WEBVTT 00:00:00.000 --> 00:00:08.000 [Sure.] 00:00:08.000 --> 00:00:16.000 Hello everyone. I am happy to bring us back for the most exciting part of the workshop, the Thunder Talks. 00:00:16.000 --> 00:00:25.000 Part 1. And to bring us through this exciting journey, we have the incomparable Grace Parker. [WooHoo!] 00:00:25.000 --> 00:00:33.000 Thanks for joining us in today's Thunder Talk session. We have 13 talks lined up for you. 00:00:33.000 --> 00:00:39.000 And this will likely take the entire time we've allotted for this session. So I encourage you to use the chat. 00:00:39.000 --> 00:00:45.000 To ask any questions you might have for the speakers. And speakers, remember to keep an eye on the chat. 00:00:45.000 --> 00:00:56.000 To see if anything comes up about your talk. And so with that, I'll turn it over to our first speaker, Max Schneider. 00:00:56.000 --> 00:01:07.000 Hi, everyone. So I'm from USGS. And I've been testing the USGS Aftershock forecasting system. I don't see my slides yet. 00:01:07.000 --> 00:01:15.000 [John's computer. Yeah. 00:01:15.000 --> 00:01:18.000 Almost. Hold, please. 00:01:18.000 --> 00:01:31.000 I think we don't have, here's the slides.] Awesome. Alright. So I'm looking at our forecasting system that after large earthquakes We calculate the probability of aftershocks. 00:01:31.000 --> 00:01:48.000 In particular, particular durations and magnitude levels following that earthquake, as well as the probability distribution of the count of aftershocks. So we now have an interactive graphic to show that information 00:01:48.000 --> 00:01:58.000 for different magnitudes and forecasts. And all this comes from a statistical model that has parameters for productivity. 00:01:58.000 --> 00:02:22.000 Which is the expected number of aftershocks for how those aftershocks are expected to decay with time, that the model lets every aftershock be a mainshock that can trigger its own earthquakes in the sequence. And our model is updated following a regular schedule to reflect the changes of the ongoing sequence. 00:02:22.000 --> 00:02:30.000 So our forecasting system has a few features to account for issues in the aftershock phenomenon. 00:02:30.000 --> 00:02:48.000 So one is that it allows us to have not one productivity parameter for all of the earthquakes in the sequence, but two parameters. So one productivity for the mainshock and one for the subsequent aftershocks that's been found to lead to less bias aftershock forecasts. 00:02:48.000 --> 00:03:05.000 We know that after large earthquakes, our aftershock catalogs are incomplete to low magnitudes. So rather than assuming constant magnitude of completeness that's quite high. Our system allows for the magnitude of completeness to start high and then go down with time since mainshock. 00:03:05.000 --> 00:03:27.000 And our system also allows to forecast parameters themselves to be updated as well on that regular schedule in addition to the forecast itself. So what we want to understand is how these aftershock, these operational choices that we can make in our forecasting system 00:03:27.000 --> 00:03:42.000 lead to different forecasts and how good those different choices are as far as the forecast skills. So we're going to be evaluating all of these forecasts made under different operational choices. 00:03:42.000 --> 00:03:52.000 Against observations to understand forecast skills. And we're going to be looking at this from a user-centered perspective. So looking at how forecast scale changes. 00:03:52.000 --> 00:04:06.000 from all of those things that were changing in our forecast product. So the aftershock magnitudes, the forecast durations, and over time throughout the sequence. 00:04:06.000 --> 00:04:27.000 So we're going to be doing this with several different metrics because while we want accurate forecasts, we also want forecasts that are precise. So for forecast accuracy, we're going to be using continuous strength probability scores to evaluate these probability distribution forecasts against observations. This is a standard metric from 00:04:27.000 --> 00:04:41.000 W eather forecast literature. We're also going to be looking at the width of the middle 95% forecast interval to understand forecast precision and tracking rates of forecast over and under prediction. 00:04:41.000 --> 00:04:57.000 So what we've done so far is run a simulation study to understand how these different metrics assess aftershock forecasts. So we start by simulating aftershock sequences with a particular mainshock and get simulated catalogs for aftershock sequences. 00:04:57.000 --> 00:05:08.000 And then we calculate forecasts made under different operational choices at all of these different time points in our forecast update schedule. 00:05:08.000 --> 00:05:15.000 And for each of those different possible forecasts that we could have issued, we calculate our testing metrics. 00:05:15.000 --> 00:05:24.000 Assessing different components of forecast skill. So we repeat those three steps many times for simulated sequences. 00:05:24.000 --> 00:05:34.000 Under different mainshock magnitudes, and many, many, many simulations to quantify the uncertainty of our testing strategy. 00:05:34.000 --> 00:05:42.000 And then we summarize those testing metrics across forecasts for different forecast events or different aftershock magnitudes, different time durations 00:05:42.000 --> 00:05:49.000 A nd across all those simulated sequences. So I'm going to show you some very first preliminary results from our simulation study. 00:05:49.000 --> 00:05:58.000 What we're looking at here are the CRPS scores. So the lower the CRPS, the better the forecast accuracy. 00:05:58.000 --> 00:06:05.000 We're looking at those over time since the mainshock. So what we can see, and for two different operational choices. 00:06:05.000 --> 00:06:17.000 So what we can see is that if we have a single productivity parameter, we have more accurate forecasts especially early on in the sequence. And that can be explained if we also look at the forecast range. 00:06:17.000 --> 00:06:28.000 So those one productivity forecasts tend to be much wider, which means that they are more likely to capture the observed number of aftershocks. 00:06:28.000 --> 00:06:39.000 So we can see this bias and precision trade-off in these early preliminary results. 00:06:39.000 --> 00:06:59.000 That's it for me. Thank you. 00:06:59.000 --> 00:07:07.000 Okay. Hi, everyone. I'm Tara with the USGS, and today I'm presenting on regional faulting and path effects in the San Francisco Bay Area. 00:07:07.000 --> 00:07:20.000 So we all know that the Bay Area has high seismic potential, but as we cannot predict when, where, or what the next earthquake will look like, we have to rely on these probabilistic models of shaking to guide our seismic resilience efforts. 