WEBVTT Kind: captions Language: en 00:00:01.500 --> 00:00:05.420 [ Silence ] 00:00:07.480 --> 00:00:10.960 Good morning, and welcome to a special Earthquake Science seminar 00:00:10.960 --> 00:00:15.730 on November 16th. We’re very pleased to have Art Frankel here. 00:00:15.730 --> 00:00:20.240 But before I formally introduce him, just remind you that next week, 00:00:20.240 --> 00:00:25.440 we won’t have a seminar, but we will have a seminar the following week 00:00:25.440 --> 00:00:29.420 by Chris Zahasky of Stanford. And he’ll be talking about some 00:00:29.420 --> 00:00:33.960 very innovative work he’s been doing on carbon sequestration. 00:00:35.300 --> 00:00:39.220 Most of you know Art. He really doesn’t need an introduction. 00:00:39.220 --> 00:00:42.379 He’s a very prominent member of our science center, 00:00:42.379 --> 00:00:48.379 and he’s located up in our Seattle office. He’s worked on a large number of strong 00:00:48.379 --> 00:00:54.219 ground motion issues over his career, including strong ground motion 00:00:54.219 --> 00:00:57.320 simulations, of which he’ll be talking about today. 00:00:57.320 --> 00:01:00.610 But he’s also done simulations in Santa – 00:01:00.610 --> 00:01:05.700 on smaller problems like Santa Clara Valley and Seattle. 00:01:06.520 --> 00:01:10.350 He was a leader and a developer of the National Seismic 00:01:10.350 --> 00:01:17.140 Hazard Map for many years and developed a Seattle hazard map as well. 00:01:17.140 --> 00:01:22.340 And he’s well-known for being able to communicate hazard to policymakers 00:01:22.340 --> 00:01:28.830 in a way that they take action on the science that we’ve developed. 00:01:28.830 --> 00:01:33.530 Art received a Distinguished Service Award in 2011, 00:01:33.530 --> 00:01:37.140 and he’s an AGU Fellow. And he started his career at 00:01:37.140 --> 00:01:42.980 Lamont working with Dr. [inaudible]. So, Art? 00:01:45.380 --> 00:01:52.700 [ Silence ] 00:01:53.640 --> 00:01:57.800 - Can you hear me okay? Okay, good. 00:02:01.760 --> 00:02:06.660 Okay, my talk concerns predicting strong ground motions for magnitude 9 00:02:06.660 --> 00:02:11.920 earthquakes on the Cascadia subduction zone using 3D simulations. 00:02:11.920 --> 00:02:16.840 And I’d like to acknowledge first my co-authors on this work – 00:02:16.840 --> 00:02:22.260 Erin Wirth, Nasser Marafi, John Vidale, and Bill Stephenson. 00:02:22.260 --> 00:02:24.920 Erin did a lot of the 3D simulations I’ll be talking about 00:02:24.920 --> 00:02:30.280 today and wrote some of the key scripts to make the rupture model. 00:02:30.280 --> 00:02:37.540 And Nasser made the broadband seismograms from those synethetics. 00:02:38.500 --> 00:02:42.840 I thought I’d start out by showing some dirt. 00:02:42.840 --> 00:02:48.320 This is Brian Atwater clearing off the side of a river bank near the coast of 00:02:48.320 --> 00:02:56.020 Washington state. And let’s see if this – I guess I don’t see – oh, here it is. 00:02:56.940 --> 00:03:02.880 So you can see these layers here – these peat-rich layers, which indicate coseismic 00:03:02.890 --> 00:03:08.140 subsidence associated with great earthquakes on the coast – off the coast. 00:03:08.140 --> 00:03:12.450 And so, by dating these layers, you get about an average repeat time 00:03:12.450 --> 00:03:17.970 of 500 years. And then you can calculate the probability of having one of these 00:03:17.970 --> 00:03:24.870 Cascadia magnitude 9 earthquakes in the next 50 years as about 10 to 14%. 00:03:24.870 --> 00:03:29.810 So we really want to get some estimate of the types of shaking 00:03:29.810 --> 00:03:33.560 we can expect from the next magnitude 9 earthquake. 00:03:34.500 --> 00:03:39.840 And it was about five years ago when I innocently suggested that 00:03:39.840 --> 00:03:45.230 we could do a large number of 3D simulations for magnitude 9 earthquakes 00:03:45.230 --> 00:03:49.120 and then hand off those synthetic seismograms to engineers so they could 00:03:49.120 --> 00:03:54.810 evaluate the responsibilities and also look at liquefaction. And this became one of 00:03:54.810 --> 00:03:59.830 the central parts of this M9 Project, which is a University of Washington 00:03:59.830 --> 00:04:04.980 project funded for four years by the National Science Foundation. 00:04:04.980 --> 00:04:10.960 And basically, the USGS is collaborating with the University of Washington 00:04:10.970 --> 00:04:16.130 on this. So the idea is to run a set of synthetic seismograms for – 00:04:16.130 --> 00:04:20.849 using the 3D velocity model and then passing on this information 00:04:20.849 --> 00:04:25.949 to engineers to evaluate tall building response and also to look at landslides 00:04:25.949 --> 00:04:30.520 and liquefaction. Also to make ShakeMaps that could be used 00:04:30.520 --> 00:04:34.680 for emergency management and emergency preparedness. 00:04:34.680 --> 00:04:41.740 And also to use these synthetics to test earthquake early warning algorithms. 00:04:42.800 --> 00:04:49.480 I should also add there is a tsunami component to the M9 Project also, 00:04:49.490 --> 00:04:52.210 which I won’t be talking about today. 00:04:52.210 --> 00:04:55.930 The outline of my talk – first I’ll talk about the 3D velocity model 00:04:55.930 --> 00:05:00.690 for Cascadia that we’re using and talk about the – and review the validation 00:05:00.690 --> 00:05:05.620 of the model for the Seattle Basin area. Then I’ll describe the compound rupture 00:05:05.620 --> 00:05:09.250 model that we’re using, and it’s based on observations 00:05:09.250 --> 00:05:14.190 and modeling of the magnitude 9.0 Tohoku, Japan, earthquake 00:05:14.190 --> 00:05:17.240 and the magnitude 8.8 Maule, Chile, earthquake. 00:05:17.240 --> 00:05:22.180 I’ll show the Cascadia M9 results for a typical run and then describe 00:05:22.180 --> 00:05:24.910 the results for 30 runs that we’ve done. 00:05:24.910 --> 00:05:29.960 And finally, I’ll talk about the range of ground motions we see 00:05:29.960 --> 00:05:34.020 and the range of basin amplification we see for the Seattle area. 00:05:34.560 --> 00:05:38.400 Okay, this shows a cutaway view of the 3D model. 00:05:38.410 --> 00:05:41.610 It extends from the south to Cape Mendocino 00:05:41.610 --> 00:05:44.680 to the north to the middle of Vancouver Island. 00:05:44.680 --> 00:05:49.710 It has a subducted slab in it, of course. And the plate interface is from 00:05:49.710 --> 00:05:56.620 the McCrory et al. 2012 model. And you can see these deep basins here. 00:05:56.620 --> 00:06:00.210 This is one of the critical things and really why – and some reasons why 00:06:00.210 --> 00:06:04.210 we’re doing the 3D simulations is to get the response of these sedimentary 00:06:04.210 --> 00:06:10.650 basins such as the Seattle Basin, Tacoma Basin, the Georgia Basin, 00:06:10.650 --> 00:06:14.960 which reach about depths of 7 or 8 kilometers. And the shallower basins 00:06:14.960 --> 00:06:18.260 around Portland – the Portland and Tualatin Basins. 00:06:19.640 --> 00:06:23.660 This 3D model – okay, Bill Stephenson developed it. 00:06:23.669 --> 00:06:26.660 It contains a lot of information. Some of it is from tomography 00:06:26.660 --> 00:06:31.789 results from the SHIPS study that Tom Brocher led a number of years ago. 00:06:31.789 --> 00:06:40.220 Also, it has the Moschetti et al. model from the – for the crustal velocities. 00:06:40.220 --> 00:06:45.660 And the finite difference code we’re using was written by 00:06:45.669 --> 00:06:50.350 Pengcheng Liu of the U.S. Bureau of Reclamation. It’s a finite difference. 00:06:50.350 --> 00:06:55.780 It’s 4th order accurate in space and 2nd order accurate in time. 00:06:55.780 --> 00:06:58.040 And everything I’ll show is, basically we’re specifying 00:06:58.040 --> 00:07:04.130 a kinematic source model. So the grid we’re using – 100-meter grid 00:07:04.130 --> 00:07:09.099 spacing in the top 5 kilometers and 300-meter horizontal spacing below that. 00:07:09.099 --> 00:07:13.290 And we use a minimum shear wave velocity of 600 meters a second, which is 00:07:13.290 --> 00:07:18.560 a typical shear wave velocity for the surficial glacial sediments in Seattle. 00:07:19.460 --> 00:07:23.400 Now, the finite difference code doesn’t handle water, so we had to replace that 00:07:23.400 --> 00:07:28.729 with material with fairly low shear wave velocity, and I used a very low Q 00:07:28.729 --> 00:07:31.930 to damp out any kind of reverberations in that layer. 00:07:31.930 --> 00:07:36.500 And I did some tests to show that the synthetics on shore were not – 00:07:36.500 --> 00:07:40.800 were not sensitive to what we had chose to replace the water with. 00:07:42.300 --> 00:07:45.870 So I’ll start with the Puget Sound area and just review some work we did 00:07:45.870 --> 00:07:49.820 to validate our model there. This shows the Puget Sound area. 00:07:49.820 --> 00:07:53.860 Seattle is located here. The Seattle Basin is here. 00:07:53.870 --> 00:07:56.699 These dots are basically events that we’ve looked at 00:07:56.700 --> 00:08:00.060 the amplification in the Seattle Basin. 00:08:01.360 --> 00:08:05.840 And also, the dots with the focal mechanisms are events that we modeled 00:08:05.840 --> 00:08:10.419 using the 3D finite difference code, including the Nisqually earthquake, 00:08:10.419 --> 00:08:13.910 which was a magnitude 6.8. So we’ve compared the observed 00:08:13.910 --> 00:08:19.060 amplification at 1 hertz with that predicted by the 3D model here. 00:08:19.060 --> 00:08:24.711 So the horizontal axis is the observed amplification for sites in the Seattle 00:08:24.711 --> 00:08:29.930 Basin versus a site just outside the basin – a rock site just outside the basin. 00:08:29.930 --> 00:08:33.969 And I should say that the Vs30 for sites in the basin and for the rock site 00:08:33.969 --> 00:08:39.449 outside the basin are very similar. So this is all caused by the deeper 00:08:39.449 --> 00:08:46.070 portion of the Seattle Basin that’s driving these amplifications. 00:08:46.070 --> 00:08:50.350 And the vertical axis is a predicted amplification from the 3D model. 00:08:50.350 --> 00:08:53.560 So you see it basically follows a one-to-one relation. 00:08:53.560 --> 00:08:57.140 It also shows that events from the southwest 00:08:57.140 --> 00:09:00.820 are producing more amplification in the Seattle Basin, 00:09:00.820 --> 00:09:04.820 and the 3D simulations match that observation. 00:09:05.800 --> 00:09:09.580 Also looked at modeling waveforms at longer periods. 00:09:09.589 --> 00:09:14.890 This is basically 2- to 4-second period. And the red is the synthetics, and the 00:09:14.890 --> 00:09:19.430 black is the data for a number of sites. And generally, we’re doing a fairly 00:09:19.430 --> 00:09:23.790 good job of matching the observed waveforms and capturing some of this 00:09:23.790 --> 00:09:28.580 basin edge generated surface wave that we see in a lot of the records. 00:09:30.140 --> 00:09:33.620 So that’s just a bit of what we’ve been doing for the validation. 