WEBVTT 00:00:00.000 --> 00:00:03.000 [Here we go now.] 00:00:03.000 --> 00:00:09.000 Hello, everyone and now it is time for the thunderous Thunder Talks. 00:00:09.000 --> 00:00:17.000 And to bring us through the lightning and the thunder is our fearless leader Grace Parker. [Woo!] 00:00:17.000 --> 00:00:24.000 Hi, everyone. Thanks for being here today for the second Thunder Talk session of the workshop. 00:00:24.000 --> 00:00:37.000 Today we'll have nine Thunder Talks, and just like yesterday I encourage everyone to ask questions in the chat and for the speakers to keep an eye on the chat to answer any questions about their talks. 00:00:37.000 --> 00:00:46.000 And with that, I'll pass it to the first speaker, Mike Blanpied. 00:00:46.000 --> 00:00:59.000 Good afternoon. Good morning. My goal today is to point you to a newly published plan for the coordination of post-earthquake investigations conducted by the NEHERP agencies and scientific and engineering communities. 00:00:59.000 --> 00:01:13.000 As you know, post-earthquake scientific and engineering investigations can substantially further the achievement of many important goals, including improved societal preparedness, enhancing emergency response and recovery, 00:01:13.000 --> 00:01:23.000 better characterization of earthquake hazards safer and more resilient built environment, including improved land use practices that support community resilience. 00:01:23.000 --> 00:01:32.000 More cost-effective construction and retrofitting of structures and of course, advancing our understanding of earthquake processes and effects. 00:01:32.000 --> 00:01:42.000 In other words, achieving a more resilient nation demands that we understand what happens to the earth, to the built environment, and to society when an impactful earthquake occurs. 00:01:42.000 --> 00:01:59.000 Each earthquake provides a rare and precious opportunity to learn. Therefore, post-earthquake investigations must be well coordinated to maximize learning and to translate lessons quickly and effectively into practices that mitigate the effects of future earthquakes. 00:01:59.000 --> 00:02:10.000 The purpose of this new plan is to facilitate the activation and coordination of investigations that are undertaken by NEHRP agencies and the many partner agencies and organizations. 00:02:10.000 --> 00:02:22.000 NEHRP, the National Earthquake Hazards Reduction Program is I think you know the federal consortium of USGS, FEMA, NIST, and NSF that's charged with improving the nation's earthquake resilience. Within NEHRP 00:02:22.000 --> 00:02:33.000 the USGS, it's charged with leading post-earthquake investigations or at least coordinating them. 00:02:33.000 --> 00:02:42.000 So the new plan, Circular 1542, updates and supersedes a plan that you may be familiar with, Circular 1242 00:02:42.000 --> 00:02:49.000 that was published in 2003. Since then, Circular 1242 has guided NEHERP investigations. 00:02:49.000 --> 00:02:57.000 The purpose of that plan was to outline procedures for how the agencies and their partners should interact and coordinate when investigating earthquakes. 00:02:57.000 --> 00:03:11.000 But since the 25+ years or so since that plan was was published a lot of conditions have changed, including technology, practices and the roles of many of the partners. 00:03:11.000 --> 00:03:20.000 That plan, the 2003 plan focused mainly on the federal agencies. The new plan focuses on the entire community. 00:03:20.000 --> 00:03:26.000 And the new plan was developed by the Applied Technology Council under contract with USGS. 00:03:26.000 --> 00:03:39.000 And the development of the plan, although it's brief, the plan development involved input from a vast array of organizations, individuals, and so forth. 00:03:39.000 --> 00:04:00.000 So features of the new plan, a focus on the coordination doesn't tell us exactly what to do. It gives us guidance on how to coordinate. So coordinating between the many organizations involved in these investigations and how to deliver and archive the geoscience engineering and data 00:04:00.000 --> 00:04:11.000 and so forth. So some key features are it identifies areas of responsibility. It encourages that investigations be conducted in an ethical manner, 00:04:11.000 --> 00:04:20.000 and with a very flexible menu of activities that can be right-sized to whatever earthquake it is that demands a response. 00:04:20.000 --> 00:04:31.000 It also emphasizes the need for pre-event activities so that we're prepared to effectively coordinate and investigate when an earthquake opportunity presents itself. 00:04:31.000 --> 00:04:46.000 This is just a diagram showing the large community involved in working with the NEHERP agencies This is state agencies, private organizations, universities, foreign organizations, and so forth and so on. 00:04:46.000 --> 00:05:07.000 The plan outlines a three-phase approach. So in addition to the pre-event activities like planning and exercising and so forth, when an earthquake happens, the new NEHRP agencies will quickly convene to decide whether this earthquake warrants declaring an official NEHRP post-earthquake investigation. 00:05:07.000 --> 00:05:13.000 If so, then immediately we'd move into phase one in the hours and initial day or so. 00:05:13.000 --> 00:05:19.000 Which is establishing a technical clearinghouse. Maggie Ortiz-Milan talked about clearinghouses earlier in this meeting. 00:05:19.000 --> 00:05:32.000 And appointing an investigations committee, an earthquake investigations committee that will have representatives of the NEHRP agencies and perhaps of other organizations that are playing key roles. 00:05:32.000 --> 00:05:42.000 Then we'll move into phase two, which is the very familiar suite of activities that have to do with data collection in the immediate reconnaissance. 00:05:42.000 --> 00:06:12.000 This is both scientific and engineering investigations as well as social science data and so forth. And then in the longer term, we move into a suite of various phase three activities that have to do with longer term studies, archiving, dissemination of data. Perhaps that would involve workshops and publications and so forth. Thanks for your time. 00:06:16.000 --> 00:06:23.000 I am Michael Barall, and I will describe the 2024 SCEC/USGS Dynamic Earthquake Rupture Simulation. 00:06:23.000 --> 00:06:35.000 Mock exercise. The SCEC/USGS Dynamic Rupture Code Verification Group is an international collaboration of dynamic rupture modelers that has been ongoing for 20 years. 00:06:35.000 --> 00:06:40.000 Our goal is to help modelers test their computer codes to verify that they work correctly. 00:06:40.000 --> 00:06:51.000 To do this, we publish carefully designed benchmark exercises. Modelers run the exercises and submit their results to the group where they are compared to results submitted by other modelers. 00:06:51.000 --> 00:06:56.000 If the results agree, then there is assurance that the codes are working correctly. 00:06:56.000 --> 00:07:06.000 Over the years, we have published benchmarks that cover a wide range of fault geometries, friction laws, rock properties, and initial stresses as shown. 