WEBVTT

4b3088d1-ddb2-4d20-8354-c3e4192820fb/665-0
00:00:06.760 --> 00:00:07.820
According to testing.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/674-0
00:00:10.570 --> 00:00:13.040
Alright, folks, welcome.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/691-0
00:00:13.330 --> 00:00:16.836
I thank you all for attending
the earthquake Science seminar

4b3088d1-ddb2-4d20-8354-c3e4192820fb/691-1
00:00:16.836 --> 00:00:17.640
weekly series.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/696-0
00:00:17.790 --> 00:00:18.960
If you're new, welcome.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/713-0
00:00:19.070 --> 00:00:22.291
If you'd like to be added to the
email distribution group, please

4b3088d1-ddb2-4d20-8354-c3e4192820fb/713-1
00:00:22.291 --> 00:00:23.120
send us an email.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/719-0
00:00:23.930 --> 00:00:25.400
Seminars are recorded mostly.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/740-0
00:00:25.410 --> 00:00:29.163
I'll talks or posted on the USGS
Earthquake Science Center web

4b3088d1-ddb2-4d20-8354-c3e4192820fb/740-1
00:00:29.163 --> 00:00:29.460
page.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/761-0
00:00:29.510 --> 00:00:33.052
Closed captioning can be turned
on by clicking on the CC icon in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/761-1
00:00:33.052 --> 00:00:34.250
the more tab in teams.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/800-0
00:00:34.800 --> 00:00:38.452
Attendees please mute your mics
and turn off your cameras until

4b3088d1-ddb2-4d20-8354-c3e4192820fb/800-1
00:00:38.452 --> 00:00:41.932
the Q&A session at the end
and submit your questions via

4b3088d1-ddb2-4d20-8354-c3e4192820fb/800-2
00:00:41.932 --> 00:00:43.130
the chat at any time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/817-0
00:00:43.540 --> 00:00:46.466
Or wait until the end and then
turn on your camera and we can

4b3088d1-ddb2-4d20-8354-c3e4192820fb/817-1
00:00:46.466 --> 00:00:46.890
call you.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/821-0
00:00:46.900 --> 00:00:48.890
Then a few announcements today.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/865-0
00:00:49.000 --> 00:00:52.814
First, we'd like to congratulate
the USGS earthquake Science

4b3088d1-ddb2-4d20-8354-c3e4192820fb/865-1
00:00:52.814 --> 00:00:56.439
Center doctor, Brian Atwater,
scientist emeritus, who was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/865-2
00:00:56.439 --> 00:01:00.127
honored as the 2023 Marcus
Milling, legendary geoscientist

4b3088d1-ddb2-4d20-8354-c3e4192820fb/865-3
00:01:00.127 --> 00:01:02.690
by the American Geoscientists
Institute.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/902-0
00:01:02.700 --> 00:01:05.853
AGI the Marcus Milling,
legendary geoscientist metal is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/902-1
00:01:05.853 --> 00:01:09.005
a lifetime achievement award
given in recognition of an

4b3088d1-ddb2-4d20-8354-c3e4192820fb/902-2
00:01:09.005 --> 00:01:12.045
individual's high quality,
basic, and applied science

4b3088d1-ddb2-4d20-8354-c3e4192820fb/902-3
00:01:12.045 --> 00:01:14.240
science achievements in Earth
science.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/906-0
00:01:14.430 --> 00:01:15.380
Way to go, Brian.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/911-0
00:01:15.390 --> 00:01:16.110
Thank you.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/920-0
00:01:16.470 --> 00:01:19.660
To read more, look at the intake
you newsletter.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/929-0
00:01:22.690 --> 00:01:24.520
Tomorrow is the ESC.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/931-0
00:01:24.530 --> 00:01:25.520
All hands meeting.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/935-0
00:01:25.570 --> 00:01:25.800
Uh.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/946-0
00:01:25.810 --> 00:01:28.420
See your inbox or teams calendar
for the link.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/961-0
00:01:29.470 --> 00:01:33.935
Uh, next week there are two
special seminars on Monday and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/961-1
00:01:33.935 --> 00:01:34.540
Tuesday.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/988-0
00:01:34.810 --> 00:01:38.992
On Monday, August 21st,
Catherine Materna will be giving

4b3088d1-ddb2-4d20-8354-c3e4192820fb/988-1
00:01:38.992 --> 00:01:43.613
a talk titled faults that Slip
slowly, quickly or somewhere in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/988-2
00:01:43.613 --> 00:01:44.200
between.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1004-0
00:01:44.210 --> 00:01:47.416
Understanding the spectrum of
slip behavior on California's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1004-1
00:01:47.416 --> 00:01:49.820
faults from space geodesy,
that's on Monday.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1038-0
00:01:49.830 --> 00:01:53.247
On Tuesday, Catherine Guns from
Scripps Institution of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1038-1
00:01:53.247 --> 00:01:56.974
Oceanography will be giving a
talk titled investigating the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1038-2
00:01:56.974 --> 00:01:58.900
earthquake cycle in California.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1055-0
00:01:59.460 --> 00:02:03.589
GPS and INSAR data sets, so we
should look forward to those are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1055-1
00:02:03.589 --> 00:02:04.750
both can be great.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1066-0
00:02:04.990 --> 00:02:07.850
Umm additional announcements for
this week.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1088-0
00:02:07.860 --> 00:02:11.429
The student presentations, the
final presentations for the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1088-1
00:02:11.429 --> 00:02:15.300
summer are this Thursday at one
in this room from one to three.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1115-0
00:02:15.310 --> 00:02:19.512
Please tune into that and then
also next week because we have

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1115-1
00:02:19.512 --> 00:02:23.578
special seminars on Monday and
Tuesday, we are coordinating

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1115-2
00:02:23.578 --> 00:02:24.730
with the speaker.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1125-0
00:02:24.970 --> 00:02:26.720
There may or may not postpone.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1149-0
00:02:26.730 --> 00:02:29.805
We wanna make sure that their
audience is maximized as well,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1149-1
00:02:29.805 --> 00:02:31.620
so we will keep you posted on
that.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1165-0
00:02:32.010 --> 00:02:35.072
In the meantime, I'd like to
bring up Anne Marie to introduce

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1165-1
00:02:35.072 --> 00:02:35.960
our speaker, Jess.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1179-0
00:02:41.660 --> 00:02:44.902
Alright, well, it's my pleasure
really to introduce Jess Marie

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1179-1
00:02:44.902 --> 00:02:45.210
today.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1204-0
00:02:45.220 --> 00:02:48.717
And probably many of you know
Jess and have worked with much

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1204-1
00:02:48.717 --> 00:02:50.150
more closely than I have.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1214-0
00:02:50.160 --> 00:02:52.190
But it's really it's always nice
to be here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1256-0
00:02:52.370 --> 00:02:56.126
And so Jess and got her
undergrad at Dartmouth and went

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1256-1
00:02:56.126 --> 00:03:00.285
on to Stanford for Masters and
PhD, and then came to the USGS

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1256-2
00:03:00.285 --> 00:03:03.370
as the men and Hall, and has
been here since.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1291-0
00:03:03.960 --> 00:03:07.634
Yes, and has been a huge
contributor as you can see from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1291-1
00:03:07.634 --> 00:03:11.823
the number of collaborators she
has simply on this talk, she led

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1291-2
00:03:11.823 --> 00:03:13.370
the deformation project.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1318-0
00:03:13.380 --> 00:03:18.335
She was the GPS crystal string
task leader and most recently is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1318-1
00:03:18.335 --> 00:03:22.360
the JAVADOC coordinator for and
shackler import HP.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1343-0
00:03:22.430 --> 00:03:26.368
I think for Shingler, OK, and so
she's worked on many projects in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1343-1
00:03:26.368 --> 00:03:27.620
her time at the USGS.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1361-0
00:03:27.890 --> 00:03:30.092
Most recently, I've gotten to
work with Jess on some of the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1361-1
00:03:30.092 --> 00:03:31.560
work that she's gonna talk about
today.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1406-0
00:03:31.900 --> 00:03:35.170
Looking at shake alert and
trying to incorporate GNSS and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1406-1
00:03:35.170 --> 00:03:38.440
Jess has more recently gotten
interested in ground motion

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1406-2
00:03:38.440 --> 00:03:41.766
modeling, which has been really
really fun and and so it's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1406-3
00:03:41.766 --> 00:03:43.400
really fun to work with Josh.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1451-0
00:03:43.410 --> 00:03:46.981
She's really dedicated, very
detailed, oriented and just

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1451-1
00:03:46.981 --> 00:03:50.928
works incredibly fast, so it's
gonna be sure to be a good talk

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1451-2
00:03:50.928 --> 00:03:54.688
today and she has been leading
this huge effort recently to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1451-3
00:03:54.688 --> 00:03:56.880
incorporate GNSS into Shake
alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1484-0
00:03:56.890 --> 00:04:00.413
So this is the culmination or an
update perhaps on where that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1484-1
00:04:00.413 --> 00:04:03.822
stands and you can see all the
collaborators that have been

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1484-2
00:04:03.822 --> 00:04:04.560
working hard.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1492-0
00:04:04.910 --> 00:04:07.590
So, thanks so much, Jess. Henry.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1504-0
00:04:09.160 --> 00:04:12.310
OK, so it's my pleasure to to
speak today.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1533-0
00:04:12.320 --> 00:04:15.555
It's been a long time since I've
been given a seminar in this

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1533-1
00:04:15.555 --> 00:04:17.590
series, so this is kind of fun
for me.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1564-0
00:04:18.020 --> 00:04:22.230
As Emery said, I'll be talking
about how we're including real

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1564-1
00:04:22.230 --> 00:04:26.508
time GNSS data and the models
that can be derived from them in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1564-2
00:04:26.508 --> 00:04:28.070
the Shake alert system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1594-0
00:04:28.160 --> 00:04:31.742
And there are a lot of people
involved in this effort in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1594-1
00:04:31.742 --> 00:04:34.570
various parts of what I'll be
showing today.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1604-0
00:04:34.580 --> 00:04:37.770
And they're listed down here at
the bottom.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1609-0
00:04:38.960 --> 00:04:39.990
Anti uh.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1645-0
00:04:40.080 --> 00:04:43.768
None of this could have been
done without the help of all

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1645-1
00:04:43.768 --> 00:04:47.774
these people, so I think many of
you know about scheduler, but

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1645-2
00:04:47.774 --> 00:04:49.300
maybe not everyone does.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1674-0
00:04:49.310 --> 00:04:53.474
So Sheckler is an earthquake
early warning system that's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1674-1
00:04:53.474 --> 00:04:57.711
aiming to alert users who will
experience MI 6 or greater

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1674-2
00:04:57.711 --> 00:05:00.340
shaking for magnitude 5 or
greater.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1697-0
00:05:00.350 --> 00:05:04.284
Earthquakes in California,
Oregon and Washington before the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1697-1
00:05:04.284 --> 00:05:08.415
shaking reaches them and also
aims to minimize alerts to users

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1697-2
00:05:08.415 --> 00:05:10.250
who experience weak shaking.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1755-0
00:05:10.680 --> 00:05:15.715
So there's a balance there that
the system needs to to find and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1755-1
00:05:15.715 --> 00:05:20.434
and the way it does this is to
and come up with a real time

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1755-2
00:05:20.434 --> 00:05:25.311
source model and use that to
calculate the anticipated ground

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1755-3
00:05:25.311 --> 00:05:29.874
motion using GMP's and ground
motion intensity conversion

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1755-4
00:05:29.874 --> 00:05:30.660
equations.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-0
00:05:30.670 --> 00:05:35.449
In an example is shown in the
maps here, the left hand map

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-1
00:05:35.449 --> 00:05:40.308
shows the anss location is the
black circle and the chiclet

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-2
00:05:40.308 --> 00:05:45.167
location of a small earthquake
that happened last fall, and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-3
00:05:45.167 --> 00:05:50.269
then this cyan Polygon around it
is the extent of the MI three

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-4
00:05:50.269 --> 00:05:55.128
felt shaking on the right is
what shake alert generated, so

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-5
00:05:55.128 --> 00:05:59.906
the polygons that you see here
are alert regions that were

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-6
00:05:59.906 --> 00:06:04.604
generated from the real time
source models, and these are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-7
00:06:04.604 --> 00:06:09.301
then used by various alert
providers such as the wireless

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-8
00:06:09.301 --> 00:06:14.322
emergency alert system and to
distribute the shake alerts and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1876-9
00:06:14.322 --> 00:06:19.100
those are distributed based on
some threshold for warning.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1895-0
00:06:19.150 --> 00:06:22.895
For example, everyone within the
MMI for region would receive an

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1895-1
00:06:22.895 --> 00:06:23.240
alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1959-0
00:06:25.370 --> 00:06:29.445
So currently the production
system is using two algorithms

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1959-1
00:06:29.445 --> 00:06:33.520
based on seismic data and the
first of these epic uses the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1959-2
00:06:33.520 --> 00:06:37.111
amplitude of the peak
displacement during the first

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1959-3
00:06:37.111 --> 00:06:41.048
few seconds of the P wave and
uses that and it's scaling

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1959-4
00:06:41.048 --> 00:06:45.122
relationship to get magnitude
treating the earthquake as a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1959-5
00:06:45.122 --> 00:06:46.020
point source.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1988-0
00:06:46.070 --> 00:06:49.053
And because it's using very
little data at the very

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1988-1
00:06:49.053 --> 00:06:52.150
beginning of the earthquake,
it's known to experience

4b3088d1-ddb2-4d20-8354-c3e4192820fb/1988-2
00:06:52.150 --> 00:06:54.960
magnitude saturation beyond
about a magnitude 7.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2056-0
00:06:55.850 --> 00:06:59.895
The other algorithm is called
Finder and it converts the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2056-1
00:06:59.895 --> 00:07:04.224
observed high frequency ground
motion to a binary map that's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2056-2
00:07:04.224 --> 00:07:08.411
then correlated with a library
of precomputed templates in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2056-3
00:07:08.411 --> 00:07:12.881
order to get the location, the
length and the orientation of a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2056-4
00:07:12.881 --> 00:07:16.926
line source, and then the
magnitude is obtained from the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2056-5
00:07:16.926 --> 00:07:20.900
line source via magnitude length
scaling relationships.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2097-0
00:07:21.330 --> 00:07:25.239
So those both produce magnitude
estimates that are then combined

