An earthquake forecasting model has pegged the odds of a San Andreas quake happening within the next year at 1.15%. The model is trained to give accurate probabilities of when earthquakes are likely to occur. The California-based catastrophe modeling company Temblor, Inc. developed the model using machine learning. So far, Temblor says the model’s earthquake forecast for the next year in southern California has been accurate.
A large #SanAndreas #earthquake is 3-5 times more likely over the next 12 months than in the years before the 2019 #Ridgecrest shocks, a new study by @Temblor scientists indicates. Read more in Temblor’s latest article. https://t.co/vYerx9Irjp— temblor (@temblor) July 13, 2020
While that 1.15%—or roughly 1 in 87—chance of the San Andreas Fault experiencing an earthquake over the next year may seem slim, it’s a significant increase over previous predictive odds. Temblor said on Twitter that this figure actually stands as a three to five-fold increase in the odds of the quake happening over those given in the years before 2019.
For those unfamiliar, the San Andreas Fault is a fracture in the Earth’s crust that extends roughly 750 miles through California. The Fault forms a boundary between two tectonic plates: the Pacific Plate and the North American Plate. The Fault is exceptionally dangerous because it’s long and capable of generating large earthquakes. Plus, it’s close to major cities like San Francisco.
Temblor CEO Ross Stein spoke with the
This increase, Temblor states, is because the Ridgecrest fault lines are connected to the San Andreas Fault (red in below picture) by the Garlock Fault (yellow). The Garlock Fault “is now about 100 times more likely to rupture in a large quake than it was prior to the Ridgecrest events.”
Stein, a scientist emeritus of the U.S. Geological Survey and adjunct professor of geophysics at Stanford, outlined his company’s model’s forecast in a recent paper. The
Temblor shared a post describing the paper and noted, “These forecasts are not earthquake predictions, which have so far proven impossible.” Temblor says the model was built based on how earthquakes occur based on their seismicity, how much stress recent earthquakes imparted, and the equations that govern how fault friction takes place.
They tuned the model using machine learning, which allowed it to make “retrospective forecasts” of quakes that have already occurred. In other words, Temblor tested the model by plugging in real-life parameters that were in place prior to previous, recorded earthquakes. Then they had the model generate forecasts to see if the forecasts aligned with what happened in reality.
The possibility of a San Andreas quake AT ANY TIME should already be part of your planning or you shouldn't be in California.— Dr. Lucy Jones (@DrLucyJones) July 13, 2020
Dr. Lucy Jones, a seismologist and public voice for earthquake safety in California not involved with Temblor, wrote on Twitter that even though the model is elegant, it “assumes a reason for quake triggering that is not consensus.” She added in a follow-up tweet that “The possibility of a San Andreas quake AT ANY TIME should already be part of your planning or you shouldn’t be in California.”