00:07:20.000 --> 00:07:27.000 Now, there are a lot of factors or ingredients that control the shaking we feel from earthquakes. These include those unique source characteristics. 00:07:27.000 --> 00:07:31.000 Properties along the source to site path and the near surface geology. 00:07:31.000 --> 00:07:38.000 Now, although we know a lot of these ingredients, we still don't know the full recipe, which can lead to increased uncertainty and are hazardousness. 00:07:38.000 --> 00:07:49.000 This is especially true when looking at the path component, where we still don't have a very good understanding of those physical processes and properties along that source to site path and how that relates to our ground motions. 00:07:49.000 --> 00:07:55.000 But here I'm going to present a couple new path ingredients to help us better understand what is actually controlling the shaking. 00:07:55.000 --> 00:08:01.000 And specifically, these focus on regional faulting in the Bay Area. 00:08:01.000 --> 00:08:10.000 So we know that these fault zones tend to be highly damaged and so one would expect these damaged conditions to contribute to the attenuation of ground motions. 00:08:10.000 --> 00:08:17.000 We investigate this by looking at several different fault metrics, including the misalignment and the density along our fault path. 00:08:17.000 --> 00:08:22.000 Along array paths, as well as the angle at which our paths are crossing the faults. 00:08:22.000 --> 00:08:26.000 We evaluate these relationships by looking at ground motion model residuals. 00:08:26.000 --> 00:08:37.000 These residuals are the difference between our observed ground motions and our model predictions, which tells us how much of our ground motions are we currently not capturing in our models. 00:08:37.000 --> 00:08:49.000 And if we break these residuals down into a source path and site component, this tells us how much of that unmodeled ground motion can we attribute to these unique repeatable source path and side effects? 00:08:49.000 --> 00:08:55.000 So the highlighted path term is what we relate to our different fall metrics. 00:08:55.000 --> 00:09:10.000 The first metric that we looked at is the fault misalignment or essentially how disordered our faults are. We took the Bay Area region, divided it into different cells, and for each cell we calculated a misalignment value and this is based off of surface traces of faults. 00:09:10.000 --> 00:09:18.000 We calculate a misalignment value for an entire ray path by summing the values for each of the cells that a path travels through. 00:09:18.000 --> 00:09:24.000 So here in the scatter plot, we have these ray path misalignment values normalized by the epicentral distance. 00:09:24.000 --> 00:09:30.000 That's on the bottom. Our path term is on the left. And what we find is this negative correlation. 00:09:30.000 --> 00:09:36.000 So path was traveling through more disordered fault zones tend to have more attenuation of ground motions. 00:09:36.000 --> 00:09:44.000 And when we look at our map, we notice the greatest misalignment in these regions dominated by these older pre-quaternary faults . 00:09:44.000 --> 00:09:51.000 So although they may not be as seismically active, that damage structure is still contributing to the attenuation of the ground motions. 00:09:51.000 --> 00:10:00.000 Another metric that we looked at is the fault density. And for this, we simply just took the number of faults that a path crosses and divided by the distance. 00:10:00.000 --> 00:10:20.000 And when we look at this correlation, we also observe this negative relationship, which is suggesting that paths c rossing through a higher density of faults also tend to observe greater attenuation of the ground motions. So the more abundance of the damage structure that you have, the more attenuation you're observing. 00:10:20.000 --> 00:10:37.000 And the last metric that we've looked at so far is what we call the fault cross angle. So we look at the angle at which our paths are crossing all the different fault segments. We take the average A zero degree would be a ray path running parallel to our faults. 90 degrees is perpendicular. 00:10:37.000 --> 00:10:57.000 And when we look at this scatter plot, we see that as we increase that angle or as our paths are more perpendicular to the faults, we also observe more attenuation of the ground motions. And so there must be something about the damage zone fabric orientation that's also controlling the attenuation of the ground motions. 00:10:57.000 --> 00:11:13.000 And so far this has been a very simple analysis so far. We've been looking at simple 2D geometries of these pads and these faults. But what this indicates so far is that there is some relationship about the increased complexity of faults along the paths. 00:11:13.000 --> 00:11:26.000 An increased attenuation of our ground motions. There's still a lot more that we can do to really dig into the more complex relationship between the faults and the paths, but that's still future work that we're working on. 00:11:26.000 --> 00:11:32.000 But this is really interesting work to help us really understand what is controlling that shaking that we're feeling from their needs. 00:11:32.000 --> 00:11:48.000 So that's it. Thank you. 00:11:48.000 --> 00:11:56.000 Thank you, everybody. I'm John Louie. I'm at the Terean. 00:11:56.000 --> 00:12:03.000 And I'm retired from the University of Nevada, Reno? Are my slides coming up? 00:12:03.000 --> 00:12:05.000 Or should I give it off the cuff? 00:12:05.000 --> 00:12:13.000 [Hold on one second.] 00:12:13.000 --> 00:12:26.000 So what you're going to see is an animation of Eckert's Reno area scenario adjacent to Northern California that he published in 2021. 00:12:26.000 --> 00:12:34.000 And what it shows is that the 00:12:34.000 --> 00:13:04.000 Is that the even very shallow basins as you've been hearing about can do five times amplification of shaking. So even though the ruptures occurring in this red box there's still from this relatively small magnitude M6.3. 00:13:05.000 --> 00:13:19.000 There's still severe shaking more than 30 centimeters per second occurring around basins that are and sub-basins that are more than a rupture length away. 00:13:19.000 --> 00:13:28.000 Now, so we've been working on Reno, central Reno here for quite a bit of time. 00:13:28.000 --> 00:13:39.000 And we've encapsulated the basin information, the basin geometry information into Z1.0 and Z2.5. 00:13:39.000 --> 00:13:49.000 To contribute that to the National Seismic Hazard Mapping Program. 