00:07:06.000 --> 00:07:14.000 Last year, we published two new benchmarks which use a thrust fault dipping at a shallow angle of 15 degrees, as shown in the diagram. 00:07:14.000 --> 00:07:25.000 We expected these benchmarks to be challenging because of the narrow 15 degree angle and because of the expected strong interactions between the propagating rupture and the Earth's surface. 00:07:25.000 --> 00:07:34.000 We did these benchmarks in collaboration with crescent because this geometry is of interest for subduction zone models and tsunami models. 00:07:34.000 --> 00:07:39.000 When we publish benchmarks, we have to specify all the details of geometry, friction, and so on 00:07:39.000 --> 00:07:44.000 as indicated on this slide, so that all modelers are doing the same problem. 00:07:44.000 --> 00:07:48.000 The two benchmarks differ only in the amount of frictional cohesion. 00:07:48.000 --> 00:07:53.000 For benchmark TPV36, the rupture is able to reach the Earth's surface. 00:07:53.000 --> 00:08:02.000 For benchmark, TPV37, the frictional cohesion makes the rupture stop spontaneously before it reaches the Earth's surface. 00:08:02.000 --> 00:08:09.000 For the first time, we asked modelers to compute the final deformation of the Earth's surface at the end of the simulation. 00:08:09.000 --> 00:08:17.000 That's important for tsunami models because the Earth's surface is also the seafloor and the deformation of the seafloor triggers the tsunami. 00:08:17.000 --> 00:08:23.000 This slide shows the vertical component of surface deformation as computed by seven modelers. 00:08:23.000 --> 00:08:29.000 Each image shows the Earth's surface in map view with red being uplift and blue being subsidence. 00:08:29.000 --> 00:08:37.000 These benchmarks produce about one meter of uplift. All seven modelers computed similar patterns of uplift and subsidence. 00:08:37.000 --> 00:08:43.000 demonstrating good agreement between the codes. 00:08:43.000 --> 00:08:48.000 Another way we compare results is to look at rupture front contours. 00:08:48.000 --> 00:08:51.000 The figure shows the fault surface with the fault trace at the top. 00:08:51.000 --> 00:08:57.000 The curves show the progression of the rupture front across the fault surface in intervals of one half second. 00:08:57.000 --> 00:09:04.000 The figure combines results from eight modelers, with each modeler's rupture front contours plotted in a different color. 00:09:04.000 --> 00:09:07.000 You can see that all the different colored curves are very similar 00:09:07.000 --> 00:09:21.000 demonstrating a good agreement between all eight modelers. The numerical table in the upper right is a metric that measures quantitatively how well each pair of modelers agree with each other. 00:09:21.000 --> 00:09:32.000 We also compared the simulated ground motions. This figure shows the vertical component of ground motion at a single point on the Earth's surface near the location of maximum uplift. 00:09:32.000 --> 00:09:36.000 The top plot shows vertical velocity as a function of time. 00:09:36.000 --> 00:09:39.000 And the bottom plot shows vertical displacement as a function of time. 00:09:39.000 --> 00:09:46.000 Again, the figure combines results from eight modelers with each modeler's waveform plotted in a different color. 00:09:46.000 --> 00:09:52.000 Six modelers produce waveforms that agree well with each other throughout the simulation. 00:09:52.000 --> 00:09:56.000 The other two produce waveforms that deviate from the rest in the second half of the simulation 00:09:56.000 --> 00:10:04.000 showing where there may be room for improvement. Both of those modelers have indeed made improvements since this comparison was done. 00:10:04.000 --> 00:10:14.000 Demonstrating how our benchmarks can provide valuable feedback to modelers and lead to improvements in dynamic rupture modeling codes and techniques. 00:10:14.000 --> 00:10:19.000 If you visit our website, you can find many more comparisons. 00:10:19.000 --> 00:10:23.000 You can also find information about our group and our complete set of benchmarks. 00:10:23.000 --> 00:10:31.000 In summary, we published two new dynamic rupture benchmarks for thrust fall dipping at a shallow 15 degree angle. 00:10:31.000 --> 00:10:38.000 Eight modelers ran the benchmarks. We compared surface deformation, rupture contours, slip histories, 00:10:38.000 --> 00:10:57.000 adnd ground motions. Comparison shows that most results were in good agreement But some disagreements were noted. Thank you. 00:10:57.000 --> 00:11:11.000 Hello, everyone. The San Francisco Bay Area has a 73% probability of producing one or more large, what we call large magnitude greater than or equal to M6-M7 earthquakes in the next 30 years. 00:11:11.000 --> 00:11:16.000 So we know that many of the faults have had earthquakes in the past, mostly from geologic studies. 00:11:16.000 --> 00:11:22.000 But the faults with the highest probabilities are the Rodgers Creek, Hayward, and Calaveras faults. 00:11:22.000 --> 00:11:29.000 So we know that at least the Hayward had a large earthquake in 1868, but we don't know what it looked like, 00:11:29.000 --> 00:11:42.000 and we have paleoseismic evidence for a large earthquake on the Rodgers Creek, and once again, we don't know what it looked like. The Calaveras has had slightly smaller earthquakes but we don't know what a bigger one will look like. 00:11:42.000 --> 00:11:51.000 So this sends us over to dynamic or spontaneous earthquake rupture simulations, which is what Michael was just talking about in the last talk. 00:11:51.000 --> 00:11:57.000 So for these, we need to know the fault geometry, rock properties, initial stresses, and friction. 00:11:57.000 --> 00:12:05.000 We put all these into our well-tested code, hopefully tested in our code comparison exercise. For this, we use Michael Barall's code fault MOD, 00:12:05.000 --> 00:12:12.000 and out of this we get earthquake source scenarios, and then also ground motions. 00:12:12.000 --> 00:12:17.000 For this particular study, the fault geometry we take the fault traces. 00:12:17.000 --> 00:12:26.000 from the San Francisco Bay Area geologic model, and we focus here on the Rodgers Creek, Hayward, and Calaveras faults. 00:12:26.000 --> 00:12:32.000 Here's a side view. We are using 3D fault geometry. 00:12:32.000 --> 00:12:46.000 For the rock properties, we also take this from the 3D San Francisco Bay Area geologic model with cross sections, map view cross sections on the left and then should I view the long strike sections on the right. 00:12:46.000 --> 00:12:52.000 We take these rock units and then we convert them into VPVS and density using the Brocher relations. 00:12:52.000 --> 00:13:06.000 We also need to know friction for the study, we assume slip weakening friction, sometimes incorporating cohesion in the upper key kilometers, sometimes not, but we need to know our coefficients of friction. We're lucky in the San Francisco Bay Area 00:13:06.000 --> 00:13:11.000 that there have been lab studies in our own USGS rock mechanics lab. 00:13:11.000 --> 00:13:23.000 On samples from the Hayward fault surface of the Hayward fault, and then also we relate some of that information to studies done in the San Andreas fault Observatory depth on the San Andreas in the creeping section 00:13:23.000 --> 00:13:29.000 and this is work done by Carolyn Morrow, Diane Moore, and Dave Lockner. 