ALM: An Asperity-based Likelihood Model for California

We propose a quasi-stationary earthquake likelihood model for California based on the assumption that small-scale (5 – 15 km) spatial variations in b-value of the Gutenberg-Richter law are meaningful for forecasting future seismicity. Because the b-value is inverse proportional to differential stress, low b-value patches (b < 0.7) in our model are much more likely to produce large ruptures than high b-value regions (b> 1.0). We calibrated our model based on the micro-seismicity from 1984 – 2005.5, using all events above the locally determined magnitude of completeness. At each node, Akaike’s Information Criterion is used to decide if a local b-value estimate or a global estimate of b = 0.82 is used in conjunction with the local a-value estimate to produce a forecast.