An earthquake instability model is used to compute theoretical fault slip and ground surface deformation before possible future earthquakes on the San Andreas fault in southern California. The quasistatic model represents the earth's crust by an elastic half-space and the San Andreas, Imperial, Cerro Prieto, and San Jacinto faults by flat rectangular surfaces whose strikes approximate the actual local strikes. The seismogenic zone of the San Andreas fault is represented by a brittle, strain-softening patch extending from the ground surface to 11 km depth and from Parkfield to the Salton Sea (500 km). The strength of the patch varies along strike, and unstable failures of parts of the patch are earthquake analogs; failed areas heat immediately after instability. For the model to reproduce the lengths, slips, and times of earthquakes since 1080 A.D. reported by K. Sieh and others, the patch must be divided into five sections along strike. The sections have alternately low and high strength, with the two strongest sections being roughly coincident with the major bends of the San Andreas fault. Ten to twenty years before unstable slippage of patch areas, portions of the patch bottom begin to fail stably (aseismically).

The stable failure causes accelerating fault slip at depth and at the fault trace, as well as anomalous changes in lengthening rates of trilateration lines. In the model simulation the next two sections to fail unstably in the future are just north of the Salton Sea and just south of Parkfield. Rupture lengths are about 100 km and 40 km. At favorable locations, precursory anomalies of fault slip and trilateration line length become large enough to be detected in field data a few years before each earthquake.