S12A-01
Seismic hazard assessment in central Greece from a new developed seismogenic source model
In the late 90's, an active seismic survey was carried out in central Greece. The sediments and crustal structure were mapped by observing two 120 Km long seismic profiles, using 14 ocean bottom seismographs, and 10 stand alone seismic stations. The local microseismicity was also observed by operating a combined on/offshore seismic array, consisting of 7 ocean bottom seismographs and 23 land stations, for a period of four months. The recorded microseismicity was well correlated with tectonic faults mapped by the active seismic experiment and enabled us to delineate a new seismogenic source model. The assessment of the regional seismic hazard was obtained by using the hypothesis of tectonic lines supported by seismic activity. The model has the advantage of recognizing rupture length as the most significant parameter in seismic hazard assessment. Probabilistic uncertainties in the maximum possible magnitude and in the expected ground motion are accounted for explicitly. The results of the analysis are significantly different than those obtained from previous studies based on large generalized seismic sources, emphasizing the importance of including local active and seismic observations for a reliable seismic hazard assessment.
S12A-02
Evaluations of M>=4 Earthquake Probability Forecasts for California and Western Nevada From 2005 to 2008
Since the beginning of 2005, we have been generating three kinds of 5-day earthquake probability forecasts following the occurrences of M≥4 mainshocks in California and Nevada. One forecast model assumes that all future M≥4 mainshocks in California and Nevada are random, uncorrelated events, and the forecast probabilities are based on Poisson distributions with fixed rates of mainshock occurrences. In the second model, the short-term temporal clustering of M≥4 mainshocks in California and Nevada from 1932 to 2004 is used as the basis of the short-term forecasts of future events. The third forecasts are based on a hidden Markov model analysis that uses the past earthquake history to estimate the short-term probability of future earthquakes in the region. The second model is not adaptive, while the third model is since each forecast depends on the seismic history available at that time. Separate forecasts are issued for the eastern and the western parts of the study area. Relative evaluations of the forecasts are made using the likelihood of the observed results under the three models. For the first and second models, the events are independent and the likelihoods can be multiplied. For the third model, successive events are not independent and a joint likelihood must be computed. After almost 40 forecasts since 1/1/2005, both the second and the third models have somewhat outperformed the Poisson model, and the second model has a slightly better likelihood score than the third.
S12A-03 [WITHDRAWN]
Exploring new Strategies to Generalize Branching Models for Earthquake Forecasting
Branching models are probably the most used models for earthquake forecasting at different time-space- magnitude windows. The most popular one is the so-called ETAS (Epidemic-Type Aftershock Sequences) model that assumes a constant and stationary seismic background intensity with superimposed the effects of short-term seismic triggering, modeled through different flavor of the Omori law and a power-law decay with distance. Nonetheless, recent studies have shown that the background may vary significantly at different time scales (few days up to decades or centuries), and also the asymmetrical distribution of aftershocks around the mainshock epicenter call for a generalization of the classical ETAS modeling. In this work, we present some empirical pieces of evidence supporting the necessity of such a generalization, and some possible strategies to achieve this goal. In particular, we show different methods to account for time variations of the seismic background, and a non-parametric technique to adjust branching modeling to explain possible preferential spatial propagation of triggered earthquakes, embedded in a general branching modeling. The preliminary comparison of these new models with classical ETAS model in different time-space-magnitude windows (i.e., in different seismic catalogs) are encouraging, since new models provide a better fit of independent datasets.
