FIGURE 4.2 Topography and bathymetry of the West Sunda subduction zone showing the location of the trench and the earthquake fracture zones (colored lines). The red dot shows the epicenter of the M 9.1 December 2004 earthquake that ruptured the India/Eurasia plate boundary (area between the trench and Sumatra) and caused the devastating Indian Ocean tsunami. Rupture propagation was primarily northward, toward the lower left of the figure, extending hundreds of kilometers beyond the figure. Subsequent earthquakes have ruptured parts of the plate boundary to the south. SOURCE: Courtesy of Mohamed Chlieh, Caltech. See also Chlieh et al. (2007) and <http://today.caltech.edu/gps.sieh/>. Used with permission.

Just as plate tectonics explains where most earthquakes occur, it has much to say about how often they occur. The velocity of relative motion across plate boundaries is known to within several millimeters per year, which provides a boundary condition on fault-slip rates, and thus on how frequently earthquakes must occur over the long term. The utility of this boundary condition for earthquake prediction is confounded by the fact that plate boundaries typically comprise complex fault systems, and the partitioning of slip among faults is difficult to unravel, even in well-studied systems such as the San Andreas. In some cases, slip rates are well known, but even then the irregular recurrence of earthquakes makes forecasting difficult.

Earthquake predictions are commonly classified by time frame. The types of predictions discussed below are:

1. Long-term forecasts of events of an uncertain magnitude that have a low probability of occurrence over a large window of time. Long-term forecasts based on probabilistic methods are an active area of research.

2. Short-term prediction of events of a specific size that have a high probability of occurrence within a narrow range of space and time, weeks or months in advance. There is currently no way to predict the days or months when