Non-Poissonian Earthquake Clustering and the Hidden Markov Model as Bases for Earthquake Forecasting in California

The quest to find successful methods to forecast future earthquakes has proven to be very challenging. Useful earthquake forecasts require detailed specification of a number of variables, namely the epicenter, depth, time and magnitude of the coming earthquake. While forecasting the times of strong aftershocks within the rupture zone of a strong earthquake has been developed with some success (e.g., Reasenberg and Jones, 1989, 1994), forecasting the times of future strong earthquakes, even when their locations are known to occur within broad geographic areas, has not been very successful. The apparent success of the M8 algorithm in forecasting the 2003 M6.7 San Simeon earthquake (Keilis-Borok et al., 2004) followed by the failure of this same algorithm after it mistakenly forecast a strong earthquake in southern California before September 2004 shows the promise and disappointment of the current state of earthquake forecasting.