00:09:33.620 --> 00:09:36.880 We’ve been working on that for a number of years, and you can 00:09:36.880 --> 00:09:39.820 look at that 2009 paper if you want more information. 00:09:39.820 --> 00:09:44.570 Now, turning to the source model, a lot – the source model we’re using for 00:09:44.570 --> 00:09:51.589 Cascadia was inspired by observations from the Tohoku and Maule earthquakes. 00:09:51.589 --> 00:09:55.690 So for frequencies of engineering interests – 0.1 hertz and greater – 00:09:55.690 --> 00:10:00.360 the Tohoku earthquake was basically a series of magnitude 8 sub-events, 00:10:00.360 --> 00:10:04.399 or magnitude 8 earthquakes on a deeper part of the 00:10:04.399 --> 00:10:06.820 subduction zone with high stress drop. 00:10:06.820 --> 00:10:12.269 So Kurahashi and Irikura call this the strong motion generation areas. 00:10:12.269 --> 00:10:15.560 And they’re located here, and you can locate these from looking at 00:10:15.560 --> 00:10:19.190 the acceleration waveforms. Obviously you had a lot of slip out by 00:10:19.190 --> 00:10:24.030 the trench, and that generated tsunami. But as far as the engineering strong 00:10:24.030 --> 00:10:27.110 ground motions, it’s really dominated by the these sub-events. 00:10:27.110 --> 00:10:31.350 And I located the sub-events by looking at the arrival time of these pulses of 00:10:31.350 --> 00:10:37.540 acceleration. Pretty much got the same locations as Kurahashi and Irikura. 00:10:39.730 --> 00:10:43.140 So this is what these sub-events look like in acceleration records. 00:10:43.140 --> 00:10:48.760 They’re pretty distinct, and you can pick off the arrival times and locate these. 00:10:50.500 --> 00:10:55.060 And I modeled the major sub-event, sub-event 3, 00:10:55.060 --> 00:11:00.760 just using a flat-layered velocity model. This is from my 2013 paper. 00:11:00.760 --> 00:11:07.420 And so the black is the observed records. This is filtered 0.1 to 0.5 hertz. 00:11:07.430 --> 00:11:09.610 The red are the synthetics. I wasn’t trying to get things 00:11:09.610 --> 00:11:13.220 wiggle-for-wiggle here, so I did forward modeling to capture the 00:11:13.220 --> 00:11:16.590 peak amplitudes and also the short durations. 00:11:16.590 --> 00:11:21.350 It’s really striking how short the duration is. Some of this is due to directivity. 00:11:21.350 --> 00:11:26.520 But generally, you can model these with a magnitude 8.0 sub-event. 00:11:29.700 --> 00:11:32.360 Now, I’ve looked a lot at the Maule, Chile, earthquake. 00:11:32.360 --> 00:11:37.060 This is the source model I used to model the waveforms for the strong 00:11:37.060 --> 00:11:42.720 motions at these sites with the triangles here for the Maule earthquake. 00:11:42.720 --> 00:11:45.730 And I used what I call a compound rupture model. 00:11:45.730 --> 00:11:51.529 And it consists of this background slip, and I chose a random field that had its 00:11:51.529 --> 00:11:55.250 peak in the northwest corner here that matches what’s seen from 00:11:55.250 --> 00:11:58.940 long-period teleseismic inversions for this event. 00:11:58.940 --> 00:12:02.450 So you have a lot of slip at the shallow portion of the zone. 00:12:02.450 --> 00:12:05.600 And the key point here is, where the high frequencies 00:12:05.600 --> 00:12:09.980 come from is not the same location as where the peak slip is. 00:12:09.980 --> 00:12:14.460 So superimposed on this background slip are these sub-events. 00:12:15.280 --> 00:12:18.260 Not saying that we know exactly what the slip distribution 00:12:18.270 --> 00:12:19.579 is in these sub-events. 00:12:19.579 --> 00:12:24.020 I varied the random seeds for these, and it didn’t make much difference. 00:12:24.020 --> 00:12:27.420 But this is what I used to model the strong motion records basically 00:12:27.420 --> 00:12:33.620 sub-events ranging from magnitude 7.9 in the south to 8.2 in the north. 00:12:33.620 --> 00:12:36.480 And these sub-events are in the deeper part of the rupture zone. 00:12:36.480 --> 00:12:42.220 They have a much higher average slip velocity. I used 5.4 meters a second. 00:12:42.220 --> 00:12:47.300 And basically, this implies – they have a higher dynamic stress drop as opposed – 00:12:47.300 --> 00:12:52.769 and a very short rise time, about a couple of seconds, as opposed to 00:12:52.769 --> 00:12:56.089 the background slip, which has a rise time of – 00:12:56.089 --> 00:13:01.040 at any given point of up to 15 or 20 seconds. 00:13:01.040 --> 00:13:03.920 So it has a lower dynamic stress drop. 00:13:06.800 --> 00:13:08.800 This shows what some of these sub-events look like 00:13:08.800 --> 00:13:12.440 in actual records from Maule. You can see – especially to the northeast 00:13:12.440 --> 00:13:16.540 where there’s forward directivity, you see these distinctive pulses. 00:13:16.540 --> 00:13:20.940 When you’re basically right above these, they’re more drawn out in time. 00:13:22.490 --> 00:13:25.380 So why do we need these sub-events anyways? 00:13:25.380 --> 00:13:30.640 And one of the key things is looking at the Fourier spectra of the observed 00:13:30.640 --> 00:13:34.480 records, particularly between 0.1 hertz and 1 hertz. 00:13:34.480 --> 00:13:39.139 I found this a very difficult frequency band to really get a match 00:13:39.139 --> 00:13:42.980 to the strong motion records unless I used the sub-events. 00:13:42.980 --> 00:13:49.760 So the black is the Fourier spectra for the observed record at one site for Maule. 00:13:49.760 --> 00:13:52.780 Then these other spectra are from synthetics I made 00:13:52.780 --> 00:13:57.440 with a 1D velocity model. So just using that background zone, 00:13:57.440 --> 00:14:02.660 which has a long rise time, about 10 seconds, in green, 00:14:02.660 --> 00:14:04.750 you just can’t get that 1 hertz energy. 00:14:04.750 --> 00:14:09.959 There’s not enough roughness in the – in the – in the rupture propagation. 00:14:09.959 --> 00:14:16.899 So I added a variable rupture speed, and you get – you pick up some of 00:14:16.899 --> 00:14:19.870 the high frequency there – by high-frequency, I mean 1 hertz. 00:14:19.870 --> 00:14:22.480 But still, it’s way under what’s observed. 00:14:22.480 --> 00:14:27.460 So I had to use an asperity with a short rise time of a couple of seconds – 00:14:27.460 --> 00:14:32.910 and that’s the red line – to start to get matches to the 1 hertz 00:14:32.910 --> 00:14:36.450 and also matching the 0.1 hertz. So it’s really critical to match 00:14:36.450 --> 00:14:39.779 the observed spectra to have some component of the rupture 00:14:39.779 --> 00:14:45.769 having these short rise times. And I put these in these sub-events. 00:14:45.769 --> 00:14:48.709 And this just shows, for Maule, some of the matches of the 00:14:48.709 --> 00:14:54.740 observed records and the synthetics. The top is the observed. 00:14:54.740 --> 00:14:56.820 Let’s see. 00:14:56.820 --> 00:15:00.840 And in this procedure here, I did the 1D synthetics up to 1 hertz, 00:15:00.850 --> 00:15:04.690 and I combined them with stochastic synthetics for greater than 1 hertz. 00:15:04.690 --> 00:15:07.480 And I’ll show that procedure more later. 00:15:07.480 --> 00:15:11.580 But for these sub-events – and I’m only assuming that the high frequency 00:15:11.580 --> 00:15:15.420 is coming only from the sub-events, not from the background zone. 00:15:15.420 --> 00:15:19.910 These are the stress drops I used for the stochastic part – between 200 – 00:15:19.910 --> 00:15:24.920 these are for a stress drop of 200 bars in the sub-events, and these are the 00:15:24.920 --> 00:15:30.899 synthetics for a stress drop of 350 bars. See, we’re bracketing the PGAs in 00:15:30.899 --> 00:15:36.930 the synthetics. We’re bracketing the observations of the PGAs in the data. 00:15:36.930 --> 00:15:40.110 And also the duration, and you can see these two sub-events 00:15:40.110 --> 00:15:44.080 are pretty distinct in the synthetics and in the data. 00:15:45.440 --> 00:15:49.820 Looking at the spectral response, the bias – so this shows you how much, 00:15:49.820 --> 00:15:54.850 on average, our predictions are – match the amplitudes – 00:15:54.850 --> 00:15:58.740 the spectral accelerations of the data. So we can see we’re bracketing 00:15:58.740 --> 00:16:04.329 the zero bias from 0.1-second period to 3-second period with this 00:16:04.329 --> 00:16:09.160 200- to 350-bar range for the sub-event stress drop. 00:16:09.160 --> 00:16:12.930 Over the – if you average the stress drop over the whole fault plane, it would be much 00:16:12.930 --> 00:16:17.060 less than 200 bars. It would be more like 100 bars. 00:16:19.110 --> 00:16:21.240 So we had to come up with an approach for these 00:16:21.250 --> 00:16:24.560 Cascadia magnitude 9 simulations. 00:16:24.560 --> 00:16:30.000 So basically, I started with the rupture parameters that worked for modeling 00:16:30.000 --> 00:16:38.170 the Maule data and used those in our first Cascadia 9 3D simulations. 00:16:38.170 --> 00:16:42.399 And I looked for the – at the non-basin sites and compared 00:16:42.399 --> 00:16:46.410 the Cascadia runs with the Maule observations and also with 00:16:46.410 --> 00:16:50.020 the BC Hydro ground motion prediction equations. 00:16:50.020 --> 00:16:55.750 Now, these are based on recordings of magnitude 5.0 to 8.4 subduction zone 00:16:55.750 --> 00:17:01.339 earthquakes, and they basically extrapolate up to magnitude 9.0. 00:17:01.339 --> 00:17:05.079 They didn’t use Maule and Tohoku data in the actual study, 00:17:05.079 --> 00:17:08.209 but they did look at event terms, so we know how the event terms 00:17:08.209 --> 00:17:11.939 compare to the BC Hydro GMPEs. 00:17:11.939 --> 00:17:18.740 So I made modifications to the rupture parameters for Cascadia to lower the bias 00:17:18.740 --> 00:17:23.600 with respect to the BC Hydro GMPEs at long periods greater than 6 seconds. 00:17:23.600 --> 00:17:28.299 So we ran 30 simulations with varying hypocenters, slip distributions, 00:17:28.299 --> 00:17:31.980 and down-dip rupture edge. And then we ran 20 other simulations 00:17:31.980 --> 00:17:37.310 for sensitivity study to investigate how this response spectra 00:17:37.310 --> 00:17:41.740 were sensitive to changes in the rupture parameters. 00:17:41.740 --> 00:17:46.240 And even before these 30 simulations were run, we did probably about 30 00:17:46.240 --> 00:17:52.059 simulations just to get a handle on, what were the important things in the process. 00:17:52.059 --> 00:17:55.480 So this shows one of the source models we’re using. 00:17:55.480 --> 00:17:59.360 Again, this compound model where we have a background slip 00:17:59.360 --> 00:18:04.230 and a set of sub-events. So the background slip – in this case, 00:18:04.230 --> 00:18:08.000 we have the peak slip up to the north. We vary this with different runs, 00:18:08.000 --> 00:18:13.510 where the peak is. This has relatively large spatial correlation distance. 