00:13:49.000 --> 00:14:06.000 This motivated NEHRP to fund UNR and the University of Cincinnati to extend our Z measurements to industrial areas that are off this map at both north and east of the Truckee Meadows here. 00:14:06.000 --> 00:14:15.000 That you're looking at. [And there is the control. Okay. 00:14:15.000 --> 00:14:21.000 And I don't Okay, there we go. 00:14:21.000 --> 00:14:51.000 Alright. I'll get used to it.] We put out arrays of Pascal nodals and also measurements with UNR's CG6 Grovimeter and combining these gravity and seismic techniques, we have developed, you see here shear velocity cross-sections for a couple of the profiles, 00:14:51.000 --> 00:14:59.000 at the lower left is the 20 km long north-south profile through the what we call the North Valleys locally. 00:14:59.000 --> 00:15:12.000 Where you can identify Z1.0. The shallowest depth at which the shear-wave velocity reaches 1.0 km per second and likewise, Z2.5. 00:15:12.000 --> 00:15:34.000 For 2.5 kms per second. Here's a much at the upper right is the much more shorter profile near the Tesla Gigafactory east of Reno, and likewise, we could identify both of those Z's in this velocity section. 00:15:34.000 --> 00:16:01.000 So putting our result, looking closely first at the profile through the North Valleys, 20 km long. The gravity in blue you can see where the bottom of the basin is volcanics that are not actually 2.5 km per second shear wave velocity. 00:16:01.000 --> 00:16:30.000 There may be 1.5 to 2.2 that's where the gravity shows the basin thickness to follow more or less the Z1.0 contour. Whereas, the HVSR side of our study showed that the Nakamura h-over-v identifies with the Z2.5, which is really where you'll find 00:16:30.000 --> 00:16:43.000 Mesozoic Sierra and basement. Sometimes right up against the Quaternary or Late Tertiary sediments and sometimes below the volcanic stack. 00:16:43.000 --> 00:16:53.000 Let's see. [And I guess I hit it too many times.] Alright. 00:16:53.000 --> 00:16:59.000 What we've built here is a 00:16:59.000 --> 00:17:15.000 detailed map north of Reno Where you can see we've added to what we've been doing the last couple of decades in Reno proper. 00:17:15.000 --> 00:17:34.000 Between Pancha et al., published in 2017, and Abbott at Louie way back in 2000, All of that is detailing on top of the USGS's original work by Saltus and Jachens, basin and range wide in 1995. 00:17:34.000 --> 00:17:53.000 And so our new project has gotten further details on this is the gravity map essentially for The North Valleys gotten in 2023 and also at the Tahoe-Reno Industrial Complex. 00:17:53.000 --> 00:18:00.000 We have this in blue. This blue square Z1.0 map, 00:18:00.000 --> 00:18:09.000 Against Saltus and Jachens, 1995, more generalized Z2.5 map. 00:18:09.000 --> 00:18:25.000 So what we'd like to do is get your comments and suggestions on how we can make sure these results most useful to the development of the further versions of the National Seismic Hazard Map. 00:18:25.000 --> 00:18:46.000 And those of you who want to use the map as well, how can we make sure that this information goes to the people who who need to know. Also, I'd certainly like to discuss more with anyone about how to best inform 00:18:46.000 --> 00:18:56.000 building code provisions and regional authorities. I'd welcome any contact to louie@unr.edu. 00:18:56.000 --> 00:19:14.000 And also there's the website where you can see all these results and get them in detail. Thank you very much for your attention. 00:19:14.000 --> 00:19:19.000 Hi, so I guess I'm up. Sue Hough here, U.S. Geological Survey, Pasadena. 00:19:19.000 --> 00:19:26.000 You're going to be hearing a couple of talks at this meeting on the Fickle Hill earthquake of 1954. 00:19:26.000 --> 00:19:35.000 Peggy Hellweg in particular got a number of the usual suspects together recently to take a closer look at the instrumental data from that event. 00:19:35.000 --> 00:19:47.000 I volunteered a Thunder Talk specifically not to talk about the earthquake so much, but about the concept of legacy macro seismic data for which this earthquake is a very interesting case study. 00:19:47.000 --> 00:20:11.000 Next slide. So the U.S. Coast and Geodetic Survey going back to 1930ish distributed questionnaires on postcards, mostly to postmasters with the standard questionnaire for macro seismic earthquake effects? Did you feel it? Did things shake and so forth. Those postcards were 00:20:11.000 --> 00:20:32.000 collected for a period of some years for this earthquake and other earthquakes, typically you might have media reports f or 1954, Laurie Dingler set out to collect actual eyewitness reports from people who still remember it. But for 1954 the U.S. Coast and Geodetic Survey data 00:20:32.000 --> 00:20:35.00 are the bulk of the data that we have to study the earthquake. 00:20:35.000 --> 00:20:50.000 Next slide. So those postcards made their way to the Coast and Geodetic Survey, they made their way into a series of publications called United States Earthquakes, which you may be familiar with. 00:20:50.000 --> 00:21:07.000 Summarize macro seismic effects and then later some instrumental observations for notable earthquakes. You probably aren't aware that there was a preliminary publication series known as the abstracts of earthquake reports for the Pacific Coast and blah, blah, blah. 00:21:07.000 --> 00:21:16.000 That was sort of a gray literature publication. It circulated mostly privately, it's a little bit hard to find, 00:21:16.000 --> 00:21:22.000 but nowadays you can find almost all of the copies online if you know where to look. 00:21:22.000 --> 00:21:34.000 Next slide. And the abstract series was less polished, but it was more complete. They culled the information from the postcards, and they summarized it. So this is, if you can see it 00:21:34.000 --> 00:21:39.000 a page for the 1954 earthquake, and if you can see the details 00:21:39.000 --> 00:21:49.000 it includes all the familiar indicators that you know, dishes rattled and up to chimneys were cracked and so forth. 00:21:49.000 --> 00:21:59.000 Sometimes the locations are given with cities, sometimes they're actual street addresses. Next slide. 00:21:59.000 --> 00:22:09.000 So for Fickle Hill, there are over 500 separate accounts for the city and streets, a lot of them have city and streets 00:22:09.000 --> 00:22:30.000 and with the traditional macro seismic indicators, which forms a tremendous legacy data set, the data are arguably more complete than Did You Feel It? with yourself selected, you have locations, it's tedious, but you can go figure out geo-reference, figure out where they are. 00:22:30.000 --> 00:22:43.000 You actually have over 100 not felt reports, so you're able to delineate the felt area very precisely, as you'll see in a second. Next slide. 00:22:43.000 --> 00:22:49.000 So these are very modern looking ShakeMaps for a 70-year old earthquake. 