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2097-1
00:07:25.239 --> 00:07:29.028
by a third algorithm, which is
called the solution aggregator,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2097-2
00:07:29.028 --> 00:07:32.936
which does a weighted average of
those where the weighting based

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2097-3
00:07:32.936 --> 00:07:34.740
on the reported uncertainties.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2112-0
00:07:37.230 --> 00:07:40.840
So the talk that I'm giving
today has two parts to it.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2146-0
00:07:40.850 --> 00:07:44.645
The first part is I'm going to
describe how we've been

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2146-1
00:07:44.645 --> 00:07:48.854
incorporating a method based on
real time GNSS data that can

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2146-2
00:07:48.854 --> 00:07:51.890
provide non saturating magnitude
estimates.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2172-0
00:07:52.900 --> 00:07:56.927
And the second part is going to
explore the potential

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2172-1
00:07:56.927 --> 00:08:01.624
contribution that real time GNSS
based distributed slip models

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2172-2
00:08:01.624 --> 00:08:04.010
might make to improved alerting.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2190-0
00:08:06.490 --> 00:08:12.310
So in the first part and to
obtain non saturating magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2190-1
00:08:12.310 --> 00:08:13.280
estimates.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2212-0
00:08:13.890 --> 00:08:17.089
What we're using is the peak
ground displacement that's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2212-1
00:08:17.089 --> 00:08:18.460
computed from GNSS data.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2243-0
00:08:18.470 --> 00:08:23.308
So the GNSS, UMM, system can
record the the ongoing ground

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2243-1
00:08:23.308 --> 00:08:28.638
motion during an earthquake and
some examples are shown here for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2243-2
00:08:28.638 --> 00:08:30.770
the Ridgecrest main shock.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-0
00:08:30.840 --> 00:08:35.009
The north and East components
and you can see the displacement

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-1
00:08:35.009 --> 00:08:39.111
at a number of GNSS stations
that were in the region and then

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-2
00:08:39.111 --> 00:08:42.948
that information is then used to
compute the total ground

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-3
00:08:42.948 --> 00:08:46.587
displacement, which is just the
the vector norm of the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-4
00:08:46.587 --> 00:08:50.094
horizontal of the three
component displacements as a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-5
00:08:50.094 --> 00:08:53.997
function of time during that
earthquake and the peak round

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-6
00:08:53.997 --> 00:08:57.570
displacement is then the maximum
of that total ground

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-7
00:08:57.570 --> 00:09:01.606
displacement from the origin
time of the earthquake up until

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2341-8
00:09:01.606 --> 00:09:04.980
whatever observation,
observation, epic you're at.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2357-0
00:09:04.990 --> 00:09:08.252
So these figures on the right
show some examples from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2357-1
00:09:08.252 --> 00:09:09.580
different earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2409-0
00:09:09.850 --> 00:09:14.014
These stations were all about
100 kilometers from the source,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2409-1
00:09:14.014 --> 00:09:18.311
but because of the size of the
earthquake, the the rupture that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2409-2
00:09:18.311 --> 00:09:22.474
happened the the displacement
seem different and you can also

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2409-3
00:09:22.474 --> 00:09:26.570
note that the Y scales are quite
different for the different

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2409-4
00:09:26.570 --> 00:09:27.040
events.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2472-0
00:09:28.710 --> 00:09:33.197
So Brendan Crawl and others
developed a magnitude scaling

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2472-1
00:09:33.197 --> 00:09:38.071
relationship that relates the
peak ground displacement and the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2472-2
00:09:38.071 --> 00:09:42.945
distance from the station to the
source to obtain a magnitude,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2472-3
00:09:42.945 --> 00:09:46.968
and this is one part of an
earthquake early warning

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2472-4
00:09:46.968 --> 00:09:51.300
algorithm called G Fast which
uses real time GNSS data.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2499-0
00:09:51.410 --> 00:09:55.351
It also includes two other
modules which do a fast CMT

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2499-1
00:09:55.351 --> 00:09:58.790
solution as well as umm
distributed slip model.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2521-0
00:09:58.800 --> 00:10:02.531
But today I'm just focusing on
the part of Gefest that uses the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2521-1
00:10:02.531 --> 00:10:05.270
peak ground displacement to
compute magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2608-0
00:10:05.860 --> 00:10:10.695
So the bottom figure here shows
the scaling relationship for 33

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2608-1
00:10:10.695 --> 00:10:15.606
earthquakes from magnitude 6 to
9.1, and this is from a study by

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2608-2
00:10:15.606 --> 00:10:20.290
Goldberg and others who used
this expanded data set to update

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2608-3
00:10:20.290 --> 00:10:24.672
the coefficients that were
derived initially by crawl and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2608-4
00:10:24.672 --> 00:10:29.356
others and have an example of an
application of this that was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2608-5
00:10:29.356 --> 00:10:33.058
done in a near real time
implementation that was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2608-6
00:10:33.058 --> 00:10:34.720
incorporated onto you.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2681-0
00:10:34.730 --> 00:10:38.821
Incorporated onto you navco's
web pages is shown here for the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2681-1
00:10:38.821 --> 00:10:43.043
two largest earthquakes of the
Ridgecrest sequence, and So what

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2681-2
00:10:43.043 --> 00:10:47.067
you're seeing is on the X axis
time since the origin time of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2681-3
00:10:47.067 --> 00:10:51.157
the earthquake, and then the
magnitude estimate which evolves

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2681-4
00:10:51.157 --> 00:10:55.314
with time as more data come in
and each of the great traces is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2681-5
00:10:55.314 --> 00:10:58.810
the magnitude determined just
from a single station.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2683-0
00:10:58.820 --> 00:10:59.240
Station.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2710-0
00:10:59.690 --> 00:11:04.048
But the black heavier trace is
the best fitting magnitude from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2710-1
00:11:04.048 --> 00:11:08.060
all the data that are available
at each time step in the.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-0
00:11:08.070 --> 00:11:11.596
The Little Cross is marked the
first data point for each

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-1
00:11:11.596 --> 00:11:15.370
station, so the reason that
we're particularly interested in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-2
00:11:15.370 --> 00:11:19.329
incorporating this approach into
Sheckler is that it offers non

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-3
00:11:19.329 --> 00:11:22.916
saturating magnitude estimates
for magnitude 7 or greater

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-4
00:11:22.916 --> 00:11:26.257
earthquakes and this is
important, especially when we

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-5
00:11:26.257 --> 00:11:30.216
think about the Cascadia region
where we anticipate that in the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-6
00:11:30.216 --> 00:11:34.113
future there may be very large
earthquakes and we want to make

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2796-7
00:11:34.113 --> 00:11:37.020
sure that we get accurate
magnitude estimates.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2822-0
00:11:38.730 --> 00:11:43.733
So it's it's been a long process
and I was just trying to think

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2822-1
00:11:43.733 --> 00:11:46.860
of how long we've been working
on this.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2825-0
00:11:46.870 --> 00:11:47.680
It's been a while.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2851-0
00:11:48.050 --> 00:11:53.474
A few years probably towards
integrating this algorithm into

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2851-1
00:11:53.474 --> 00:11:57.920
shake alert for in the future of
production used.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2873-0
00:11:57.930 --> 00:12:01.892
So basically these are some of
the steps that that were

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2873-1
00:12:01.892 --> 00:12:03.660
required in this process.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2911-0
00:12:03.670 --> 00:12:07.252
We had to connect first the real
time data streams from the GNSS

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2911-1
00:12:07.252 --> 00:12:10.338
stations that are in shakers
boundaries to at first the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2911-2
00:12:10.338 --> 00:12:13.258
development system and
ultimately we've done that to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2911-3
00:12:13.258 --> 00:12:14.470
the production system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2985-0
00:12:15.990 --> 00:12:19.447
We prepared a test suite
consisting of Genesis S data in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2985-1
00:12:19.447 --> 00:12:23.207
the corresponding seismic data
for a number of real events in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2985-2
00:12:23.207 --> 00:12:26.725
the US and Japan, and then we
implemented this G fast PGD

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2985-3
00:12:26.725 --> 00:12:30.121
algorithm first in the
development system in a way that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2985-4
00:12:30.121 --> 00:12:34.003
allowed it to run jointly with
the seismic algorithms, which is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/2985-5
00:12:34.003 --> 00:12:37.520
what was required to actually
test it in a realistic way.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3028-0
00:12:37.670 --> 00:12:41.488
So it has two modes real time
mode where it's ingesting the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3028-1
00:12:41.488 --> 00:12:45.307
position time series that are
coming in all the time and it

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3028-2
00:12:45.307 --> 00:12:49.125
this is triggered by the seismic
portion of the Shake Alert

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3028-3
00:12:49.125 --> 00:12:49.570
system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3052-0
00:12:50.060 --> 00:12:53.581
And there's also a replay mode,
which allows the retrospective

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3052-1
00:12:53.581 --> 00:12:56.430
analysis of the historic events
in the test suite.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-0
00:12:57.310 --> 00:13:01.721
And in sort of the facts of it
was developing a methodology for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-1
00:13:01.721 --> 00:13:05.512
combining the magnitude
estimates that come from the G

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-2
00:13:05.512 --> 00:13:09.717
fast algorithm with those from
the other algorithms and that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-3
00:13:09.717 --> 00:13:13.576
required estimating the
uncertainty of those magnitudes

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-4
00:13:13.576 --> 00:13:17.574
that could be used in the
weighted average and developing

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-5
00:13:17.574 --> 00:13:21.571
logic that would determine when
the chief asked magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-6
00:13:21.571 --> 00:13:25.362
estimates would actually be
passed on to this solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3131-7
00:13:25.362 --> 00:13:26.120
aggregator.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3180-0
00:13:26.130 --> 00:13:31.216
And I'm enabling that solution
aggregator algorithm to actually

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3180-1
00:13:31.216 --> 00:13:36.302
use all the data to all the the
source magnitudes together, and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3180-2
00:13:36.302 --> 00:13:41.228
then we carried out three plays
of the test suite data and it

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3180-3
00:13:41.228 --> 00:13:46.155
with all the algorithms running
together and fine tune things

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3180-4
00:13:46.155 --> 00:13:46.870
that way.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3225-0
00:13:46.880 --> 00:13:50.801
So these last several bullets
are what I'll be talking about

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3225-1
00:13:50.801 --> 00:13:54.594
today and ultimately at the end
of all this process, we've

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3225-2
00:13:54.594 --> 00:13:58.001
submitted the code for the
formal system testing and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3225-3
00:13:58.001 --> 00:14:00.700
performance evaluation, which is
ongoing.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3283-0
00:14:02.410 --> 00:14:07.585
So it turns out that in order to
properly use real time DNS data

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3283-1
00:14:07.585 --> 00:14:12.601
for earthquake early warning,
you actually need to think a lot

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3283-2
00:14:12.601 --> 00:14:17.457
about the noise characteristics
of the data, and this figure

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3283-3
00:14:17.457 --> 00:14:22.233
shows a zooming in from longer
time periods into short time

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3283-4
00:14:22.233 --> 00:14:22.870
periods.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-0
00:14:23.790 --> 00:14:28.109
Some of the characteristics of
noise that are is seen in in the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-1
00:14:28.109 --> 00:14:32.090
GNSS data, and in particular
you'll see that the real time

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-2
00:14:32.090 --> 00:14:36.544
Genos time series can have large
outliers as well as time varying

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-3
00:14:36.544 --> 00:14:40.188
noise due to a variety of
factors including satellite

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-4
00:14:40.188 --> 00:14:43.292
clock errors and failed
ambiguity resolution,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-5
00:14:43.292 --> 00:14:47.679
atmospheric effects, multipath,
and all of these are things that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-6
00:14:47.679 --> 00:14:51.997
are affecting the processing of
the data that takes it from the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-7
00:14:51.997 --> 00:14:56.181
raw measurements to positions
and focusing in on time periods

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3388-8
00:14:56.181 --> 00:14:59.420
that are relevant for earthquake
early warning.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3424-0
00:14:59.430 --> 00:15:03.519
You can see that there's, uh
colored noise in the data that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3424-1
00:15:03.519 --> 00:15:07.744
causes the position to drift
with time, and that's evident in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3424-2
00:15:07.744 --> 00:15:09.720
in these example time series.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3505-0
00:15:09.730 --> 00:15:14.145
You can also see by looking at
the how about ballistic power

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3505-1
00:15:14.145 --> 00:15:18.631
spectral density, which is shown
in the bottom plots that the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3505-2
00:15:18.631 --> 00:15:23.191
noise level increases with time,
particularly over this period

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3505-3
00:15:23.191 --> 00:15:27.388
from 10 to about 60 seconds,
which is of course period of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3505-4
00:15:27.388 --> 00:15:31.296
that we're particularly
interested in for earthquake,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3505-5
00:15:31.296 --> 00:15:35.421
early warning and so this is
something that we needed to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3505-6
00:15:35.421 --> 00:15:39.690
account for when incorporating
these data into the system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3562-0
00:15:41.250 --> 00:15:45.092
So as you can imagine, since at
each time step following an

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3562-1
00:15:45.092 --> 00:15:49.126
earthquake origin time the peak
displacement is calculated and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3562-2
00:15:49.126 --> 00:15:52.904
within that window from the
origin time to the observation

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3562-3
00:15:52.904 --> 00:15:56.106
time, this drift in the
positions accumulates and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3562-4
00:15:56.106 --> 00:15:59.884
generally causes the peak ground
displacement estimates to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3562-5
00:15:59.884 --> 00:16:01.100
increase with time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3614-0
00:16:02.400 --> 00:16:06.886
And so we when about exploring
how this was going to affect the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3614-1
00:16:06.886 --> 00:16:11.021
magnitude estimates by looking
at real time data that were

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3614-2
00:16:11.021 --> 00:16:15.227
collected at the GNSS stations
in the Shaklee region, which

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3614-3
00:16:15.227 --> 00:16:18.240
were shown in this map is the
red circles.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3618-0
00:16:19.300 --> 00:16:20.420
And then what we did was?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3656-0
00:16:20.480 --> 00:16:25.063
Those to find a a grid of
hypothetical epicenters across

