ALBERT

Description: We present a new approach for event detection and accurate determination of the onset time of seismic phases. We use a simple time-domain function (T (super pd) ), which is readily and efficiently calculated in real time. Our function is based on the T (super p) function, which has been used to characterize the predominant period of a waveform and which was originally introduced by Nakamura (1988) for use in real-time magnitude estimation. We find that while T (super p) itself produces changes at the onset of a seismic signal, it is not suitable for detecting seismic onsets. However, our modified damped predominant period function, T (super pd) , has considerable potential for detecting such onsets. We show examples of waveforms with their corresponding T (super pd) functions and demonstrate how the choice of parameters influences the function. The examples given show clear increases in T (super pd) for both the P- and S-wave onsets. We implement an approach to determining seismic onsets based on the T (super pd) function and call it the T (super pd) picker. The T (super pd) picker is trained, and we evaluate its performance on four days of continuous data. The T (super pd) picker compares favorably with a trained short-term average/long-term average picker--there are improvements in the number of accurate picks but, more notably, there is a significant reduction in missed picks. We analyze the signal-to-noise ratios of the waveforms and find that the improvements are mainly, but not exclusively, for cases of high noise. The aforementioned facts, combined with the simplicity and efficiency of the T (super pd) calculation, implies that there is considerable potential for its use as an automatic picker. The results also suggest that the T (super pd) picker could be further improved by refining the position of the pick after a seismic phase has been detected.

Description: Source, site, and propagation parameters are inverted from a U.K. database of weak-motion events (2.0〉M (sub L) 〉4.7). This results in the complete spectral parameterization of over 3200 velocimetric records of 273 events from the year 1992 to 2006. The S wave is extracted from the vertical records and is processed using a multitaper Fourier transform. We initially use a nonlinear least-squares log-space optimization to obtain estimates of the attenuation parameter for each spectrum. The estimates of t (super *) are then used to geometrically constrain a depth-dependent Q model using a technique adapted from velocity tomography. We then invert for the remaining frequency-dependent parameters and a collective amplitude parameter from the velocity spectra while fixing the newly computed attenuation parameters based on raytracing through our Q model. The resultant amplitude parameters are then split into source moments, apparent geometrical spreading, and site correction factors. We find a frequency-independent depth-dependent Q structure. A linear relationship proportional to 0.7 M (sub L) between moment magnitude (M (sub w) ) and local magnitude (M (sub L) ) is found in the range of 2-4.7 M (sub L) . The majority of stress drops are found to range on the order of 0.1-10 MPa. A multiple segment apparent geometrical spreading model is found to best describe the amplitude decay with distance, accounting for factors such as geometrical spreading and scattering, along with multiple phase interference in the analysis window. Site response functions are found to broadly correlate with regional geology, mean amplification occurring in the Cenozoic sedimentary rock sites to the southeast of England relative to the harder Palaeozoic rock sites of Wales and Scotland. We use a bootstrap analysis technique to analyze the dependence of our results on the data in order to estimate the variance of the results and check the robustness of different inversions. Synthetic spectra are also computed in order to obtain minimum variance and bias of model parameters associated with the method. In applying a geometrical Q constraint, through the use of Q tomography, we find that the robustness of the results is significantly increased. A thorough analysis of the trade-offs involved in the inversion is performed using synthetic datasets. We find strong trade-offs between the parameters, but we are able to show that this covariance is reduced when adopting the Q-tomography approach.

Description: Attenuation relations are derived for central Japan (broadly spanning the Kanto, Tokai, and Chubu regions) using recordings of small earthquakes (2.0〈 or =M (sub JMA) 〈 or =4.0 [Japan Meteorological Agency magnitude]) and moderate- to large-magnitude earthquakes (3.0〈 or =M (sub JMA) 〈 or =7.2). We independently analyze data from both small-magnitude and moderate- to large-magnitude earthquakes to provide an insight into the use of attenuation relations derived in regions of low seismic activity. A strong correlation is found between the attenuation parameters derived from each dataset. We find that Q is strongly depth dependent and that apparent geometrical decay increases with increasing hypocentral distance. This is modeled by using a three segment decay function, with the initial decay forced to 1/R. Moment magnitudes are close to the published M (sub JMA) magnitude, but are, on average, slightly higher. An increase in stress drop with magnitude is required in order to model both the small-magnitude and moderate- to large-magnitude datasets. Alternatively we show that a constant stress-drop model is suitable to model the response spectra of both small-magnitude and moderate- to large-magnitude earthquakes when considering the saturation of the source-corner frequency due to a static site filter such as f (sub max) or kappa . We test our ability to predict strong ground motion by using our attenuation and source-scaling relations derived from the small-magnitude recordings to stochastically simulate peak ground acceleration, peak ground velocity, and 5% damped response spectral ordinates over a range of magnitudes and distances. The residuals of this simulation are found to be largely independent of distance and magnitude. We compare our attenuation relations against other relations derived for Japan. The residuals of these relations are analyzed and compared against those obtained from the model found in this study. We find that, in this study, the prediction of strong ground motions is possible using small-magnitude data and that the validity of the prediction extends across all magnitudes available for comparison (2.0〈 or =M (sub JMA) 〈 or =7.8). On the other hand, by using an alternative published predictive relation for Japan, derived using large-magnitude events, peak ground acceleration is significantly overestimated for small-magnitude earthquakes.

Description: We use an aftershock dataset of over 1500 events (M (sub L) 0.7-5.8) to study the relationship between magnitude and the predominant period calculated from the initial P-wave arrival. We calculate Formula (Nakamura, 1988; Allen and Kanamori, 2003) and find that there is a trend between Formula and magnitude, as reported by previous authors. However, the trend is weaker than expected. We calculate an alternative predominant period function, tau (sub c) (Kanamori, 2005), and find virtually no relationship to magnitude for these data. We therefore implement a modified, damped version of the T (super p) function, which we term T (super pd) . The T (super pd) function introduces an additional term, D (sub s) , aimed at stabilizing the predominant period function in the transition between noise and signal. We show that T (super pd) (sub Max) has an improved relationship to magnitude, with the average coefficient of determination (R (super 2) ) increasing from 0.15 for T (super pd) (sub Max) to 0.5 for T (super pd) (sub Max) . This improvement is consistent for all stations. We then apply the T (super pd) function to the displacement waveforms, calling the associated function T (super pd_D) . The trend in the T (super pd_D) (sub Max) versus magnitude relationship is superior to that of tau (sub c) . Analyzing the T (super pd) function, we conclude that improvements result from damping large values in the noise region, or reducing spikes during the noise-to-signal transition, thus preventing incorrect maxima from being selected. We attempt to optimize the T (super p) (sub Max) and tau (sub c) results, and find that although the results improve, they are still significantly worse than for T (super pd) (sub Max) . The T (super pd) (sub Max) performance is shown to be robust and less dependent on the choice of parameters than T (super p) (sub Max) . We then apply T (super p) (sub Max) and T (super pd) (sub Max) to estimating magnitudes. Average errors are 20% smaller for T (super pd) (sub Max) estimates compared with optimized T (super p) (sub Max) results, with greater improvement for unoptimized parameters. We conclude that the performance of T (super pd) (sub Max) is superior to T (super p) (sub Max) and tau (sub c) and should be considered for real-time magnitude estimation.