Towards more stable time varying ambient noise empirical Green's functions

This study examines under which input parameters the cross correlation of ambient seismic noise generates stable empirical Green’s functions (EGFs) with a minimum quantity of data. We evaluate the effects of varying the duration of correlated ambient data windows, the over-sampling, and the threshold for automatically removing noise data “contaminated” by earthquakes. We computed daily EGFs for 20 USArray stations near Yellowstone National Park in 2009. For each station, all available daily EGFs are stacked to generate a year stack. Stacks for subsets of daily stacks (2,3,4,5,6,10,15,20,25,30,40,50,100 days) are correlated with the yearlong stack to estimate how quickly or slowly the stacked EGFs converge. By examining how these stacked EGFs converge across different parameter combinations for a specific station-station pair, an optimal parameter combination is determined. We examine the variations in optimal parameterization as a function of inter-station distance and azimuth change. As is expected, the EGFs converge faster when the causal and acausal sides are averaged. Excellent correlations (R>0.9) are often achieved within 25-35 days and good correlations (R>0.7) are typically achieved with as little as 6-10 days. The examination of time varying seismic structures will be more viable with more stable EFGs calculated using a more optimal set of input parameters.