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3656-1
00:16:25.063 --> 00:16:30.128
the Shake Alert region and it
temporal spacing of hypothetical

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3656-2
00:16:30.128 --> 00:16:34.228
origin times and each
combination of epicenter and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3656-3
00:16:34.228 --> 00:16:34.790
origin.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3682-0
00:16:34.800 --> 00:16:38.356
Time defined a hypothetical
earthquake that might happen

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3682-1
00:16:38.356 --> 00:16:41.787
within the Shake Alert region,
and from those we could

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3682-2
00:16:41.787 --> 00:16:42.410
calculate.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3709-0
00:16:42.420 --> 00:16:45.777
Then the the peak ground
displacement from the actual

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3709-1
00:16:45.777 --> 00:16:49.258
real time data that were
recorded in the absence of any

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3709-2
00:16:49.258 --> 00:16:51.060
genus detectable earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3754-0
00:16:51.070 --> 00:16:54.776
So these are time series of
noise only and we can calculate

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3754-1
00:16:54.776 --> 00:16:58.359
the peak ground displacements
and also the magnitude that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3754-2
00:16:58.359 --> 00:17:01.941
would be derived from them at
any time step following the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3754-3
00:17:01.941 --> 00:17:02.620
earthquake.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3794-0
00:17:03.950 --> 00:17:09.209
And So what we found was looking
at the PGD estimates that with

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3794-1
00:17:09.209 --> 00:17:14.303
increasing time after the origin
time going from 203050 to 90

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3794-2
00:17:14.303 --> 00:17:19.479
seconds, we we saw that the peak
ground displacement generally

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3794-3
00:17:19.479 --> 00:17:20.300
increased.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3821-0
00:17:20.310 --> 00:17:24.897
So what you're seeing in these
histograms is the daily 99th

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3821-1
00:17:24.897 --> 00:17:29.560
percentile PGD calculated from
these noise only time series.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3851-0
00:17:30.110 --> 00:17:33.834
And then as you might imagine,
the magnitudes that were

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3851-1
00:17:33.834 --> 00:17:38.155
calculated from these tended to
also drift to larger values with

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3851-2
00:17:38.155 --> 00:17:39.950
time passed the earthquake.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3885-0
00:17:39.960 --> 00:17:44.107
So essentially what you're
seeing in this lower figure is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3885-1
00:17:44.107 --> 00:17:48.396
the magnitude estimates obtained
from these noise only time

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3885-2
00:17:48.396 --> 00:17:53.043
series at different time windows
following a hypothetical origin

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3885-3
00:17:53.043 --> 00:17:53.400
time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3947-0
00:17:53.670 --> 00:17:57.506
And so, for example, with 20
seconds following the origin

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3947-1
00:17:57.506 --> 00:18:01.541
time just based on noise, you're
gonna end up with magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3947-2
00:18:01.541 --> 00:18:05.443
estimates that are in the in mid
6 range often, and so the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3947-3
00:18:05.443 --> 00:18:09.412
minimum reliable magnitude
estimate is increasing with time

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3947-4
00:18:09.412 --> 00:18:13.248
since the origin time and this
underpins our strategy for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/3947-5
00:18:13.248 --> 00:18:16.290
incorporating the algorithm into
shake alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4000-0
00:18:17.550 --> 00:18:21.461
So we developed a strategy for
doing this, which is spelled out

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4000-1
00:18:21.461 --> 00:18:25.371
in a paper that came out earlier
this year and was described in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4000-2
00:18:25.371 --> 00:18:29.098
this paper's essentially the
framework that's being used now

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4000-3
00:18:29.098 --> 00:18:32.703
in the testing system, with a
few modifications that we've

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4000-4
00:18:32.703 --> 00:18:33.680
made since then.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4011-0
00:18:33.850 --> 00:18:35.560
But I'll summarize it here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4045-0
00:18:35.570 --> 00:18:39.062
So as a reminder, the solution
aggregator is computing a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4045-1
00:18:39.062 --> 00:18:42.555
weighted average of the
magnitudes that it receives from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4045-2
00:18:42.555 --> 00:18:45.740
whatever algorithms contributed
a given time stuff.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4065-0
00:18:45.750 --> 00:18:49.181
So that would be potentially up
to three algorithms up big

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4065-1
00:18:49.181 --> 00:18:50.460
Finder and cheap fast.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4081-0
00:18:51.320 --> 00:18:55.584
And I think it's important to
remember that Jeff has just is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4081-1
00:18:55.584 --> 00:18:58.310
triggered by the solution
aggregators.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4096-0
00:18:58.320 --> 00:19:01.550
So it does not try to detect the
earthquakes itself.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4123-0
00:19:01.600 --> 00:19:05.896
It, but once the seismic portion
of the system has determined the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4123-1
00:19:05.896 --> 00:19:08.890
intake is happening, then G fast
can kick in.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4173-0
00:19:08.900 --> 00:19:12.965
So it's triggered when there's
seismic evidence of an ongoing

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4173-1
00:19:12.965 --> 00:19:16.243
earthquake, and it can
contribute to the solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4173-2
00:19:16.243 --> 00:19:19.651
aggregator if the solution
aggregators sizmek based

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4173-3
00:19:19.651 --> 00:19:23.454
magnitude is at least six and
pretty fast magnitude is at

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4173-4
00:19:23.454 --> 00:19:24.240
least seven.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4187-0
00:19:24.250 --> 00:19:28.480
So really we're using G fast to
focus on the large events.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4193-0
00:19:28.490 --> 00:19:30.730
That's where we think it can
contribute.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4228-0
00:19:31.560 --> 00:19:36.451
Uh, we're using the filter that
eliminates noisy data based on

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4228-1
00:19:36.451 --> 00:19:41.418
the reported GNSS position and
certainties that come out of the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4228-2
00:19:41.418 --> 00:19:42.660
data processing.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4297-0
00:19:42.910 --> 00:19:47.733
So that prevents data that are
likely to be not reliable from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4297-1
00:19:47.733 --> 00:19:51.777
ever getting into the
calculation, and then we have

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4297-2
00:19:51.777 --> 00:19:56.366
two what we call throttling
criteria, which are that the G

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4297-3
00:19:56.366 --> 00:20:00.799
fast PGD algorithm only sends
magnitude estimate to this

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4297-4
00:20:00.799 --> 00:20:05.544
solution aggregator if at least
three stations within G fast

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4297-5
00:20:05.544 --> 00:20:08.810
travel time radius have enough
dissenter.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-0
00:20:08.820 --> 00:20:13.154
Have PGD observed that is at
least a threshold value where

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-1
00:20:13.154 --> 00:20:17.782
that threshold is time varying
in based on the analysis of the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-2
00:20:17.782 --> 00:20:22.190
noise time series that I was
showing earlier and the second

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-3
00:20:22.190 --> 00:20:26.818
criteria is that the estimated
uncertainty on the magnitude is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-4
00:20:26.818 --> 00:20:31.593
less than half a magnitude unit
and when when PGD is included in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-5
00:20:31.593 --> 00:20:36.368
the in the solution aggregator,
it's weighted by its uncertainty

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-6
00:20:36.368 --> 00:20:40.849
which decreases with increasing
magnitude and increases with

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-7
00:20:40.849 --> 00:20:45.183
time since the earthquake
because of how we saw that those

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4403-8
00:20:45.183 --> 00:20:49.370
magnitude estimates will tend to
drift higher with time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4439-0
00:20:50.360 --> 00:20:54.334
So just to reiterate the message
throttling parameters, the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4439-1
00:20:54.334 --> 00:20:58.043
thresholds in the magnitude
uncertainty are empirically

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4439-2
00:20:58.043 --> 00:21:02.348
derived through analysis of the
effect of GNSS time series noise

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4439-3
00:21:02.348 --> 00:21:03.540
on the magnitudes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4464-0
00:21:05.430 --> 00:21:11.033
So as I mentioned earlier, we're
now in the formal testing phase

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4464-1
00:21:11.033 --> 00:21:12.240
of G fast PGD.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4475-0
00:21:12.250 --> 00:21:16.220
As in incorporated into the
Shake Alert system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4506-0
00:21:16.230 --> 00:21:20.254
And so this is running on test
servers and once that's done, if

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4506-1
00:21:20.254 --> 00:21:23.900
the IT shows that it improves
the system then it would be

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4506-2
00:21:23.900 --> 00:21:26.100
included in the production
system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4539-0
00:21:26.550 --> 00:21:30.782
So this testing started in
February and it has sort of two

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4539-1
00:21:30.782 --> 00:21:34.870
components to it, real time
operation and also replay of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4539-2
00:21:34.870 --> 00:21:37.380
data from the historic test
suite.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4569-0
00:21:38.100 --> 00:21:41.795
So this figure is summarizing
what we've observed with the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4569-1
00:21:41.795 --> 00:21:43.360
real time implementation.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4638-0
00:21:43.780 --> 00:21:47.202
And this is from a A the an
instance running on the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4638-1
00:21:47.202 --> 00:21:51.151
development system, but using
the same settings as what are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4638-2
00:21:51.151 --> 00:21:54.770
being used in the the test
system, but the development

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4638-3
00:21:54.770 --> 00:21:58.916
system has results that go back
in time further, so I chose to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4638-4
00:21:58.916 --> 00:22:03.128
show this one and the top plot
shows the number of triggers per

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4628-0
00:22:00.670 --> 00:22:02.890
Next Tuesday. Ohh.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4638-5
00:22:03.128 --> 00:22:07.010
day that Gefasst was receiving
from the seismic algorithm.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4678-0
00:22:07.020 --> 00:22:10.834
So for example, on the day of
the Ferndale earthquake, as as

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4678-1
00:22:10.834 --> 00:22:14.898
you might imagine, there were a
lot of triggers and occasionally

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4678-2
00:22:14.898 --> 00:22:18.650
there are other days where maybe
there was a small swarm or

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4678-3
00:22:18.650 --> 00:22:19.900
something like that.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4699-0
00:22:19.910 --> 00:22:22.624
So there are some days that have
quite a few triggers, others

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4699-1
00:22:22.624 --> 00:22:23.630
that have not too many.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4722-0
00:22:25.720 --> 00:22:31.375
The middle plot is showing how
gene fast message throttling

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4722-1
00:22:31.375 --> 00:22:37.030
criteria affected what was sent
to the solution aggregator.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4774-0
00:22:37.040 --> 00:22:40.209
So every time that she faced was
triggered, then it needed to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4774-1
00:22:40.209 --> 00:22:43.174
meet those two criteria of
having at least three stations

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4774-2
00:22:43.174 --> 00:22:45.423
with observed ground
displacement exceeding

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4774-3
00:22:45.423 --> 00:22:48.591
thresholds and then having a
magnitude estimate that was less

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4774-4
00:22:48.591 --> 00:22:49.460
than half a unit.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4797-0
00:22:49.830 --> 00:22:55.898
And so the days said, had some
events where either of those

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4797-1
00:22:55.898 --> 00:22:59.640
were those two criteria were
evaded.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4821-0
00:22:59.650 --> 00:23:03.289
Essentially, I've shown with the
green Fern for magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4821-1
00:23:03.289 --> 00:23:06.300
uncertainty and the blue for the
PGD threshold.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4833-0
00:23:06.590 --> 00:23:13.440
With the this Y axis indicating
how many seconds.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4844-0
00:23:13.450 --> 00:23:15.380
Basically, how many time steps
had that?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4856-0
00:23:15.520 --> 00:23:19.703
And then there were three times
where both of those criteria

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4856-1
00:23:19.703 --> 00:23:20.320
were met.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4899-0
00:23:20.330 --> 00:23:23.676
So there were no times when
actually PGD magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4899-1
00:23:23.676 --> 00:23:27.612
instrument was sent on to the
solution anchor Gator because

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4899-2
00:23:27.612 --> 00:23:31.482
they the solution aggregator
itself did not think that the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4899-3
00:23:31.482 --> 00:23:34.040
earthquake was at least a
magnitude 6.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4926-0
00:23:34.050 --> 00:23:38.391
We don't want too fast to send
its magnitude estimate, which

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4926-1
00:23:38.391 --> 00:23:40.170
may be just due to noise.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4951-0
00:23:40.530 --> 00:23:44.128
Then you have 6 1/2 if the size
being part of the system thinks

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4951-1
00:23:44.128 --> 00:23:45.140
it's only a 3 1/2.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4966-0
00:23:45.150 --> 00:23:50.580
So so that criteria prevented
these ones from being sent in.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4996-0
00:23:51.170 --> 00:23:55.549
The other thing I would point
out is that we did, based on the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4996-1
00:23:55.549 --> 00:23:59.650
results of the very earliest
real time testing, adjust the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/4996-2
00:23:59.650 --> 00:24:01.040
throttling criteria.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5032-0
00:24:01.050 --> 00:24:05.003
So you see a lot of times early
on when it was exceeding the PGD

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5032-1
00:24:05.003 --> 00:24:08.651
threshold and then we adjusted
those parameters at the time

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5032-2
00:24:08.651 --> 00:24:09.320
shown here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5064-0
00:24:09.670 --> 00:24:14.735
And finally, just have is
another piece of of information,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5064-1
00:24:14.735 --> 00:24:20.316
and the bottom you're seeing the
number of stations per day that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5064-2
00:24:20.316 --> 00:24:22.290
had any amount of data.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5077-0
00:24:22.840 --> 00:24:26.647
Who that exceeded the
uncertainty threshold that's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5077-1
00:24:26.647 --> 00:24:28.140
applied to the data.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5120-0
00:24:28.150 --> 00:24:32.314
So these are the the data that
would be filtered out and you

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5120-1
00:24:32.314 --> 00:24:36.274
can see that for example, on
these days where we had some

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5120-2
00:24:36.274 --> 00:24:40.165
cases where the magnitude
estimate was exceeding some of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5120-3
00:24:40.165 --> 00:24:41.120
the throttles.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5167-0
00:24:41.130 --> 00:24:44.489
It are days when there happened
to be a lot of noisy data, so

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5167-1
00:24:44.489 --> 00:24:47.630
these these data were weeded
out, but it would not be and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5167-2
00:24:47.630 --> 00:24:50.609
expected that there would be
other data that maybe had

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5167-3
00:24:50.609 --> 00:24:52.180
slightly lower uncertainties.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5188-0
00:24:52.190 --> 00:24:56.834
But we're still not reliable, so
that's what we have been seeing

