Testing the Classic 1988 Forecast

The Working Group on California Earthquake Probabilities published its first 30-yr forecast in 1988. They estimated magnitudes and conditional rupture probabilities for 16 segments of the San Andreas, Hayward, San Jacinto, and Imperial faults; these were based on the last rupture date of each of these faults and a quasi-periodic recurrence model. I applied the numbers and the space tests, developed by the Collaboratory for Study of Earthquake Predictability (CSEP), to the forecast (Working Group on California Earthquake Probabilities [WGCEP], 1988). Although only later earthquakes were used in the application, the tests are not considered to be prospective because the test metrics were not agreed upon in advance. Forecasts and test protocols are best when concurrently formalized. Association, that is, attaching a prescribed probability to any relevant future earthquake, challenges formal testing. Only the 2004 magnitude 6 Parkfield earthquake can be unambiguously associated with an anticipated outcome. Association of the 1989 magnitude 6.9 Loma Prieta earthquake is ambiguous. Only if it is included will the forecast agree acceptably with the number of events. A hindcast in 1988 would have shown that the 1988–2018 forecast would likely overpredict the earthquake rate. A simple alternate forecast that was constructed and tested this year passes both numbers and space tests without the time-dependent recurrence assumptions of the Working Group forecast. The comparison shows the importance of calibrating a forecast on past earthquake rates; it also shows that a 30-yr forecast test based on California alone is unlikely to effectively test the inherent assumptions. The results here have not been, nor could be, done via the CSEP process, which requires prescribed protocols and authoritative data sources.