00:13:29.000 --> 00:13:42.000 For initial stresses, we use geodetically inferred and microseismically inferred slip-rate patterns for the Central Calaveras Hayward and Rogers Creek faults and then also northern Calaveras are from studies by Estelle Chaussard, 00:13:42.000 --> 00:13:47.000 And then also Gareth Funning, who was a co-author on this earthquake source studies work. 00:13:47.000 --> 00:14:03.000 Where the faults are slipping faster, have faster slip rates we assume that faults are slightly farther from failure, so lower initial shear stress and where they are more locked. We assume that we are closer to failure. 00:14:03.000 --> 00:14:11.000 So now we have our fault geometry, rock properties, initial stresses and friction. We have our excellent fault MOD code by Michael Barall. 00:14:11.000 --> 00:14:15.000 And then let's go look at what some of our earthquake sources look like. 00:14:15.000 --> 00:14:21.000 And this is from our JGR 2021 paper, which has lots more information too. 00:14:21.000 --> 00:14:28.000 So here are some two samples. We have in the top, very top, we have initiation on the Rodgers Creek fault. 00:14:28.000 --> 00:14:40.000 And then we see that it gets all the way, almost all the way down to the central Calaveras. In the next row, the second row, we have Hayward nucleation, and then we see it gets all the way to the end of our Rodgers Creek sections. 00:14:40.000 --> 00:14:47.000 But makes it most of the way, but not all the way down to the Calaveras, the central Calaveras. 00:14:47.000 --> 00:14:57.000 And that's where we did not include cohesion. When we did include cohesion, we got slightly smaller earthquakes. Rodgers Creek making it not quite as far, but still quite a big earthquake, magnitude 7.9, 00:14:57.000 --> 00:15:05.000 and Hayward also doesn't get quite as far, but still quite sizable earthquake of magnitude 6.9. 00:15:05.000 --> 00:15:09.000 So then we looked at the ground motions in that 2021 paper and we saw they're really high. 00:15:09.000 --> 00:15:14.000 But then we realized, oh, that's just really high frequency stuff that we can't actually resolve. 00:15:14.000 --> 00:15:28.000 So for our new study. We then filtered that data, but using the same source model. So once again, we go back to our magnitude 7.19 and magnitude 6.94 for Rodgers Creek earthquake and a Hayward earthquake. 00:15:28.000 --> 00:15:41.000 And now here we're looking at how our ground motion now reasonably filtered at one second how it matches or does not match a ground motion model, and this is from Barall et al., 2014. 00:15:41.000 --> 00:15:47.000 Should we get an okay match for Rodgers Creek? If we do include cohesion. 00:15:47.000 --> 00:15:54.000 This is long period ground motion. Hayward is a little bit low if we include cohesion looking at PGV and 5%. 00:15:54.000 --> 00:16:00.000 Damp three second PSA when we don't include cohesion. 00:16:00.000 --> 00:16:05.000 Then we get slightly bigger earthquakes, then we get slightly too high, Rodgers Creek, 00:16:05.000 --> 00:16:09.000 but we get a good match with Hayward. 00:16:09.000 --> 00:16:15.000 So this is showing that sometimes we should use cohesion, sometimes not, and we can think about the physical reasons for doing that. 00:16:15.000 --> 00:16:20.000 But we can make a really good match to the long-period ground motions. 00:16:20.000 --> 00:16:28.000 In summary, if we have adequate geology and geophysics data, we can use dynamic rupture simulations to create scenarios and produce ground motions that look okay. 00:16:28.000 --> 00:16:46.000 For here, we focused on creeping faults into San Francisco Bay Area, but we could also use the same approach in other locations. And thank you. 00:16:46.000 --> 00:17:03.000 All right. So we're going to keep talking about creeping faults, but in this case rather than focusing on ground motion, I'm using dynamic simulations to try and get at passing probabilities for rupture going through a creeping patch on a partially creeping fault. 00:17:03.000 --> 00:17:14.000 And so why are we doing this? I mean, because a lot of faults in Northern California and elsewhere have patches of a seismic creep. 00:17:14.000 --> 00:17:43.000 Within otherwise locked faults or sometimes vice versa and a seismic creep in this case can act as a barrier both because it is frictionally resistant. It has a rate strengthening property rather than a weight weakening property, but also that interseismic creep can release a lot of the shear stress that would otherwise be accumulating on a locked fault. So there's both a frictional resistance issue and a less shear stress issue. 00:17:43.000 --> 00:17:51.000 And in this case, we know that some historic earthquakes have stopped at creeping sections, like in 1906, stopped at the creeping section at its southern end; 00:17:51.000 --> 00:18:03.000 1857 at its northern end. We also know of some historic events or recent events like Parkfield or possibly 1868 that were on lock patches completely confined by creep. 00:18:03.000 --> 00:18:14.000 But there are really very few or like no instrumental examples of this, but from a hazard standpoint, knowing what could possibly happen 00:18:14.000 --> 00:18:30.000 with these creeping patches affecting earthquake size is important. And so we want to get at passing probabilities to hopefully have some sort of physics-based input for probabilistic hazard calculations, or maybe for deterministic models as well. 00:18:30.000 --> 00:18:51.000 I'm using dynamic modeling for this because it lets me test a wide range of parameters, just completely Just go all over parameter space using, in this case, a very simple geometry where I have a planar right-lateral strike-slip fault with a creeping patch in the middle that has a variable length, 00:18:51.000 --> 00:19:21.000 a variable amount of frictional resistance and a variable amount of shear stress release. And I also tested the whole thing at different areas in stress space, like high or low stress. And the idea is just to do a ton of them because dynamic models are fast enough now that I can do that. You can do a really complex model or you can do a gazillion simple ones. So far, I've actually done more than 1474. That's just what's on this figure. 00:19:21.000 --> 00:19:50.000 And so on this figure, every single individual symbol is a single simulation. I'll zoom in on some of these in a second. But the important part is red means stop and blue means go in terms of showing patterns of rupture behaviors. So the ones at the top are lower stress state, intermediate stress state, and high stress state; sorry, I got it reversed; high stress at the top, low stress at the bottom. But anyway, you can see there's some patterns going on here. And I'm going to highlight a few. 00:19:50.000 --> 00:20:20.000 So I found that the amount of shear stress release actually matters a whole lot more than the degree of frictional contrast. The amount of frictional contrast affects the details of the slip distribution, but fundamentally for all of these, we have the percentage of shear stress rate release on the y-axis and the length of the creeping patch on the x-axis and each individual plot is the amount of frictional resistance. And these all look very, very similar to each other for this stress state, really showing that the frictional contrast doesn't matter too much, 00:20:22.000 --> 00:20:39.000 but the stress contrast does. Meanwhile, the rupture velocity actually matters quite a lot. So all of these are in the same overall stress state but have a sort of very sub-shear or a very super-shear or sort of moderate 00:20:39.000 --> 00:20:53.000 rupture velocity and in this case, the slower the rupture is going, the more likely, or rather the smaller of a creeping patch is likely to prevent the rupture from reaching the other end of the fault. 