Electronic Theses and Dissertations



Document Type


Degree Name

Doctor of Philosophy


Earth Sciences

Committee Chair

Charles Langston

Committee Member

Mitch Withers

Committee Member

Eunseo Choi

Committee Member

Robert Smalley


Part 1: Crustal structure in the New Madrid Seismic Zone (NMSZ) is investigated through a detailed study of explosion data obtained from the Embayment Seismic Excitation Experiment. The data show a distinct anisotropy in distance attenuation for both P and S waves in the range from 0 to 200km distance. Waves that propagate northward from the 1,134kg Marked Tree, Arkansas, explosion attenuate quickly with distance until a range of about 100km from the source where high-amplitude, high-phase velocity critical reflections from the boundary between the middle crust and rift pillow structure produce high amplitude waves. Propagation southward from the 2,268kg Mooring, Tennessee blast shows less distance attenuation compared to northward propagation. Reflections from the middle crust-lower crust boundary occur but do not significantly increase in amplitude with distance and travel with slower apparent phase velocity than observed for the northward propagation data set. A smooth velocity model is developed using a stabilized Weichert-Herglotz travel time inversion using first arrival travel times. Then an inversion using the travel time of both direct and middle crustal reflected waves is developed to obtain a 2D inhomogeneous-layered isotropic crustal model. The result reveals that there is a significant southwest dip to the top of the middle crust interface in the vicinity of the NMSZ, consistent with previously inferred changes in the thickness of the rift pillow model. This 2D feature characterizes the local wave propagation along the Reelfoot Rift and demonstrates the need for an improvement of the current Central United States velocity model.Part 2: Obtaining reliable empirical Greens functions (EGFs) from ambient noise by seismic interferometry requires homogenously distributed noise sources. However, it is difficult to attain this condition since ambient noise data usually contains highly correlated signals from earthquakes or other transient sources from human activities. Removing these transient signals is one of the most essential steps in the whole data processing flow to obtain EGFs. We propose to use a denoising method based on the continuous wavelet transform to achieve this goal. The noise level is estimated in the wavelet domain for each scale by determing the 99% confidence level of the empirical probability density function of the noise wavelet coefficients. The correlated signals are then removed by an efficient soft thresholding method. The same denoising algorithm is also applied to remove the noise in the final stacked cross-correlogram. A complete data processing workflow is provided with the overall data processing procedure divided into four stages: (1) single station data preparation, (2) removal of earthquakes and other transient signals in the seismic record, (3) spectrum whitening, cross-correlation and temporal stacking, and (4) remove the noise in the stacked cross-correlogram to deliver the final EGF. The whole process is automated to make it accessible for large datasets. Synthetic data constructed with a recorded earthquake and recorded ambient noise is used to test the denoising method. We then apply the new processing workflow to data recorded by the USArray Transportable Array stations near the New Madrid Seismic Zone where many seismic events and transient signals are observed. We compare the EGFs calculated from our workflow with commonly used time domain normalization method and our results show improved signal-to-noise ratios. The new workflow can deliever reliable EGFs for further studies.Part 3: We incorporate seismic ambient noise data recorded by different temporary and permanent broadband stations around the northern Mississippi Embayment from 1990 to 2018 to develop a crustal shear wave velocity (Vs) model for this area with full waveform ambient noise tomography. Empirical Greens functions at periods between 8 and 40s for all the possible pairs of stations are extracted by using a new seismic ambient noise data processing flow based on the continuous wavelet transform. Synthetic waveforms are then calculated through a heterogeneous Earth model using a GPU-enabled collocated finite-difference code. The cross-correlation time shifts between the synthetic waveforms and the extracted empirical Greens functions are used to construct the velocity updated kernel by using the adjoint method. Starting from the Central United States Velocity Model, the shear wave velocity model is then iteratively updated with the Vs kernel calculated in each iteration. Checkerboard tests show that perturbations in the top 30km of the crust are well recovered but amplitude recovery ability gradually decreases for deeper structure. We find that velocity lows characterize the Reelfoot Rift Graben and Rough Creek Graben separated by a high velocity crust. High velocity anomalies are observed under the Ozark Uplift and Paducah Gravity Lineament. A low velocity area previously interpreted as the Missouri Batholith is observed between them. A massive high velocity body in the southeast Mississippi Embayment is observed and is explained by the faulting as well as partly mafic intrusion. The Ouachita-Appalachian Thrust Front is clearly observed with a thinner crustal layer underneath. The rift pillow is well observed in the final tomography model along the Reelfoot Rift in the lower crust. The final inverted velocity model is consistent with local geological features and can be used for other seismological studies such as earthquake source determination and earthquake hazard assessment.


Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest