Western US Ambient Noise Cross-Correlations
This page describes the ambient noise cross-correlation product generated at the Center for Imaging the Earth's Interior at the University of Colorado in partnership with the IRIS DMC. 171,120 vertical component (LHZ) ambient noise cross-correlation waveforms, or empirical Green's functions, are available using data from 622 USArray Transportable Array (TA) stations west of 105° W operating between 2005/1/1 - 2010/12/31. Traditional ambient noise data processing procedures (Benson et al., 2007, Lin et al., 2008) were used, but refinements to the method (Lin et al., 2011) were applied to facilitate the interpretation of amplitude measurements. The stacked cross-correlation waveforms (in SAC format) and associated metadata are available in a 5 GB tarball.
Citation for the product:
IRIS DMS Product, Western US Ambient Noise Cross-Correlations, by Mikhail Barmine and Michael Rtizwoller, published electronically June 2012, Incorporated Research Institutions for Seismology, Last accessed CURRENT_DATE, http://www.iris.edu/dms/products/ancc-ciei
Map of 622 USArray virtual array (NET="_US-TA") stations west of 105° W for which cross-correlations are calculated.
ProcessingBensen et al. (2007) and Lin et al. (2008) described what we refer to as the traditional method of ambient noise data processing and this data product uses these procedures. The method breaks into four principal stages: (1) single station data preparation, (2) cross-correlation and temporal stacking, (3) measurement of dispersion curves (performed with frequency-time analysis for both group and phase speeds) and (4) quality control, including error analysis and selection of the acceptable surface wave dispersion measurements. The procedures have been applied to broad-band seismic data around the world and have been shown to produce robust, largely unbiased measurements of Rayleigh and Love wave phase velocities (e.g., Shapiro et al. 2005; Moschetti et al. 2007; Bensen et al. 2008; Lin et al. 2009; Lin and Ritzwoller 2011a; amongst many others). Simulations and, most importantly, comparisons with earthquake data also establish the reliability of the dispersion measurements and maps (e.g., Lin et al. 2008; Yang and Ritzwoller 2008; Ritzwoller et al. 2011). They have been used to produce 3D models of the crust and uppermost mantle for isotropic shear velocity, radial anisotropy, and azimuthal anisotropy (e.g., Yang et al. 2008; Bensen et al. 2009; Moschetti et al. 2010a, 2010b; Lin et al. 2011a). The ambient noise cross-correlations, or Empirical Green's functions, which define the substance of the data product, arise entirely from the first two stages of this procedure. The principal step in single-station data preparation is 'temporal normalization' which is designed to ameliorate the contamination of the ambient noise signals by earthquakes, instrumental irregularities, and non-stationary noise sources near to stations (such as passing storms and high local sea heights). This has been done by applying 'running absolute-mean' normalization, which we prefer to sign-bit normalization because it allows for tuning to regional earthquake conditions. In particular, for each day of data we filter the data set between 15 and 50 sec period, compute the absolute mean of the data in a running 80 sec long time window, and normalize the central data point of the unfiltered data by the reciprocal of this mean. In addition, we apply a smooth spectral whitening procedure in a band between 5 and 170 sec prior to cross-correlation in order to minimize contamination from spatially localized microseisms and to broaden the measurement band. Instrument responses are removed and a notch filter around the 26 sec microseism has been applied. Inter-station distances are a useful quality control metric that exists after the second stage of data processing. We find that inter-station distances greater than 2-3 wavelengths are needed to produce reliable dispersion measurements and this distance is included in the metadata for each cross-correlation. Data gaps introduce the need to define an "effective" time series length. Lin et al. (2011b) show that the square of the rms amplitude of trailing noise provides an accurate proxy for time series length. Here, however, we compute the time series length in hours and include it in the metadata. Cross-correlations are not normalized by this length but if ambient noise amplitudes across inter-stations pairs are to be compared, dividing by this quantity (tsnorm) is recommended.
Example waveform
|