00:18:13.510 --> 00:18:19.200 We used basically a random number – a filtered random number seed – 00:18:19.200 --> 00:18:23.520 field, I should say – and then we draped this on the configuration 00:18:23.520 --> 00:18:26.320 of the plate interface from McCrory et al. 00:18:27.340 --> 00:18:32.120 So this has a relatively long rise time of about – maximum rise time of about 00:18:32.120 --> 00:18:39.340 35 seconds. Slow slip velocity – in other words, a low dynamic stress drop. 00:18:39.350 --> 00:18:42.160 And we concentrate this slip out in the 00:18:42.160 --> 00:18:45.460 shallower portions of the subduction zone. 00:18:45.460 --> 00:18:50.380 We taper this off in the areas of the sub-events that we superimpose on this. 00:18:50.380 --> 00:18:55.340 We chose to have five sub-events on the Cascadia subduction zone. 00:18:55.340 --> 00:19:00.620 And the reason here is that, by using five, the spacing between the sub-events 00:19:00.620 --> 00:19:05.820 is about the same as the spacing that was observed for the Tohoku sub-events. 00:19:05.820 --> 00:19:09.900 Obviously, you could use fewer or more, but this was what we chose, 00:19:09.900 --> 00:19:12.990 and that’s the reason. So each one of these sub-events, 00:19:12.990 --> 00:19:15.559 the slip is specified with a filtered random field, 00:19:15.559 --> 00:19:20.090 a Von Karman correlation function. We have a higher slip velocity – 00:19:20.090 --> 00:19:25.620 higher dynamic stress, in other words – a short rise time – a couple of seconds. 00:19:25.620 --> 00:19:30.700 And both for the slip – the hypocenters are the same. 00:19:30.700 --> 00:19:34.320 These models are basically added together for each run. 00:19:34.320 --> 00:19:36.460 And so the hypocenter is the same. 00:19:36.460 --> 00:19:41.680 The average slip velocity is the same for both of these parts of the rupture. 00:19:42.680 --> 00:19:45.120 And I should say, for the high-frequency part, 00:19:45.130 --> 00:19:49.080 we’re assuming, above 1 hertz, that that’s generated only by these 00:19:49.080 --> 00:19:53.500 sub-events. And, again, that’s what worked for the Maule earthquake. 00:19:53.500 --> 00:19:57.630 So we do 3D finite difference simulations up to 1 hertz. 00:19:57.630 --> 00:20:01.230 Again, use the code of Pengcheng Liu. 00:20:01.230 --> 00:20:05.030 Each run took about two days on the Pacific Northwest National Lab’s 00:20:05.030 --> 00:20:10.170 supercomputer with 576 cores and took only 10 hours on the 00:20:10.170 --> 00:20:15.740 Texas Advanced Computer Center – Computing Center supercomputer. 00:20:15.740 --> 00:20:19.860 Then, above 1 hertz, we used the stochastic synthetics where sum point 00:20:19.860 --> 00:20:25.760 source synthetics from the SMSIM program of Dave Boore. 00:20:25.760 --> 00:20:28.480 And, again, we assumed the sub – only the sub-events radiate 00:20:28.480 --> 00:20:31.640 the high-frequency energy. And we used matched filters 00:20:31.640 --> 00:20:34.700 to add these together to make broadband synthetics. 00:20:34.700 --> 00:20:40.210 So here’s a flow chart of this. At the top, we have the background 00:20:40.210 --> 00:20:44.100 slip model and the sub-event slip model. These are both added together 00:20:44.100 --> 00:20:49.640 then run in a finite difference run. And these go up to 1 hertz. 00:20:50.380 --> 00:20:54.860 The stochastic synthetics – basically, we sum together these point source 00:20:54.860 --> 00:21:01.490 stochastic Green’s functions with proper delay for rupture and propagation time. 00:21:01.490 --> 00:21:07.700 We convolved this sum with a function that’s a relative slip velocity function to 00:21:07.700 --> 00:21:13.020 basically get a flat acceleration spectrum out to the frequencies of the sub-event. 00:21:13.020 --> 00:21:17.440 So this is a method I described in the 1995 paper. 00:21:18.750 --> 00:21:21.740 So what this looks like, this middle trace is a 00:21:21.740 --> 00:21:27.010 finite difference synthetic that’s low-pass filtered at 1 hertz. 00:21:27.010 --> 00:21:33.000 The top here, the green, is a stochastic S wave synthetic filtered 00:21:33.000 --> 00:21:37.790 from – basically from 1 to 10 hertz. I should say high-passed at 1 hertz. 00:21:37.790 --> 00:21:42.570 And we have our stochastic P wave, again high-passed at 1 hertz. 00:21:42.570 --> 00:21:46.740 And we basically add these together after doing the match-filtering to get 00:21:46.740 --> 00:21:50.419 this broadband seismogram, which is good from zero to 10 hertz. 00:21:50.419 --> 00:21:51.880 And I should also say the dC – 00:21:51.880 --> 00:21:57.140 the coseismic offset is contained in these synthetics, and you can use that too. 00:21:57.140 --> 00:21:59.480 Now, one of the things that people don’t talk a lot about, 00:21:59.490 --> 00:22:05.040 but is a critical aspect of these kinematics simulations, 00:22:05.040 --> 00:22:09.830 is the variability of the rupture front, or rupture velocity. 00:22:09.830 --> 00:22:15.000 And we use a variation on the secant rupture velocity approach to do this. 00:22:15.000 --> 00:22:20.470 Basically, the idea here is that the effective rupture velocity across areas of 00:22:20.470 --> 00:22:26.620 high coseismic slip is going to be higher than areas of low coseismic slip. 00:22:26.620 --> 00:22:28.770 And this is consistent with observations from some 00:22:28.770 --> 00:22:32.830 large earthquakes, including the 2002 Denali earthquake 00:22:32.830 --> 00:22:37.700 where you saw higher rupture velocity across areas with higher slip. 00:22:37.700 --> 00:22:43.530 So basically, if this is one of the sub-event outlines, we look at the – 00:22:43.530 --> 00:22:47.120 for the timing to get to the nearest point to the hypocenter, 00:22:47.120 --> 00:22:50.940 we use the average rupture velocity. And then, from this distance on, 00:22:50.940 --> 00:22:58.990 for this cell, the timing is given by this equation here so that areas of higher slip, 00:22:58.990 --> 00:23:03.770 basically, will have higher effective rupture velocities, so they’ll trigger 00:23:03.770 --> 00:23:07.950 earlier than areas of lower slip, and they’ll have higher rupture velocity 00:23:07.950 --> 00:23:10.750 across these areas of higher slip. 00:23:10.750 --> 00:23:12.820 So this is an important thing, especially when you’re 00:23:12.820 --> 00:23:17.070 looking at frequencies of about – periods like 1 to 5 seconds. 00:23:17.070 --> 00:23:20.880 This rupture front variability is an important thing. 00:23:21.830 --> 00:23:26.880 And this bottom figures here show the rupture initiation times. 00:23:26.890 --> 00:23:31.110 It’s a little hard to see here, but you can see some of the variability of that within 00:23:31.110 --> 00:23:34.820 the sub-events and also the main shock. So this is an important thing. 00:23:34.820 --> 00:23:38.950 And we use a 10% standard deviation in the rupture velocity. 00:23:38.950 --> 00:23:43.820 We don’t allow things to go super shear in the cases we’ve been doing. 00:23:43.820 --> 00:23:48.900 But we have a 10% rupture speed variation for the sub-events 00:23:48.900 --> 00:23:54.850 and 5% for the background slip. And for the sub-events, that 10% 00:23:54.850 --> 00:23:59.460 standard deviation is what worked, again, to model the Maule observations. 00:24:00.240 --> 00:24:04.720 Another important point I want to bring up is that, for areas of sediment, 00:24:04.730 --> 00:24:08.789 like the Puget Lowlands, I added a small, random component to 00:24:08.789 --> 00:24:13.490 the shear wave velocity in the 3D model. Because I thought these are realistic 00:24:13.490 --> 00:24:19.480 small-scale variations that are important to include in the 3D model, so you can 00:24:19.480 --> 00:24:24.970 see the variations here. I chose a 5% standard deviation of the shear wave 00:24:24.970 --> 00:24:31.580 velocity for these areas, and it’s just in the top 1.3 kilometers of the model. 00:24:31.580 --> 00:24:34.020 And it turns out it does make a significant difference 00:24:34.040 --> 00:24:38.900 to the response spectra for sites in these sedimentary basins. 00:24:38.900 --> 00:24:41.970 I also used it for the Portland area basins. 00:24:41.970 --> 00:24:45.350 So if you’re going to talk about 3D models, you have to show a movie. 00:24:45.350 --> 00:24:55.080 So here’s one of – one of the 3D runs. This one starts off the coast of Oregon. 00:24:55.080 --> 00:24:58.600 Bilateral rupture. You can see – when it hits some of 00:24:58.600 --> 00:25:01.490 these sub-events, you see these energy radiating out from that. 00:25:01.490 --> 00:25:04.000 And there’s a sub-event down here too. 00:25:04.000 --> 00:25:07.460 You can see, as it reaches the Puget Lowland area in the Seattle-Tacoma 00:25:07.460 --> 00:25:12.270 Basin, things start to light up there and keep reverberating there. 00:25:12.270 --> 00:25:16.750 And eventually, it propagates off the edge of the model here. 00:25:16.750 --> 00:25:22.520 So this corresponds to a 300-second duration in real time for this simulation. 00:25:23.970 --> 00:25:26.480 So here I show the results of one of these simulations. 00:25:26.490 --> 00:25:29.830 I’m just showing the sub-event slip. We also had the background slip. 00:25:29.830 --> 00:25:33.200 But the background slip doesn’t really do much for periods shorter than 00:25:33.200 --> 00:25:38.289 about 7 seconds. So it’s really dominated by the sub-events. 00:25:38.289 --> 00:25:42.680 So this shows a 0.2-second spectral acceleration derived from the synthetics. 00:25:42.680 --> 00:25:45.440 So this is basically from the stochastic part. 00:25:45.440 --> 00:25:47.770 You can see the high ground motions right above 00:25:47.770 --> 00:25:50.440 the sub-events – not a big surprise here. 00:25:50.440 --> 00:25:55.310 But the biggest implication is that there’s a lot of variability of the high frequency 00:25:55.310 --> 00:25:58.840 along the coast because it depends on how close you are to one of these 00:25:58.840 --> 00:26:02.100 sub-events, how close you are to the peak slip in these sub-events. 00:26:02.100 --> 00:26:05.040 And I could – I should say that the correlation distance we chose 00:26:05.040 --> 00:26:08.320 for the magnitude 8 sub-events is similar to what people see 00:26:08.320 --> 00:26:13.679 in actual slip inversions for magnitude 8 earthquakes. 00:26:13.679 --> 00:26:19.240 At the longer period – and I should also add that, for the stochastic part, for now, 00:26:19.240 --> 00:26:22.770 we’re using only one type of site condition corresponding to a 00:26:22.770 --> 00:26:27.840 Vs30 of 500 meters a second. In the future, we will use maps 00:26:27.840 --> 00:26:31.789 of surficial geology and customize the shear wave velocity 00:26:31.789 --> 00:26:35.850 for the stochastic part with that. 00:26:35.850 --> 00:26:39.800 For the finite difference simulation result, this is for 3-second spectral 00:26:39.800 --> 00:26:46.140 acceleration. Again, you see the highest values are above the sub-events, 00:26:46.140 --> 00:26:49.170 but you see these streaks of high ground motions extending inland. 