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5188-1
00:24:56.834 --> 00:24:58.620
in the real time testing.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5223-0
00:24:58.630 --> 00:25:02.459
And I think the important things
to point out, however, that no

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5223-1
00:25:02.459 --> 00:25:05.689
erroneous magnitudes have been
passed to the solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5223-2
00:25:05.689 --> 00:25:08.560
aggregator with the criteria
that we've set up.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5244-0
00:25:09.250 --> 00:25:13.091
Umm, there were very few
instances where both the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5244-1
00:25:13.091 --> 00:25:16.240
throttling criteria where evaded
and it.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5287-0
00:25:16.290 --> 00:25:21.050
It's interesting to note that no
magnitude was reported for the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5287-1
00:25:21.050 --> 00:25:25.661
Ferndale earthquake, and that
was as expected, because that's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5287-2
00:25:25.661 --> 00:25:30.421
really within sort at the limit
of what's reliable to report on

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5287-3
00:25:30.421 --> 00:25:32.280
using the G fast systems.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5353-0
00:25:32.290 --> 00:25:36.681
So umm, the other thing I would
point out is that we had we did

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5353-1
00:25:36.681 --> 00:25:41.071
notice in the real time testing
that one thing that can lead to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5353-2
00:25:41.071 --> 00:25:45.393
potentially to overlearning is
when steps are excursions occur

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5353-3
00:25:45.393 --> 00:25:49.509
in the GPS time series and is
often has to do with problems

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5353-4
00:25:49.509 --> 00:25:53.350
with the clock corrections that
are used in processing.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5423-0
00:25:53.360 --> 00:25:56.657
So Central Washington
University, who processes the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5423-1
00:25:56.657 --> 00:26:00.652
data, are developing a real time
binary data quality flag that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5423-2
00:26:00.652 --> 00:26:04.456
would, umm, take information
that's known to the processing

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5423-3
00:26:04.456 --> 00:26:08.641
software about things like clock
corrections, satellite elevation

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5423-4
00:26:08.641 --> 00:26:12.573
angles, things like this, and
use that to weed out data which

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5423-5
00:26:12.573 --> 00:26:16.440
we think will also mitigate the
potential for over alerting.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5434-0
00:26:17.740 --> 00:26:20.290
So that's looking at the real
time results.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5492-0
00:26:20.560 --> 00:26:24.597
It's also interesting to look at
replaying data for past events

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5492-1
00:26:24.597 --> 00:26:28.445
in a simulated real time mode,
so I'll show you some results

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5492-2
00:26:28.445 --> 00:26:32.293
from the paper that came out
earlier this year and these are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5492-3
00:26:32.293 --> 00:26:36.014
historic events that are also
being run through the formal

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5492-4
00:26:36.014 --> 00:26:36.960
testing system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5519-0
00:26:37.810 --> 00:26:40.170
But the results that I'm showing
you here are the ones from the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5519-1
00:26:40.170 --> 00:26:42.160
paper with the settings that
were used in that paper.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5553-0
00:26:42.220 --> 00:26:46.611
So and so the size we can gain
access data are replayed

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5553-1
00:26:46.611 --> 00:26:51.316
simultaneously in a simulated
real time mode and that makes

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5553-2
00:26:51.316 --> 00:26:53.590
them run through epic Finder.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5600-0
00:26:54.040 --> 00:26:58.398
The solution aggregator in G
Fast PGD all running together

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5600-1
00:26:58.398 --> 00:27:02.239
jointly and as I've already
discussed, the solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5600-2
00:27:02.239 --> 00:27:06.449
aggregator was modified to
include G fests magnitude and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5600-3
00:27:06.449 --> 00:27:10.438
the GEFEST algorithm is
triggered by the the solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5600-4
00:27:10.438 --> 00:27:11.250
aggregator.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5647-0
00:27:11.840 --> 00:27:15.828
So let's start with an example
from Kumamoto, which I was a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5647-1
00:27:15.828 --> 00:27:19.949
magnitude 7 earthquake and I'll
I'll explain this plan detail

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5647-2
00:27:19.949 --> 00:27:23.670
because you'll see similar ones
in the next few slides.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-0
00:27:23.680 --> 00:27:28.284
So basically what we're seeing
here is the time past origin

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-1
00:27:28.284 --> 00:27:33.118
time and the estimated magnitude
from the different algorithms

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-2
00:27:33.118 --> 00:27:38.029
and the shaded regions are the
reported uncertainties from each

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-3
00:27:38.029 --> 00:27:42.709
algorithm and the two there are
two versions of the solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-4
00:27:42.709 --> 00:27:47.159
aggregator that are shown, one
that is what's the current

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-5
00:27:47.159 --> 00:27:51.916
production system with no G,
fast included and then the black

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-6
00:27:51.916 --> 00:27:56.674
dots, which are what, what it
would be if you included G fast

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-7
00:27:56.674 --> 00:28:01.201
and so uh in this case what
where you can see if you stare

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-8
00:28:01.201 --> 00:28:05.497
at it for a bit is that these
blue boxes are the D fast

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-9
00:28:05.497 --> 00:28:10.101
magnitude and it's magnitude
never really got as high as it

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5781-10
00:28:10.101 --> 00:28:13.170
should to match the observed
magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5790-0
00:28:13.410 --> 00:28:18.080
The the Nic magnitude and so
when.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5826-0
00:28:18.150 --> 00:28:22.761
So in fact, although finders
magnitude estimates started out

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5826-1
00:28:22.761 --> 00:28:27.448
low, eventually a climbed up to
what was close to the correct

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5826-2
00:28:27.448 --> 00:28:31.606
value, and in this case,
including GST was not helpful

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5826-3
00:28:31.606 --> 00:28:32.210
because.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5845-0
00:28:34.480 --> 00:28:37.594
Because it was lower, it
actually pulled the solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5845-1
00:28:37.594 --> 00:28:39.670
aggregator magnitude estimate
down.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5891-0
00:28:39.840 --> 00:28:43.447
So this would be an example
where this this earthquake is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5891-1
00:28:43.447 --> 00:28:47.177
probably on the low end of what
we would want to be fast to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5891-2
00:28:47.177 --> 00:28:50.535
contribute to, and in this
particular case it was not

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5891-3
00:28:50.535 --> 00:28:51.840
improving the system.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5903-0
00:28:53.090 --> 00:28:57.060
However, we applied it then to
the Ridgecrest main shock.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5953-0
00:28:57.070 --> 00:29:02.125
So 7.1 and here you can see this
is a similar kind of plot and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5953-1
00:29:02.125 --> 00:29:07.340
the blue squares again are the G
fast magnitude estimates and in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5953-2
00:29:07.340 --> 00:29:12.154
this case they reached a level
closer to the true magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5953-3
00:29:12.154 --> 00:29:14.160
sooner than did find her.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5982-0
00:29:14.170 --> 00:29:19.351
So they actually increased the
solution aggregator magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5982-1
00:29:19.351 --> 00:29:24.531
compared to what it would have
been without having it having

4b3088d1-ddb2-4d20-8354-c3e4192820fb/5982-2
00:29:24.531 --> 00:29:25.380
breakfast.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6011-0
00:29:25.390 --> 00:29:29.662
And this system and what you're
seeing on the right is that how

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6011-1
00:29:29.662 --> 00:29:33.266
does that translate into the
alerting region for that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6011-2
00:29:33.266 --> 00:29:34.000
earthquake?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6030-0
00:29:34.010 --> 00:29:40.204
So for example, the red polygons
are what would be based only on

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6030-1
00:29:40.204 --> 00:29:40.680
epic.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6053-0
00:29:40.990 --> 00:29:44.720
The green would be based on
Finder etcetera.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6075-0
00:29:44.780 --> 00:29:49.612
And because this solution
aggregator updates its magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6075-1
00:29:49.612 --> 00:29:54.444
in sort of 1 cadence and G fast
in, in another, they don't

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6075-2
00:29:54.444 --> 00:29:57.310
necessarily update
simultaneously.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6113-0
00:29:57.320 --> 00:30:01.359
So you're seeing here that the
solution aggregator and Finder

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6113-1
00:30:01.359 --> 00:30:05.333
are essentially the same, and
that's because and just by the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6113-2
00:30:05.333 --> 00:30:08.850
timing of the the 2G fast versus
solution aggregator.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6146-0
00:30:08.860 --> 00:30:12.976
What's plotted here is the first
time step at which gefest

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6146-1
00:30:12.976 --> 00:30:16.883
provided a non throttled
magnitude estimate and at that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6146-2
00:30:16.883 --> 00:30:20.650
point the solution aggregator
hadn't quite caught up.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6189-0
00:30:21.020 --> 00:30:24.689
But you can see that if we had
waited another second and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6189-1
00:30:24.689 --> 00:30:28.679
plotted this one, it would be
much more similar to G Fest and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6189-2
00:30:28.679 --> 00:30:32.604
would have alerted a somewhat
larger area in this particular

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6189-3
00:30:32.604 --> 00:30:32.990
event.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6202-0
00:30:33.000 --> 00:30:37.100
So and so those first two were
crustal earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6221-0
00:30:37.110 --> 00:30:40.669
The other two I want to talk
about are subduction zone,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6221-1
00:30:40.669 --> 00:30:42.130
megathrust earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6241-0
00:30:42.140 --> 00:30:46.543
And again, I would say this is
more where we think that G Fest

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6241-1
00:30:46.543 --> 00:30:48.570
could really be most helpful.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6271-0
00:30:48.940 --> 00:30:54.052
And so this is the Tohoku
earthquake, and you can see here

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6271-1
00:30:54.052 --> 00:30:59.510
that Epic was not able to get
anywhere near to this magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6295-0
00:30:59.520 --> 00:31:04.232
And Finder slowly climbed up,
but once and GT fast started

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6295-1
00:31:04.232 --> 00:31:08.703
reporting magnitudes, it
immediately was much closer to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6295-2
00:31:08.703 --> 00:31:10.220
the true magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6359-0
00:31:10.230 --> 00:31:15.132
So when it became included in
the solution aggregator, that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6359-1
00:31:15.132 --> 00:31:20.197
magnitude estimate jumped up and
so you can see then that the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6359-2
00:31:20.197 --> 00:31:25.588
solution aggregator polygons are
much larger, recover larger area

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6359-3
00:31:25.588 --> 00:31:30.653
and more accurately alert people
who felt shaking compared to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6359-4
00:31:30.653 --> 00:31:35.555
what was available with Just
Finder, which are the green or

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6359-5
00:31:35.555 --> 00:31:37.760
certainly compared to epic.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6411-0
00:31:37.810 --> 00:31:42.037
So if you look at the purple
diamonds, which are solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6411-1
00:31:42.037 --> 00:31:46.629
aggregator without G fast, you
can see that at the time when G

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6411-2
00:31:46.629 --> 00:31:51.075
fast came into it, they were
down near the epic magnitude so

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6411-3
00:31:51.075 --> 00:31:55.156
that they were really just
alerting and this small area

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6411-4
00:31:55.156 --> 00:31:55.520
here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6456-0
00:31:55.530 --> 00:31:59.783
So this is an example where
bringing in the the Genesis data

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6456-1
00:31:59.783 --> 00:32:04.174
actually made a big difference
and that was just to show which

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6456-2
00:32:04.174 --> 00:32:07.660
time step it was and then
finally the Tokachi Oki

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6456-3
00:32:07.660 --> 00:32:09.890
earthquake was somewhat similar.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6486-0
00:32:09.900 --> 00:32:15.333
In this entity G fast magnitude
estimate was much higher from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6486-1
00:32:15.333 --> 00:32:20.502
the gecko, and in this case
Finder happened to have a hard

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6476-0
00:32:16.590 --> 00:32:16.880
Umm.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6486-2
00:32:20.502 --> 00:32:20.940
time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6489-0
00:32:21.550 --> 00:32:22.280
It took longer.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6504-0
00:32:22.290 --> 00:32:26.205
I should say to get to the
something approaching the true

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6504-1
00:32:26.205 --> 00:32:26.880
magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6522-0
00:32:26.890 --> 00:32:31.059
So again, Gefest made a big
difference in the solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6522-1
00:32:31.059 --> 00:32:32.650
aggregator magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6549-0
00:32:32.660 --> 00:32:37.169
Once it was available and
increased the alerting region to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6549-1
00:32:37.169 --> 00:32:41.830
more accurately, umm, cover the
the people who felt shaking.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6582-0
00:32:43.090 --> 00:32:47.454
So so to summarize, the first
part of the talk of we've

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6582-1
00:32:47.454 --> 00:32:52.442
adapted the TGD based magnitude
estimation module of G fast for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6582-2
00:32:52.442 --> 00:32:54.000
use and shake alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6622-0
00:32:54.050 --> 00:32:57.364
And we have message throttling
criteria that respond to the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6622-1
00:32:57.364 --> 00:33:00.622
unique noise characteristics of
real time GNSS and they're

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6622-2
00:33:00.622 --> 00:33:03.825
empirically based and they help
ensure that only reliable

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6622-3
00:33:03.825 --> 00:33:06.420
magnitude estimates contribute
to shake alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6654-0
00:33:07.660 --> 00:33:11.110
The real time testing has helped
us a lot in fine tuning the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6654-1
00:33:11.110 --> 00:33:14.617
throttling criteria and it also
has demonstrated the expected

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6654-2
00:33:14.617 --> 00:33:18.010
behavior in the absence of
Guinness detectable earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6705-0
00:33:18.550 --> 00:33:22.956
And we're developing a binary
data quality flag that will also

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6705-1
00:33:22.956 --> 00:33:26.941
help to further reduce the
chance of over alerts and the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6705-2
00:33:26.941 --> 00:33:31.416
replays of the historic events
have shown chief SPG's potential

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6705-3
00:33:31.416 --> 00:33:35.262
to improve alerts, especially
for the large subduction

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6705-4
00:33:35.262 --> 00:33:36.870
megathrust earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6729-0
00:33:37.720 --> 00:33:41.593
And we submitted this algorithm
for the formal testing and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6729-1
00:33:41.593 --> 00:33:44.350
evaluation in February, which is
ongoing.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6787-0
00:33:44.500 --> 00:33:49.100
And I guess I'd make one other
comment here, which is that what

