The intellectual merit of this study involved the investigation of local anthropogenic noise and P-wave coda from teleseismic events to investigate sources and shallow and deep crustal structure from array-based seismic data. The data are from a large-N array in Oklahoma and from the 17-station Autocorr Seismic Array.  The AutoCorr array was located in the midwestern United States and consisted of a 10 km diameter subarray in the south and extends to the north with a linear array component. The full north-south extent of the combined array was about 30 km. The array was deployed from August 2019 to July 2020, which included the initial months of the Covid-19 pandemic. The northernmost seismic stations of the array are located within a wind farm with more than 1,100 wind turbines, one of the largest onshore wind farms in the world.

Ambient noise from wind turbines and other anthropogenic sources were recorded by the array. Even during the times when trains were present, the time-frequency signatures of the wind turbines were dominant over much of the array, including seismic stations well to the south of the wind farm. When utilizing seismic interferometry, a coherent Rayleigh wave signal was observed even for short time windows of ambient noise. For seismic stations within the wind farm, both north and south propagating Rayleigh waves were observed in the correlations. However, for seismic stations to the south of the wind farm, only south propagating waves were observed, which are inferred to be coming from the wind turbines. We had concurrent measurements of average hourly wind speeds and peak hourly wind gusts interpolated at the locations of the seismic stations. These data show that for one-hour ambient noise correlations, clear south propagating Rayleigh waves are observed for moderate to large average hourly wind speeds. For lower wind speeds, less coherent Rayleigh wave signals were observed in the one-hour ambient noise correlations.

We then investigated seismic array processing for a large-N array in Oklahoma using shallow waste-water injection events for the estimation of crustal and Moho structure beneath the array.  For smaller seismic arrays, machine learning techniques were applied to convert small and sparse seismic arrays to larger-N synthetic aperture arrays.  Convolutional neural networks (CNNs) with a U-net architecture were then used to train smaller seismic arrays to have responses more like larger seismic arrays.

We investigated P-wave coda recorded by the AutoCorr array, as well as single regional stations, for crustal structure.  We applied correlation analysis of the P-wave coda of selected teleseismic events recorded by the AutoCorr Array to estimate the incident ray parameter across the array and crustal reflectivity using both autocorrelation and cross-correlation analysis. Stacking was applied across the array to enhance signal to noise, and machine learning was tested for the 10 km aperture sub-array to construct a larger virtual aperture array.  We first applied correlation analysis to several deep focus earthquakes in Central and South America to compare different correlation windowing and frequency filtering to enhance crustal reflectivity.  We then applied this to select shallow and deep-focus events at different distance ranges for the combined estimation of crustal reflectivity and slowness.  We then constructed an east-west transect of seismic stations from western Iowa to eastern Ohio for teleseismic events to investigate the variability of the correlation crustal results across the north-central midwestern US. 

The broader impacts of the proposal were first in the identification of anthropogenic sources of seismic signals, in particular from the large wind farm to the north of the AutoCorr array, where the wind turbine signals dominated over other anthropogenic seismic signals across the 30 km length of the array even for moderate wind speeds.  This has important implications for the future placement wind farms. The use of machine learning was then used to convert smaller seismic arrays to larger synthetic aperture arrays, and this has important implications for the deployment of seismic arrays in the future. The array-based study of P-wave coda correlations then allowed for the estimation of P-wave crustal structure even for a smaller number of teleseismic events and this can be important from a complementary standpoint with standard receiver functions.  Finally, during this study a number of undergraduate and graduate students were involved with the field deployments and data processing, and this is important in the training of future scientists.  Also, several student and teacher outreach videos were made for the K-12 SuperHeroes of Science youtube video series.

Last Modified: 03/14/2022
Modified by: Robert L Nowack