00:26:49.170 --> 00:26:54.400 And these are basically from forward rupture directivity from the – from the 00:26:54.400 --> 00:26:59.180 nearest asperity. So you see streaks coming in here from that sub-event. 00:26:59.180 --> 00:27:02.750 The hypocenter is right here. So you see that streak. 00:27:02.750 --> 00:27:04.929 You also see the high ground motions 00:27:04.929 --> 00:27:08.600 in the Puget Sound region and the Seattle Basin. 00:27:08.600 --> 00:27:12.460 This shows some of the broadband acceleration synthetics. 00:27:12.460 --> 00:27:16.590 Obviously, when you’re close to the coast, Seaside and Newport, 00:27:16.590 --> 00:27:22.660 you see large peak ground – PGAs of about half a g or so. 00:27:22.660 --> 00:27:25.490 As you go inland – Seattle’s about 100 kilometers away – 00:27:25.490 --> 00:27:28.990 it’s more like 2/10 of a g. And Portland’s about the 00:27:28.990 --> 00:27:34.960 same distance – about 80 kilometers away – and about 15%g in this case. 00:27:34.960 --> 00:27:38.720 But what’s interesting, again, is comparing Seaside, Oregon, 00:27:38.720 --> 00:27:43.789 with a site 50 kilometers to the south and 100 kilometers to the south here. 00:27:43.789 --> 00:27:46.830 You can see the variability in the peak ground acceleration and the 00:27:46.830 --> 00:27:52.320 waveform character. So it really – highly variable along the coast. 00:27:53.800 --> 00:27:56.700 If you zoom in on the Puget Sound region, 00:27:56.700 --> 00:28:01.060 this shows the 3-second spectral acceleration there. 00:28:01.060 --> 00:28:05.140 And I should say, this is based on 25,000 00:28:05.140 --> 00:28:07.360 synthetic seismograms in this box here. 00:28:07.360 --> 00:28:11.890 We saved the synthetics for the whole model on a 1-kilometer grid. 00:28:11.890 --> 00:28:15.770 And then we’re more selective in how often – the grid size we use 00:28:15.770 --> 00:28:19.700 to make the actual broadbands. But we have synthetics on basically 00:28:19.700 --> 00:28:23.900 a couple hundred thousand synthetics – sets of synthetics for each run. 00:28:24.700 --> 00:28:29.100 So here’s the outline of the Seattle Basin provided by Rick Blakely. 00:28:29.100 --> 00:28:32.610 These contours are depths to a shear wave velocity 00:28:32.610 --> 00:28:38.690 of 2-1/2 kilometers a second. So that’s Z2.5, and so the depth 00:28:38.690 --> 00:28:44.590 to crystalline basement under Seattle is about – as much as 7 kilometers. 00:28:44.590 --> 00:28:46.429 So you can see, within the Seattle Basin, 00:28:46.429 --> 00:28:51.559 you have a large amplitudes, big amplification relative to sites outside 00:28:51.559 --> 00:28:55.950 the basin, and also, to some extent, the Tacoma Basin does the same thing. 00:28:55.950 --> 00:29:00.000 You see the highest values are towards the western edge of the Seattle Basin. 00:29:00.000 --> 00:29:05.120 Here’s some velocity synthetics. The site within the basin has 00:29:05.120 --> 00:29:09.520 much larger PGVs than the two sites outside the basin here. 00:29:11.720 --> 00:29:15.100 And I should say that engineers in Seattle are very concerned about 00:29:15.100 --> 00:29:19.419 the Seattle Basin amplification because it affects the tall buildings. 00:29:19.419 --> 00:29:22.960 As you know, there’s a lot of tall buildings being built in Seattle 00:29:22.960 --> 00:29:28.789 right now above 10 stories or more. And so it’s a really important issue 00:29:28.789 --> 00:29:32.180 to quantify the amplification of the Seattle Basin. 00:29:32.180 --> 00:29:40.380 This shows the results from 10,000 on-land sites for one of these 3D runs. 00:29:40.380 --> 00:29:42.200 You can see these high values here are for 00:29:42.200 --> 00:29:45.049 sites in the Seattle and Tacoma Basins. 00:29:45.049 --> 00:29:47.590 You see more variability as you get close in, again, 00:29:47.590 --> 00:29:51.160 from directivity and where the sub-events are located. 00:29:53.470 --> 00:30:00.140 So what we decided to do is basically have a logic tree – 00:30:00.159 --> 00:30:04.640 although I put this in quotes because there’s aleatory and epistemic 00:30:04.640 --> 00:30:09.539 components in this logic tree, so we want to be precise in our language here. 00:30:09.539 --> 00:30:12.549 So we did 30 rupture scenarios. We kept everything at 00:30:12.549 --> 00:30:16.110 a moment magnitude 9.0. So we vary the rupture – 00:30:16.110 --> 00:30:19.549 the down-dip position of the rupture. 00:30:19.549 --> 00:30:22.529 We have three choices consistent to what was used in the 00:30:22.529 --> 00:30:27.640 2014 National Seismic Hazard Maps. We varied the hypocenter. 00:30:27.640 --> 00:30:31.799 We varied the slip distributions of the sub-events and the background slip. 00:30:31.800 --> 00:30:35.640 And we vary where the sub-events were located. 00:30:38.070 --> 00:30:44.640 So this – on the left, this picture shows the hypocenters from these 30 runs. 00:30:44.640 --> 00:30:48.320 And the stars are color- coded based on which of these down-dip ruptures 00:30:48.320 --> 00:30:53.250 we use. So this shows the three rupture – down-dip ruptures we used. 00:30:53.250 --> 00:30:58.980 We always assume rupture on the west end was extended to this point here. 00:30:58.980 --> 00:31:02.610 So the blue line is the top of the tremor zone. 00:31:02.610 --> 00:31:07.100 The red line is the 1-centimeter-a-year locking depth found 00:31:07.100 --> 00:31:11.460 from inversion of GPS and uplift data. And the green line is basically the 00:31:11.460 --> 00:31:18.179 midpoint of the locking position from thermal models and the red line here. 00:31:18.179 --> 00:31:22.679 So we had big discussions about these things in a workshop – I think it was 00:31:22.680 --> 00:31:29.320 about 2011 or so. And these were the major choices people came up with. 00:31:29.320 --> 00:31:33.020 And the weighting – the middle one gets the most weight. 00:31:33.020 --> 00:31:37.179 And so this is what was used in the 2014 map. 00:31:37.179 --> 00:31:44.160 And notice a number of runs were proportional to the weighting of 00:31:44.160 --> 00:31:49.809 that subduction – of that west – I should say of the down-dip interface. 00:31:49.809 --> 00:31:56.780 So we did more runs with this middle trace than with the ones to either side. 00:31:56.780 --> 00:32:00.430 Here shows the sub-event rupture zones, and they vary between runs. 00:32:00.430 --> 00:32:05.270 We always kept these five sub-events, but they moved along strike, 00:32:05.270 --> 00:32:07.840 and to some extent, moved along dip too. 00:32:07.840 --> 00:32:11.799 We try to keep them in the – towards the bottom of the subduction zone. 00:32:11.799 --> 00:32:15.990 And so, when we do this kind of eastern edge, 00:32:15.990 --> 00:32:19.680 the sub-events move to the east a bit, and they get deeper. 00:32:21.360 --> 00:32:24.160 So here shows the log average spectral accelerations 00:32:24.160 --> 00:32:28.289 at different periods from these 30 scenarios. 00:32:28.289 --> 00:32:31.390 Basically – so these are dominated by the stochastic part. 00:32:31.390 --> 00:32:36.289 This is a mix of the stochastic and deterministic at 1 second, and 3 seconds 00:32:36.289 --> 00:32:40.980 and 7-1/2 seconds are dominated by the finite difference synthetics. 00:32:40.980 --> 00:32:45.100 You can see, for all these cases, the highest values of the average 00:32:45.100 --> 00:32:48.130 are in the southern portion of the rupture zone. 00:32:48.130 --> 00:32:51.140 So what’s going on here is the – as you go to the south, 00:32:51.140 --> 00:32:54.610 the rupture zone gets narrower, and the sub-events are pushed up 00:32:54.610 --> 00:32:59.750 to basically a shallower depth because – and they basically fill the whole width 00:32:59.750 --> 00:33:03.510 of the subduction zone in the south. So perhaps in the future, we might 00:33:03.510 --> 00:33:06.820 want to make the sub-events narrower in the south than in the north. 00:33:06.820 --> 00:33:10.860 But that’s what’s causing these higher ground motions to the south. 00:33:11.500 --> 00:33:15.780 If you look at the Puget Lowland area for the average of these 00:33:15.789 --> 00:33:19.220 30 scenarios, 2-second spectral acceleration, again, you see the 00:33:19.220 --> 00:33:24.640 large amplification in the Seattle Basin and also in the Tacoma Basin, 00:33:24.640 --> 00:33:27.809 and the whole Puget Sound – Puget Lowland area is 00:33:27.809 --> 00:33:32.820 higher than the areas around it. See the same thing at 5 seconds. 00:33:33.500 --> 00:33:36.620 I think maybe things are concentrated in the western part because there’s 00:33:36.620 --> 00:33:40.640 a big velocity contrast between the Olympic Mountains and 00:33:40.640 --> 00:33:45.260 the Seattle Basin velocities that could be causing more focusing 00:33:45.260 --> 00:33:48.559 and more basin edge converted phases. 00:33:48.559 --> 00:33:50.779 So one of the key things we did in this whole process was 00:33:50.779 --> 00:33:56.559 compare our spectral response values from the synthetics with that of 00:33:56.559 --> 00:34:01.350 BC Hydro’s GMPEs and also with the Maule earthquake observations. 00:34:01.350 --> 00:34:02.940 And these are for non-basin sites. 00:34:02.940 --> 00:34:06.659 And everything I’m showing is for onshore sites. 00:34:06.659 --> 00:34:10.130 So these are the results at 3-second period for each one 00:34:10.130 --> 00:34:15.110 of these runs – the 30 runs. Each one of these is the – runs is 00:34:15.110 --> 00:34:20.680 represented by one of the black lines here. And I binned these by distances. 00:34:21.970 --> 00:34:27.200 The black dots are the median – or, log average for each 00:34:27.210 --> 00:34:31.810 distance bin from the simulations for the 30 runs. 00:34:31.810 --> 00:34:37.030 The red ticks are the inter-event sigma or standard deviation. 00:34:37.030 --> 00:34:43.389 The black error bars are the total standard deviation of all the – 00:34:43.389 --> 00:34:48.120 all these 30 runs and all the sites used in each of the 30 runs. 00:34:49.360 --> 00:34:51.580 These green lines are the predictions from the 00:34:51.580 --> 00:34:55.290 BC Hydro relation, plus or minus 1 sigma. 00:34:55.290 --> 00:35:00.890 The blue diamonds are the observations from the Maule earthquake. 00:35:00.890 --> 00:35:04.460 So the first thing to note is that we’re – our medians 00:35:04.460 --> 00:35:10.220 are generally within 1 sigma of the BC Hydro GMPEs. 00:35:10.220 --> 00:35:15.540 And we’re falling right in line with the Maule observations at 3 seconds. 00:35:16.380 --> 00:35:18.040 You can see, as you get closer in, 00:35:18.040 --> 00:35:23.680 there’s more variability because of the rupture directivity in the sub-events. 00:35:23.680 --> 00:35:26.640 This shows the results at other periods. 00:35:27.570 --> 00:35:31.320 For 0.2 seconds and 1 second, since you have the stochastic part, 00:35:31.339 --> 00:35:34.190 is dominated at 0.2 seconds. 