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6787-1
00:33:49.100 --> 00:33:53.340
I've shown you has taken a
fairly conservative approach in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6787-2
00:33:53.340 --> 00:33:57.293
terms of when the defects
magnitudes will actually get

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6787-3
00:33:57.293 --> 00:34:01.462
sent to the solution aggregator
and that's why by design,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6787-4
00:34:01.462 --> 00:34:05.630
because we can always relax
those criteria in the future.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6795-0
00:34:05.740 --> 00:34:08.380
But it's better to start
conservatively.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6831-0
00:34:08.450 --> 00:34:12.158
Make sure we don't have over
alerts and make sure that all

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6831-1
00:34:12.158 --> 00:34:16.180
the pieces are working together
and and then start to see where

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6831-2
00:34:16.180 --> 00:34:19.070
we might change things to allow
it to report.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6857-0
00:34:19.120 --> 00:34:23.807
For example, on smaller
earthquakes, so the second part

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6857-1
00:34:23.807 --> 00:34:28.829
of the talk is going to start to
think about what can we do

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6857-2
00:34:28.829 --> 00:34:30.670
beyond a point source.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6884-0
00:34:30.680 --> 00:34:35.330
So Epic provides appoints source
and G fast, provides a point

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6884-1
00:34:35.330 --> 00:34:38.930
source, have approximation to to
an earthquake.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6905-0
00:34:39.620 --> 00:34:44.344
But what can we do if we use
real time TSS data to develop a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6905-1
00:34:44.344 --> 00:34:46.280
a distributed slip model?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6913-0
00:34:46.290 --> 00:34:48.530
And can that lead to better
alerting?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6928-0
00:34:49.000 --> 00:34:53.559
So in an earlier paper, we
looked at this using uh

4b3088d1-ddb2-4d20-8354-c3e4192820fb/6928-1
00:34:53.559 --> 00:34:56.330
synthetic models of earthquake.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7024-0
00:34:56.340 --> 00:35:00.989
So we evaluated whether a 3D
finite fault model improves

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7024-1
00:35:00.989 --> 00:35:05.720
alert triality hi enabling
better ground motion estimates

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7024-2
00:35:05.720 --> 00:35:10.614
compared to point and line
sources, and an example is shown

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7024-3
00:35:10.614 --> 00:35:15.589
from that paper in the figure
here and what we were doing is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7024-4
00:35:15.589 --> 00:35:20.157
we had as a reference sources
which were 3D rectangular

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7024-5
00:35:20.157 --> 00:35:25.459
sources and then we developed a
bunch of test sources which were

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7024-6
00:35:25.459 --> 00:35:30.760
either 3D or line sources where
some parameter was misestimated.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7038-0
00:35:30.770 --> 00:35:34.801
So in the figure here we're
showing examples where length

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7038-1
00:35:34.801 --> 00:35:36.260
was was misestimated.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7076-0
00:35:36.270 --> 00:35:39.978
So if you have a source model
that's it's still 3D finite

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7076-1
00:35:39.978 --> 00:35:43.430
fault, but you under
overestimate the length and then

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7076-2
00:35:43.430 --> 00:35:46.370
what if you try to represent
that earthquake?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7099-0
00:35:46.380 --> 00:35:49.782
Actually, with just aligned
source that may have it length

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7099-1
00:35:49.782 --> 00:35:52.030
and its magnitude under
overestimated.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7156-0
00:35:52.040 --> 00:35:56.918
And I think the take home from
this figure for the purpose of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7156-1
00:35:56.918 --> 00:36:01.874
this talk is that in almost all
cases, the line sources, which

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7156-2
00:36:01.874 --> 00:36:06.909
are the triangles and the stars
and had a higher false negative

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7156-3
00:36:06.909 --> 00:36:10.370
rate compared to the 3D finite
fault model.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7180-0
00:36:10.380 --> 00:36:14.259
And even when that model had
incorrect parameters and false

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7180-1
00:36:14.259 --> 00:36:17.750
negative rate basically
translates to under alerting.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7198-0
00:36:17.760 --> 00:36:20.389
So there are people who would
have felt shaking would not get

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7198-1
00:36:20.389 --> 00:36:20.770
an alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7237-0
00:36:20.780 --> 00:36:24.431
And so our conclusion, I think
the the important conclusion

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7237-1
00:36:24.431 --> 00:36:28.446
from that paper was that using a
3D geometry could offer the most

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7237-2
00:36:28.446 --> 00:36:31.610
improvement in the case of a
megathrust earthquake.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7262-0
00:36:31.880 --> 00:36:35.206
But that study was using
hypothetical source geometries

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7262-1
00:36:35.206 --> 00:36:38.889
that were very simplified, and
they didn't consider spatially

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7262-2
00:36:38.889 --> 00:36:39.720
variable slip.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7294-0
00:36:39.850 --> 00:36:44.697
But there are real time Genesis
as based algorithms that have

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7294-1
00:36:44.697 --> 00:36:49.230
been published and they do
produce real time slip models.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7317-0
00:36:49.240 --> 00:36:52.899
And so here we're exploring the
potential impact of one of those

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7317-1
00:36:52.899 --> 00:36:55.770
algorithms using data models for
real earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7359-0
00:36:56.890 --> 00:37:00.581
So the algorithm that I'll be
talking about today is called

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7359-1
00:37:00.581 --> 00:37:04.209
befores and it was something
that Sarah Minson developed a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7359-2
00:37:04.209 --> 00:37:07.285
number of years ago and it
provides a real time 3

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7359-3
00:37:07.285 --> 00:37:09.560
dimensional distributed slip
models.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7405-0
00:37:09.750 --> 00:37:14.899
So the input our real time GNSS
displacement time series and it

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7405-1
00:37:14.899 --> 00:37:19.807
does a Bayesian model class
selection for the strike and the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7405-2
00:37:19.807 --> 00:37:25.117
depth of a fault plane, and then
using the the most likely strike

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7405-3
00:37:25.117 --> 00:37:25.760
and dip.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7429-0
00:37:25.770 --> 00:37:28.851
It then does an analytical these
in inversion for the slip

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7429-1
00:37:28.851 --> 00:37:29.530
distribution.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7462-0
00:37:29.540 --> 00:37:35.202
So the length and and width of
the the rupture are determined

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7462-1
00:37:35.202 --> 00:37:41.046
by where the inversion puts slip
with the idea that if you make

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7462-2
00:37:41.046 --> 00:37:42.050
this fixed.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7468-0
00:37:42.100 --> 00:37:44.070
Uh, fault dimension.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7525-0
00:37:44.280 --> 00:37:47.607
If you make it big enough, then
the inversion will put slip

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7525-1
00:37:47.607 --> 00:37:51.211
where it needs to be to fit the
data and that will actually tell

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7525-2
00:37:51.211 --> 00:37:54.593
you that the actual length and
width and then as I mentioned

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7525-3
00:37:54.593 --> 00:37:58.086
before, the strike and dip our
from the Bayesian inversion for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7525-4
00:37:58.086 --> 00:37:59.250
those two parameters.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7546-0
00:37:59.260 --> 00:38:03.074
And here are two examples from
that paper for the Tokachi Oki

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7546-1
00:38:03.074 --> 00:38:05.350
earthquake in the Tohoku
earthquake.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7565-0
00:38:06.060 --> 00:38:09.933
So I think it's useful to look
in a little more detail what's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7565-1
00:38:09.933 --> 00:38:10.870
happening here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7596-0
00:38:10.880 --> 00:38:15.008
So this is where Tokachi, Oki
and it's showing how the strike

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7596-1
00:38:15.008 --> 00:38:18.270
the dip in the magnitude are
evolving with time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7618-0
00:38:18.980 --> 00:38:22.853
Since origin time and so you can
see that initially when you

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7618-1
00:38:22.853 --> 00:38:26.090
don't have much data, it's kind
of jumping around.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7668-0
00:38:26.370 --> 00:38:30.864
And it's not very stable with
the magnitude is is increasing

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7668-1
00:38:30.864 --> 00:38:35.064
and eventually the strike and
dip do kind of settle down

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7668-2
00:38:35.064 --> 00:38:39.705
towards something like the the,
the true value that that I was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7668-3
00:38:39.705 --> 00:38:41.620
determined after the fact.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7688-0
00:38:41.830 --> 00:38:45.822
This map at the right is showing
what things look like at 30

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7688-1
00:38:45.822 --> 00:38:47.720
seconds past the origin time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7726-0
00:38:47.730 --> 00:38:52.418
So where this pink bar is so you
can see there are very few

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7726-1
00:38:52.418 --> 00:38:57.341
stations reporting and and the
best fitting strike and dip are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7726-2
00:38:57.341 --> 00:39:00.700
chosen based on the on maximum
likelihood.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7761-0
00:39:00.710 --> 00:39:04.006
So this is dip and this is
strike and you can see that the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7761-1
00:39:04.006 --> 00:39:07.582
the best fitting is sort of the
reddest of the red part of that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7761-2
00:39:07.582 --> 00:39:08.140
color map.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7813-0
00:39:08.150 --> 00:39:13.165
But there are other orientations
that might sort of have a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7813-1
00:39:13.165 --> 00:39:18.690
similar fit by the time you get
to 60 seconds after origin time,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7813-2
00:39:18.690 --> 00:39:24.130
it's getting a more reasonable
strike dip and in magnitude, but

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7813-3
00:39:24.130 --> 00:39:29.400
you can see still that that,
umm, strike in particular is not

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7813-4
00:39:29.400 --> 00:39:31.440
really well constrained.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7844-0
00:39:31.450 --> 00:39:34.148
So there are a lot of
orientations that might affect

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7844-1
00:39:34.148 --> 00:39:35.370
the data similarly well.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7854-0
00:39:35.380 --> 00:39:40.257
So there's some instability here
that would then map into

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7854-1
00:39:40.257 --> 00:39:45.470
changing source parameters and
intern ground motion estimates

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7854-2
00:39:45.470 --> 00:39:46.310
over time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7896-0
00:39:47.850 --> 00:39:51.682
Umm, so I think that most people
would would think that the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7896-1
00:39:51.682 --> 00:39:55.514
generally in a retrospective
sense using a distributed slip

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7896-2
00:39:55.514 --> 00:39:59.282
model would improve ground
motion prediction compared to a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7896-3
00:39:59.282 --> 00:40:01.900
point source for very large
earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7914-0
00:40:01.910 --> 00:40:06.637
And so here is an example for
Tokachi Oki where I'm showing

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7914-1
00:40:06.637 --> 00:40:08.370
the point source from.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7932-0
00:40:08.380 --> 00:40:11.160
This is from the egg's
subduction flat file.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7968-0
00:40:11.170 --> 00:40:16.026
That would be considered a sort
of final point source, magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7968-1
00:40:16.026 --> 00:40:20.209
and location, and many of the
locations are being under

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7968-2
00:40:20.209 --> 00:40:24.690
predicted in terms of in in this
case peak ground velocity.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/7982-0
00:40:24.700 --> 00:40:29.230
So red would mean under
predicted and here on the right.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8002-0
00:40:29.240 --> 00:40:32.629
What I'm showing is the ground
motion, the meeting ground

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8002-1
00:40:32.629 --> 00:40:35.550
motion that was predicted from
several published.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8017-0
00:40:35.810 --> 00:40:39.466
So again, retrospective
distributed slip models where

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8017-1
00:40:39.466 --> 00:40:41.970
the residuals are generally
smaller.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8038-0
00:40:41.980 --> 00:40:45.780
So so we think that having that
finite fall information can be

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8038-1
00:40:45.780 --> 00:40:47.710
important for large earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8054-0
00:40:47.720 --> 00:40:51.345
But the question that we need to
look at is whether this is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8054-1
00:40:51.345 --> 00:40:52.190
actually true.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8075-0
00:40:52.200 --> 00:40:54.712
When you look at real time
models, please, that's the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8075-1
00:40:54.712 --> 00:40:57.130
situation that will be in for
quick early warnings.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8113-0
00:40:57.140 --> 00:41:02.071
So these first two maps are the
same as in the previous figure,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8113-1
00:41:02.071 --> 00:41:06.925
and the third one is what you
would get if you used the before

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8113-2
00:41:06.925 --> 00:41:10.160
model from 65 seconds passed
origin time.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8140-0
00:41:10.790 --> 00:41:15.024
You can see that and in this
case it's it's tending to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8140-1
00:41:15.024 --> 00:41:19.180
overpredict, so you know it's
it's the ground motion.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8176-0
00:41:19.240 --> 00:41:22.401
Residuals are probably similar
in terms of absolute value

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8176-1
00:41:22.401 --> 00:41:25.563
between the point source in this
real time source and the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8176-2
00:41:25.563 --> 00:41:29.160
question really is how does this
translate into who would receive

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8176-3
00:41:29.160 --> 00:41:29.650
an alert?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8216-0
00:41:30.930 --> 00:41:34.858
So what I've done is to use
observed and predicted ground

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8216-1
00:41:34.858 --> 00:41:38.921
motion to explore the alert
outcomes using different source

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8216-2
00:41:38.921 --> 00:41:43.052
models and the procedure try to
walk you through is starting

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8216-3
00:41:43.052 --> 00:41:44.000
with observed.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8285-0
00:41:44.010 --> 00:41:48.095
In this case, I'm showing peak
ground velocity, but could be

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8285-1
00:41:48.095 --> 00:41:52.246
peak ground acceleration from
the NGA subduction database and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8285-2
00:41:52.246 --> 00:41:56.130
then getting predicted round
motion at the same locations

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8285-3
00:41:56.130 --> 00:42:00.214
using subduction zone ground
motion lookup tables that Amory

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8285-4
00:42:00.214 --> 00:42:04.365
and others have been developing
for shake alert and those are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8285-5
00:42:04.365 --> 00:42:07.780
based on the inja subduction
ground motion models.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8319-0
00:42:08.210 --> 00:42:13.705
And so here I'll be showing
today examples for PGB and PGA

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8319-1
00:42:13.705 --> 00:42:19.480
and then you need to choose some
alert thresholds that you're

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8319-2
00:42:19.480 --> 00:42:20.970
going to to use.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8355-0
00:42:20.980 --> 00:42:25.917
So based on work by Jesse
Saunders and others, it's been

