Earthquakes in seismically-active regions, such as in California, present a significant human and financial risk to communities. Ground motion models (GMMs) have been developed to account for earthquake impacts when infrastructures are designed. However, GMMs assume ergodic hypothesis that is ground motions behave the same globally over time. The assumption was made as the global data had to be combined for modeling due to very limited available data. As more and more data collected, it was realized that spatial variations of ground motions are too large to be neglected. This study proposes a Bayesian hierarchical model to extract seismic site terms, the spatial site response bias from ergodic GMMs, to develop non-ergodic seismic site responses. The model was then implemented on the data from earthquake stations in Los Angeles County. The Kriging prediction is also conducted to generate heat map for seismic site responses visualization.