The 1989 Loma Prieta, Calif., earthquake caused spectacular damage to structures up to 100 km away in the San Francisco Bay sedimentary basin, including the Cypress Street viaduct overpass, the Bay Bridge, and buildings in the San Francisco Marina district. Although the few mainshock ground motions recorded in the northern San Francisco Bay area were “significantly larger … than would be expected from the pre-existing data set,” none were recorded at the sites of these damaged structures [Hanks and Krawinkler, 1991].
Loma Prieta aftershocks produced order-of-magnitude variations of ground motions related to sedimentary basin response over distances of 1–2 km and less [Cranswick et al., 1990]. In densely populated neighborhoods, these distances can encompass the residences of thousands of people, but it is very unlikely that these neighborhoods are monitored by even one seismograph. In the last decade, the complexity of computer models used to simulate high-frequency ground motions has increased by several orders of magnitude [e.g., Frankel and Vidale, 1992], but the number of seismograph stations—hence, the spatial density of the sampling of ground motion data—has remained relatively unchanged. Seismologists must therefore infer the nature of the ground motions in the great unknown regions between observation points.