Survival in Seattle: Magnitude estimation, survival analysis, and hazard and loss from Puget Lowland paleoearthquakes

I presented a talk yesterday at the USGS Earthquake Science Center in Menlo Park, California, on statistical paleoseismology in the Puget Lowland region of Washington State. I was a little nervous going into the belly of the beast to deliver this talk, with some of the world's experts on statistical seismology and paleoseismology, and the collectors of some of the data that I'm analyzing, in the audience. But it went well and there was a lot of fantastic discussion in one on one meetings afterwards with several scientists. So many sharp people there!


The paleoseismic record from crustal faults in the Puget Lowland in western Washington State contains ~30 earthquakes that cut the late Pleistocene glacial surface. This rich dataset, produced by decades of scientific effort by the USGS and collaborators, has dramatic implications for seismic hazard and risk in this densely-populated and economically-vital area, and also provides insight into earthquake processes and fault interaction in crustal fault arrays. However, before the fruits of this labor can be snacked upon, more thorough characterization of these paleoearthquakes is necessary. To these ends, I present a three-pronged approach of ongoing research into this characterization:

  1. Magnitude estimation: Paleoseismological techniques measure earthquake displacement at a study site, which is then used in scaling relationships to infer plausible magnitude ranges for each event. However, the fault geometry can add additional constraints to the magnitude as well. By using Bayesian methods that incorporate both rupture length and point displacement, we decrease both the magnitude and the uncertainty in the magnitude estimation for Puget Lowland earthquakes.

  2. Earthquake recurrence and survival analysis: The expected time until the next earthquake may depend on the time since the last earthquake, but the form of this dependence is not always obvious. Using the tools of survival analysis (borrowed from biology and engineering), we can calculate the probability of earthquake occurrence in western Washington given the time since the last event based on the statistics of previous events, rather than by assuming a standard statistical model for earthquake recurrence. This empirical technique lets us directly incorporate phenomena such as earthquake triggering and clustering (which are clearly observed in the Puget Lowland) that standard recurrence models cannot.

  3. Time-dependent seismic hazard and loss analysis: The paleoearthquake record from the Puget Lowland provides quite strong evidence for earthquake triggering and fault interaction to a degree that invalidates the assumptions of time- and space-independence used in classical PSHA. In particular, records of clustered M7+ events within a few years to decades imply that communities with greatly damaged infrastructure from one event may face successive episodes of strong ground shaking. I present very preliminary work describing how expected seismic hazard and loss within a 50-year window may be strongly affected by earthquake clustering, and plans for incorporating cumulative infrastructure damage and repair into PSHA.

A video of the presentation is here.