Twenty-seven earthquake forecasting models generated 51,625 next-day seismicity forecasts in California between August 2007 and August 2018, according to an evaluation published in Nature Communications. The models were operated prospectively and benchmarked against 597 recorded earthquakes with magnitudes of Mw 3.95 or higher using community-vetted statistical methods.
The evaluation, conducted through the Collaboratory for the Study of Earthquake Predictability, found that approximately 95 percent of the models produced forecasts consistent with the total number of observed earthquakes. Another 95 percent aligned with observed earthquake magnitudes, while two-thirds of the models matched earthquake locations. Individual versions of well-established, operational-type models demonstrated sustained consistency across the testing timeframe.
Researchers from the University of Bristol, the Statewide California Earthquake Center, the University of Naples Federico II, the GFZ German Research Centre for Geosciences, and the University of Nevada, Reno took part in the study. The authors published benchmarking datasets, software tools, and technical recommendations to assist operational earthquake forecasting systems worldwide.
Grants from the U.S. Geological Survey and the Leverhulme Trust Early Career Fellowship program funded the research alongside the Statewide California Earthquake Center, which operates under National Science Foundation and USGS cooperative agreements. Model descriptions were contributed by researchers David Rhoades, Frederic Schoenberg, Jiancang Zhuang, and Matthew Gerstenberger.