00:22:49.000 --> 00:22:56.000 Constrained by about 500 intensity values. There's two instrumental intensities that come from strong motion data 00:22:56.000 --> 00:23:06.000 recorded at Ferndale and Eureka, those instrumental intensities are a little bit high, but mostly consistent with the directly estimated intensity values. 00:23:06.000 --> 00:23:15.000 Next slide. You can compare the intensities to a modern intensity prediction equation. Atkinson et al., 2014. 00:23:15.000 --> 00:23:41.000 The intensities are nicely consistent with the predictions. They support the instrumental magnitude 6.5. They also support a relatively deeper crustal depth around 18 kms. You don't have observations supporting very high intensity, suggesting it was not an especially shallow spruce. At the higher end, the values are below the IPE, 00:23:41.000 --> 00:23:58.000 and that's because work by Jack Boatwright and Eleyne Phillips years ago showed that Did You Feel It? overestimates the felt extent because of the way that it estimates the lower intensities, but with the postcard, you can delineate the felt extent location. 00:23:58.000 --> 00:24:21.000 Next slide. So for the macro seismic data for this event confirm the preferred instrumental location near Fickle Hill, the largest town of any size was Blue Lake. It confirms the 18 km depth. There's some indication even that supporting the high stress drop that 00:24:21.000 --> 00:24:32.000 comes out of analysis of strong motion. Data. But the point that I wanted to make is that there are more data to be mined. And arguably, let's call it Did They Feel It? 00:24:32.000 --> 00:24:36.000 it's data is very analogous to Did You Feel It? 00:24:36.000 --> 00:24:44.000 The same questionnaire, it's the same level of detail, the data are out there. 00:24:44.000 --> 00:24:55.000 So maps like this can be put together for any number of notable 20th century earthquakes. So I wanted to end also with an invitation, also a thank you to the organizers. I forgot to say that 00:24:55.000 --> 00:25:15.000 for pulling this meeting together, but an invitation, if there are 20th century earthquakes in the west that people think are notable for which a more detailed ShakeMap would be of interest, please feel free to reach out to me. 00:25:15.000 --> 00:25:45.000 Thank you. 00:25:45.000 --> 00:25:59.000 Sorry, I thought I was unmuted. Thank you, Sue. For introducing this very intriguing earthquake here on the North Coast, we seem to have a fondness for December quakes. 00:25:59.000 --> 00:26:09.000 So I'm going to look at modern, in other words responses over the past year. 00:26:09.000 --> 00:26:31.000 About this earthquake that occurred 70 years ago. Are they of any use? And I really do want to thank Sue and our big crew of folks working on this earthquake, please tune in to Peggy's talk this afternoon. 00:26:31.000 --> 00:26:39.000 [Okay, somebody needs to advance my slide because now it is not working.] Oh, thank you. 00:26:39.000 --> 00:26:45.000 So in order to solicit comments. 00:26:45.000 --> 00:26:55.000 I used a number of platforms. First. I read a weekly column in our regional newspaper, and this was last July. 00:26:55.000 --> 00:27:03.000 Where I tried to explain why this earthquake is of interest, and if anyone recalled it 00:27:03.000 --> 00:27:19.000 to please let me know. I also had a small article in our Senior News Monthly Publication and I utilize the Humboldt County History Facebook page. 00:27:19.000 --> 00:27:24.000 Next. 00:27:24.000 --> 00:27:33.000 What I've come back with so far is 30 reports from Humboldt County. I've got about another 25 that I need to get back to. 00:27:33.000 --> 00:27:40.000 Each one, I mean, basically people say give me a call I'll tell you about it. 00:27:40.000 --> 00:27:47.000 And if you've ever asked people about their earthquake experiences, you know that each call can last a long time. 00:27:47.000 --> 00:27:55.000 We see a lot of responses in Eureka and then scattered throughout the rest of the area. 00:27:55.000 --> 00:28:09.000 I've just color coded these for intensity M7.0 and larger in red, 5 to 6 in yellow, less than 5 in green. 00:28:09.000 --> 00:28:17.000 Next. So here's just a little sample of some of these stories. 00:28:17.000 --> 00:28:30.000 I'll start with the really low shaking one. This was a good friend of mine who's lived here all his life in my dog walking group. So I asked him about this earthquake. 00:28:30.000 --> 00:28:41.000 He was four years old at the time and distinctly remembers a cup and saucer slowly vibrating along the counter of his home and it didn't fall off. 00:28:41.000 --> 00:28:47.000 Obviously, there was a lot more going on, but based on that, I would say this was less than a five. 00:28:47.000 --> 00:29:08.000 On the stronger end, the account of someone whose father worked at a lumber company. He was sitting at his desk and the shaking slammed him in his chair across the room banged into the wall. Our assumption is that his chair probably didn't have rollers. 00:29:08.000 --> 00:29:26.000 But we don't know that for sure. Many accounts of people observing ground roll. In fact, pretty much everyone outside mention seeing the ground rolling. 00:29:26.000 --> 00:29:50.000 The cattle running across the street and seeing the wave, the ground rolling like an ocean The 11 and 12 year old girls were riding their bicycles in Eureka suddenly couldn't move at all, watched the ground rolling and saw chimneys toppling. 00:29:50.000 --> 00:29:58.000 Gives you a feel for what that experience was. Next one. 00:29:58.000 --> 00:30:08.000 If we compare it to the ShakeMap. Clearly, we're not seeing a lot of detail, but there's probably more detail in the actual shaking. 00:30:08.000 --> 00:30:15.000 Then you see in the ShakeMap. Next. 00:30:15.000 --> 00:30:24.000 So how reliable are these accounts? These people were all children in 1954. 00:30:24.000 --> 00:30:32.000 And they're memories are very much fixated on things that were unusual to them. 00:30:32.000 --> 00:30:39.000 So seeing the ground roll, seeing chimneys topple. 00:30:39.000 --> 00:30:55.000 Seeing a woman racing out of their house wearing curlers in their hair. That was actually one of the main things one of these respondents thought. It was really weird to see someone in the street with their curlers. 00:30:55.000 --> 00:31:06.000 Many of the factors that we normally consider with intensity. They didn't notice. So very few of them mentioned noise. 00:31:06.000 --> 00:31:18.000 When I asked, well, was there a lot of stuff off of shelves often the response was, well, I wasn't quite sure. Some did say that there were. 00:31:18.000 --> 00:31:23.000 So it's a very selective memory. 00:31:23.000 --> 00:31:39.000 A number of these people still live on the North Coast today and described more recent earthquakes as being far stronger, in particular December 2022, I would argue that that's a specious comparison. 