COMPREHENSIVE GROUND MOTION SIMULATION AND ITS PREDICTION IN WESTERN HIMALAYAN REGION

Abstract

Strong ground motion records are used as primary inputs for the development of a ground motion attenuation model, seismic hazard analysis, and structural response analysis. Observed records from the past earthquakes have conventionally been marked as the best version of seismic input to these analyses. However, the frequency of occurrence of small and moderate magnitude earthquakes is very high relative to major seismic events, with most of the strong motion dataset used in the development of ground motion models consisting primarily of small and moderate earthquake records. Consequently, as the magnitude increases, the ground motion datasets become sparse and poorly sampled in the case of the short source to site distance ranges. To circumvent this issue, ground motion scaling and spectral matching are commonly used to modify the recorded time history and make it the best sample of input for specific analysis conditions. The time history simulated by this approach is highly inconsistent with the recorded data due to ground motion parameters being scaled to large factors and not being able to match data for various combinations of source and site characteristics. This has led to steps towards simulation techniques that preserve both the physical condition and characteristics of the observed data. Regarding the frequency range and existing advanced solution techniques, simulation methods are divided into three main parts: Deterministic or Physical-based, stochastic, and hybrid methods. Among these techniques, stochastic simulation is widely used in the development of the ground motion attenuation models as well as linear response-history analysis due to very few input parameters required in comparison to the other simulation techniques. This present study proposes some modifications to the classical stochastic method and then uses it to develop a ground motion prediction model for the western Himalayan region. This study firstly presents a compilation of the strong-motion records and algorithms to process those uniformly. The existing metadata of all records are comprehensively compiled and reappraised. Furthermore, the available metadata information of the site condition is highly unbalanced, since the site characterization of most of the stations has been carried out in terms of surface geology and only a very small number of the stations having measured shear wave velocity profiles. To remedy this problem, the Vs30 of the remaining sites has been based on the method of Ghofrani and Atkinson (2014), who have developed empirical relationships for VS30 in terms of the peak amplitude, Apeak, and the associated frequency, fpeak, of the average curve for HVRS for NGA–West2 and Japanese strong-motion data. An empirical site amplification model has been then developed for the western Himalayan region as a function of Vs30 and soil depth (Z1.0). The primary purpose of developing such a model is to use it in simulations of ground motion. The second objective of this thesis is to develop a versatile, nonstationary stochastic simulation technique that envisions the temporal variation of both intensity and the frequency content akin to the observed ground motion. In the classical stochastic simulation method, the ground acceleration time histories generated using a uniformly random phase spectrum are unable to model the nonstationarity in both the amplitudes and the frequency contents in a physically realistic way. Though the nonstationarity in amplitudes can be assigned easily using a deterministic envelope function, it is necessary to define a realistic phase spectrum to achieve the nonstationary frequency contents. Frequency nonstationarity may significantly influence the fatigue and nonlinear behavior of structures. This thesis proposes a method to define the Fourier phase spectrum of ground acceleration using empirically simulated equivalent group velocity dispersion curves. The empirically predicted dispersion curves are shown to be in reasonably good agreement with the actual dispersion curves for several real accelerograms with widely differing characteristics. The empirical dispersion curves are used to simulate the Fourier phase spectra for the generation of synthetic accelerograms. The synthetic accelerograms generated using Fourier amplitude spectra (FAS) of several real accelerograms are shown to have physically realistic nonstationary characteristics, as evidenced by comparisons between the peak ground accelerations, Husid plots, strong-motion durations, and elastic pseudo acceleration response spectra of the real and the synthetic accelerograms. The proposed method can thus be considered to provide a useful practical approach for simulation of nonstationary design accelerograms in conjunction with the FAS defined from seismological source model approach. The other aspects of the success of the stochastic simulation, both stochastic point source simulation, and finite source simulation, mainly depend on the accurate specification of Fourier acceleration amplitude spectra, which in reality has both far-field and near-field contributions in proportion to the source-to-site distance relative to the fault rupture dimension. To incorporate this modification, the Fourier acceleration site spectrum is defined as a weighted combination of the Brune’s far-field and near-field source spectra. Following the idea from Trifunac(1993), the weights are so defined that as the distance increases from zero to four times the rupture dimension, the contribution of near-field spectrum decreases from 100% to only 5%. From the comparison of stochastic point simulation based on equivalent distance with the proposed spectrum, it is seen that the weighted site spectra are quite different at closer distances, and they converge to the far-field spectrum as the distance increases to very large values. Also, the rate of fall of amplitudes below the corner frequency in the weighted spectrum is slower as compared to the far-field spectrum. This effect is not possible to be generated realistically by ad hoc measures commonly adopted in stochastic simulation with the far-field spectrum. Furthermore, 278 archival data of 80 events recorded at 92 strong motions sites have been analyzed to estimate the earthquake source parameters (viz., stress drop and seismic moment) and attenuation parameters (quality factor and kappa) for the western Himalayan region. The isolation of the seismic wave attenuation due to anelastic medium from the observed Fourier spectra is a perdurable problem in seismology. To resolve these issues in the present study, a modified spectral ratio technique has been employed for deconvolution of the source and the site effect simultaneously. The spectral ratios are defined as the ratios of the observed Fourier spectrum amplitudes to the amplitudes determined from an empirical scaling model for western Himalaya (Gupta et al., 2017). The significant advantage of this approach over the basic spectral technique is that it does not require the seismic stations to align with a common source. Further, the kappa factor has been obtained by fitting a least-square line between Fourier Amplitude spectra and frequency on a log-linear scale. The average kappa value for Western Himalaya has been estimated as 0.04s. After deriving the regional Quality factor and Kappa, the stress drop is determined individually for all the events. The grid search algorithm was utilized to estimate the best fit of stress drop for each earthquake by minimizing the root mean square error between the theoretical weighted site spectrum and the observed spectrum. Next, an empirical prediction model has been developed for horizontal 5% damped response spectra (PSA) by combining the recorded and simulated data. The simulated ground acceleration time histories are generated using theoretical Fourier amplitude spectra based on weighted far-field and near-field Brune’s source model and the phase spectrum from the empirical group velocity dispersion model. The regression coefficients of the model developed have been derived based on the mixed-effect regression analysis for 21 periods ranging from 0.03 to 5 s. The performance of the model has been established by physically realistic behavior as a function of magnitude and distance, and perfect matching with the observed PSA spectra for widely differing magnitude and distance combinations. Illustrative numerical results show that observed response spectra are well-bounded within a median plus / minus one logarithmic standard deviation of the predicted response spectra, divulging that the proposed model correctly handles observed natural ground motion variability In summary, the research work carried out throughout the study has been able to upgrade the existing simulation approach, estimation of input seismology parameters, and the development of the attenuation models. The new knowledge and information contributed to the western Himalayan area would pave the way for improved quantification of seismic hazard and resulting dynamic response of structures in the region.