from seismic ambient noise recorded in horizontal and vertical seismometers, which may be produced due to various types of seismic sources. The main content of these noises is seismic surface waves, with Rayleigh wave ellipticity showing a peak at fundamental resonance frequency on the HVSR curve. A common method for generation of HVSR curves relies on robust measurement of predominant frequencies in multiple time windows of ambient noises. However, reliable HVSR estimates may not be possible without careful picking of absolute values of predominant frequencies. A new technique called Multi-Window Cross-Correlation (MWCC) will replace the initial picking part for computing HVSR curves. This technique will simultaneously find depth to major impedance contrasts and 3D shear wave velocity models (or Vs30 maps) to determine the site response for future earthquakes and generate more accurate seismic hazard maps to mitigate the risk associated with devastating future earthquakes.