00:35:34.190 --> 00:35:39.380 You see much lower inter-event variability – this is much tighter here. 00:35:39.380 --> 00:35:42.470 And that’s because we haven’t considered a lot of things, 00:35:42.470 --> 00:35:47.140 like variations in the crustal path. We haven’t considered 00:35:47.140 --> 00:35:49.869 that in the stochastic part. Obviously that’s the key to the 00:35:49.869 --> 00:35:54.210 3D model, but the 3D model’s not used in the stochastic part. 00:35:54.210 --> 00:35:56.560 So let’s see. 00:35:56.560 --> 00:36:00.600 We also haven’t considered variations in the shear wave velocity profile 00:36:00.600 --> 00:36:05.050 for a given Vs30 or variations in stress drop or anything like that. 00:36:05.050 --> 00:36:10.940 So it’s much tighter. You get much – you get large variability when you’re close in. 00:36:10.940 --> 00:36:14.960 And some of the Maule sites do have higher ground motions 00:36:14.960 --> 00:36:16.510 than the synthetics. 00:36:16.510 --> 00:36:21.950 Some of these sites have lower Vs30 than we used in the – for the synthetics. 00:36:21.950 --> 00:36:23.920 But in general, for all these periods, 00:36:23.920 --> 00:36:28.670 we’re within plus or minus 1 sigma of the BC Hydro. 00:36:28.670 --> 00:36:31.070 You can see, at 7-1/2 seconds, something interesting is happening. 00:36:31.070 --> 00:36:34.530 It looks like the distance decay in our synthetics is different 00:36:34.530 --> 00:36:40.100 from that in the BC Hydro relation. So this is a interesting question. 00:36:40.100 --> 00:36:43.000 Is this a Q issue? We’re basically assuming 00:36:43.000 --> 00:36:48.849 that the Q in the mid-crust is about 300, and we use a frequency-independent Q 00:36:48.849 --> 00:36:51.829 in the finite difference. So is that too high? 00:36:51.829 --> 00:36:57.260 I mean, there are some studies using noise correlation at fine Q’s 00:36:57.260 --> 00:37:02.710 at long periods lower than 300 – maybe as low as 70 or so. 00:37:02.710 --> 00:37:06.720 But a Q of 70 would not explain this. That would only make a factor 00:37:06.720 --> 00:37:11.160 of 1-1/2 difference at 200 kilometers, and we see more than that. 00:37:11.160 --> 00:37:14.900 One interesting thing is that the sources of – the sub-event sources 00:37:14.900 --> 00:37:21.020 in Cascadia have depths between about 15 and 30 kilometers, which is shallower, 00:37:21.020 --> 00:37:25.750 generally, than the sub-event depths seen in the Tohoku earthquake 00:37:25.750 --> 00:37:27.670 and also the Maule earthquake. 00:37:27.670 --> 00:37:30.849 So I’m wondering if this source depth effect could be responsible for this. 00:37:30.849 --> 00:37:34.750 We could have more surface wave generated by the shallower sources 00:37:34.750 --> 00:37:37.339 and produce a different distance decay. 00:37:37.339 --> 00:37:40.810 But this is something we need to look at in the future. 00:37:40.810 --> 00:37:45.100 Okay, so this summarizes the comparisons of the non-basin sites 00:37:45.100 --> 00:37:50.740 with the BC Hydro GMPEs. This shows the bias and the 00:37:50.740 --> 00:37:53.990 standard deviation of the spectral accelerations from 00:37:53.990 --> 00:37:59.339 zero to 10-second period relative – of the synthetics relative to BC Hydro. 00:37:59.339 --> 00:38:04.070 So zero would be that, on average, we’re matching the spectral accelerations 00:38:04.070 --> 00:38:09.831 of the BC Hydro GMPEs. So we can see, between zero and 00:38:09.831 --> 00:38:15.740 6 seconds, we’re within plus or minus 0.5 natural log units of BC Hydro. 00:38:15.740 --> 00:38:19.470 And that’s generally considered a fairly good fit in the 00:38:19.470 --> 00:38:24.180 ground motion modeling world. But for periods greater than 6 seconds, 00:38:24.180 --> 00:38:27.440 we’re overestimating things compared to BC Hydro. 00:38:27.440 --> 00:38:31.190 A lot of that was that distance dependence difference that 00:38:31.190 --> 00:38:34.860 I showed you in the last slide. Some of this could also be – 00:38:34.860 --> 00:38:39.780 even at close distances, there could be a small bias there too. 00:38:43.680 --> 00:38:48.619 This shows the various standard deviations – the total from the 00:38:48.619 --> 00:38:52.770 synthetics and the inter-event and intra-event at 3 seconds. 00:38:52.770 --> 00:38:56.790 So the blue is the total sigma, and these are natural log units. 00:38:56.790 --> 00:38:58.950 It’s interesting that, from the synthetics – 00:38:58.950 --> 00:39:02.930 the long-period synthetics, the sigma for the total variability 00:39:02.930 --> 00:39:07.670 is very similar to the sigma that’s quoted in the BC Hydro GMPEs, 00:39:07.670 --> 00:39:10.400 which is based on a global set of data. 00:39:10.400 --> 00:39:17.190 So our synthetics seem to have already captured the total sigma seen in – 00:39:17.190 --> 00:39:23.200 I guess you would call it the ergodic sigma from the BC Hydro GMPEs. 00:39:23.200 --> 00:39:29.020 You can see that – and we haven’t even tried to vary the stress drop or some – 00:39:29.020 --> 00:39:32.050 or slip velocity or any of those other parameters that you would 00:39:32.050 --> 00:39:36.760 want to add to get a true measure of the total variability. 00:39:36.760 --> 00:39:39.830 You see, as you get close in, again, you have a higher variability 00:39:39.830 --> 00:39:43.380 because of the location of the sub-events. 00:39:43.380 --> 00:39:46.780 If you look at the short period, again, we’re very – we’re underestimating 00:39:46.790 --> 00:39:50.490 the variability of the short period, particularly the inter-event, 00:39:50.490 --> 00:39:54.720 because we’re not considering these lateral variations in crustal 00:39:54.720 --> 00:40:00.579 propagation effects or variations in the shear wave profile. 00:40:00.580 --> 00:40:05.020 So we’re clearly underestimating things on the stochastic part. 00:40:05.020 --> 00:40:08.280 Now I’d like to turn to amplification of the Seattle Basin. 00:40:08.280 --> 00:40:11.010 And I mentioned before that we added this random component 00:40:11.010 --> 00:40:16.390 of shear wave velocity. Only 5% standard deviation. 00:40:16.390 --> 00:40:18.780 And this makes an effect on the response spectra. 00:40:18.780 --> 00:40:23.190 So this shows the response spectra for a Seattle Basin site from the synthetics. 00:40:23.190 --> 00:40:27.730 The black line is for the case with no randomness. 00:40:27.730 --> 00:40:33.380 And notice the large amplification relative to a site outside the basin. 00:40:33.380 --> 00:40:39.020 If we add just this 5% randomness, we get this red response spectra here. 00:40:39.030 --> 00:40:43.470 So it really – it drops things about 30% in the response spectra. 00:40:43.470 --> 00:40:46.490 And this is because adding the randomness breaks down 00:40:46.490 --> 00:40:50.849 some of the shear wave velocity resonances and also cuts down 00:40:50.849 --> 00:40:54.380 on some of these basin edge generated surface waves a bit. 00:40:54.380 --> 00:40:58.520 So it’s an important thing to consider. I think it’s a realistic thing to consider. 00:40:58.520 --> 00:41:02.240 And if we don’t add it in, I think we can have – 00:41:02.240 --> 00:41:07.150 we might overestimate things as far as the amplification. 00:41:07.150 --> 00:41:12.349 So this shows the response spectra for these 30 scenarios for different sites. 00:41:12.349 --> 00:41:16.710 So red is for a site in the Seattle Basin, plus or minus 1 sigma. 00:41:16.710 --> 00:41:21.280 Black is a site just outside the basin. You can see the large amplification 00:41:21.280 --> 00:41:27.120 of the Seattle Basin from the finite difference synthetics, 00:41:27.120 --> 00:41:31.200 so greater than 1 second. Here’s Portland versus Newport. 00:41:31.200 --> 00:41:33.460 You can see, obviously Newport’s closer in. 00:41:33.460 --> 00:41:38.200 It has higher short-period spectral accelerations. 00:41:38.200 --> 00:41:41.270 But Portland Basin is starting to amplify things, and you see 00:41:41.270 --> 00:41:46.060 that effect in the simulations, particularly at 1 to 2 seconds. 00:41:46.060 --> 00:41:50.880 Okay, so now I’m focusing on the Seattle Basin sites. 00:41:51.830 --> 00:41:54.900 So I did synthetics – looked at synthetics at a site 00:41:54.900 --> 00:41:58.180 within the Seattle Basin versus one just outside the basin. 00:41:58.180 --> 00:42:02.270 This is the southern edge of the Seattle Basin in red here. 00:42:02.270 --> 00:42:06.270 So this black line is that amplification 00:42:06.270 --> 00:42:09.630 averaged over 30 runs, plus or minus 1 sigma. 00:42:09.630 --> 00:42:11.250 So you see the amplification is about 00:42:11.250 --> 00:42:15.480 a factor of 2 to 3 periods of 1 to 10 seconds. 00:42:16.380 --> 00:42:18.920 Now, these colored lines are from actual data. 00:42:18.920 --> 00:42:22.260 So these are from this – data from this array that we’ve been operating 00:42:22.260 --> 00:42:27.690 in Seattle for about 20 years now. And I used an earthquake – 00:42:27.690 --> 00:42:32.630 the Satsop earthquake, which is about 120 kilometers to the west of Seattle. 00:42:32.630 --> 00:42:34.800 It’s located at a depth of 40 kilometers. 00:42:34.800 --> 00:42:38.140 It’s located just below the Cascadia interface. 00:42:38.140 --> 00:42:40.099 So the angles of incidence to Seattle 00:42:40.099 --> 00:42:44.610 should be similar to the Cascadia M9 simulations. 00:42:44.610 --> 00:42:48.760 And so these colored lines are the observed response 00:42:48.760 --> 00:42:52.920 spectral amplitudes referenced to a site outside the Seattle Basin. 00:42:52.920 --> 00:42:56.360 And I should also again say that the Vs30s inside and outside the 00:42:56.360 --> 00:43:01.230 basin are about the same, on average. So we can see that the observed – 00:43:01.230 --> 00:43:05.491 we have two spectral peaks, but if you take this 1-second to 4-second 00:43:05.491 --> 00:43:10.940 average here, it’s in the same ballpark as the M9 simulations. 00:43:10.940 --> 00:43:17.260 Very comparable – factors of 2 to 3 averaged amplification from the data. 00:43:17.260 --> 00:43:19.079 Above 5 seconds, we don’t have good 00:43:19.079 --> 00:43:22.660 signal-to-noise for the Satsop earthquake. 00:43:22.660 --> 00:43:28.900 This dash-dot line is the predicted basin amplification term from the Campbell 00:43:28.900 --> 00:43:35.620 and Bozorgnia NGA-West2 GMPEs, which are based on crustal earthquakes. 00:43:35.630 --> 00:43:39.671 And you can see that’s only about a factor of 1-1/2 – if you plug in a 00:43:39.671 --> 00:43:44.650 Z2.5 of 7 kilometers, you only get a amplification of 1-1/2. 00:43:44.650 --> 00:43:48.069 So it’s much lower than the M9 synthetics show 00:43:48.069 --> 00:43:51.950 and lower than the observations from the Satsop earthquake. 00:43:51.950 --> 00:43:53.950 If you plug in the Chou and Youngs, 00:43:53.950 --> 00:44:01.350 you only get a factor of 1.2 amplification using a Z1.0 of about a kilometer. 00:44:01.350 --> 00:44:07.020 So clearly these crustal earthquakes in the GMPEs are underestimating 00:44:07.020 --> 00:44:14.290 the amplification of the Seattle Basin. This could be because the – both the 00:44:14.290 --> 00:44:16.420 M9 simulation and the Satsop earthquake, 00:44:16.420 --> 00:44:19.960 the angles of incidence to the basin are very shallow. 