The accuracy of stationary mathematical models of seismicity for calculating probabilities of damaging shaking is examined using the history of earthquakes in China from 1350 A.D. to 1949 A.D. During this time, rates of seismic activity varied periodically by a factor of 10. Probabilities of damaging shaking are calculated in 62 cities in North China using 50 yr of earthquake data to estimate seismicity parameters; the probabilities are compared to statistics of damaging shaking in the same cities for 50 yr following the data window. These comparisons indicate that the seismic hazard analysis is accurate if: (1) the maximum possible earthquake size in each seismogenic zone is determined from the entire seismic history rather than from a short-time window; and (2) the future seismic activity can be estimated accurately. The first condition emphasizes the importance of realistically estimating the maximum possible size of earthquakes on faults. The second indicates the need to understand possible trends in seismic activity where these exist, or to develop an earthquake prediction capability with which to estimate future activity. Without the capability of estimating future seismicity, stationary models provide less accurate but generally conservative indications of seismic ground-shaking hazard.
In the United States, the available earthquake history is brief but gives no indication of changing rates of activity. The rate of seismic strain release in the Central and Eastern United States has been constant over the last 180 yr, and the geological record of earthquakes on the southern San Andreas Fault indicates no temporal trend for large shocks over the last 15 centuries. Both observations imply that seismic activity is either stationary or of such a long period that it may be treated as stationary for seismic hazard analyses in the United States.