Annual Meeting - SCEC.org

Poster Abstracts
Although substantial progress has been made in physics-based ground motion simulations in the recent years, the engineering
community is still reluctant to use simulated time series for design. One of the reasons for this is a lack of understanding of
how simulated ground motions compare to recorded ground motions, especially when it comes to their impact on structural
response. There are on-going efforts of validation and verification of simulated ground motions, but these tend to be focused
on record properties or on the response of single-degree-of-freedom systems. We are proposing a different approach: we plan
to compare the nonlinear structural response of buildings subjected to recorded and simulated ground motions, given that
both sets have similar spectral shapes.
The responses of buildings to recorded motions have already been processed in a project recently completed by the PIs. The
recorded set is representative of a magnitude 7 earthquake, rupturing within 20 km from a site with a Vs30 of 400 m/s
(average shear wave velocity of the upper 30 meters of the soil column). We selected SCEC simulated records (for the same
type of event and site conditions) with spectral shapes that were consistent with the recorded set previously used by the PIs.
We then performed the structural simulations and compared the response results from both sets of time series (recorded and
simulated). We present the summary of the approach used and the structural response results we have obtained for recorded
and simulated ground motion records.
DETECTION OF ANOMALOUS STRAIN TRANSIENTS USING PRINCIPAL COMPONENT ANALYSIS AND
COVARIANCE DESCRIPTOR ANALYSIS METHODS (A-064)
R.A. Granat, J.W. Parker, S. Kedar, and Y. Bock
We have tested two classes of anomalous transient detectors that are completely independent of each other in origin, logic and
implementation. Principal Component Analysis (PCA) has been successfully implemented in tectonic geodesy for regional
filtering and common mode removal. Covariance Descriptor based detection methods are relatively recent technologies that
have seen considerable success in the field of computer vision, and are predominantly used to detect statistically significant,
spatially correlated anomalies in images. The two techniques examine the data in a fundamentally different and
complementary way. The one (PCA) detecting common fundamental modes of ground motion across the array, and the other
(CDA) detecting statistically significant common anomalies in a space-time “image” composed of deformation time series.
The advantage of the PCA technique lies in its ability to resolve temporal transients that although common to a particular
region, are not necessarily of the same amplitude across it. In addition, examination of higher residual modes highlights
whether or not the transient has propagated across the network in time.
Covariance descriptor methods are typically used in image analysis applications, where a set of filters is applied to an image,
and the outputs of these filters form a feature vector for each pixel. The covariance matrices of the feature vectors for different
image regions are calculated (these are termed the covariance descriptors for the regions), and then compared using a distance
metric. The generalized covariance distance between regions that resemble one another (in the chosen feature space) will be
small, while that between regions that are dissimilar will be relatively large. In this manner anomalies or matches between
regions can be identified. This method can be readily extended to tectonic time series analysis, where the “image” regions are
simply time windows, while the feature vectors are the tectonic displacement time series measurements themselves. The
covariance descriptor of the time series as a whole is calculated and compared with covariance descriptors of localized time
windows. Descriptors for localized windows that are far from the descriptor for the entire time series are likely anomalous.
We present results from synthetic cases, where these methods were tested using an iterative approach, in which the output of
one technique was used to fine-tune the input to the other.
TESTING SCEC 3D SEISMIC VELOCITY MODELS IN THE SAN BERNARDINO AND LOS ANGELES BASIN
REGIONS (B-018)
R.W. Graves, A. Plesch, and J. Shaw
The 2001 Mw 4.6 Big Bear Lake earthquake was recorded at over 180 broadband sites of the Southern California Seismic
Network throughout the greater Los Angeles region of southern California. At periods longer than about 1 sec, the ground
motions in the San Bernardino, San Gabriel and Los Angeles basins have significantly larger amplitudes and extended
durations relative to sites outside the basins. To model the longer period (T > 1 sec) near-source motions, Graves (2008)
developed a 3D representation of the basin structure in the San Bernardino region based on the potential field modeling of
Anderson et al. (2000). This study extends the previous analysis by modeling the motions across the San Gabriel, Chino and
Los Angeles basins using two Southern California Earthquake Center three-dimensional seismic velocity models (CVM-S4 and
2011 SCEC Annual Meeting | 171