00:20:53.000 --> 00:21:01.000 So the very subshear ruptures can maybe go through a one kilometer patch the very super sheer ones so far have just blasted through everything. 00:21:01.000 --> 00:21:15.000 To compare that to the effects of stress amplitude, maybe if the slide advances. 00:21:15.000 --> 00:21:29.000 [I think I accidentally put in the wrong slide.] This is basically the overall stress amplitude and oh yeah, these are all different rupture velocities with the same stress different stress amplitudes with the same rupture velocity. 00:21:29.000 --> 00:21:42.000 Whereas these are different uh rupture velocities with the same stress amplitude. So in this case, there is a little bit more of a shift towards more things rupturing all the way through, more conditions rupturing all the way through. 00:21:42.000 --> 00:21:48.000 With the high stress case, but overall, these are still much more like each other. 00:21:48.000 --> 00:21:55.000 Then these ones where it's the same stress taste but very different rupture velocities. 00:21:55.000 --> 00:22:15.000 So what this says is that very much that the creeping patches can stop ruptures and be accounted for in hazard forecasts, but really the rupture velocity and the stress release are the really key things going on here are the main parameters driving this. Meanwhile, the frictional resistance is, I mean, it's a thing, 00:22:15.000 --> 00:22:33.000 but less than you'd think. So that kind of goes that, hey, slip weakening actually still probably totally fine, even though we were using rate state here. But we're going to keep going with this and hopefully expand this with some more lengths, some different nucleation locations, and eventually actually come up with some 00:22:33.000 --> 00:22:48.000 equations, some probability equations and rules. So thank you. 00:22:48.000 --> 00:23:09.000 Hi, everybody. I'm just going to give a quick overview of the fun earthquake that we had in early December. It was a magnitude 5.7 earthquake in a little place called Parker Butte, which is in the northern Mason Valley, just for reference, we're in the Walker Lane Zone here. You might think of this as the San Andreas's greedy little sister that's trying to steal some of the slip, but 00:23:09.000 --> 00:23:20.000 after I actually moved to Reno about six months ago, I found out from reading that the Walker Lane is as old as the San Andreas. There's evidence of dextral offset here going back to the early Miocene. 00:23:20.000 --> 00:23:32.000 So the Walker Lane is characterized by a bunch of different fault orientations, we have these like east-northeast striking, sinistral faults, we have northwest striking. 00:23:32.000 --> 00:23:37.000 Textural faults, and then we have paranormal faults, mostly paranormal faults 00:23:37.000 --> 00:23:53.000 that have either a north-south strike or the edge of the Walker Lane is defined by the transition to a northeast strike, which you can see kind of in the area of the 1932 Cedar Mountain earthquake. So this is the location of our mainshock. 00:23:53.000 --> 00:24:04.000 It's closely associated with some of these known kind of tectonic lineaments, in this case, the Wassuk lineaments where there was a 1933 earthquake of probably similar magnitude. 00:24:04.000 --> 00:24:13.000 We had 8,400, Did you feel it for the USGS website for this earthquake going all the way out to the coast. 00:24:13.000 --> 00:24:29.000 Okay. So zooming in, this is the actual Parker Butte. There's a monument there that's why we named it that. The town of Yarrington is down to the southwest this area, especially around the river, is part of the Walker River Paiute 00:24:29.000 --> 00:24:46.000 lands and their primary community is down the river to the southeast here at [indiscernible]. So the mainshock occurred here a little bit south of the known mapped faults. We see aftershock cluster here that basically is everything from December 9th on the mainshock up to December 31st 00:24:46.000 --> 00:25:08.000 and the aftershocks align pretty closely with the sinistral focal plain solution. So we consider that to be likely the mainshock rupture, the little white triangles are the stations that we rapidly deployed after the earthquake. There's a fifth one out there now. 00:25:08.000 --> 00:25:13.000 Okay, so we went out the morning after the earthquake. The earthquake was at 3.08 p.m. 00:25:13.000 --> 00:25:20.000 And during that time, that was about the highest temperature of the day, which is eight degrees Celsius. There was a rapid freeze going into sunset. 00:25:20.000 --> 00:25:36.000 So we had a lot of really fun shaking evidence that we observed in the surface. We explored the area where the aftershocks were starting to come in along that trend. So this is just a couple of examples. We have sand blow here. There was quite a bit of liquefaction evidence along the Walker River and on the floodplain. 00:25:36.000 --> 00:25:43.000 And then on the right here, we have, this is actually an irrigation channel that drains into the Walker River. 00:25:43.000 --> 00:25:51.000 And in the channels, we saw cracks. In this case, they're aligned with the banks, but in the channels that were not parallel. 00:25:51.000 --> 00:26:05.000 we had the set, two sets and they're either aligned with the banks or they were aligned with what we think is the rupture plane. So I think we have some very transient shaking evidence that was preserved in the overnight freeze in 00:26:05.000 --> 00:26:11.000 frazzle ice, which is kind of a fun thing, like little chips of ice that form when the air is colder than the water and the water is moving fast. 00:26:11.000 --> 00:26:18.000 So it's quite a delicate structure that was able to preserve this deformation. 00:26:18.000 --> 00:26:31.000 [Oops, I might have clicked twice. What's going to happen? Yeah, okay.] So this is some of our field observations. We had evidence of shaking. We had minor sedimentary collapses. We had sand blows. 00:26:31.000 --> 00:26:39.000 About eight days after the mainshock was the first time that the Sentinel satellite passed over the mainshock. 00:26:39.000 --> 00:26:45.000 And so we have a trend where we observe shaking that extends a long strike from the mainshock. 00:26:45.000 --> 00:26:56.000 But two lineaments were noted in the inSAR after that second sentinel pass. So one of them is directly south and parallel in the mainshock. 00:26:56.000 --> 00:27:03.000 And the other one is perpendicular, basically parallel to the other focal plane of the moment tensor solution. 