00:44:19.960 --> 00:44:23.040 And the crustal earthquake database maybe be dominated by cases 00:44:23.040 --> 00:44:27.640 where the earthquake is directly below the basin and things are coming up 00:44:27.640 --> 00:44:31.840 relatively vertically into the basin. So you don’t have the surface 00:44:31.840 --> 00:44:37.380 wave amplification that you would get from sources off to the side. 00:44:37.390 --> 00:44:39.701 You don’t get the basin edge converted phases where the 00:44:39.701 --> 00:44:43.150 S wave converts to a basin surface wave at the side. 00:44:43.150 --> 00:44:46.760 So this could explain the difference in the amplification. 00:44:46.760 --> 00:44:51.340 Now, this site just outside the Seattle Basin also is amplified relative 00:44:51.349 --> 00:44:55.390 to a site outside the Puget Lowlands. So here we calculate the 00:44:55.390 --> 00:45:02.589 basin amplification factor for sites within this thick line here. 00:45:02.589 --> 00:45:07.660 So this is a Z2.5 of 5 kilometers. 00:45:07.660 --> 00:45:11.760 And we referenced it to the sites around the perimeter of the 00:45:11.760 --> 00:45:15.990 Puget Lowlands here. So this is the amplification we get. 00:45:15.990 --> 00:45:19.680 Between factors of 2 to 5, peaking at about 3 seconds. 00:45:19.680 --> 00:45:24.580 Again, far higher than what was seen – what’s found in the – 00:45:24.580 --> 00:45:27.860 in the NGA-West2 crustal database. 00:45:28.560 --> 00:45:32.680 If we look at the overall duration of shaking, this is largely driven by 00:45:32.690 --> 00:45:35.950 the stochastic part because we’re using the Arias Intensity – 00:45:35.950 --> 00:45:40.900 the 5th to 95th percentile. You see the duration increases with distance. 00:45:40.900 --> 00:45:45.540 For Seattle, it’s about 100 seconds of significant shaking. 00:45:45.540 --> 00:45:49.060 And you can compare that with the Nisqually earthquake, which was 6.8, 00:45:49.060 --> 00:45:52.630 which had about 20 seconds. So the duration of shaking in Seattle 00:45:52.630 --> 00:45:57.500 should be about 4 to 5 times of what we saw in the Nisqually earthquake. 00:45:57.500 --> 00:46:01.510 How am I on time here? Guess I’d better – so just show 00:46:01.510 --> 00:46:05.410 a couple of sensitivity studies. We wanted to see how the 00:46:05.410 --> 00:46:09.440 eastern edge of the rupture zone affected the ground motions. 00:46:09.440 --> 00:46:12.600 So here are those three choices for the eastern edge. 00:46:13.600 --> 00:46:18.220 This shows the synthetics for a case where I used the same hypocenter. 00:46:18.220 --> 00:46:21.290 Of course, as we do the deeper choice, 00:46:21.290 --> 00:46:25.050 the sub-events are moved farther to the east. 00:46:25.050 --> 00:46:28.440 So you can see this is a case where rupture stops offshore. 00:46:28.440 --> 00:46:31.060 This is where it reaches onshore, basically. 00:46:31.060 --> 00:46:33.790 And this is where it gets east of the coast. 00:46:33.790 --> 00:46:37.030 So you can see that, along the coast, it’s pretty sensitive to whether 00:46:37.030 --> 00:46:41.230 rupture was stopped offshore or reaches the shore. 00:46:41.230 --> 00:46:45.090 You see these higher ground motions along the coast for this case. 00:46:45.090 --> 00:46:48.849 As things go further inland, though, there’s not that big of a change, 00:46:48.849 --> 00:46:51.079 generally, in the ground motions. 00:46:52.020 --> 00:46:56.000 If you look at the response spectra for a site in the Seattle Basin, 00:46:56.000 --> 00:46:59.530 you can see that the values are much higher for the rupture to the 00:46:59.530 --> 00:47:04.450 coast than a rupture offshore, at least at these periods of 1 to 5 seconds. 00:47:04.450 --> 00:47:09.410 So obviously we don’t know which of these is the proper position 00:47:09.410 --> 00:47:16.360 for the eastern edge, so this adds to the epistemic uncertainty here. 00:47:16.360 --> 00:47:19.740 I like to show sort of a relatively bad and relatively good case 00:47:19.740 --> 00:47:25.390 for Seattle using the same background slip in both. 00:47:26.000 --> 00:47:32.260 Here’s the sub-events used in a relatively bad scenario, 00:47:32.260 --> 00:47:35.610 relatively good scenario for Seattle. The rupture starts here. 00:47:35.610 --> 00:47:39.790 And in this case, you have forward rupture directivity from this sub-event 00:47:39.790 --> 00:47:42.859 towards the Puget Sound region, so that’s relatively bad. 00:47:42.859 --> 00:47:48.530 In this case, rupture starts here and essentially ruptures away from the – 00:47:48.530 --> 00:47:52.900 from the Puget Sound area, so it’s relatively good. 00:47:52.900 --> 00:47:57.140 So this shows a response spectra for the Seattle Basin for the 00:47:57.140 --> 00:48:01.730 relatively bad scenario. You can see the response spectra 00:48:01.730 --> 00:48:04.200 much higher than for the relatively good scenario. 00:48:04.200 --> 00:48:08.720 It’s about a factor of 4 difference. What’s interesting is, it seems like 00:48:08.720 --> 00:48:13.350 the amplification of the Seattle Basin is coupled to the rupture directivity. 00:48:13.350 --> 00:48:17.550 So you see higher amplification in the Seattle Basin for ruptures 00:48:17.550 --> 00:48:21.060 heading towards it than you see for ruptures heading away. 00:48:21.060 --> 00:48:23.930 And this has some similarity to what we’ve seen in the ShakeOut 00:48:23.930 --> 00:48:28.510 scenario for San Andreas Fault. So this – and we also have seen 00:48:28.510 --> 00:48:32.240 this in our simulations of Seattle Fault earthquakes. 00:48:34.620 --> 00:48:37.849 And you notice this peak here in about – 00:48:37.849 --> 00:48:42.020 between 1 and 2 seconds for the basin response spectra. 00:48:42.020 --> 00:48:46.720 And you can understand this better by looking at the synthetic seismograms. 00:48:46.730 --> 00:48:52.190 Here’s the velocity waveform for a site just outside the Seattle Basin. 00:48:52.190 --> 00:48:54.450 You see the S wave. Notice – this is one with 00:48:54.450 --> 00:48:58.760 forward-rupture directivity, so the durations are very short here. 00:48:58.760 --> 00:49:01.510 And then you see the surface wave – the fundamental mode surface waves 00:49:01.510 --> 00:49:05.200 coming in behind there. When you go into the basin, S waves, 00:49:05.200 --> 00:49:09.290 amplitude, now you see this phase, which is a basin edge converted phase. 00:49:09.290 --> 00:49:13.369 So this is like a schematic cross-section of the Seattle Basin, 00:49:13.369 --> 00:49:18.020 going from southwest to north – going from south to north. 00:49:18.020 --> 00:49:22.839 So this is – a S wave hits the side of the basin, gets multiply reflected, 00:49:22.839 --> 00:49:27.950 gets critically trapped in the sediments – the Quaternary sediments. 00:49:27.950 --> 00:49:30.630 In this case, it becomes a Rayleigh wave, 00:49:30.630 --> 00:49:34.120 and that’s what you see on the north-south component here. 00:49:34.900 --> 00:49:39.540 And so this is a basin edge – and this is very important for the amplification, 00:49:39.550 --> 00:49:44.220 and this is actually dominated at periods of about 1 to 2 seconds. 00:49:44.220 --> 00:49:48.520 You also see the amplification of the fundamental mode surface waves. 00:49:48.520 --> 00:49:52.780 So if you band-pass filter these synthetics at 1 to 2 seconds, 00:49:52.780 --> 00:49:55.890 for the basin site, the S wave is actually quite small, 00:49:55.890 --> 00:49:58.750 and this basin edge converted phase is the largest phase. 00:49:58.750 --> 00:50:02.079 Of course, you don’t see it for the site outside the basin. 00:50:02.079 --> 00:50:04.349 At 3-second period, they’re about the same amplitude – 00:50:04.349 --> 00:50:09.360 the surface wave and the S wave. And we saw this in the Nisqually earthquake. 00:50:09.360 --> 00:50:13.710 We saw a very clear case of this basin edge generated surface wave. 00:50:13.710 --> 00:50:17.970 So this is real data from the Nisqually earthquake, filtered to 1 and 2 hertz. 00:50:17.970 --> 00:50:21.970 That basin edge surface wave again dominates between 1 and 2 seconds. 00:50:21.970 --> 00:50:24.470 So I think it’s something in the eigenfunctions of these 00:50:24.470 --> 00:50:28.230 higher-mode Rayleigh waves in the basin that causes this 00:50:28.230 --> 00:50:33.520 higher amplification at 1 to 2 seconds in the basin edge phase. 00:50:34.800 --> 00:50:39.820 The final thing I’d like to just show is a set of M9 synthetics. 00:50:39.820 --> 00:50:45.400 And I filtered these between 0.25 and 1 hertz – so 1 to 4 seconds. 00:50:45.400 --> 00:50:50.420 And this is for a firm rock site outside of the Seattle Basin but near Seattle. 00:50:50.430 --> 00:50:54.660 And I’ve also shown – put on here two real records – one from – 00:50:54.660 --> 00:50:58.060 well, from the Tohoku earthquake and from the Maule earthquake. 00:50:58.060 --> 00:51:01.680 I like to do these sanity checks to see if anything – you know, 00:51:01.680 --> 00:51:04.349 we’re in the right ballpark on these things, of course. 00:51:04.349 --> 00:51:06.420 And so you can guess which ones are the real ones 00:51:06.420 --> 00:51:08.840 and which ones are the synthetic. 00:51:09.589 --> 00:51:12.460 These top four are the synthetics. 00:51:13.210 --> 00:51:15.460 You can see this rupture directivity effect. 00:51:15.460 --> 00:51:20.940 This is obviously ruptured towards Seattle – pretty compact pulse. 00:51:20.940 --> 00:51:25.609 This is rupture away from Seattle. Smaller amplitude, longer duration. 00:51:25.609 --> 00:51:30.090 This is a record from the Maule earthquake – a site near Santiago. 00:51:30.090 --> 00:51:31.930 And this is a site from the Tohoku earthquake, 00:51:31.930 --> 00:51:36.059 and these stations are about the same distance from the sub-events 00:51:36.059 --> 00:51:40.060 as the Cascadia M9 cases I’m showing here. 00:51:40.060 --> 00:51:43.109 Now, the Tohoku earthquake, this site is more of a thin soil 00:51:43.109 --> 00:51:46.890 over hard rock, so perhaps the amplitude would be different 00:51:46.890 --> 00:51:50.880 if we had exactly the same site conditions, but generally, I think 00:51:50.880 --> 00:51:55.940 the character of the M9 synthetics is similar to what we see in the data. 00:51:55.940 --> 00:51:58.840 And I think this is pretty encouraging. 00:51:59.940 --> 00:52:02.080 So take-home messages. 00:52:03.100 --> 00:52:07.319 We think that we’ve captured a range of plausible ground motions 00:52:07.319 --> 00:52:09.309 for the magnitude 9 earthquakes. 00:52:09.309 --> 00:52:13.599 And these are being used to evaluate tall building response and ground failure. 00:52:13.599 --> 00:52:18.940 From 0.1 to 6 seconds, these synthetics are similar, on average, 00:52:18.940 --> 00:52:25.190 to BC Hydro GMPEs, but exceed them at periods greater than 6 seconds. 00:52:25.190 --> 00:52:28.329 The synthetic spectra show large variability, especially for 00:52:28.329 --> 00:52:33.450 close-in distances from the distance of the sub-events. 