Quantifying sensitivity of local site response models to statistical variations in soil properties

We perform a combined stochastic-deterministic analysis of local site response using two computer codes, an equivalent linear analysis program SHAKE and a fully nonlinear finite element code SPECTRA. Our goal is to compare the relative sensitivity of the two codes to statistical variations in soil properties. For the case studies, we re-analyze two ground motion records in Lotung, Taiwan, and one ground motion record in Gilroy, California, utilizing the recorded ground motions at the site deterministically as input into the two codes while treating the uncertain soil parameters as random variables. We then obtain empirical cumulative distribution functions of the Arias intensity and acceleration spectrum intensity, two measures of cumulative damage, to compare the relative sensitivity of the two codes to variations in model parameters. We show that the two codes exhibit comparable sensitivities to statistical parameter variations, indicating that even in the presence of fluctuations in the soil parameter values it is possible to pursue a fully nonlinear site response analysis with SPECTRA and benefit from its superior accuracy.