<br><font size=2 face="sans-serif">Seth,</font>
<br>
<br><font size=2 face="sans-serif"> I have some quick comments on
your paper with James Hebden that I think would also be of interest to
members of the CEUS hazards bulletin board.</font>
<br>
<br><font size=2 face="sans-serif">1) On page 3 of your paper, you
mistakenly claim that the national seismic hazard maps (i.e., Frankel et
al. 1996, 2002) define "the hazard" at 2% probability of exceedance
in 50 years. You state "Frankel et al. (1996; 2002) define the
hazard as the maximum shaking predicted at a geographic point with 2% probability
of exceedance in 50 years, or about once in 2,500 years." Actually,
the USGS makes national seismic hazard maps at a variety of probability
levels, based on scientific information such as earthquake recurrence rates
and ground-motion attenuation relations. In fact, we release seismic hazard
curves for a grid of sites across the nation, so that users can calculate
the ground motions at any probability level they choose. I assume
you are referring to the 2/3 times the 2% probability of exceedance
in 50 year level that is used in seismic DESIGN maps in the NEHRP Recommended
Provisions written by the Building Seismic Safety Council, published by
FEMA, and adopted in the International Building Code (IBC) and the ASCE
standards. This probability level for design was not decided by
the U.S. Geological Survey. This probability level and design
procedure were decided by a group of engineers under the Building Seismic
Safety Council (funded by FEMA) and voted on and approved by a wide set
of engineers and engineering groups. It is based on their engineering
judgement of acceptable risk. It's also important to note that in
some areas of the country the design maps are based on a deterministic
calculation of the median ground motions for a characteristic earthquake
on a specific fault. In the 2006 IBC, for example, the design values
around the New Madrid area are based on the median ground motions calculated
for a M7.7 earthquake, averaging five different attenuation relations.</font>
<br>
<br><font size=2 face="sans-serif">2) In the same sentence of your paper
you say "the maximum shaking... with 2% probability of exceedance
in 50 years..." This is not correct. It is not the maximum
shaking. Probabilistic ground motions are the ground motions with
a specified probability of being exceeded. They are not the maximum
shaking. It should also be reiterated that the national seismic
hazard maps are based on the average hazard curves from a variety of input
models and attenuation relations; they are not worst-case maps.</font>
<br>
<br><font size=2 face="sans-serif">3) There seems to be something wrong
with some of your calculations. In your Figure 7, you show significant
changes to the seismic hazard in the northeast U.S. and southeast Canada,
compared to the USGS map, when you change the magnitude and add time dependence
for the New Madrid and Charleston sources. The changes in your hazard
maps extend past 1000 km from these sources. It is very unlikely that the
changes you made in New Madrid and Charleston would significantly affect
the hazard at these distances. As you probably know, we use a 1000
km maximum distance when calculating the hazard in the CEUS for the national
maps, so there is no way changes in New Madrid and Charleston would affect
the hazard calculated for the northeast U.S. </font>
<br>
<br><font size=2 face="sans-serif">4) You use a Gaussian distribution of
recurrence times, rather than the log-normal distribution or Brownian Passage
Time model that are typically used in modern earthquake probability studies,
such as the Working Group on California Earthquake Probabilities (WGCEP,
1995 and 2002). The coefficient of variation (COV; standard deviation divided
by the mean recurrence time) is very important in calculating time-dependent
probabilities and is a source of uncertainty. Values centered at 0.5 are
often assigned the highest weight in California probability studies (e.g.,
WGCEP, 1995. 2002), reflecting the substantial variation in recurrence
times that are observed in many areas that have long enough paleo-event
chronologies.</font>
<br>
<br><font size=2 face="sans-serif">5) Using a time-dependent model with
a log-normal distribution of recurrence times with a COV of 0.5, the USGS
calculated a 7% probability of a 1811-12 type New Madrid earthquake in
the next 50 years, as opposed to the 10% probability found from the time-independent
model. This probability range (7-10%) was stated in the USGS fact
sheet on New Madrid (FS-131-02).</font>
<br>
<br><font size=2 face="sans-serif">6) Of course, key questions are whether
a time-dependent model is appropriate for an intraplate area and what distribution
of recurrence times and COV to use in a probability calculation for these
areas. As many studies have shown, when a large earthquake occurs
on one fault it can increase the stress on nearby faults and increase the
probability of having an earthquake on these faults. So a time dependent
model where the hazard in a region is zero right after a large earthquake
is very naive (it also ignores aftershocks). We know the New Madrid source
zone is actually a fault system rather than a single fault and we might
expect a complicated pattern of loading and unloading not described by
the simple time dependent model used in your paper. In addition,
intraplate fault systems are not loaded in the same way as faults along
plate boundaries, which are being continually loaded by the displacements
of tectonic plates. </font>
<br>
<br>
<br><font size=2 face="sans-serif">Art Frankel<br>
U.S. Geological Survey<br>
MS 966, Box 25046<br>
DFC<br>
Denver, CO 80225<br>
phone: 303-273-8556<br>
fax: 303-273-8600<br>
email: afrankel@usgs.gov</font>