00:52:33.450 --> 00:52:37.300 We see large amplification factors in the Seattle Basin, 00:52:37.300 --> 00:52:42.500 which are much larger than are shown in the GMPEs for crustal earthquakes. 00:52:42.500 --> 00:52:46.490 And we’re going to post these synthetics on the DesignSafe website. 00:52:46.490 --> 00:52:51.119 The ShakeMaps will be posted on the USGS Scenario ShakeMap site. 00:52:51.119 --> 00:52:54.450 And these basin amplifications may go into a future round 00:52:54.450 --> 00:52:57.670 of the National Seismic Hazard Maps. 00:52:57.670 --> 00:53:01.320 So how do we improve the simulations? We have to look at improving our 3D 00:53:01.320 --> 00:53:08.220 velocity model, especially for urban areas in the Seattle Basin and Portland Basins. 00:53:08.230 --> 00:53:12.030 We also need better shear wave velocity structure offshore. 00:53:12.030 --> 00:53:15.940 That would – that’s important on these simulations. 00:53:15.940 --> 00:53:17.640 Can we say anything about these – 00:53:17.640 --> 00:53:21.080 where these future sub-events will be located? 00:53:21.080 --> 00:53:24.500 Are there places where the upper plate is composed of continental crust, 00:53:24.500 --> 00:53:29.599 as proposed by Ray Wells and others? Can GPS tell us something about 00:53:29.600 --> 00:53:35.600 where the fault zone is locked that could correspond to these sub-event locations? 00:53:35.600 --> 00:53:42.140 Are they – correspond to structural highs off the coast of the Pacific Northwest? 00:53:42.150 --> 00:53:44.780 So we have a lot of work left to do, but we have completed 00:53:44.780 --> 00:53:48.660 this stage of the M9 Project. Thanks. 00:53:48.660 --> 00:53:53.480 [ Applause ] 00:53:53.480 --> 00:53:56.560 - Thank you, Art, for that really interesting summary of your work. 00:53:56.560 --> 00:53:58.320 Have some questions? 00:54:00.100 --> 00:54:02.760 - Art, excellent talk. 00:54:02.760 --> 00:54:08.300 While you have that slide up here, though, could you prioritize those? 00:54:08.300 --> 00:54:13.360 It’s kind of hard, you know, from your – I mean, your sensitivity analysis clearly 00:54:13.360 --> 00:54:18.060 showed each one was important, but are some more important than others? 00:54:18.740 --> 00:54:21.820 - Well, there’s – can you hear me okay? 00:54:21.820 --> 00:54:26.290 There’s also the question of, what’s easier to do? You know, 00:54:26.290 --> 00:54:31.390 what’s more likely to make progress in a short period of time, I guess. 00:54:31.390 --> 00:54:35.550 I think it would be nice to know where these sub-events are, 00:54:35.550 --> 00:54:38.080 but I think that’s going to be difficult to predict them. 00:54:38.080 --> 00:54:41.680 I think we’re probably going to have to live with this randomness for a while. 00:54:43.280 --> 00:54:46.359 We probably should look back at Tohoku and Maule to see if we 00:54:46.359 --> 00:54:50.829 could have seen where those sub-events were located in the 00:54:50.829 --> 00:54:54.119 interseismic period before those events. I think some of that’s done, 00:54:54.120 --> 00:54:57.520 but I don’t think it’s conclusive one way or the other. 00:54:59.700 --> 00:55:05.340 In my thinking, it – really the highest priority is improving the velocity model. 00:55:05.349 --> 00:55:08.480 I think, in the Seattle Basin, one of the key things is the 00:55:08.480 --> 00:55:12.750 depth to bedrock and what the shear wave velocity contrast is 00:55:12.750 --> 00:55:16.570 from the base of the Quaternary sediments to the top of bedrock, 00:55:16.570 --> 00:55:20.960 how sharp that interface is, how gradual it is. 00:55:20.960 --> 00:55:24.480 And we have a very poor understanding of that in the Seattle – 00:55:24.480 --> 00:55:26.060 in these urban areas. 00:55:26.060 --> 00:55:30.760 SHIPS got us some information on that, but I think we need to do more work on 00:55:30.760 --> 00:55:34.980 the shear wave profiles in the top 2 kilometers in these urban basins. 00:55:34.980 --> 00:55:37.030 And I think that’s quite doable. 00:55:37.030 --> 00:55:40.640 And that’s one thing Bill Stephenson, I know, is concentrating on. 00:55:41.340 --> 00:55:44.380 You know, our 3D model as we have now does a pretty good job 00:55:44.390 --> 00:55:48.320 of modeling the data, but I want to have more information 00:55:48.320 --> 00:55:52.150 about the shallow structure of the Seattle Basin and these other basins. 00:55:52.150 --> 00:55:55.609 So that’s doable, and I think that’s a high priority. 00:55:55.609 --> 00:56:01.250 The sub-events – you know, you may need to have offshore GPS to really – 00:56:01.250 --> 00:56:05.260 because these sub-events are mainly offshore if they are there. 00:56:05.260 --> 00:56:08.130 You know, one question is, on the Cascadia subduction zone, is it 00:56:08.130 --> 00:56:12.240 going to be like the Maule earthquake? Is it going to be like Tohoku? 00:56:12.240 --> 00:56:15.680 The age of the subducting slab is much younger. 00:56:15.680 --> 00:56:19.400 So it’s much warmer than those other two places. 00:56:20.960 --> 00:56:24.820 So there is a possibility there won’t be sub-events for Cascadia. 00:56:24.829 --> 00:56:27.819 I have to admit that. But I think it’s prudent to assume 00:56:27.820 --> 00:56:33.500 it’s going to look like Maule and Tohoku and also the Sumatra event – 00:56:33.500 --> 00:56:38.700 2004 Sumatra event, which also had high frequencies coming from deeper parts. 00:56:38.710 --> 00:56:44.700 But I think, you know, having more GPS – denser GPS array 00:56:44.700 --> 00:56:49.049 along the coast might be able to help us tell where the lock – 00:56:49.049 --> 00:56:51.980 where these sub-events – because they should be expressed 00:56:51.980 --> 00:56:57.500 in the interseismic period. Their signature should be in the 00:56:57.500 --> 00:57:03.220 deformation field between, you know, times of the great earthquakes. 00:57:03.220 --> 00:57:06.310 So I would hope more dense GPS data, 00:57:06.310 --> 00:57:08.970 offshore GPS, could tell you something about that. 00:57:08.970 --> 00:57:11.430 But those two things – you know, it’s not clear 00:57:11.430 --> 00:57:15.080 whether we have the resources to do that right now. 00:57:18.040 --> 00:57:19.020 Yes? 00:57:19.020 --> 00:57:22.040 - That was a really interesting and comprehensive talk. 00:57:22.040 --> 00:57:26.040 At the beginning, you alluded to the use of these synthetics in the 00:57:26.040 --> 00:57:28.140 earthquake early warning context. - Yes. 00:57:28.140 --> 00:57:30.920 - I wondered if you might just say a few more words about that? 00:57:31.780 --> 00:57:36.800 - Well, I haven’t done too much of that, but the question is, in this early warning, 00:57:36.810 --> 00:57:41.940 how do you quickly recognize the magnitude of the event? 00:57:41.940 --> 00:57:44.599 And one of the key things, as you know, with early warning, 00:57:44.599 --> 00:57:49.819 is you – you know, it starts out – a rupture could start out – 00:57:49.819 --> 00:57:51.690 if you think of that movie, rupture could start out 00:57:51.690 --> 00:57:55.680 off the coast of Oregon and then just stop, right? 00:57:55.680 --> 00:57:58.400 So if you’re sitting in Seattle, what do you tell people? 00:57:58.400 --> 00:58:01.200 You know, well – it’s sort of like Sarah Minson’s talk. 00:58:01.200 --> 00:58:05.680 Like, you have to be willing to accept a lot of false alarms 00:58:05.680 --> 00:58:08.710 in some of these places. So we can’t – but maybe you could – 00:58:08.710 --> 00:58:12.280 by looking at how the long period builds up in the beginning, 00:58:12.280 --> 00:58:17.240 you can say earlier whether this is going to be a larger event. 00:58:17.240 --> 00:58:20.680 So in other words, a slip at any – in the beginning, once it builds up 00:58:20.680 --> 00:58:26.569 to a certain point, that gives you some clues to the – I think it 00:58:26.569 --> 00:58:29.950 gives you some – a little bit of advanced information about 00:58:29.950 --> 00:58:35.559 whether this is going to grow. But I just think they need to look at 00:58:35.560 --> 00:58:40.560 that and, you know, real – the synthetic data, in order to get 00:58:40.560 --> 00:58:44.940 a handle on how soon they can make pronouncements on – 00:58:44.940 --> 00:58:49.920 or lower their uncertainties on how far this thing’s going to propagate. 00:58:49.920 --> 00:58:51.420 And what … - So you see these synthetics 00:58:51.420 --> 00:58:53.580 as being useful for that … - Right. 00:58:53.580 --> 00:58:55.600 - [inaudible] - Right. 00:58:55.600 --> 00:59:02.460 And one interesting point is the dynamics of this kind of process. 00:59:02.460 --> 00:59:05.480 Are these great earthquakes really driven by these 00:59:05.490 --> 00:59:08.960 high stress drop sub-events or not? 00:59:08.960 --> 00:59:13.830 And so maybe – you know, how far this thing grows depends on whether 00:59:13.830 --> 00:59:18.329 these sub-events are there to sort of power them up the strike of a fault. 00:59:18.329 --> 00:59:21.840 So I think, at some point, obviously it’d be useful to 00:59:21.840 --> 00:59:26.049 try dynamic rupture simulations for these kind of scenarios. 00:59:26.049 --> 00:59:29.040 The one thing I’m concerned about – you see a lot of literature about 00:59:29.040 --> 00:59:35.070 subduction zone earthquakes, but I think this idea that you have these deeper 00:59:35.070 --> 00:59:41.630 sub-events is relatively limited to, like, the strong ground motion community. 00:59:41.630 --> 00:59:44.549 And a lot of the literature about the subduction zone earthquakes 00:59:44.549 --> 00:59:48.260 is based on long-period teleseismic inversions. 00:59:48.260 --> 00:59:53.200 And they sort of – you know, and these kind of sub-events are very small glitch – 00:59:53.210 --> 00:59:57.390 you know, perturbations on that long-period – where most of 00:59:57.390 --> 01:00:03.740 the slip is the shallower part of the fault. So I hope we start to think holistically 01:00:03.740 --> 01:00:08.040 about these subduction zone earthquakes and look at this range of frequencies 01:00:08.040 --> 01:00:14.640 and realize that, you know, subduction zone has different character with depth. 01:00:14.640 --> 01:00:18.530 Thorne Lay has a model for this, and other people have models for this, 01:00:18.530 --> 01:00:24.940 but I hope that we get a more broadband picture of the subduction zone than 01:00:24.940 --> 01:00:29.940 what’s been maybe dominating the literature – most of the literature. 01:00:30.960 --> 01:00:32.580 Yeah, Brad? 01:00:32.580 --> 01:00:35.040 - Two questions. The first one goes back to your Maule model 01:00:35.040 --> 01:00:39.599 and your actual rupture model. It looks like, in your sort of 01:00:39.599 --> 01:00:43.960 high stress drop areas, you’re allowing slip to go 01:00:43.960 --> 01:00:46.520 all the way to the edge of your rectangle, 01:00:46.520 --> 01:00:49.040 and then you have an abrupt … - It’s – it’s – yeah, 01:00:49.040 --> 01:00:51.980 that was a – it’s tapered. - Oh, okay. So you’re tapering off. 01:00:51.980 --> 01:00:55.380 Okay, so then the second question is, for that Satsop earthquake, 01:00:55.380 --> 01:00:59.980 where you’re looking at inside the basin, and you had sort of two peaks … 01:00:59.980 --> 01:01:01.190 - Yeah. - And you compared that to 01:01:01.190 --> 01:01:06.130 sort of the average over all – I think it was 30 simulations. 