Estimating Near-Surface Elastic Structure from Low-Frequency Seismic Noise

When large atmospheric pressure variations occur at the Earth’s surface, the solid Earth deforms elastically in response. Low-frequency seismic data record such deformations. This thesis provides a better understanding of this near-surface deformation and its applications with data from the USArray Transportable Array (TA) stations after 2012. The two main sets of data we use are: low-frequency seismic signals, recorded on the stations’ broadband seismometers, and atmospheric pressures on collocated pressure sensors. We observe high coherence between seismic noise signals and pressure variations in their power spectral densities (PSDs) in the frequency band between 0.01 and 0.05 Hz. When atmospheric pressure variations are sufficiently large, recorded seismic signals varies alongside pressure data. The ratios between seismic and pressure data can be formulated to solve for near-surface elastic structure, especially the shear-modulus. Knowledge of the near-surface elastic structure is important for its application on seismic hazard and ground motion prediction studies.We analyze data from over nine hundred TA stations from 2012 to 2019, spanning across the Central and Eastern US (CEUS) and Alaska. The two interconnected methods we develop and use for this dataset are the half-space method and the layered inversion method. Both methods, founded with similar fundamental backgrounds, regard the excitation source as a propagating plane pressure wave on the surface that generates ground deformation. The half-space model assumes a homogeneous medium under the surface, whereas the layered model assumes a vertically heterogeneous medium. In essence, the half-space model is a unique case of the layered model. The former permits straightforward estimations of near-surface shear-modulus at fixed frequencies, but it lacks depth resolution to constrain important parameters of seismic hazards such as Vs30—the averaged shear-wave velocity down to 30 meters deep. In contrast, the layered inversion allows to estimate Vs30 by utilizing data at multiple frequencies from 0.01 to 0.05 Hz and resulting layered velocity models. Our results for the TA stations indicate that the two methods can resolve unique geological regions such as the Appalachian Mountains, and the Mississippi Alluvium Plain. For the Appalachian Mountains, where the bedrock is likely to be near the surface, our results indicate fast Vs30 at stations located in the area. For the Mississippi Alluvium Plain, known for thick alluvial sediments near the surface, our results demonstrate slow Vs30 at stations there. Large-scale geology maps and Vs30 models also corroborate our results. Our estimated Vs30s agree with mapped Quaternary sediment depths: fast Vs30 stations are often associated with areas categorized as thin sediment, while slow Vs30 stations are observable at locations with thick underlying sediment.