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8355-1
00:42:25.917 --> 00:42:31.460
shown that in order to alert the
people that you want to alert,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8355-2
00:42:31.460 --> 00:42:35.530
you need to choose an alert
issuing threshold.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8373-0
00:42:35.610 --> 00:42:39.214
We'll call it the alert
threshold that's lower than the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8373-1
00:42:39.214 --> 00:42:41.980
target shaking that you want to
alert for.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8409-0
00:42:41.990 --> 00:42:46.846
So if you want to alert people
who will feel a make MI six, you

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8409-1
00:42:46.846 --> 00:42:51.627
should be using an alert issuing
threshold that's around in my

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8409-2
00:42:51.627 --> 00:42:51.930
MI4.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8437-0
00:42:51.940 --> 00:42:55.717
So how so you set your tube
thresholds here and then

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8437-1
00:42:55.717 --> 00:42:59.351
identify the stations with
predicted ground motion

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8437-2
00:42:59.351 --> 00:43:02.130
exceeding the alert issuing
threshold?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8451-0
00:43:02.460 --> 00:43:05.450
Those would be the ones that are
going to receive alerts.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8474-0
00:43:05.920 --> 00:43:10.057
So so now you have observed and
predicted Grandma motion and you

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8474-1
00:43:10.057 --> 00:43:13.876
need to categorize whether
people were correctly alerted or

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8474-2
00:43:13.876 --> 00:43:14.130
not.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8507-0
00:43:14.200 --> 00:43:20.191
So and this is then the map
showing for the same earthquake

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8507-1
00:43:20.191 --> 00:43:26.582
and the point source model and
how the different locations were

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8507-2
00:43:26.582 --> 00:43:27.380
alerted.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8525-0
00:43:27.390 --> 00:43:30.017
So true negative is pretty
straightforward and the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8525-1
00:43:30.017 --> 00:43:32.540
predicted ground motion was
below the threshold.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8528-0
00:43:32.550 --> 00:43:33.660
So it was the observed.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8539-0
00:43:33.870 --> 00:43:36.500
So you don't need to worry about
that.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8578-0
00:43:36.870 --> 00:43:40.725
A false negative would be that
it was not alerted, so it was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8578-1
00:43:40.725 --> 00:43:44.643
below the alert threshold, but
these are was greater than the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8578-2
00:43:44.643 --> 00:43:48.182
threshold and then there are two
here which I'm calling

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8578-3
00:43:48.182 --> 00:43:48.940
conditional.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8624-0
00:43:48.950 --> 00:43:52.145
There may be better terminology
that I could use but, but

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8624-1
00:43:52.145 --> 00:43:55.506
basically what I'm calling the
conditional false negative is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8624-2
00:43:55.506 --> 00:43:58.535
just to to get at the fact that
you have now these two

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8624-3
00:43:58.535 --> 00:44:01.620
thresholds, the alert issuing
and the target threshold.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8644-0
00:44:01.630 --> 00:44:04.997
So these are the ones that did
not receive an alert because the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8644-1
00:44:04.997 --> 00:44:07.890
predicted ground motion was
below the alert threshold.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8666-0
00:44:08.730 --> 00:44:12.118
Umm, but they actually did
observe ground motion higher

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8666-1
00:44:12.118 --> 00:44:15.445
than that threshold, but just
not exceeding the target

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8666-2
00:44:15.445 --> 00:44:16.050
threshold.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8714-0
00:44:16.060 --> 00:44:20.148
So these are locations that
would have been nice to be able

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8714-1
00:44:20.148 --> 00:44:24.440
to alert them because they did
feel light to moderate shaking,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8714-2
00:44:24.440 --> 00:44:28.460
but they weren't in the sort of
priority group of alerting

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8714-3
00:44:28.460 --> 00:44:30.980
people who felt many MI 6 or
higher.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8746-0
00:44:31.470 --> 00:44:34.638
And likewise it conditional true
positive would be the ones who

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8746-1
00:44:34.638 --> 00:44:37.855
did receive an alert, but their
ground motion was in between the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8746-2
00:44:37.855 --> 00:44:39.340
alert threshold in the target.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8755-0
00:44:39.350 --> 00:44:41.310
So that kind of in some sense, a
bonus.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8784-0
00:44:41.320 --> 00:44:44.746
We're glad that we were also
able to alert them because they

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8784-1
00:44:44.746 --> 00:44:47.892
did feel something and then the
true positive is fairly

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8784-2
00:44:47.892 --> 00:44:48.790
straightforward.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8817-0
00:44:48.800 --> 00:44:52.963
It received an alert and was
above the target threshold and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8817-1
00:44:52.963 --> 00:44:57.195
the false positive would be
those who received an alert, but

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8817-2
00:44:57.195 --> 00:44:59.970
actually felt less than MI for
shaking.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8862-0
00:45:00.800 --> 00:45:04.301
So you can categorize the
locations this way and then you

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8862-1
00:45:04.301 --> 00:45:07.801
can also look at statistics like
in this case 100% of the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8862-2
00:45:07.801 --> 00:45:11.422
locations that should have been
alerted based on the target

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8862-3
00:45:11.422 --> 00:45:12.810
threshold were alerted.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8878-0
00:45:12.820 --> 00:45:18.060
Those are the dark green or 5.8%
of the alerts that were issued.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8901-0
00:45:18.070 --> 00:45:21.438
So all of the, yeah, the ones
that exceeded the alert

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8901-1
00:45:21.438 --> 00:45:23.870
threshold happened to be false
alerts.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8913-0
00:45:24.120 --> 00:45:25.860
So you can start to look at
those things.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8939-0
00:45:26.870 --> 00:45:30.851
And one thing that we had to
consider in trying to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8939-1
00:45:30.851 --> 00:45:35.379
incorporate distributed slip
models is how do you measure

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8939-2
00:45:35.379 --> 00:45:36.940
that source station?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8955-0
00:45:36.950 --> 00:45:39.980
Distance is not as obvious, it's
the point source, for example.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8994-0
00:45:39.990 --> 00:45:43.025
So the ground motion lookup
tables use the joiner board

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8994-1
00:45:43.025 --> 00:45:46.059
distance, which is the shortest
distance to the surface

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8994-2
00:45:46.059 --> 00:45:49.581
projection of the rupture, which
makes it zero for stations that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/8994-3
00:45:49.581 --> 00:45:51.260
are directly above this source.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9049-0
00:45:51.990 --> 00:45:55.756
So for the distributed slip
models that I'm showing today,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9049-1
00:45:55.756 --> 00:45:59.585
we're using the outline of the
slip region to calculate the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9049-2
00:45:59.585 --> 00:46:03.605
joiner board distance where the
outline is defined as the area

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9049-3
00:46:03.605 --> 00:46:07.690
surrounding everything that had
at least 15% of the peak slept,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9049-4
00:46:07.690 --> 00:46:09.540
which is an arbitrary choice.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9058-0
00:46:09.550 --> 00:46:11.880
But that's how the outline was
defined here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9081-0
00:46:14.720 --> 00:46:18.566
I'll make one side note about
the rupture distance for the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9081-1
00:46:18.566 --> 00:46:20.130
distributed slip models.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9097-0
00:46:20.700 --> 00:46:24.189
As I just commented, you know
taking the outline of of the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9097-1
00:46:24.189 --> 00:46:24.780
slip area.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9105-0
00:46:25.900 --> 00:46:27.850
Uh, it's kind of an arbitrary
choice.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9176-0
00:46:28.300 --> 00:46:32.175
Comes in and Balti developed a
different, a different kind of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9176-1
00:46:32.175 --> 00:46:35.487
distance measure called the
generalized mean rupture

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9176-2
00:46:35.487 --> 00:46:39.612
distance, which it offers a more
natural way to include spatially

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9176-3
00:46:39.612 --> 00:46:43.486
variable slip because you can
use it as a weighting that when

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9176-4
00:46:43.486 --> 00:46:47.048
you're finding the source
station distance would perhaps

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9176-5
00:46:47.048 --> 00:46:50.673
put more weight on parts of the
fault where this clip was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9176-6
00:46:50.673 --> 00:46:51.110
higher.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9221-0
00:46:51.190 --> 00:46:56.562
So it's a takes out sort of the
arbitrary choice of 15% and it

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9221-1
00:46:56.562 --> 00:47:02.018
allows you some flexibility in
how you weight the, the relative

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9221-2
00:47:02.018 --> 00:47:07.560
importance of parts of the file
that are close and farther away.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9251-0
00:47:07.730 --> 00:47:12.609
And you can see if you squint,
that the ground motion residuals

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9251-1
00:47:12.609 --> 00:47:17.640
are different if you use the two
the joiner board distance versus

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9251-2
00:47:17.640 --> 00:47:20.460
the mean rupture distance. And
what?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9309-0
00:47:20.510 --> 00:47:24.366
I'm still exploring is whether
they're different in a way that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9309-1
00:47:24.366 --> 00:47:28.161
improves alerting, and that's
less clear, whether it's really

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9309-2
00:47:28.161 --> 00:47:31.650
whether using this distance
measure, although physically

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9309-3
00:47:31.650 --> 00:47:35.262
speaking, it feels like a a
better choice, it may not be a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9309-4
00:47:35.262 --> 00:47:37.220
better choice for alert outcome.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9330-0
00:47:37.230 --> 00:47:40.385
So in in the results I'm showing
today, I'm using the joiner boy

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9330-1
00:47:40.385 --> 00:47:41.550
distance for everything.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9344-0
00:47:42.870 --> 00:47:47.140
So as I said, I wanted to look
at real earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9371-0
00:47:47.150 --> 00:47:51.169
And So what I did was find a set
of earthquakes where I could

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9371-1
00:47:51.169 --> 00:47:54.280
find a this a published
distributed slip model.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9396-0
00:47:54.290 --> 00:47:59.332
I could find real time GNSS data
to run before us and that I

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9396-1
00:47:59.332 --> 00:48:04.290
could find observed ground
motion and some self consistent.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9418-0
00:48:05.000 --> 00:48:09.722
So what I have are 10
earthquakes here and then I have

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9418-1
00:48:09.722 --> 00:48:13.070
all the the different source
modeling.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9499-0
00:48:13.180 --> 00:48:17.390
Pardon me, source modeling
methods and that I've applied

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9499-1
00:48:17.390 --> 00:48:21.601
and so some of these epic G
Fest, Finder and befores are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9499-2
00:48:21.601 --> 00:48:25.663
going to give you a time
evolution of the source and I

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9499-3
00:48:25.663 --> 00:48:30.169
only have the results for Epic G
Fest and Finder for the two

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9499-4
00:48:30.169 --> 00:48:34.822
events I I showed you earlier to
Kachidoki and Tohoku from the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9499-5
00:48:34.822 --> 00:48:37.850
test suite and but then I've run
before.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9502-0
00:48:37.860 --> 00:48:38.720
He's on all of them.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9521-0
00:48:38.730 --> 00:48:42.685
And then I have these two
columns that are called our best

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9521-1
00:48:42.685 --> 00:48:43.020
case.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9568-0
00:48:43.030 --> 00:48:47.200
So these are the retrospective
models, either retrospective

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9568-1
00:48:47.200 --> 00:48:50.883
point source or published
distributed slip models to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9568-2
00:48:50.883 --> 00:48:55.261
basically say well OK before if
the real time approximation is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9568-3
00:48:55.261 --> 00:48:59.569
not going to be as good as an
approximation is what you might

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9568-4
00:48:59.569 --> 00:49:00.820
do after the fact.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9587-0
00:49:00.830 --> 00:49:06.381
If we could have a perfect real
time distributed slip algorithm,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9587-1
00:49:06.381 --> 00:49:08.260
would that be helpful?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9617-0
00:49:08.270 --> 00:49:13.814
So those are the things that
I'll be comparing and so I'll

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9617-1
00:49:13.814 --> 00:49:19.640
start with looking at the PGV
alert outcomes for and the best

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9617-2
00:49:19.640 --> 00:49:20.110
case.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9622-0
00:49:20.230 --> 00:49:20.710
Please sorry.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9669-0
00:49:20.720 --> 00:49:24.567
So this is coming the location
and magnitude from the NG's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9669-1
00:49:24.567 --> 00:49:28.674
subduction database, and so you
can, you know, it's useful and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9669-2
00:49:28.674 --> 00:49:32.715
and informative to look at these
maps of where people receive

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9669-3
00:49:32.715 --> 00:49:34.410
different kinds of alerts.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9694-0
00:49:34.420 --> 00:49:38.607
But it's also useful to look at
all the earthquakes on one flat,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9694-1
00:49:38.607 --> 00:49:40.410
so that's what's shown here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9720-0
00:49:40.540 --> 00:49:45.899
And what you're able to see is
that basically this area up here

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9720-1
00:49:45.899 --> 00:49:49.750
is where the granulation was
under predicted.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9743-0
00:49:50.150 --> 00:49:53.831
You're able to see that some
places were not alerted, so

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9743-1
00:49:53.831 --> 00:49:57.770
you've under alerting and then
also you had false positives.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-0
00:49:57.780 --> 00:50:02.008
So there were some locations
where Overlearning happened and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-1
00:50:02.008 --> 00:50:05.680
you can look at this then for
best case in real time

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-2
00:50:05.680 --> 00:50:10.115
distributed slip models and so
in the published slip models for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-3
00:50:10.115 --> 00:50:14.343
each earthquake there may be
one, there may be 5 and publish

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-4
00:50:14.343 --> 00:50:18.293
slip models and so for each
earthquake I took the median

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-5
00:50:18.293 --> 00:50:22.381
ground motion predicted for
whatever models were available

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-6
00:50:22.381 --> 00:50:26.747
and for before it's I'm looking
at what one might consider the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-7
00:50:26.747 --> 00:50:30.905
final time step, which is the
time step where it's solution

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9847-8
00:50:30.905 --> 00:50:32.360
started to stabilize.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9875-0
00:50:32.920 --> 00:50:38.707
And so you can see what these
models you have minimal under

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9875-1
00:50:38.707 --> 00:50:44.590
learning, but you also have more
overpredicted found motion.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9932-0
00:50:44.600 --> 00:50:49.126
So you have an increased
likelihood of of over alerting