00:31:39.000 --> 00:31:49.000 Because as an adult, feeling an earthquake in the middle of the night is quite different than feeling an earthquake in the middle of the day when you're a child. 00:31:49.000 --> 00:31:55.000 But it's kind of interesting. These accounts really do add texture. 00:31:55.000 --> 00:32:00.000 To other accounts from the event. 00:32:00.000 --> 00:32:17.000 I'd argue maybe most useful as an education tool. Thank you. That's it. 00:32:17.000 --> 00:32:29.000 Hello. Yeah, good evening, which feels a bit weird to say but it's definitely not the well, good morning. It's the evening here. 00:32:29.000 --> 00:32:43.000 It's good meet you all and I'm really excited to give a five minute talk. And yeah, if we just go on to the next slide, slightly different format. So I'll just be talking through a poster presentation. 00:32:43.000 --> 00:32:51.000 But I'm joining this workshop on behalf of Brit Insurance, which is a global insurer and free insurer based in London. 00:32:51.000 --> 00:33:00.000 The post I'm in front of you shares the process I've taken to evaluate wood frame vulnerability for the purpose of earthquake insurance pricing in California. 00:33:00.000 --> 00:33:12.000 For those who don't know, building vulnerability is one of the components of a catastrophe model Which in simple terms is a financial loss model that sits atop of a probabilistic hazard model. 00:33:12.000 --> 00:33:29.000 The U.S. Earthquake Catastrophe Model that we license to Brit is several years old now, and so one of the reasons why we evaluate these models is to assess the performance of the model against alternate and perhaps more recent models and data ingest new research findings and ultimately determine whether it is fit for use. 00:33:29.000 --> 00:33:40.000 So where do I start? Using open source data from the Global Earthquake Model wood frame construction constitutes 98% of residential buildings, which is certainly not a new finding. 00:33:40.000 --> 00:33:51.000 Therefore, understanding the vulnerability of this exposure type is hugely important when considering an aggregate view of building vulnerability and loss in California. 00:33:51.000 --> 00:33:59.000 When talking about building vulnerability, I'm explicitly referring to the damage ability or the damage ratio as a function of some ground motion intensity measure. 00:33:59.000 --> 00:34:07.000 In developing vulnerability functions, typically for high frequency perils such as hurricanes or floods, insurance claims are used. 00:34:07.000 --> 00:34:12.000 However, given the infrequency of lost earthquakes in the U.S. and globally even. 00:34:12.000 --> 00:34:20.000 We instead use engineering models such as single degree of freedom oscillation models to model the building response to ground shaking. 00:34:20.000 --> 00:34:33.000 Despite this, the catastrophe models that we license to Brit is developed largely from claims by the 1994 Northridge event due to widespread damage to wood frame and residential buildings. 00:34:33.000 --> 00:34:44.000 However, I believe this approach over calibrates the vulnerability to an event that is constrained by the ground motion characteristics exhibited by the event. 00:34:44.000 --> 00:34:58.000 So just moving up to point three. Findings in a recent paper from Robert Chase and others identified a reduction in the performance of building vulnerability when subjected to ground motions from subduction earthquakes rather than crustal earthquakes, which is attributed 00:34:58.000 --> 00:35:04.000 To the longer durations and elevated frequency contents at moderate to long-periods. 00:35:04.000 --> 00:35:11.000 Furthermore, even considering locations where seismic hazard is only dominated by crusted events. 00:35:11.000 --> 00:35:19.000 Variability in source or hazard characteristics may still alter the frequency content or duration patterns in ground motions. 00:35:19.000 --> 00:35:27.000 It's also worth noting that building standards and codes don't generally consider ground motion duration or spectral shape in the design of new buildings. 00:35:27.000 --> 00:35:42.000 So there's ultimately concern that our internally licensed catastrophe model is understating the building vulnerability in Northern California when considering the different ground motion characteristics exhibited by large crustal and subduction events. 00:35:42.000 --> 00:35:54.000 The global earthquake model develops vulnerability curves that employ the FEMA P58 engineering approach which exhibits a greater damage ratio, high-return period ground motion intensities when compared to our licensed model. 00:35:54.000 --> 00:36:04.000 Therefore, one approach we can use is to adjust the tail vulnerability to that of GEM to reduce the performance of wood-frame buildings in the model. 00:36:04.000 --> 00:36:12.000 So for point four, this is another piece of work. We explored the vulnerability of contents as a function of building vulnerability. 00:36:12.000 --> 00:36:29.000 I mean, just referring back to point one you can see that content is a significant proportion of the total building value for residential structures. So it's really important to understand the vulnerability of each of those construction or components. 00:36:29.000 --> 00:36:41.000 By using a component-based analysis rather than a more simple function derived from the building vulnerability, we can model the non-linearity of contents vulnerability at low mean damage ratios. 00:36:41.000 --> 00:36:53.000 By doing so, recent papers show an increase in the damageability and therefore potential under appreciation of the contents vulnerability in our licensed Catastrophe model. 00:36:53.000 --> 00:37:07.000 So there's a few different ways in which we can evaluate Castrophe model components within insurance. And so I really hope this is like an insightful piece of information and how we bridge the gap between academia and industry. 00:37:07.000 --> 00:37:21.000 So if any of this work aligns with some of the work that you're doing, then I'll be really happy to talk to you and would love to collaborate. 00:37:21.000 --> 00:37:32.000 And some references. There you go. Thank you. 00:37:32.000 --> 00:37:40.000 Hi, my name is Theresa Sawi. I'm a postdoc at the USGS, and today I'll be talking about rapid magnitude estimation using strain data from borehole strain meters. 00:37:40.000 --> 00:37:48.000 And also from DAS, distributed acoustic sensing instrumentation for earthquake early warning. 00:37:48.000 --> 00:38:08.000 So when it comes to warning populations about offshore earthquakes, it's all about locations and the problem with earthquake early warning in particular ShakeAlert is that all of the instrumentation is onshore and that leads to consistent mislocations, as we can see on the figure on the right. 