00:27:03.000 --> 00:27:10.000 So a couple of us went back out immediately after that to go check those lineaments. 00:27:10.000 --> 00:27:16.000 On the presumed sinister one parallel to the mainshock, there were very small opening mode crack features. 00:27:16.000 --> 00:27:29.000 But sheer mode cracks were found on this other lineament where, if you'll notice in the aftershocks there are on the map on the right, we don't really have any aftershocks there and we had no indication of co-seismic slip 00:27:29.000 --> 00:27:37.000 along that structure, but in fact, we have field evidence for displacement along that structure. 00:27:37.000 --> 00:27:52.000 So these are some of the cracks. These photographs are taken by Rich Khaler he is going to be leading the paper. We'll submit in a couple of weeks, that's going to summarize all of this data and including some relocation of the aftershocks to better refine 00:27:52.000 --> 00:28:00.000 the mainshock plane and the aftershock distribution. So basically we have the mainshock and a previously unmapped sinistral fault. 00:28:00.000 --> 00:28:07.000 Close to known faults and close to historic events of similar magnitude, we found shaking evidence in sediment and ice. 00:28:07.000 --> 00:28:19.000 We found evidence for deformation along this fault that was buried in some of the Lahontan Lake beds, so this predates Lahontan but wasn't mapped at surface because of that recent resurfacing. 00:28:19.000 --> 00:28:34.000 The surface rupture was found not on the mainshock, but on this other orthogonal plane and this is kind of a thing that the Walker Lane does, like two ruptures at 90 degrees to each other, Ridgecrest being the best known, Monte Cristo also had some features like this. 00:28:34.000 --> 00:28:46.000 We are not sure at this point if those surface deformation formed concurrent with the mainshock, if it was a triggered slip, it does not appear to have been seismic after the earthquake, so it could also be a triggered 00:28:46.000 --> 00:28:59.000 slip. So that's kind of fun to think about. Thank you very much. 00:28:59.000 --> 00:29:15.000 All right, cool. I'm Travis. So today I'm going to be talking about how some fault systems can be to first order probably determined in a semi-automated way from earthquake patterns and how we might assess the quality of those results. 00:29:15.000 --> 00:29:18.000 This talk highlights the core method that I developed during the first year in my Mendenhall. 00:29:18.000 --> 00:29:26.000 The workflow mimics how a human expert might approach the problem of mapping faults, but hopefully in a more scalable and less subjective way. 00:29:26.000 --> 00:29:31.000 So we begin by clustering seismicity according to density, which is shown by the colors, 00:29:31.000 --> 00:29:35.000 and then we calculate the orientations and merge those clusters that are coplanar. 00:29:35.000 --> 00:29:46.000 And then we go fitting a flexible regression method that avoids relying on any strict functional form assumptions to fit the fault surfaces. 00:29:46.000 --> 00:29:55.000 Here's just a few more details. We're using HDB Scan to identify the clusters. This is pretty cool because we don't need to predefine the number of clusters a priority. 00:29:55.000 --> 00:30:07.000 Once we have those clusters, then we do the merging by looking at the nearest neighbor orientations and the direction vectors and do a calculation with a series of dot products, 00:30:07.000 --> 00:30:16.000 and finally, the 3D surfaces fit. And it's actually guided by seismicity that's outside of the fault core. 00:30:16.000 --> 00:30:26.000 So this is an example, one example from San Juan Bautista, and it's pretty nice because we're capturing the salient features of the San Andreas-Calaveras Fault Junction, 00:30:26.000 --> 00:30:32.000 and we're able to identify all of the latest Quaternary active fault structures in the Q-fault database. 00:30:32.000 --> 00:30:38.000 Additionally, you'll notice that this approach maps nonlinear geometries of the San Andreas fault. 00:30:38.000 --> 00:30:44.000 On the right-side of the panel, you'll probably see that the dip shallows on the San Andreas fault as you move to the north- 00:30:44.000 --> 00:31:02.000 west, which has been documented in other studies. Okay, so let's actually compare our results to those that have been previously published by the SCEC CFM model that Scott talked about a few days ago. What stands out when looking at these is that 00:31:02.000 --> 00:31:10.000 our results actually closely match those that are generated by experts, even though we're only relying on the seismicity patterns. 00:31:10.000 --> 00:31:18.000 These figures show the distribution of distances between our results and the San Andreas fault on the bottom and the Calaveras fault on the top. 00:31:18.000 --> 00:31:23.000 And you'll find that in both cases, the models agree to within about a kilometer, which is impressive 00:31:23.000 --> 00:31:31.000 considering that we don't actually have any shallow constraints, because we don't really have any shallow-size missing to guide the faults. 00:31:31.000 --> 00:31:44.000 Okay, so this seems promising for well-documented faults. That's great. But can we actually come up with a way of assessing the reliability in cases that are a little bit less certain than things like the San Andreas fault to the Calaveras fault? 00:31:44.000 --> 00:31:56.000 So one way that we might think about evaluating this method is by comparing the earthquake density distribution around the modeled f ault to observe patterns like previously shown. 00:31:56.000 --> 00:32:07.000 So observations show that there's a really a nonlinear decrease in the earthquake density with increasing lateral distance from the fault. 00:32:07.000 --> 00:32:13.000 This decrease in the earthquake density is often described using power law, which is pretty common in the earthquake world. 00:32:13.000 --> 00:32:22.000 But in this formulation here, they've introduced this parameter D, which represents the distance away from the fault where saturation of these things occurs. 00:32:22.000 --> 00:32:29.000 So many of you have probably seen something like this before and thought a little bit about this, but we've taken it maybe a step further 00:32:29.000 --> 00:32:37.000 and basically shown that this D parameter can help us understand how much unmodeled complexity exists in our results. 00:32:37.000 --> 00:32:43.000 So we show this by generating some synthetic tests where we're actually prescribing a power law. 00:32:43.000 --> 00:32:57.000 And then testing the outcomes of errors on the fit. So like on the top panel, we're miscalculating the dip by a couple of degrees. And in the bottom panel, we're just like adding some sine wave roughness to otherwise flat faults. 00:32:57.000 --> 00:33:04.000 And in both cases, as you kind of like increase the misfit this pushes the deed to larger and larger values. 00:33:04.000 --> 00:33:20.000 So this suggests to us that the D and other properties of the seismicity can serve as a useful metric to help assess the fit quality and whether it's actually appropriate to represent some volume of earthquakes at the single fault. 