Earthquake forecasting from paleoseismic records.

Area of Science:

• Geophysics
• Seismology
• Earthquake Science

Background:

• Accurate forecasting of large earthquakes is crucial for seismic hazard assessment.
• Statistical models using paleoseismic data are key tools for earthquake recurrence interval analysis.
• Existing methods often struggle with measurement errors and model uncertainties inherent in paleoseismic records.

Purpose of the Study:

• To compare five statistical models for forecasting large earthquakes.
• To implement Bayesian model-averaging for time-dependent, probabilistic earthquake forecasts.
• To assess the benefits of model-averaging for paleoseismic data with significant uncertainties.

Main Methods:

• Compilation and analysis of paleoseismic data from 93 active fault segments globally.
• Comparison of five distinct statistical models for earthquake recurrence.
• Application of Bayesian model-averaging to integrate model predictions and account for measurement errors.

Main Results:

• The study produced time-dependent, probabilistic earthquake rupture forecasts for the next 50 years for 93 fault segments.
• A single best model was favored for 65 fault segments, while 28 segments benefited from the model-averaging approach.
• Forecasts included probabilities of rupture and predicted occurrence times for the next large earthquake on each segment.

Conclusions:

• No single statistical model universally explains large earthquake recurrence.
• An ensemble forecasting approach, like Bayesian model-averaging, is advantageous for paleoseismic data, especially with limited data and large measurement errors.
• The developed methodology enhances seismic hazard assessment by providing more robust earthquake forecasts.