01:01:06.130 --> 01:01:12.160 If you look at, say, events that have a similar location for, say, 01:01:12.160 --> 01:01:15.859 one of the high stress drop areas, have you looked to see if you 01:01:15.859 --> 01:01:17.930 get a similar double-peak? 01:01:17.930 --> 01:01:21.839 Or did they remain generally smooth across that period band? 01:01:21.839 --> 01:01:24.530 - That’s a good idea. I haven’t looked at that. 01:01:24.530 --> 01:01:27.750 That peak is – there’s a couple things going on there. 01:01:27.750 --> 01:01:32.980 I looked those two peaks – one that – the hole between the two peaks 01:01:32.980 --> 01:01:37.990 is actually a peak in the denominator spectrum at that rock site, 01:01:37.990 --> 01:01:42.109 which I don’t understand why that has a peak there and you don’t see it 01:01:42.109 --> 01:01:47.220 in the basin site. So part of that is due to the denominator spectrum. 01:01:47.220 --> 01:01:51.220 And, yeah, so a question is, do you see that peak in M9 01:01:51.220 --> 01:01:56.180 simulations for that site? I don’t – that would be a good idea to see 01:01:56.180 --> 01:02:00.740 some of the individual sites if we see that same signature in the synthetics. 01:02:00.740 --> 01:02:03.460 That’s a good – I’ll look at that. 01:02:05.320 --> 01:02:12.300 - Hey, Art. Nice talk. For the sub-events and these high rise 01:02:12.300 --> 01:02:18.940 time events, these are seen very clearly in the seismic back projection … 01:02:18.940 --> 01:02:21.000 - Right. - … that were started in 2004 – 01:02:21.000 --> 01:02:26.200 after the 2004 event. John Vidale and others worked on that. 01:02:26.200 --> 01:02:32.789 So did you get your sub-events from looking at back projections of – 01:02:32.789 --> 01:02:35.900 for Maule? - Well, in the Maule case, 01:02:35.900 --> 01:02:38.569 you see these sources of high-frequency – 01:02:38.569 --> 01:02:44.010 I didn’t use it explicitly, but, you know, when I looked at – 01:02:44.010 --> 01:02:49.630 so the Maule back projection results – well, they vary with investigators, 01:02:49.630 --> 01:02:53.710 but they generally saw the high-frequency, say, 1-second energy 01:02:53.710 --> 01:02:57.400 coming from the deeper part. Some actually had two sources. 01:02:57.400 --> 01:03:03.210 I think some of the back projections had that, but I didn’t look at that explicitly. 01:03:03.210 --> 01:03:07.150 I just looked at the strong motion records, and you see these two pulses. 01:03:07.150 --> 01:03:08.280 - Right. - On the stations. 01:03:08.280 --> 01:03:09.839 So that’s what sort of drove my thinking. 01:03:09.839 --> 01:03:14.259 - But you’re right. It is consistent with a lot of the back projection results. 01:03:14.259 --> 01:03:17.960 - Yeah. Because the back projections use the P wave, and so they’re more 01:03:17.960 --> 01:03:20.390 sensitive to the high frequencies. - But it’s interesting. 01:03:20.390 --> 01:03:24.089 When you look at the Tohoku back projection results, you see all these 01:03:24.089 --> 01:03:26.609 circles – because it shows you where the coherent energy 01:03:26.609 --> 01:03:32.000 is coming from at that period. And you see, like, five or six circles 01:03:32.000 --> 01:03:35.961 spread out over 50 kilometers. When you look over the strong motion 01:03:35.961 --> 01:03:40.560 records, you just see it’s this little – this narrow-duration pulse. 01:03:40.560 --> 01:03:44.619 It’s obviously coming from a small location on the rupture – 01:03:44.620 --> 01:03:48.620 much smaller than the area implied by the back projection. 01:03:48.620 --> 01:03:50.100 - Right. 01:03:50.100 --> 01:03:55.420 Question number two is, maybe the – the BC Hydro prediction is not – maybe 01:03:55.430 --> 01:04:00.130 your calculation is right, and BC Hydro is wrong on that 7.5-second long period. 01:04:00.130 --> 01:04:06.730 - Well, we also – well, we also compared it to a different GMPE 01:04:06.730 --> 01:04:14.160 which included Tohoku earthquake. And we still saw the same trend, so – 01:04:14.160 --> 01:04:18.270 and it is higher than what we’ve seen for the Tohoku earthquake at that distances. 01:04:18.270 --> 01:04:22.869 So it’s an interesting thing. And, you know, if you’re talking 01:04:22.869 --> 01:04:26.810 about Q differences, then you’re saying, well, is the scattering – do we have 01:04:26.810 --> 01:04:31.640 the scattering Q wrong, perhaps? You know, are we – we have not 01:04:31.640 --> 01:04:35.680 enough heterogeneity at these length scales than is in the real Earth? 01:04:35.680 --> 01:04:40.599 Or is it really a frictional – an error in the frictional Q or the intrinsic Q? 01:04:40.599 --> 01:04:45.490 So we’ve tried to look at Cascadia and see – Erin’s been looking at 01:04:45.490 --> 01:04:48.000 regional earthquakes, say from California, 01:04:48.000 --> 01:04:51.579 recorded up the coast of Pacific Northwest to see how 01:04:51.579 --> 01:04:57.109 our Q model explains that, to see if we’re maybe missing something, but – 01:04:57.109 --> 01:05:02.060 so we’re looking more at this Q at long periods in the Pacific Northwest. 01:05:04.020 --> 01:05:07.440 - Yeah. Wayne. - Terrific talk. I learned a lot, 01:05:07.440 --> 01:05:11.400 and I have a lot more to learn, I think, from all of the work that you showed. 01:05:11.410 --> 01:05:15.200 I have one suggestion – a quick one – and then a question. 01:05:15.200 --> 01:05:18.970 My suggestion is, on all of those ground motion prediction maps, 01:05:18.970 --> 01:05:23.770 you put Seattle on it. [chuckles] Because I had a hard time locating 01:05:23.770 --> 01:05:26.530 where the areas of real interest were. - Okay. 01:05:26.530 --> 01:05:29.920 - My question was, in your simulations, 01:05:29.920 --> 01:05:36.310 where you have the sub-events and the – sort of the smooth background slip, 01:05:36.310 --> 01:05:45.340 is the static slip at the end of the rupture, is it greater at your sub-event locations? 01:05:45.340 --> 01:05:47.000 - Yes. 01:05:47.000 --> 01:05:55.120 - Okay, then a suggestion. If – to test whether GPS could help locate those 01:05:55.120 --> 01:06:00.120 sub-events, if you just assume that the locking was the inverse … 01:06:00.120 --> 01:06:03.820 - Right. - … then – and do all your simulations, 01:06:03.820 --> 01:06:06.890 you might get a kind of a – well, an individual event and 01:06:06.890 --> 01:06:11.480 an average idea of how much particularly coastal stations might 01:06:11.480 --> 01:06:16.569 contribute because they’re cheap, and seafloor GPS is expensive. 01:06:16.569 --> 01:06:22.060 But just because it’s expensive, please don’t fail to emphasize it. 01:06:22.060 --> 01:06:28.200 - Right. And Erin Wirth, who is a new USGS hire, has been looking at this. 01:06:28.200 --> 01:06:35.241 And she gave a talk to a few people here back in June or May, and she’s been 01:06:35.241 --> 01:06:40.540 looking at the implications for coastal subsidence from these synthetics. 01:06:40.540 --> 01:06:47.549 And basically, you would find that, where these sub-events are, there 01:06:47.549 --> 01:06:49.700 should be more coastal subsidence … - Yeah. 01:06:49.700 --> 01:06:53.760 - … along strike, so a lot of people have been trying to model 01:06:53.760 --> 01:06:57.800 the coastal subsidence of the 1700 Cascadia earthquake. 01:06:57.800 --> 01:07:01.000 And people argue about whether there’s any good resolution 01:07:01.000 --> 01:07:05.289 of that with the diatom data, et cetera. 01:07:05.289 --> 01:07:07.390 But they claim there are these variations 01:07:07.390 --> 01:07:12.380 of the subsidence from 1700 along the coast. 01:07:12.380 --> 01:07:17.420 And people – I think Kelin Wang and his colleagues have been looking at that. 01:07:17.420 --> 01:07:21.880 And they’ve been modeling this as differences in the offshore slip – 01:07:21.880 --> 01:07:25.839 what we would call the background, so – but Erin has been finding that 01:07:25.840 --> 01:07:29.840 this could be signatures of a sub – of the sub-events. 01:07:29.840 --> 01:07:32.440 That’s why I say we need to look holistically at this stuff. 01:07:32.440 --> 01:07:36.700 Because if you ignore that part, it is significant along the coast, and there – 01:07:36.700 --> 01:07:39.160 I think there should be signal. You’re right. 01:07:39.160 --> 01:07:40.760 - Thanks. 01:07:42.480 --> 01:07:46.540 - Any more questions for Art? - Yeah, I have one more comment. 01:07:46.559 --> 01:07:52.240 In the recent earthquake in Mexico – I think it was roughly September 17 – 01:07:52.240 --> 01:07:56.370 I’ve forgotten – some of the buildings that collapsed were survivors 01:07:56.370 --> 01:08:03.829 from the 1985 subduction zone event. Because the recent earthquake in 01:08:03.829 --> 01:08:06.670 Mexico was an intra-plate event, as you know very well. 01:08:06.670 --> 01:08:10.280 It’s probably at a depth of maybe 60 kilometers. 01:08:10.280 --> 01:08:13.890 So I’d like to suggest that you’re actually working on the wrong problem. 01:08:13.890 --> 01:08:17.319 [laughter] That you should be modeling an inter-plate – 01:08:17.319 --> 01:08:21.770 a magnitude 8 inter-plate event that would have – be very enriched in 01:08:21.770 --> 01:08:25.770 high frequencies, and that would be the one that would cause the most damage. 01:08:25.770 --> 01:08:31.290 - Well, it’s also the question of what’s the likelihood of these things. 01:08:31.290 --> 01:08:34.960 You know, obviously, for – if you look at Seattle, the most likely 01:08:34.960 --> 01:08:39.780 thing to happen in the next 30 years is another Nisqually deep event. 01:08:39.780 --> 01:08:43.040 But we’re talking about magnitude 6-1/2 to 7.0. 01:08:43.049 --> 01:08:46.609 I think the National Seismic Hazard Maps do allow these inter-plate 01:08:46.609 --> 01:08:49.850 earthquakes to get up to – maybe up to magnitude 8.0 01:08:49.850 --> 01:08:52.390 with a very low weight to that. 01:08:52.390 --> 01:08:57.279 So I – you know, but what’s the probability of that happening? 01:08:57.279 --> 01:09:00.690 We haven’t – you know, we’ve seen a series of three 01:09:00.690 --> 01:09:06.799 deep earthquakes in Seattle – 1949, 1965, Nisqually 2001, 01:09:06.799 --> 01:09:11.569 which were between 6-1/2 to 7 on the deep part of the subduction zone. 01:09:11.569 --> 01:09:17.230 So we can’t rule out one of these type of earthquakes, you know, in the high 7s 01:09:17.230 --> 01:09:22.940 or so, but what’s the likelihood? I agree we should model it, sure. 01:09:25.820 --> 01:09:28.180 - Well, let’s thank Art again for coming to give us 01:09:28.180 --> 01:09:29.900 a great presentation. Thanks, Art. 01:09:29.900 --> 01:09:35.260 [ Applause ] 01:09:37.520 --> 01:09:43.620 [ Silence ]