00:38:08.000 --> 00:38:23.000 We have two earthquake locations from the ShakeAlert System and a star you can't see my pointer, I believe, but one of them is further offshore, one of them is further onshore, and that can lead to a lack of alerts in this case. 00:38:23.000 --> 00:38:31.000 And so distributed acoustic sensing offers opportunities to improve earthquake early warning because it allows for offshore instrumentation. 00:38:31.000 --> 00:38:37.000 By repurposing telecom cables for kms long arrays of strain meters 00:38:37.000 --> 00:38:53.000 and so can DAS seismology help fill this instrumentation gap? In this study, we aim to estimate earthquake magnitudes using borehole strain meters and DAS instrumentation actually without the need for locations at all. 00:38:53.000 --> 00:39:01.000 DAS measures strain and can capture low frequency waveforms associated with large earthquakes rupture larger patches of the Earth. 00:39:01.000 --> 00:39:10.000 Which can generate low frequency waveform signals. As seen in the figure below, while typical seismometers shown on the right-hand side. 00:39:10.000 --> 00:39:16.000 May not be able to capture low frequency signals, strain instrumentation shown on the left-hand side can. 00:39:16.000 --> 00:39:31.000 We can see that this P-wave first arrival is shown in the vertical light dotted blue line where low frequency signals are shown at the upper portion of the figures and we can see that there's a lot more bright signal energy coming in. 00:39:31.000 --> 00:39:39.000 And for the strain meter data than for the seismometer shown on the right. 00:39:39.000 --> 00:39:55.000 And so in this work, we're going to train a machine learning model to predict whether an earthquake is greater than or equal to magnitude 5.4 Based on just the first four seconds of strain meter data with no earthquake location. 00:39:55.000 --> 00:40:01.000 Our DAS cable is onshore, but we're hoping that the results here can be applied to offshore cables as well. 00:40:01.000 --> 00:40:07.000 It stretches 15 kms from Arcata to Eureka, California, with 3,000 channels recording strain. 00:40:07.000 --> 00:40:12.000 Because we do not have many large magnitude earthquakes in our data set on which to train our machine learning model. 00:40:12.000 --> 00:40:19.000 We use another strain instrument for whole strain meters to do so. 00:40:19.000 --> 00:40:27.000 We use borehole strain meter data from three regions in California, the Mendocino Triple Junction, Napa, and Ridgecrest from 2008 to 2018. 00:40:27.000 --> 00:40:32.000 Which includes 35 magnitude 5 or larger earthquakes and about 2,000 waveforms. 00:40:32.000 --> 00:40:41.000 Which is sufficient for training our simple model. This is an example on the lower left-hand side of a borehole strain meter. 00:40:41.000 --> 00:40:49.000 Next, we take four seconds after the P-wave of our strain waveforms and 10 seconds before them, then calculate statistics from those waveforms. 00:40:49.000 --> 00:40:54.000 Some are simple statistics like maximum amplitude shown on the y-axis of the plots below. 00:40:54.000 --> 00:41:02.000 And some are based on the low frequency content of the waveform, which are calculated from continuous wavelet transforms shown on the x-axis. 00:41:02.000 --> 00:41:08.000 In the plots on the bottom, each data point comes from one waveform and they're colored by magnitude on the left-hand side. 00:41:08.000 --> 00:41:13.000 And by whether the earthquake is magnitude greater than or equal to 5.4 on the right. 00:41:13.000 --> 00:41:21.000 We have a suite of 56 features which we use as input into our machine learning model. 00:41:21.000 --> 00:41:36.000 I'm going to skip this slide. This is our methods. We use XGBoost for our machine learning models and then SHAP scores, which we heard about yesterday, actually, to determine which of our features are most important. 00:41:36.000 --> 00:41:44.000 Our results show that the most predictive features are indeed the low frequency ones. We have these colored boxes in the middle here. 00:41:44.000 --> 00:41:54.000 And I'm not going to talk about how they work, but the information we get from them tells us that of all the waveforms from the magnitude greater than or equal to 5.4 earthquakes. 00:41:54.000 --> 00:42:01.000 We correctly predict 80% of them. And of all of our positive predictions, 79% are correct. 00:42:01.000 --> 00:42:13.000 These are really promising results for earthquake early warning given that we use no location information, only one borehole station and only four seconds of earthquake data. 00:42:13.000 --> 00:42:23.000 We then compare these results to two years of California Shake Alert system, which predicted that 10 earthquakes We're magnitude 5.4. 00:42:23.000 --> 00:42:33.000 And yet 70% of the predictions were smaller than magnitude 5.4, which shows that our method may be able to supplement the current earthquake early warning system. 00:42:33.000 --> 00:42:44.000 Finally, we can see how the DAS data compares to the borehole data. We've calculated the same statistical values for each DAS channel to take in the mean and standard deviation for three segments of the cable. 00:42:44.000 --> 00:43:02.000 Leading to 75 points for our 25 DAS earthquakes. We can see that there's a slight offset between the DAS, which is shown in squares and the borehole strain meter, which is shown in Xs, which we attribute to differing burial conditions and instrument responses. 00:43:02.000 --> 00:43:11.000 But overall, the trend between the features and between the features and magnitude seems to be the same between the DAS and the borehole strain meter. 00:43:11.000 --> 00:43:34.000 Which shows that this is a potentially good metric for measuring magnitude using these features for both DAS and for borehole strain users. Thank you. 00:43:34.000 --> 00:43:56.000 Hello, everyone. So this is Yihe Huang from the University of Michigan. Today, I will just talk about a speed value job, a good indicator of mainshocks in California and it's funded by NSF and also is primarily done by students, Sydney Gable and Francesca Burkett at Michigan. 