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9932-1
00:50:49.126 --> 00:50:53.895
here and then just to look at
the two real time and models

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9932-2
00:50:53.895 --> 00:50:58.017
that are one is in the
production system, one will

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9932-3
00:50:58.017 --> 00:51:02.705
hopefully soon be in the
production system from scheduler

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9932-4
00:51:02.705 --> 00:51:06.180
and you can see how those
results compare.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9949-0
00:51:06.190 --> 00:51:10.540
So again, you have largely under
predicted ground motion in some

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9949-1
00:51:10.540 --> 00:51:12.280
under alerting from these.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9976-0
00:51:12.700 --> 00:51:17.099
So again, these are all using on
that time step where before they

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9976-1
00:51:17.099 --> 00:51:21.030
started to stabilize, which I'm
calling sort of the final.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/9988-0
00:51:21.980 --> 00:51:27.150
And so and it's that's ohh one
way to look at it.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10009-0
00:51:27.160 --> 00:51:30.590
But then it's also useful to try
to see all the algorithms on on

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10009-1
00:51:30.590 --> 00:51:31.170
one figure.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10034-0
00:51:31.180 --> 00:51:34.930
So here what I'm looking at are
two metrics, the percent of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10034-1
00:51:34.930 --> 00:51:38.430
locations that should have
received an alert and do so.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10055-0
00:51:38.440 --> 00:51:41.853
That would be when these numbers
get small, you have under

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10055-1
00:51:41.853 --> 00:51:45.150
alerting and then the percent of
alerts that were false.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10094-0
00:51:45.160 --> 00:51:49.916
So and the bigger that is a more
over alerting you have and then

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10094-1
00:51:49.916 --> 00:51:54.305
the different algorithms are
color coded and so you can see

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10094-2
00:51:54.305 --> 00:51:58.329
that most of the algorithms
alert most of the priority

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10094-3
00:51:58.329 --> 00:51:59.060
locations.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10121-0
00:51:59.070 --> 00:52:02.958
So those exceeding the target
threshold for most of the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10121-1
00:52:02.958 --> 00:52:07.400
earthquakes, because they're all
clustering near to 90 to 100%.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10170-0
00:52:08.030 --> 00:52:12.406
But there's a lot of variation
in the amount of overlearning,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10170-1
00:52:12.406 --> 00:52:16.640
and so to to break that apart a
bit more, now I put the two

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10170-2
00:52:16.640 --> 00:52:21.016
metrics on 2 separate plots and
the percent of locations that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10170-3
00:52:21.016 --> 00:52:24.050
should receive an alert and do
shows that.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10212-0
00:52:26.930 --> 00:52:31.079
The under learning is primarily
happening with the point sources

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10212-1
00:52:31.079 --> 00:52:35.227
and particularly epic because we
know that it has a problem with

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10212-2
00:52:35.227 --> 00:52:39.120
magnitude saturation and in
terms of alerts that were false.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10257-0
00:52:39.210 --> 00:52:43.322
The point source is, as one
might expect, issue fewer false

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10257-1
00:52:43.322 --> 00:52:47.501
alerts and the over alerting
increases with lower magnitudes

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10257-2
00:52:47.501 --> 00:52:51.339
and you do see that there are
probably some earthquakes

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10257-3
00:52:51.339 --> 00:52:54.970
specific effects where you all
models for this work.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10277-0
00:52:54.980 --> 00:52:58.822
We have some higher number of
false alerts and you can see

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10277-1
00:52:58.822 --> 00:53:00.450
other questers similarly.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10307-0
00:53:00.460 --> 00:53:05.234
So I did want to point out where
Finder falls on this, because

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10307-1
00:53:05.234 --> 00:53:09.781
one question that needs to be
answered is we already have a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10307-2
00:53:09.781 --> 00:53:11.220
finite fault model.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10327-0
00:53:11.230 --> 00:53:14.348
It's a line source when it is a
finite fault model in shake

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10327-1
00:53:14.348 --> 00:53:14.660
alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10334-0
00:53:14.670 --> 00:53:16.330
Do we need something more than
that?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10346-0
00:53:16.600 --> 00:53:18.840
And you can see that here it's
doing quite well.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10383-0
00:53:19.510 --> 00:53:23.795
Now I'm looking at peak ground
acceleration and so you can see

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10383-1
00:53:23.795 --> 00:53:28.080
there's more under alerting than
we saw with PGV and perhaps a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10383-2
00:53:28.080 --> 00:53:30.800
greater tendency towards over
alerting.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10414-0
00:53:30.810 --> 00:53:36.326
These markers extend to higher
percentages here and and we find

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10414-1
00:53:36.326 --> 00:53:41.928
that the point sources generally
are causing more under learning

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10414-2
00:53:41.928 --> 00:53:45.030
than and the other kinds of
models.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10443-0
00:53:45.040 --> 00:53:48.881
However, the best case point
source, which is the the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10443-1
00:53:48.881 --> 00:53:53.291
retrospective one, is doing as
well as the finite fault model

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10443-2
00:53:53.291 --> 00:53:55.140
for several of the events.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10464-0
00:53:55.990 --> 00:54:01.199
Umm, in terms of the the overall
learning before shows the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10464-1
00:54:01.199 --> 00:54:04.730
highest tendency towards over
alerting.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10496-0
00:54:07.090 --> 00:54:11.454
And the even the sort of best
case distributed is the the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10496-1
00:54:11.454 --> 00:54:15.366
models also show more
overlearning than for example

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10496-2
00:54:15.366 --> 00:54:16.720
the Finder source.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10548-0
00:54:17.630 --> 00:54:20.916
So yeah, it's interesting with
with Finder because it's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10548-1
00:54:20.916 --> 00:54:24.377
performance in terms of those
metrics is as good or better

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10548-2
00:54:24.377 --> 00:54:27.897
than some of the distributed
slip models, which if you look

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10548-3
00:54:27.897 --> 00:54:31.123
at what the source actually
looks like, it's it's very

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10548-4
00:54:31.123 --> 00:54:33.880
different from what the actual
earthquake was.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10561-0
00:54:33.890 --> 00:54:35.660
I mean, this was a megathrust
earthquake in here.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10594-0
00:54:35.670 --> 00:54:40.197
We have a line source right
under the coast so so they these

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10594-1
00:54:40.197 --> 00:54:44.352
kind of models can enable
alerting the target locations

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10594-2
00:54:44.352 --> 00:54:47.320
despite not being physically
realistic.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10638-0
00:54:47.330 --> 00:54:50.919
And here is a similar thing town
for Tohoku, so it's a little

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10638-1
00:54:50.919 --> 00:54:54.161
dissatisfying to someone who's
spent a lot of my career

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10638-2
00:54:54.161 --> 00:54:57.460
modeling slip, but that I think
it's worth pointing out.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10677-0
00:54:58.280 --> 00:55:02.572
And so finally, I'll just show a
few slides showing how the the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10677-1
00:55:02.572 --> 00:55:06.797
time progression comes into it,
because so far we've only been

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10677-2
00:55:06.797 --> 00:55:09.680
looking at what I termed the
final models.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10687-0
00:55:10.340 --> 00:55:13.640
And so it's important to have
accurate magnitudes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10717-0
00:55:14.050 --> 00:55:17.938
But how important is it to have
a finite hit include those

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10717-1
00:55:17.938 --> 00:55:21.100
finite fault effects, and in an
evolving model?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10780-0
00:55:21.110 --> 00:55:26.056
So this is showing the temporal
evolution of the magnitude from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10780-1
00:55:26.056 --> 00:55:30.616
several algorithms and the
befores magnitude evolution was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10780-2
00:55:30.616 --> 00:55:35.408
from the published results, but
time aligned with the results

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10780-3
00:55:35.408 --> 00:55:40.431
from the Shake Alert test suite
and you can see that fluctuation

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10780-4
00:55:40.431 --> 00:55:44.682
in the befores magnitude
reflecting the fluctuation in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10780-5
00:55:44.682 --> 00:55:46.150
the source rameurs.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10812-0
00:55:47.290 --> 00:55:51.040
And I guess the important thing
to point out is that G fast,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10812-1
00:55:51.040 --> 00:55:54.606
which are these blue triangles
and B force magnitudes are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10812-2
00:55:54.606 --> 00:55:56.880
similar, but gefasst is more
stable.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10847-0
00:55:57.930 --> 00:56:01.280
Umm, so looking at those two
sort of measures of alert

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10847-1
00:56:01.280 --> 00:56:04.812
outcomes, the percent of
locations that should be alerted

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10847-2
00:56:04.812 --> 00:56:08.100
and are versus the percent of
alerts that were false.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10893-0
00:56:08.470 --> 00:56:13.416
And we see that before is
substantially outperforming

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10893-1
00:56:13.416 --> 00:56:19.004
Finder and we also see that
before and alerts more locations

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10893-2
00:56:19.004 --> 00:56:24.591
sooner than fifas PGD for PGA,
but not for PGV, for for PGV,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10893-3
00:56:24.591 --> 00:56:27.430
they're they're pretty similar.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10928-0
00:56:27.720 --> 00:56:32.199
And before it's does over alert
more locations than the other

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10928-1
00:56:32.199 --> 00:56:36.967
algorithms that particularly for
PGA and now looking at a similar

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10928-2
00:56:36.967 --> 00:56:38.050
thing for tofu.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10958-0
00:56:38.640 --> 00:56:43.539
And in this case, PFAS gave the
best magnitude along compared to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10958-1
00:56:43.539 --> 00:56:47.911
the other algorithms and even
though towards the end they

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10958-2
00:56:47.911 --> 00:56:49.870
start to to come together.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10991-0
00:56:50.800 --> 00:56:55.398
And so in this case and we see
that you faxed PGD has most

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10991-1
00:56:55.398 --> 00:57:00.463
accurate magnitude and for peak
ground velocity, it provides the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/10991-2
00:57:00.463 --> 00:57:01.710
earliest alerts.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11008-0
00:57:03.280 --> 00:57:07.362
But the finite fault models
befores and Finder ultimately

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11008-1
00:57:07.362 --> 00:57:08.840
alert more locations.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11055-0
00:57:08.850 --> 00:57:14.392
So you have less under learning
for those compared to G fast and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11055-1
00:57:14.392 --> 00:57:19.421
the distributed source models in
line sources give similar

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11055-2
00:57:19.421 --> 00:57:24.963
results for PGV, but distributed
slip alerts more for PGA at the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11055-3
00:57:24.963 --> 00:57:27.520
expense of more over alerting.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11129-0
00:57:29.010 --> 00:57:33.454
So just to wrap it up then and
the conclusions as far and these

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11129-1
00:57:33.454 --> 00:57:35.050
are pretty new results.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11161-0
00:57:35.060 --> 00:57:38.746
So there's a lot to still
consider here, but they are

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11161-1
00:57:38.746 --> 00:57:42.910
outcomes that we're seeing are
generally better for TGV than

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11161-2
00:57:42.910 --> 00:57:47.006
PGA and for earthquakes with
magnitudes lesson about an 8.5

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11161-3
00:57:47.006 --> 00:57:50.761
point source of accurate
magnitude seems to be able to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11161-4
00:57:50.761 --> 00:57:54.925
alert about 90% or more of the
target locations for these 10

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11161-5
00:57:54.925 --> 00:57:59.157
earthquakes that we've looked
at, overlearning increases with

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11161-6
00:57:59.157 --> 00:58:00.590
decreasing magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11197-0
00:58:00.820 --> 00:58:04.370
So taking those things into
consideration, we've also seen

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11197-1
00:58:04.370 --> 00:58:07.860
the finite fault models can
reduce the under alerting for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11197-2
00:58:07.860 --> 00:58:11.590
the largest events, even if it's
magnitude is underestimated.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11223-0
00:58:11.600 --> 00:58:15.646
That's what we suffer to hoku,
but they do lead to more over

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11223-1
00:58:15.646 --> 00:58:17.900
alerts, particularly for PGA
Umm.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11288-0
00:58:18.240 --> 00:58:22.098
And these results serve taken
together, I would suggest that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11288-1
00:58:22.098 --> 00:58:25.955
the real time distributed some
models could improve alerting

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11288-2
00:58:25.955 --> 00:58:29.116
for the magnitude 8 1/2 or
greater slab interface

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11288-3
00:58:29.116 --> 00:58:32.278
earthquakes, but more
comprehensive evaluation is

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11288-4
00:58:32.278 --> 00:58:35.756
likely going to require
synthetic data because we just

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11288-5
00:58:35.756 --> 00:58:39.360
don't have observations for many
very large earthquakes.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11317-0
00:58:39.770 --> 00:58:42.996
And I guess the last thing I'd
point out is that these tests

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11317-1
00:58:42.996 --> 00:58:46.010
explore the temporal variations
in performance among the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11317-2
00:58:46.010 --> 00:58:47.860
algorithms relative to each
other.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11375-0
00:58:48.190 --> 00:58:51.480
But to actually quantify
timeliness of a learning, you

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11375-1
00:58:51.480 --> 00:58:55.009
need to evaluate with respect to
the temporal evolution of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11375-2
00:58:55.009 --> 00:58:58.538
observed ground motion at the
stations and to do it in the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11375-3
00:58:58.538 --> 00:59:02.366
best way possible, we would run
all the algorithms jointly in a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11375-4
00:59:02.366 --> 00:59:04.280
simulated shared implementation.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11392-0
00:59:05.250 --> 00:59:09.159
So I'm gonna end there and I'm
happy to take questions and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11392-1
00:59:09.159 --> 00:59:10.550
thanks for listening.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11465-0
00:59:17.540 --> 00:59:21.146
We're going to now attempt to
switch audio to the Poly instead

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11465-1
00:59:21.146 --> 00:59:24.809
of through justice computer, so
give us about 20 seconds, Jess,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11465-2
00:59:24.809 --> 00:59:28.472
if we can use your speaker and
your mic through teams will then

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11465-3
00:59:28.472 --> 00:59:31.963
switch to the Poly and then
we'll can facilitate some online

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11465-4
00:59:31.963 --> 00:59:32.650
discussions.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11481-0
00:59:43.970 --> 00:59:48.870
Hey, I don't hear any echo, so I
think we're good questions in