00:33:20.000 --> 00:33:28.000 So here I'm showing some parameter fits in map view. So these I've already ran through the fitting exercises. 00:33:28.000 --> 00:33:39.000 And the symbol fill colors indicate the parameter value. Where the yellower colors indicate values that fall outside the range of previously documented results. 00:33:39.000 --> 00:33:45.000 And the outline color on the symbol is just representing the cluster family 00:33:45.000 --> 00:33:53.000 t hat contributed to generating default, and so the yellow results have potentially unrealistic values, and I pointed these out with red arrows. 00:33:53.000 --> 00:34:01.000 So these parameters along with others that I'm not talking about today, we then combine together into a general scoring system. 00:34:01.000 --> 00:34:09.000 And so this score can help us describe the quality, reliability, and representability of the faults derived from the seismicity patterns. 00:34:09.000 --> 00:34:25.000 So this is a work in progress and I'll be interested in hearing some feedback if there's any. Thanks. 00:34:25.000 --> 00:34:39.000 Hi, this is Claire Yoon from USGS for the last time. And now I will talk about fault geometry under Lake Almanor in Northern California pictured here from a 2023 nodal seismometer deployment. 00:34:39.000 --> 00:34:45.000 This is work in progress with a large USGS team and is partially funded by PG&E. 00:34:45.000 --> 00:34:53.000 And if you saw my SCEC poster last year, you've already seen everything I will present today. 00:34:53.000 --> 00:35:01.000 So we're looking at this part of Northeastern California where the fault structure and earthquake hazards are poorly understood. 00:35:01.000 --> 00:35:09.000 And the seismic instrumentation is sparse, but it is a seismically active area. 00:35:09.000 --> 00:35:15.000 In May 2023, there was a magnitude 5.5 earthquake, soon followed by a magnitude 5.2. 00:35:15.000 --> 00:35:36.000 A few hours later and there were many aftershocks near Lake Almanor, which is shown colored in blue. And then 10 years before that in 2013, There was a magnitude 5.7 earthquake just to the southeast and this is concerning because PG&E has six dams and other 00:35:36.000 --> 00:35:58.000 nearby facilities shown by the diamonds within 15 kilometers of these earthquakes. And so we'd really like to be able to resolve the fault geometry and dimensions at depth in order to better constrain site-specific earthquake hazard. So our approach here is to use a nodal seismic data set that USGS deployed starting the day after 00:35:58.000 --> 00:36:10.000 the 2023 mainshock and was recording for two months. They're shown by these blue triangles with numbers on top. So really good azimuth and distance coverage 00:36:10.000 --> 00:36:17.000 compared to the permanent, nearest permanent stations, which are over 20 kilometers away. 00:36:17.000 --> 00:36:22.000 And so we're going to use this nodal seismic data to improve earthquake locations, 00:36:22.000 --> 00:36:42.000 and focal mechanisms. So our general approach is to start with already known ComCat earthquake events, their origins, and also the face pics that were manually reviewed, and we're basically going to add in the nodal data. So we're adding in new P and S. 00:36:42.000 --> 00:36:56.000 Space packs from the nodal data, which were made automatically using the face, not deep learning picker seemed to work pretty well and then locating them with a hypoinverse and 1D velocity model used by the seismic network. 00:36:56.000 --> 00:37:20.000 And so the effect was to move the effect of adding the nodal data was to move the earthquake locations a couple kilometers to the east and 2 to 5 kilometers deeper. We also added my first motion, clarities and STP amplitude ratios, and then use SKHASH to compute focal mechanisms. 00:37:20.000 --> 00:37:31.000 And the effect of that was to reduce the uncertainty and the focal mechanism nodal plane by about 25 degrees, which is good. 00:37:31.000 --> 00:37:43.000 And then the last step was to do a precise relative double difference style location, including the permanent and the nodal data event waveforms. 00:37:43.000 --> 00:37:48.000 So here are the results that we have so far on the left. 00:37:48.000 --> 00:37:58.000 Top left in map view, I'm showing the earthquakes colored by depth and you can see the mechanisms for the magnitude 5.0 and above earthquakes. 00:37:58.000 --> 00:38:16.000 Their normal fault mechanisms, but for the smaller aftershocks, we do see some strike-slip mechanisms as well. On the top right in map view and also in the cross-sections below, the 2023 earthquake sequence is colored by time 00:38:16.000 --> 00:38:34.000 and the green to blue colors and the 2013 sequence is shown in gray and the map view and then white in the cross sections and generally what we observe is that the 2013 and the 2023 sequences, so the colors and the white or the gray 00:38:34.000 --> 00:38:50.000 they're located near each other, but they don't have too much overlap. And it looks like they might have happened on the same larger underlying fault structure that might be big enough for a larger magnitude 6.0 or above earthquake. So we're looking into that. 00:38:50.000 --> 00:39:04.000 Kind of looking at the cross-sections in more detail. We see a steeply dipping fault to the northeast, but as we go further to the southeast, the dip gets steeper, almost to near vertical. 00:39:04.000 --> 00:39:13.000 And we're investigating if this is a known fault, maybe the Skinner Flat fault, or if it's some other fault that's unknown. So that's still ongoing. 00:39:13.000 --> 00:39:37.000 There's also an intriguing question of was there some stress transfer happening over the past 10 years, it looks like the 2013 earthquake happened at the southeastern end of Lake Almanor. And then did that somehow load the area to the northwest where eventually the magnitude 5 earthquakes happened in 2023. 00:39:37.000 --> 00:39:45.000 And then its aftershocks in turn happen further northwest. So that's also something that we're currently looking into. 00:39:45.000 --> 00:40:04.000 So this is ongoing work. So there's a lot of things that we could do in the future. I think the first priority is really take advantage of the nodal data to develop and test new velocity models for the Lake Almanor area, which should help us improve our event locations and mechanisms in return. 00:40:04.000 --> 00:40:11.000 Since the nodal data is continuous, we're also looking into developing an enhanced earthquake catalog 00:40:11.000 --> 00:40:18.000 that's complete to lower magnitudes using a lot of the deep learning automated picker methods. 00:40:18.000 --> 00:40:37.000 And then ideally what we really want to do is relocate the seismicity in your Lake Almanor going back over a decade since before 2013 and do this with a consistent analysis and kind of the best velocity model that we come up with in order to really understand the underlying fault structure that's nearby and 00:40:37.000 --> 00:41:00.000 kind of stress interactions between earthquake sequences and hopefully this will all feed into site-specific earthquake hazard. So that's it. Thank you very much. Contact me at my email for any questions. 