00:43:56.000 --> 00:44:15.000 So we know that this b-value kind of quantifies the relative proportion of the small earthquakes and large earthquakes for a certain region. And in the left figure, if we plot this cumulative number of earthquakes as a function of earthquake magnitude, the slope here shows the b- value. And in this Gulia-Wiemer paper 00:44:15.000 --> 00:44:40.000 they demonstrated that the b-value can be greatly reduced during the first shock period. So for this particular case, the b-value during the first shock period is about 0.98, but during the background, it's 1.2. So they were saying like this over 10% reduction in the b-value suggests that we may have a larger earthquake coming soon. And they also observed that during the aftershock period 00:44:40.000 --> 00:45:00.000 the b-value increased much higher than the background level again. And they proposed, based on the data that we can use this traffic light system in which like when the b-value is reduced by more than 10%, there is a red light suggesting that we may have a larger earthquake coming soon. 00:45:00.000 --> 00:45:19.000 So we want to see if we can apply this method to better understand the mainshock occurrence, like in California. And if we look at the b-value calculation from the left figure, we can see that the small earthquake magnitude contribute a lot to the calculation of this b-value, especially like those magnitude 1s to magnitude 3s. 00:45:19.000 --> 00:45:30.000 But if you look at the catalog magnitudes we can see that their magnitudes actually have a pretty large uncertainty for this magnitude range. 00:45:30.000 --> 00:45:48.000 So here we just want to see like whether we can design a better method to quantify the magnitudes for the small events to the larger small events. So what we are doing is that we are just linking the magnitude of small and large earthquakes in California 00:45:48.000 --> 00:46:11.000 through recalculating their magnitude from their b-value 2 ratios. So essentially, we're just doing waveform cross-correlations and pair these earthquakes together. So for example, in this waveform, these two earthquakes, they are paired together based on a cross-relation coefficient threshold. And the MP2 ratio here is about 10 at this particular station, and there are magnitude difference is about one. 00:46:11.000 --> 00:46:26.000 So we have tested the sensitivity of this relative magnitude approach based on the stations in Ridgecrest and also like the bent press filter of the seismic data. And we published results in JGR and the BSSA last year. 00:46:26.000 --> 00:46:32.000 So after we quantify the main two differences between the large events and the small events. 00:46:32.000 --> 00:46:47.000 We then apply this V positive estimator proposed by vendor else in 2021 to calculate the b-value. So this b-positive is shown to be much more robust when there's an incompleteness in the earthquake catalog. 00:46:47.000 --> 00:47:01.000 And also for our variety magnitude calculation, it's actually designed to do this b- positive because we are just quantifying the differences of magnitudes between small and large events. 00:47:01.000 --> 00:47:19.000 So here in the end, I just want to show like a pretty new results for five mainshocks in California that are recorded by the QTM tablet matching catalog and also these foreshock sequences have been shown to be like robust for shock sequences from statistical analysis. 00:47:19.000 --> 00:47:27.000 So here we're just applying the b- positive estimator to both the original catalog and also the rarity magnitude catalog. 00:47:27.000 --> 00:47:34.000 The great line here shows the original results and the colored lines here show the relative magnitude results. 00:47:34.000 --> 00:47:41.000 So you can see that for some sequences, especially like the top two panels, the left and right. 00:47:41.000 --> 00:47:50.000 We are observing like precursory b-value reduction several days before the occurrence of the mainshock, which is shown by the dashed line here. 00:47:50.000 --> 00:47:55.000 So this is true for both the rarity magnitude and original magnitude. 00:47:55.000 --> 00:48:05.000 But then if we look at the middle panel here, we can see that although the original catalog was predicting a b-value reduction during the first shock sequence. 00:48:05.000 --> 00:48:15.000 But then after we recalculate the magnitude, we don't see that b-value reduction anymore. So this b-value reduction actually happened after the mainshock. 00:48:15.000 --> 00:48:29.000 And this is the same for the right figure. And you can also see that the amount of the b- value reduction after the magnitude calculation is usually much more mild compared to like the b-value reduction estimated from the original catalog. 00:48:29.000 --> 00:48:46.000 So right now we're applying this method to analyze like all these mainshocks in northern and southern California, and we're trying to see if there's a systematic b-value reduction for certain recalculating the moment magnitudes of the small events using our 00:48:46.000 --> 00:48:58.000 relative magnitude. Thank you. 00:48:58.000 --> 00:49:06.000 Everyone, I'm Chris Milliner. I'm a hazard analyst at Armis Moody, and I'll be talking about work I conducted while I was at Caltech. 00:49:06.000 --> 00:49:26.000 So I'm going to discuss how we can use remote sensing data to measure inelastic awful deformation on a number of historic strikes that service rupturing earthquakes. The aim of our project is to determine whether inelastic strain localizes as false mature and how this affects various earthquake properties. 00:49:26.000 --> 00:49:33.000 So to answer these questions, we measured the surface deformation of 17 continental structures at earthquakes, which is what's shown here. 00:49:33.000 --> 00:49:37.000 And to do this, we use both radar and optical satellite and aerial images. 00:49:37.000 --> 00:49:52.000 And we apply a pixel tracking approach so you can obtain coherent constraint of the near field service motion across the fall zone And so I'd like to highlight one event here, which is located in Northern California that is the 2014 Napa earthquake. 00:49:52.000 --> 00:50:01.000 And so we actually generated a new service deformation map for this event using pixel offsets with the UAV resource data. 00:50:01.000 --> 00:50:18.000 So that's generated from using two opposing loop directions so you can retrieve the full to the horizontal motion as opposed to just the surface motion projected into the line of sight direction that's currently available from standard radar interferometry of UAV InSAR data. 00:50:18.000 --> 00:50:29.000 And so from these surface deformation maps, we can then measure the magnitude of fault slip along the length of the service ruptures and secondary faults that's what's plotted here on the figure on the right. 00:50:29.000 --> 00:50:39.000 These are the long fault slip profiles for the 17 earthquakes and i've just separated them arbitrarily in the vertical direction so you can get a sense of how they compare in scale. 