The USGS Automatic Aftershock Forecasting System
Michael Barall, Andrew J. Michael, & Jeanne L. HardebeckPublished August 14, 2019, SCEC Contribution #9562, 2019 SCEC Annual Meeting Poster #029
Last year, the USGS expanded aftershock forecasts to the nation with upgraded methods and presentation on the USGS website. A forecast is probabilistic and includes information such as the probability that an M6 aftershock will occur in the next week, or the likely number of M5 aftershocks in the next month. The forecasts for the M7.1 Anchorage and M6.4 and M7.1 Ridgecrest earthquakes attracted attention from the public and the news media, and were used by FEMA, CalOES, the U.S. Navy, and others to help plan the earthquake response.
So far, the aftershock forecasts have been prepared manually. This year, the USGS is launching an automated system to generate and update the forecasts. Initially, it will produce forecasts for all M5 and larger earthquakes in the U.S.
The system runs continuously, listening to USGS PDL, which provides a near-real-time notification when an earthquake is detected. The first forecast is published 30 minutes after the earthquake occurs. Then, the system uses USGS ComCat to monitor the ongoing aftershock sequence. For one year after the original earthquake, the system periodically updates its model parameters to fit the observations and sends out revised forecasts.
There are algorithms to determine when one earthquake is an aftershock or foreshock of another, even in cases where a mainshock is at the edge of its aftershock distribution (rather than in the center). These are used to identify which earthquakes are part of the aftershock sequence, and to avoid generating forecasts for earthquakes which are themselves aftershocks or foreshocks of larger earthquakes.
Currently, forecasts are computed using a Reasenberg and Jones (Science, 1989) model as updated by Page et al. (BSSA, 2016) with California parameters from Hardebeck et al. (SRL, 2018). In this model, aftershock rates are computed by combining a Gutenberg-Richter distribution of aftershock magnitudes, modified-Omori decay of the aftershock rate with time, and productivity as a function of mainshock magnitude and location. A future version will add an ETAS model (Ogata, J. Am. Stat. 1988), which can handle more complex situations, such as when a large aftershock re-energizes the aftershock sequence.
Although the system can generate forecasts completely automatically, analysts can intervene to adjust model parameters if desired. The system will run redundantly on two servers, so that if the primary server fails, the secondary server can step in and take over.
Key Words
aftershock forecast
Citation
Barall, M., Michael, A. J., & Hardebeck, J. L. (2019, 08). The USGS Automatic Aftershock Forecasting System. Poster Presentation at 2019 SCEC Annual Meeting.
Related Projects & Working Groups
Earthquake Forecasting and Predictability (EFP)