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11481-1
00:59:48.870 --> 00:59:49.570
the room.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11505-0
00:59:54.860 --> 00:59:59.231
I've got one, so this is pretty
different from the type of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11505-1
00:59:59.231 --> 01:00:02.490
earthquake problems that I think
about and.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11548-0
01:00:02.500 --> 01:00:06.506
And you touched on it at the
very end how in in, in the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11548-1
01:00:06.506 --> 01:00:10.298
Sheckler community, how
important is a difference of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11548-2
01:00:10.298 --> 01:00:14.948
five to 10 seconds when it comes
to sending out these alerts for

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11548-3
01:00:14.948 --> 01:00:17.380
earthquakes have different
sizes?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11594-0
01:00:17.450 --> 01:00:21.001
I understand that that'll vary
as a function of distance from

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11594-1
01:00:21.001 --> 01:00:24.322
your source, but in terms of
optimizing parameters, is it

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11594-2
01:00:24.322 --> 01:00:27.987
more important to umm you know
get the right result if it comes

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11594-3
01:00:27.987 --> 01:00:28.960
10 seconds later?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11620-0
01:00:29.070 --> 01:00:33.013
Or is a variation in predicted
magnitude of about, you know,

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11620-1
01:00:33.013 --> 01:00:36.050
the factor, one with whether
it's a 7 or an 8.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11625-0
01:00:36.650 --> 01:00:38.240
Is that kind of prioritized?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11632-0
01:00:38.250 --> 01:00:40.880
Is that the discussions right?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11665-0
01:00:40.890 --> 01:00:46.653
So timeliness is that we had
priority, sure, sure, sure, that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11665-1
01:00:46.653 --> 01:00:52.416
was partly why that two tears
threshold or threshold is being

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11665-2
01:00:52.416 --> 01:00:52.880
used.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11673-0
01:00:53.450 --> 01:00:56.440
The MI4 for issuing the alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11679-0
01:00:56.730 --> 01:00:59.390
When you're really aiming to
alert the people.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11701-0
01:01:01.620 --> 01:01:07.111
Greater and that was designed to
improve the timeliness because

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11701-1
01:01:07.111 --> 01:01:09.170
for a number of reasons.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11726-0
01:01:09.180 --> 01:01:13.918
But like you're like you're
getting at at is like we saw

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11726-1
01:01:13.918 --> 01:01:15.580
some of the figures.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11733-0
01:01:15.590 --> 01:01:17.640
You can under estimate the
magnitude.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11751-0
01:01:17.650 --> 01:01:21.582
For example, and still manage to
alert many of the people that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11751-1
01:01:21.582 --> 01:01:23.080
you would like to alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11760-0
01:01:23.090 --> 01:01:25.240
Maybe you know that you would
like to.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11781-0
01:01:25.510 --> 01:01:30.076
So I think in that that's really
what sort of the trucks of we

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11781-1
01:01:30.076 --> 01:01:31.960
need a more complex model.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11788-0
01:01:32.010 --> 01:01:33.640
I mean, can we do it with
Finder?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11792-0
01:01:33.650 --> 01:01:34.120
Can we?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11837-0
01:01:36.340 --> 01:01:40.154
It may be that it's just the
very largest earthquakes that

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11837-1
01:01:40.154 --> 01:01:44.421
having that additional financial
information would be useful, and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11837-2
01:01:44.421 --> 01:01:47.330
that's where we're gonna
probably need food.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11857-0
01:01:47.340 --> 01:01:51.698
A synthetic tests because I
showed you 2 examples, but a

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11857-1
01:01:51.698 --> 01:01:54.450
data set of two is pretty small
so.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11861-0
01:01:56.750 --> 01:01:57.220
Gotcha.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11868-0
01:01:57.350 --> 01:01:57.570
Thanks.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11879-0
01:01:58.870 --> 01:02:01.900
And Henry, that's just that was
really nice.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11901-0
01:02:01.910 --> 01:02:05.064
And I know I think this should
have known a lot more of this

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11901-1
01:02:05.064 --> 01:02:07.080
stuff was really nice, but I I
didn't.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11918-0
01:02:07.090 --> 01:02:09.130
So it was really nice to see it
all kind of show in one part.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11976-0
01:02:09.760 --> 01:02:13.610
Umm, I have a lot of different
questions but what I was

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11976-1
01:02:13.610 --> 01:02:17.667
actually curious about was
towards the beginning about the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11976-2
01:02:17.667 --> 01:02:21.723
like the magnitude drift and and
this is maybe beyond like

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11976-3
01:02:21.723 --> 01:02:23.510
certain your purview, but.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11982-0
01:02:23.060 --> 01:02:26.620
I'll check with milk and Curtis
about the two special seminars.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/11994-0
01:02:29.540 --> 01:02:32.190
But so the the magnitude tends
to drift larger.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12011-0
01:02:32.640 --> 01:02:36.673
How does that factor into the
actual development of like say

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12011-1
01:02:36.673 --> 01:02:37.400
the G fast?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12013-0
01:02:38.370 --> 01:02:38.790
Alright.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12029-0
01:02:38.870 --> 01:02:42.390
Is that taken into account or is
that sort of just added?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12054-0
01:02:43.550 --> 01:02:49.134
So that's taken into account in
a way you can 13 cents attached

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12054-1
01:02:49.134 --> 01:02:51.490
to the management estimate.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12074-0
01:02:51.500 --> 01:02:55.784
So yeah, I didn't show because
of and it takes a while to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12074-1
01:02:55.784 --> 01:02:56.670
explain the.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12096-0
01:02:56.680 --> 01:02:59.844
Basically, using that same
approach it with a bunch of

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12096-1
01:02:59.844 --> 01:03:02.720
types of Caterpillar,
earthquakes and noise, only

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12096-2
01:03:02.720 --> 01:03:03.640
couldn't series.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12115-0
01:03:05.510 --> 01:03:10.957
You were able to see, but for
any given time step passed

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12115-1
01:03:10.957 --> 01:03:12.390
origin time it.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12121-0
01:03:12.540 --> 01:03:14.010
Does that distribution of
things.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12139-0
01:03:17.050 --> 01:03:20.765
10 OK, you need this maybe two
weeks higher than that before it

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12139-1
01:03:20.765 --> 01:03:22.390
starts having uncertainties.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12145-0
01:03:24.590 --> 01:03:25.040
That's right.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12179-0
01:03:25.050 --> 01:03:29.772
But additionally the the media
magnitude is biased by that drip

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12179-1
01:03:29.772 --> 01:03:33.977
tray and I've never done
chevious types or anything, but

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12179-2
01:03:33.977 --> 01:03:37.740
you don't do anything, you
already get a magnitude

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12179-3
01:03:37.740 --> 01:03:38.920
whatever, right?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12199-0
01:03:38.930 --> 01:03:42.857
So if I see a magnitude,
whatever isn't the median

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12199-1
01:03:42.857 --> 01:03:44.320
already OK, I I hi.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12203-0
01:03:44.990 --> 01:03:45.380
Yeah, maybe.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12229-0
01:03:45.390 --> 01:03:47.625
I'm not sure what the question
is, so I think it's, I think

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12229-1
01:03:47.625 --> 01:03:49.040
what you're saying is like does
that.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12242-0
01:03:50.280 --> 01:03:53.455
So let's say for example that
what you get request assistance

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12242-1
01:03:53.455 --> 01:03:54.120
today, right?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12244-0
01:03:54.350 --> 01:03:54.750
Uh.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12250-0
01:03:55.390 --> 01:03:57.080
That added on it.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12256-0
01:03:57.090 --> 01:03:58.960
Yeah, yeah, yeah, they're right.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12266-0
01:03:59.110 --> 01:04:00.640
And I I was in the answer is no.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12294-0
01:04:00.650 --> 01:04:04.154
Like once you get to how to when
the above, yeah, once you get

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12294-1
01:04:04.154 --> 01:04:06.990
above the noise, it's actually
that the languages.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12299-0
01:04:07.160 --> 01:04:08.810
Is it accurate?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12313-0
01:04:08.820 --> 01:04:12.990
And and I'm telling the slides
and maybe I should print it out.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12341-0
01:04:13.300 --> 01:04:17.430
You can see that the new fast
and then you can do this pretty

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12341-1
01:04:17.430 --> 01:04:21.160
similar results for, similar to
the NGA seduction file.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12367-0
01:04:22.530 --> 01:04:27.255
So my head that is for the
Kumamoto example, how come it so

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12367-1
01:04:27.255 --> 01:04:31.900
underestimated the bias should
always be everything right?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12389-0
01:04:33.830 --> 01:04:37.392
And in about packing actually
like the the things are going

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12389-1
01:04:37.392 --> 01:04:39.410
through distribution of
stations.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12426-0
01:04:39.890 --> 01:04:42.974
I don't have an answer for that,
but if you remember back in the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12426-1
01:04:42.974 --> 01:04:45.489
very beginning of the talk when
I showed the scaling

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12426-2
01:04:45.489 --> 01:04:48.431
relationship and there's a lot
of scattered around, right and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12426-3
01:04:48.431 --> 01:04:48.620
you.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12430-0
01:04:54.280 --> 01:04:54.580
I would.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12438-0
01:05:01.890 --> 01:05:03.030
Other questions from the root.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12453-0
01:05:05.670 --> 01:05:08.690
Looks like we've got a question
from Gary Fruit.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12462-0
01:05:08.700 --> 01:05:09.910
Do you wanna unmute yourself,
Gary?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12464-0
01:05:12.000 --> 01:05:12.250
Yeah.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12501-0
01:05:14.510 --> 01:05:18.168
Jess, that was a great talk and
congratulations on the huge

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12501-1
01:05:18.168 --> 01:05:20.972
amount of work that it
represents and all the

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12501-2
01:05:20.972 --> 01:05:23.410
thoughtfulness that you've put
into it.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12506-0
01:05:23.690 --> 01:05:26.590
I'm a simple question.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12524-0
01:05:27.240 --> 01:05:31.187
Subduction zone earthquakes
occur offshore, and presumably

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12524-1
01:05:31.187 --> 01:05:34.130
the greatest ground motions are
underwater.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12538-0
01:05:34.780 --> 01:05:37.948
What are your thoughts about
using only on land stations to

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12538-1
01:05:37.948 --> 01:05:38.740
evaluate those?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12549-0
01:05:41.140 --> 01:05:44.800
Uh, it's not ideal after so.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12551-0
01:05:44.880 --> 01:05:47.250
So I mean they.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12571-0
01:05:50.470 --> 01:05:53.470
In fact, about sure, earthquakes
are just out of network.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12586-0
01:05:53.480 --> 01:05:57.328
Earthquakes in general is is a
big topic of concern for shake

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12586-1
01:05:57.328 --> 01:05:57.700
alert.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12621-0
01:05:57.750 --> 01:06:01.874
We have seen already that the
epic and Finder algorithms do

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12621-1
01:06:01.874 --> 01:06:05.791
have more difficulty with the
offshore events, and those

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12621-2
01:06:05.791 --> 01:06:09.570
developers have been working on
ways to mitigate that.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12672-0
01:06:11.470 --> 01:06:16.633
And so I think that basically
it's at least the offshore data

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12672-1
01:06:16.633 --> 01:06:21.462
is better than no data and
particularly for the for these

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12672-2
01:06:21.462 --> 01:06:26.958
large megathrust earthquakes, me
know that the seismic algorithms

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12672-3
01:06:26.958 --> 01:06:32.120
particularly epic will saturate,
the magnitude will saturate.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12739-0
01:06:32.130 --> 01:06:35.626
So if we can include the
geodetic data which will provide

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12739-1
01:06:35.626 --> 01:06:39.483
a non saturating well because we
don't need to filter integrate

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12739-2
01:06:39.483 --> 01:06:42.798
the data, it's directly
measuring the displacement and

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12739-3
01:06:42.798 --> 01:06:46.715
doesn't put that's going to help
improve the system, even though

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12739-4
01:06:46.715 --> 01:06:50.210
you're still at the disadvantage
of having your stations.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12744-0
01:06:50.880 --> 01:06:52.120
It's one side of the source.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12746-0
01:06:56.540 --> 01:06:57.200
Thanks.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12758-0
01:07:01.090 --> 01:07:02.790
The other questions online are
in the room.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12767-0
01:07:05.660 --> 01:07:07.070
I had one last one for you.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12771-0
01:07:07.130 --> 01:07:07.550
Umm.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12793-0
01:07:08.070 --> 01:07:12.571
What kinds of delays in
timeliness do including those 3D

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12793-1
01:07:12.571 --> 01:07:14.860
fault models introduced into?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12822-0
01:07:14.870 --> 01:07:18.138
I know like it makes it a more
complex model, which must be

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12822-1
01:07:18.138 --> 01:07:21.406
more complex to solve, but like
what kind of a delay are we

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12822-2
01:07:21.406 --> 01:07:22.550
really talking about?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12833-0
01:07:22.560 --> 01:07:24.420
Is it in an order of two to
three seconds?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12843-0
01:07:24.430 --> 01:07:25.630
Is it another 5 to 10?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12853-0
01:07:26.480 --> 01:07:28.590
No, it's not in the five to 10
range.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12858-0
01:07:28.600 --> 01:07:29.070
So.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12885-0
01:07:29.840 --> 01:07:33.955
So I guess it's a little hard to
answer because the before this

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12885-1
01:07:33.955 --> 01:07:37.940
runs that I did for using MATLAB
implementation, so it's not.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12904-0
01:07:37.990 --> 01:07:41.269
It's because it for sure for me
like and I don't think it's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12904-1
01:07:41.269 --> 01:07:42.580
really been blocked for.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12906-0
01:07:43.570 --> 01:07:43.920
What?

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12935-0
01:07:43.930 --> 01:07:47.449
What it could be in those time,
but yeah, I mean I I would say

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12935-1
01:07:47.449 --> 01:07:50.744
on the order of a second not
we're not having that in five

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12935-2
01:07:50.744 --> 01:07:51.190
seconds.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12952-0
01:07:55.150 --> 01:07:59.764
If there are no other questions
going once going twice, let's

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12952-1
01:07:59.764 --> 01:08:01.550
thank our speaker again.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12961-0
01:08:06.230 --> 01:08:07.240
And we'll conclude there.

4b3088d1-ddb2-4d20-8354-c3e4192820fb/12964-0
01:08:08.500 --> 01:08:09.010
Thanks everyone.