S12A-04
Geostatistics for the Next Generation Attenuation (NGA) Flatfile Database
Modeling of strong ground motions shares with other geographically distributed attributes the problem of interpolating and extrapolating values away from seismic stations. The geostatistical methods of kriging and stochastic simulation have proved adequate both for providing interpolation estimates and assessing their reliability. In most previous studies of spatial covariance, station pairs have been binned as a function of inter-station distance and the standard deviations of observational residuals (VS30 and distance corrected) have been taken within each bin, producing the equivalent of an interpolation variance when given only a single observation point near the interpolated point. Kriging interpolation and its variance account for multiple nearby observations and their relative locations. For selected events and records from the NGA flatfile database we evaluate natural logs of PGA, PGV, and PSA at the three ShakeMap periods, 0.3, 1.0, and 3.0 s. These motions are corrected either for VS30 alone or for both VS30 and Joyner-Boore rupture distance. The latter is more promising and makes better use of seismological knowledge. We improve upon the geostatistical work of N. Jayaram and J. W. Baker (written comm., June 2008) by using UTM northing and easting in place of latitude and longitude and by taking the process beyond semivariograms. UTM coordinates approximate the Cartesian system required for geostatistics and we show that the Chi-Chi mainshock exhibits evidence of dependence upon epicenter- station azimuth for PGV and PSA at 3.0 s. Kriging (and modeling to be performed shortly) allow us to estimate interpolation variance directly. For PGA and PGV, the largest kriging standard deviation (root of kriging variance) in well-observed regions stays below about 0.28 for Chi-Chi. By assuming that log errors are normally distributed, with mean and variance equal to the kriging estimates, one can make additional inferences. In such a case, the kriging standard deviation is equivalent (at 1σ) to multiplying or dividing interpolated ground-motion values by a factor of 1.32 -- in shorthand a "32% σ". In the same terms, 1σ is about 28% midway between proximal stations in the well-observed region and 34% far from stations but still within the seismic array (e.g., sparsely instrumented mountainous terrain). By comparison, previous single-station estimates of interpolation standard deviations are ~20% for PGA midway between proximal stations and about 30% at locations equivalent to the mountainous Taiwan site (Boore et al., 2003). This result is the reverse of the hoped-for variance reduction from using multiple nearby observations. Finally, the kriging standard deviations for PSA are below about 57%, higher than error estimates for PGV and PGA, and show no clear dependence upon the period of the PSA estimate. Use of geostatistical simulation (analogous to the bootstrap method) makes it possible to eliminate the assumption of normally distributed errors as well as the preparation of maps absent the smoothing associated with any mean-square-error estimation method, including kriging. Simulations also provide maps with the same spatial covariance as the data; these will be presented at the meeting.
S12A-05
Influence Of Site Classification On Computing Empirical Ground-Motion Prediction Equations In Italy
In this study, we investigate a site classification method for stations of the Italian Accelerometric Network based on the predominant period of ground motion at the site. The site predominant period is identified from the average horizontal-to-vertical (H/V) spectral ratios of the 5%-damped response spectra of Italian earthquake records. We selected a data-set of 610 three-component analogue and digital recordings from 120 earthquakes recorded at 214 seismic stations within an hypocentral distance of 200 km. Selected events are in the moment-magnitude Mw range of 4.0 to 6.8 and the focal depth ranges from 5 to 40 km. Whenever possible, we classified each site by assigning them to one of six predominant period classes (in the range 0.05 to 2 seconds) that we propose as a modification of the Zhao et al. (2006) procedure. We then investigated the impact of this classification scheme on empirical ground-motion prediction equations. We adopted the same functional form of Fukushima et al. (2007) and we computed a nonlinear period- dependent regression that allowed us to derive site coefficients using the proposed six predominant period classes. We also derived site coefficients for a simplified classification based on the general soil conditions at each site. This classification uses two classes (which we call A-B and C-D, with Vs ≥ 360 m/s and Vs < 360 m/s, respectively) based on the four basic ground categories in the current European (CEN 2004) and Italian seismic codes. Our empirical site classification scheme based on strong-motion data provides the opportunity to explore whether we can decrease the misfit by improving the site characterization of the Italian data set. Comparison of our results with other empirical ground-motion prediction equations (GMPEs) based on conventional site classifications do not display a significant reduction of overall standard deviation. However, our site classification schemes shows promise in reducing the uncertainty in ground-motion prediction. The advantage over the simpler classification schemes used in many GMPEs is that we capture the effect of deep and shallow soil profiles and high shear-wave velocities. Another advantage of our classification scheme is that it is based on relatively quick and inexpensive measurements and interpretation.