00:41:00.000 --> 00:41:13.000 Hi, my name is Janet Watt, and today I want to give you a preview of some brand new shoreline crossing aeromagnetic data that I am very hopeful will reveal some of Northern California's faulting secrets. 00:41:13.000 --> 00:41:19.000 I'd like to start by giving a shout out to my partners in crime here, Vicki Langenheim, and Lydia Staisch. 00:41:19.000 --> 00:41:29.000 And I'd also like to thank, special thanks to DJ Bandrowski of the Yurok Tribe and both Russell Taylor Grimes and Jeff Phelps of the USGS. 00:41:29.000 --> 00:41:41.000 Next slide, please. So I'm happy to report that with funding from both PG&E and the USGS Coastal and Marine Hazards and Resources Program 00:41:41.000 --> 00:41:51.000 we were successful in partnering with the local Yurok Tribe and Excalibur Multiphysics to collect fixed wing aeromagnetic data across the shoreline in Northern California 00:41:51.000 --> 00:41:58.000 for the purpose of characterizing crush or fault geometry, slip mode, slip rate, and offset history. 00:41:58.000 --> 00:42:10.000 The Surrey area is shown in the map on the left by the click black polygon. It extends from the Eel River in the south to just north of Crescent City and includes eight tribal land areas. 00:42:10.000 --> 00:42:17.000 The survey was phoned perpendicular to shoreline crossing structures, including the Table Bluff, Little Salmon, 00:42:17.000 --> 00:42:21.000 Mad River, Bald Mountain, Big Lagoon, and the Lost Man Fault Zones. 00:42:21.000 --> 00:42:28.000 The survey was completed in September of 2024, and we just received the final process data. 00:42:28.000 --> 00:42:40.000 So this is hot off the press. For those of you not familiar with aeromagnetics, the data are collected using a fixed wing airplane like the one shown in the center of the slide. 00:42:40.000 --> 00:42:46.000 With a sensor in the tail of the airplane that passively measures the local magnetic field. 00:42:46.000 --> 00:42:57.000 And we can detect variations in the magnetic properties of the rocks below the airplane and when faults offset rocks with different amounts of, say, magnetite, we can see changes or anomalies in the magnetic field. 00:42:57.000 --> 00:43:05.000 And also note that this survey complements recent high resolution seismic imaging and coring efforts on the shelf offshore. 00:43:05.000 --> 00:43:10.000 Shown by the red lines and the green circles in the map on the left. 00:43:10.000 --> 00:43:31.000 Next slide, please. So the survey outlined in yellow on the map on the left here fills an important gap in an otherwise high resolution aeromagnetic data coverage in Northern California, covering the complex yet important Mendocino Triple Junction area that we've heard a lot about at this meeting. 00:43:31.000 --> 00:43:39.000 The image on the center shows a preliminary grid of the new shoreline crossing data with existing magnetic data coverage. 00:43:39.000 --> 00:43:46.000 And the most eye-catching anomalies here reflect the highly magnetic Klamath terrain rocks in the northeast corner of the grid. 00:43:46.000 --> 00:43:59.000 So now I would like you to try to ignore those interesting anomalies and look at the more subtle anomalies to the south and east that reflect magnetic properties within the Franciscan complex. 00:43:59.000 --> 00:44:08.000 What I hope you can see is that these subtle anomalies form distinct magnetic lineaments that follow the major fault zones that cross the shoreline in Northern California. 00:44:08.000 --> 00:44:15.000 That's good news because now we can use those anomalies to characterize the connectivity and geometry of these fault zones. 00:44:15.000 --> 00:44:24.000 We can now also continue to piece together how the faults north of the Triple Junction connect with the San Andreas Fault System 00:44:24.000 --> 00:44:30.000 to the south. And I've highlighted some possible connections in the black dashed lines on the right. 00:44:30.000 --> 00:44:45.000 Next slide, please. So next steps, more specifically our plans for these data include integrated interpretation of magnetic and gravity data with offshore geophysics and new tomography. 00:44:45.000 --> 00:44:58.000 To first characterize fault geometry and latest Pleistocene to Holocene deformation on these cross-shore structures to develop more realistic fault models and separate estimates. 00:44:58.000 --> 00:45:08.000 And second, to understand how and where strike-slip motion on the Maacama Bartlett Springs Fault System is transferred across the Triple Junction into southern Cascadia. 00:45:08.000 --> 00:45:26.000 And of course, this information can be used to improve regional models of slip deficit rate as have has been recently done by Materna and co-authors. And this information can also inform future earthquake and tsunami hazard assessments in Northern California. 00:45:26.000 --> 00:45:35.000 Thanks. 00:45:35.000 --> 00:45:43.000 Hi, everyone. My name is Keene Karlsson. I'm a geology lecturer at Sonoma State University. 00:45:43.000 --> 00:45:54.000 And today I want to share with you the results of a fault trenching campaign that I undertook with colleagues from my former employer, Earth Consultants International, 00:45:54.000 --> 00:46:03.000 along the Green Valley fault in Napa County in the Summer of 2021. 00:46:03.000 --> 00:46:10.000 So for those of you who are not familiar with the Green Valley fault, it's part of the larger San Andreas Fault System 00:46:10.000 --> 00:46:15.000 and unlike a lot of the other faults in Northern California 00:46:15.000 --> 00:46:30.000 It's paleoseismic record is not super well-constrained so this map shows you the site of several previous studies that have been done on this fault to try to better constrain its paleoseismic history. 00:46:30.000 --> 00:46:38.000 Some of those contributors include John Baldwin, James Lienkaemper, Rich Koehler, and many others. 00:46:38.000 --> 00:46:46.000 The Green Valley fault does have good geomorphic expression in the east kind of northeast Bay Area. 00:46:46.000 --> 00:46:53.000 And the most recent event on this fault is fairly well-constrained around AD 1600. 00:46:53.000 --> 00:47:01.000 But going further back into the past, its earthquake history is not well constrained. 00:47:01.000 --> 00:47:24.000 So my family as it turns out, owns a 200 acre ranch in eastern Napa County that is right smack dab on the middle of the Green Valley fault and so with some help from Eldon Gath at Earth Consultants International, we submitted a NEHRP proposal and won the grant. 00:47:24.000 --> 00:47:41.000 And we undertook a trenching campaign up there just a little bit about a previous study that was done. So Baldwin et al., 2002 trenched the fault right on the other side of this stone fence from my family's property. 00:47:41.000 --> 00:47:47.000 They found what they thought was very subtle evidence of faulting. 00:47:47.000 --> 00:48:05.000 Unfortunately, my family members, I was just a little kid at the time. My family members did not allow them access to our side of the fence and so they were limited in what they could achieve. We selected the site that we did because 00:48:05.000 --> 00:48:13.000 there is a seasonal pond there and so we thought that because it's right along mapped traces of the fault. 