[CEUS-earthquake-hazards] no "right" answer

[Seth Stein]

Chris,

I doubt that there's much to be gained by debating the "right" way to
estimate earthquake probabilities. We have lots of models that
give different numbers and there's little reason to believe that any
of them are particularly good or better than the others.

We haven't yet resolved the simplest question: whether
the probability of earthquakes is better modeled
as constant with time or changing.
Different researchers assume one or the other, and
your group - USGS for New Madrid - use both.

Since we're not even sure how useful time dependent models are,
we certainly can't say convincingly which of the time
dependent models is better.

As such, I see little case to say that any way of estimating the
parameters of any particular time dependent model from the limited
earthquake data is "right" or "wrong".

I think this will be the case until we can show that a particular 
approach best predicts the recurrence of future earthquakes. There may 
be no one best approach, and different approaches may turn out better in 
different places.

This seems unlikely to be resolved soon. Recall that Kagan and Jackson 
(1991) estimated that in California, where earthquakes are 30-100 times 
more common that in New Madrid, it would take 3000 years of data to 
distinguish whether time dependent or time independent models did better.

Hence I think it's interesting to compare the probabilities predicted by 
different models with different parameters, but wouldn't place great 
significance on any of the specific values. For New Madrid, they range 
about 0-10%, but beyond that I doubt one could convincingly say more.

I think the best overall assessment is from
Savage's (1991) criticism of earthquake probability estimates
for California, where the data are much better than for
New Madrid and the slip process resulting from steady plate motion
seems to be much less variable with time:

"The range of possible 30-year conditional probabilities for many
of the fault segments is so great due to the uncertainty in the average
recurrence time for that segment that the assigned probability is
virtually meaningless."

"This suggests that the conditional probability is not resolved beyond
broad categories like low (0-10%), intermediate (10-90%), and high
(90% or more) risk. A more quantitative assessment is not justified."

This strikes me as wise humility in the face of
the poorly understood complexity of nature.

Cheers,

Seth

Kagan, Y. Y., and D. D. Jackson, Seismic gap hypothesis: ten years 
after, J. Geophys. Res., 96, 21,419-21,431, 1991.

Savage, J. C., Criticism of some forecasts of the national earthquake
prediction council, Bull. Seismol. Soc. Am., 81, 862-881, 1991.

-- 
Seth Stein
William Deering Professor
Department of Earth and Planetary Sciences
1850 Campus Drive
Northwestern University, Evanston, IL 60208
(847) 491-5265 FAX: (847) 491-8060 E-MAIL: seth at earth.northwestern.edu
http://www.earth.northwestern.edu/people/seth