Static Coulomb stress‐based Southern California earthquake forecasts: A pseudoprospective test

Abstract Many studies support the hypothesis that where earthquakes occur, recent changes in resolved Coulomb stress tend to be positive. How about the converse hypothesis, that where resolved Coulomb stress recently increased, earthquakes are more likely to occur? Successful earthquake forecasting by Coulomb stress changes requires the converse. To test this, we calculated stress everywhere in our study area, not just at earthquake locations. We modeled stress accumulation in Southern California since 1812 both from the elastic effect of slip below locked faults and from M  ≥ 5 “source” earthquakes up to any given date. To minimize the effect of secondary aftershocks not directly related to the source earthquakes, we measured seismicity using a gridded binary map: each 0.1° × 0.1° cell is “activated” if containing one or more test events (“receiver” earthquakes) of M  ≥ 2.8. We then constructed an empirical relationship between resolved Coulomb stress and activation rate within regions with similar stress values, defining probabilities of activated cells during the “test” period, within 11 years of the M 7.1 Hector Mine earthquake. We found that Coulomb stress reliably indicates future earthquake locations at the 95% confidence interval. However, smoothed seismicity forecasts outperformed Coulomb forecasts in some areas with large earthquakes due to aftershock clustering. Most earthquakes tend to nucleate in areas with Coulomb stress changes greater than 0.5 MPa or less than −0.5 MPa. Within areas with increased Coulomb stress from older earthquakes, fewer earthquakes occurred than anticipated. After reducing stress uncertainty impact, Coulomb rate‐and‐state forecasts may also improve upon statistical earthquake forecasts.