00:48:13.000 --> 00:48:26.000 And it's a seasonal pond. We assumed that was a sag pond and that it would have an excellent stratigraphic record to help improve the paleoseismic record for the fault. 00:48:26.000 --> 00:48:34.000 So here on the left, you'll see a Google Earth screenshot showing where the USGS Q-faults layer has 00:48:34.000 --> 00:48:41.000 the Green Valley fault map as you can see, there are a few subparallel strands going through this area. 00:48:41.000 --> 00:48:49.000 And where you see seasonal pond, that whole area, that whole valley, my family calls Coal Hollow. 00:48:49.000 --> 00:49:06.000 And so our NEHRP proposal was to open trenches across both of those fault strands that go through Coal Hollow as well as through the middle of the valley to try to find a place with a great stratigraphic record 00:49:06.000 --> 00:49:29.000 right up against the fault and also on the image to the left, you can see where Baldwin et al., 2002 had their trenches just on the north side of that fence. However, their trenches were on a fairly steep slope and so did not have a good chance of being a good sediment trap and having a good, well-preserved 00:49:29.000 --> 00:49:41.000 sedimentary record. So this is a drone image taken by Chris Madugo of PG&E showing what this valley looks like. 00:49:41.000 --> 00:50:01.000 This area is very mountainous a nd so to have this fairly flat valley with an ephemeral pond is excited when we went to open our first trenches. 00:50:01.000 --> 00:50:12.000 So here's showing you some LIDAR imagery of the area and the location of Baldwin's trenches at the very northwest corner of Cole Hollow shown in dark gray. 00:50:12.000 --> 00:50:24.000 And then all of our trenches shown in blue. And you'll see that there are tons of previous studies that had faults mapped right through Cole Hollow. 00:50:24.000 --> 00:50:38.000 And those fault traces based on previous studies have now been adopted by the USGS, the CGS, and perhaps most importantly Alquist Priolo. 00:50:38.000 --> 00:50:50.000 My family doesn't have any plans to build a big house or something on this property, but that is kind of the most important point I want to get across here is the AP zone does go right through Cole Hollow, 00:50:50.000 --> 00:50:57.000 and as you'll see in the coming slides, maybe that needs to be changed. 00:50:57.000 --> 00:51:02.000 So I'm going to walk you through our trench logs. I'm not going to spend too much time on each one. 00:51:02.000 --> 00:51:08.000 Definitely not going to spend much time because I'm running out of time here. But basically what you'll see is we didn't find any faults. 00:51:08.000 --> 00:51:15.000 It was basically layer cake stratigraphy in all of these trenches in the middle of the valley. 00:51:15.000 --> 00:51:22.000 This is right near Baldwin's Trench, the closest one, right on our side of the fence. 00:51:22.000 --> 00:51:30.000 And there's a little hint, a little whisper of some offset there, but it's not. It's a contact between two units. 00:51:30.000 --> 00:51:37.000 The trenches along the side of the valley, right where the fault was mapped. Again, we didn't find any evidence of faulting. 00:51:37.000 --> 00:51:50.000 And our most important trench log is this one. Where we collected an OSL sample, we found some young ash that we didn't expect there, but again, no evidence of faulting. 00:51:50.000 --> 00:52:01.000 So the point of this is we need to figure out where the active trace of the Green Valley fault goes through this area and potentially remap the AP zone. 00:52:01.000 --> 00:52:11.000 In my last few seconds, I want to make a desperate plea for the Sonoma State University Geology Department, which is slated for elimination at the end of this semester. 00:52:11.000 --> 00:52:16.000 If any of you want to get involved in supporting our department and saving Sonoma State University. 00:52:16.000 --> 00:52:30.000 Please send me an email. Thank you so much for your time and attention. 00:52:30.000 --> 00:52:38.000 Okay, thanks to all of our speakers for their great talks. We have about 5 minutes before our scheduled break. 00:52:38.000 --> 00:52:47.000 So I'd like to open it up to any live Q&A or discussion. So if you have a question you'd like to ask any of the speakers, please raise your hand. 00:52:47.000 --> 00:52:54.000 And I noticed there was some great discussions going on in the chat. So if any of the speakers want to reply 00:52:54.000 --> 00:53:01.000 live to any of those questions, feel free. 00:53:01.000 --> 00:53:08.000 Otherwise, I have a question for Travis. Travis, I'm sorry if I missed this. 00:53:08.000 --> 00:53:16.000 Are you considering the uncertainty in your earthquake locations. 00:53:16.000 --> 00:53:17.000 In your model fitting or is that something you've thought about? 00:53:17.000 --> 00:53:24.000 Yeah, I'm currently not really taking those into account. 00:53:24.000 --> 00:53:41.000 What I'm doing to kind of like work around the uncertainty issue is to use this so-called support vector regression Which fits a surface, but does not penalize points that are near the surface. So like within some damage zone or some uncertainty 00:53:41.000 --> 00:53:52.000 Region. So I'm kind of just like, oh, we're going to just push all of the potential error into that like that epsilon value. 00:53:52.000 --> 00:53:56.000 I don't know if that's the best way to do it, but it's how I'm approaching it. 00:53:56.000 --> 00:54:02.000 And so that's like a physical length scale. 00:54:02.000 --> 00:54:10.000 Yeah, so it is a physical length scale and um it's something that we actually do cross validation for. 00:54:10.000 --> 00:54:26.000 So we kind of give it some range of like values of a typical earthquake uncertainty or damage zone dimensions and then basically like kind of like search over that space to find the surface that actually fits it the best. 00:54:26.000 --> 00:54:36.000 So those come out on the range of like usually like 100 to 400 meters kind of makes sense, right? 00:54:36.000 --> 00:54:38.000 Yeah. Thanks. 00:54:38.000 --> 00:54:43.000 Yeah. 00:54:43.000 --> 00:54:52.000 Okay, does anyone else have any questions? Please raise your hand. 00:54:52.000 --> 00:55:04.000 Or if not, we can reclaim at this point two minutes of our break, our lunch. So I will hand it back to Sarah. 00:55:04.000 --> 00:55:05.000 Thank you to Grace and all the speakers. That was amazing. 00:55:05.000 --> 00:55:07.000 Thanks again to all the speakers. 00:55:07.000 --> 00:55:13.000 Very energizing. I hope it got everyone ready for lunch or whatever meal is most appropriate for your time zone. 00:55:13.000 --> 00:55:27.000 And then come on back. Mike has… promised us that he is going to carry who on the vicky chad line of thoughts that connect through the East Bay and how they walk. And he's going to answer all of our questions. He promised. 00:55:27.000 --> 00:55:38.000 So come back after lunch at 1.15 p.m. Pacific time, unless you all in the next session, in which case please come back at 1 p.m. If you all, Mike, Cassie, Danny, Katya, Kathy. 00:55:38.000 --> 00:55:43.000 Or Chelsea, okay? Bon appetit.