The empirical prediction of ground motion

abstract Recent additions to the strong-motion data set, primarily from earthquakes in California and Italy, are responsible for a large number of papers examining the prediction of ground-motion measures using regression methods. Peak acceleration is still the most common measure being considered, but increasing attention is being given to peak velocity and spectral amplitudes. Although direct comparisons among the studies are hampered by differing definitions of distance and magnitude, in general the various studies give similar answers for peak acceleration in the region of distance and magnitude space in which most of the data are concentrated. As might be expected, the differences are most pronouced for large magnitudes and distances close to the fault, where data are few. Even so, widely differing assumptions about the form of the regression equation and differences in the composition and weighting of the data set can give similar answers. This was true in recent studies by Campbell (1981b) and Joyner and Boore (1981), where the predicted accelerations for large earthquakes at close distances differed by less than 40 per cent. This seemingly large uncertainty is small compared to the scatter in the data about the regression lines. A Monte Carlo study shows that the question of whether the shape of the attenuation curves is magnitude-dependent cannot be resolved by existing data.