S12A-06
Real-time Application of Earthquake Early Warning Considering Effects of Near-field Terms
This research analyzes strong motion records of 24 large earthquakes to investigate the relationship between τc (the average period of the first motion) and moment magnitude. The records of large earthquakes very close to the source include large long-period near-field terms. Therefore, if we do not consider the effect of the near-field term, we may overestimate the final size of the magnitude. We processed the records which do not have strong near-field terms, and compute the τc for each event. Our analysis shows the value of τc takes between the corner frequency and the period determined by the record duration to compute τc. If the magnitude is less than 6, τc becomes close to the period corresponding to the corner frequency, whereas τc for larger earthquakes depends on the rupture process and location of area of large slip. τc for large earthquakes regulates the lower bound of the magnitude estimation, and it approaches the period corresponding to the corner frequency, if a longer record duration is used. We also propose a method to classify the records with and without large near-field terms. If the Pd3 (peak displacement of the first 3 seconds) exceeds 1cm and τc > 2 seconds, the record is more likely to contain a large near-field term. For the purpose of quick onsite warnings, stations observing large near-field terms provide valuable information. Large near-field terms can be observed only if the magnitude is large and the epicentral distance is small. Therefore, if the displacement exceeds a threshold (e.g. Pd = 0.5cm), the ground motion at the site would likely become very large. This criterion will help to issue warnings to the blind zone that is close to the source. We incorporated this algorithm into the Seismic Automatic Triggering and Recording Network (SATARN) system operated by the Research Center for Earthquake Prediction, Kyoto University. We constructed a prototype system using the data recorded by stations around Kyoto city. Event triggers are picked by using ratios of the short-term and long-term averages, and then the minimum AIC is computed to search for the onset of the P-arrival. Currently, near- field terms and values of tauc are monitored, which may provide quicker warnings for large nearby earthquakes.
S12A-07
Testing the Performance of W-phase Source Inversion
We have recently developed a method to perform fast source parameter inversions for large earthquakes using the W-phase, a very long-period phase starting right after the P-wave arrival (Kanamori and Rivera, 2008). Here we report on the results of a systematic test on the performance of the method for seismic tsunami warning purposes. We tested it using a global data set of all Mw ≥ 7.0 earthquakes between 1990 and 2008 (240 events). We use the vertical VBB data from GSN and Geoscope stations (mainly STS-1 seismometers) filtered between 0.001 and 0.005 Hz. The window length, τ, used for inversion is distance dependent, τ= 15 Δ sec (Δ in degree). Typically we use stations up to 40°- 60° corresponding to a delay of 18-25 min after the origin time. To perform inversion, we need, in addition to the seismic traces, estimates of the hypocenter parameters and of the centroid delay. The PDE parameters, which are quickly available, can be used as a first approximation of the hypocenter location. The centroid source delay is first estimated from an approximate Mw determined from the W phase amplitudes. Having a database of pre-computed Green's functions, the actual inversion process is nearly instantaneous. We can improve the inversion using either grid search or parameter optimization by a non- linear least squares method. In the test reported here, we systematically use the Harvard or the Global CMT solution as reference. Specifically, we compare our solutions with the reference solutions, in terms of the seismic moment (i.e., Mw) and the orientation of the nodal planes. The results of comparison are encouraging. The differences in Mw between the W phase and the CMT inversions are 0.1 or smaller. The differences in the angular distance between the nodal planes are typically less than 5-10 deg. The solutions are in general robust. Because of the sparse station coverage, the solutions for older events (~ 1990-1992) are not constrained well, but are still adequate for tsunami warning purposes. Our tests indicate that the W phase inversion provides rapid and robust solutions useful for tsunami warning purposes. Also, since W phase covers the frequency band which has not been covered by the existing regional and global inversion methods, W phase solutions provide important additional information on long- period characteristics of earthquakes.
S12A-08
Automatic 3D Moment tensor inversions for southern California earthquakes
We present a new source mechanism (moment-tensor and depth) catalog for about 150 recent southern California earthquakes with Mw ≥ 3.5. We carefully select the initial solutions from a few available earthquake catalogs as well as our own preliminary 3D moment tensor inversion results. We pick useful data windows by assessing the quality of fits between the data and synthetics using an automatic windowing package FLEXWIN (Maggi et al 2008). We compute the source Fréchet derivatives of moment-tensor elements and depth for a recent 3D southern California velocity model inverted based upon finite-frequency event kernels calculated by the adjoint methods and a nonlinear conjugate gradient technique with subspace preconditioning (Tape et al 2008). We then invert for the source mechanisms and event depths based upon the techniques introduced by Liu et al 2005. We assess the quality of this new catalog, as well as the other existing ones, by computing the 3D synthetics for the updated 3D southern California model. We also plan to implement the moment-tensor inversion methods to automatically determine the source mechanisms for earthquakes with Mw ≥ 3.5 in southern California.