00:50:39.000 --> 00:50:48.000 And so the misassumption in our analysis here is that these remote sensing observations capture both the off-fault 00:50:48.000 --> 00:50:56.000 because with this pixel tracking approach we can measure the service motion across and around the fault zone. 00:50:56.000 --> 00:51:17.000 Okay, and so it's then estimate inelastic fault deformation. We essentially difference and compare the field survey estimates of false set from remote sensing-based observations, which is a pretty common approach. And the idea here is that the field service estimates of fault-slip or capturing the on fault component of displacement as a comedy on the primary fault structure 00:51:17.000 --> 00:51:25.000 And with the remote sensing data, we're capturing both the on foot and the distributed inelastic shear because we have these wider aperture measurements. 00:51:25.000 --> 00:51:45.000 So comparing these two data sets, that's what's turning this plot for each earthquake. The red points field measurements of false slope in the black line are the soap profile I showed before for the different earthquakes. And you can see this kind of enveloping effect where the black line for most of these events is larger than those measures in the field. 00:51:45.000 --> 00:52:05.000 And so we can compare these two data sets more directly. That's what's shown by this plot here so i'm regressing the geodetic data on the y-axis onto the field data are shown on the x-axis so you can see When we plot these measurements all together into one plot, they don't fall in this one-to-one line. 00:52:05.000 --> 00:52:20.000 Systematically larger by about 20%. So, of course, here, this is just a math compilation of all the events that we missed together. We can break this apart and estimate awful deformation for each each event, each of these earthquakes so that's just shown by the deviation between those two slopes 00:52:20.000 --> 00:52:30.000 So this plot is similar to what I showed on the previous slide, but I've just switched the axes. So the x-axis now has due date data and the y-axis is the field measurements. 00:52:30.000 --> 00:52:46.000 And so essentially what you can take from this figure is that the way I've ordered it is from top left to bottom right is increasing amount of cumulative displacement. And those values of cumulative displacement are shown in the top left of each sub panel. 00:52:46.000 --> 00:52:53.000 You might not be able to see each value, but the main thing you can take away from this is that as you go from top left to bottom right. 00:52:53.000 --> 00:52:59.000 That red slope, which is the regression between these two data sets, starts the approach and converge that one-to-one line. 00:52:59.000 --> 00:53:10.000 Interpret this as indicating there's lower and lower amounts of awful inelastic strain occurring for earthquakes that have higher amounts of cumulus displacement. 00:53:10.000 --> 00:53:24.000 So another way of showing this is you can we can plot the average amount of deformation along the surface rupture or that deviation slope that I showed in the previous slide as a function of cumulative displacement that floated in the x-axis. 00:53:24.000 --> 00:53:36.000 We find a pretty clear negative trend in linear log space. So we think this is consistent with this notion that faults indeed localize as they mature. 00:53:36.000 --> 00:53:47.000 And now we can put some numbers on this long-term localization process. And just to clarify, we use cumulative displacement here as kind of a proxy for a proxy full maturity. 00:53:47.000 --> 00:54:04.000 And we also looked at how this awful deformation varies as a function of the arm's roughness so we looked at the actual roughness of the map surface fractures for each of these earthquakes and here we see a precursive association. I mean, this is indicating a geometric effect, 00:54:04.000 --> 00:54:12.000 whereby smoother faults produce fewer awful stress concentrations and therefore lower amounts of awful inelastic strain. 00:54:12.000 --> 00:54:20.000 And we also compared alpha deformation with different metrics one of which is the after poloarity. 00:54:20.000 --> 00:54:34.000 And here we see a pretty positive trends and we interpret this as indicating again, this would be consistent with this notion that geometrically smoother faults produce fewer off fault stress concentration. 00:54:34.000 --> 00:54:51.000 So with that, I think I'll leave it there. Thank you for your attention and look forward to taking questions. 00:54:51.000 --> 00:55:00.000 Great. Good morning, everyone. So my name is Rob Skoumal. I like to talk a little bit about a new Python algorithm for computing stress tensors using focal microscopes. 00:55:00.000 --> 00:55:09.000 And this work would not have been possible without the support of Jeanne Hardebeck and Andy Michael. 00:55:09.000 --> 00:55:23.000 So the inspiration for this work actually comes from Jeanne's and Andy's previous work. So in the 80s, Andy created this really powerful technique of taking the set of focal mechanisms and determining the stress orientations that are responsible for those earthquakes. 00:55:23.000 --> 00:55:32.000 Jeanne and Andy later expanded this technique to include damping. One of the advantages of that is that you can produce or you can reduce the impact of artifacts and produce more stable solutions. 00:55:32.000 --> 00:55:37.000 So their algorithm is called SATSI, which is written in Perl and C. 00:55:37.000 --> 00:55:43.000 So the motivation for today's work is to expand the capabilities of those damp stressed immersions. 00:55:43.000 --> 00:55:50.000 Putting some new ideas we have. We'll also want to expand this to include or to address catalogs with these millions of mechanisms. 00:55:50.000 --> 00:55:56.000 So over the previous years of this workshop, we've talked about some of our work trying to improve our ability to determine mechanisms. 00:55:56.000 --> 00:56:01.000 And we're at the point now, just on the verge of producing catalogs with these millions of solutions. 00:56:01.000 --> 00:56:11.000 So we like to have this algorithm scale appropriately. And we also like to do these versions as quickly as possible. The less time you spend doing the versions, the more time interpreting the results 00:56:11.000 --> 00:56:28.000 better. I am going to do all this in Python. So we designed this algorithm to be very flexible. So using a single script should handle pretty much any kind of dimension you throw at it. So if you want to do a zero dimensional case like an indie's original work 00:56:28.000 --> 00:56:32.000 up to some kind of complex situation, they should be able to handle that. 00:56:32.000 --> 00:56:51.000 It actually adjusts based off of your inputs automatically. But one of the motivations here is to handle these non-uniform types of problems. I think even today, the standard is to do these uniform spatial grids, right? So this is a figure from Jeanne and Andy's original paper where they introduced damping. 00:56:51.000 --> 00:56:56.000 Each one of these grid cells is the same size, about 0.1 degree weapons. 00:56:56.000 --> 00:57:07.000 One of the issues with that is that although this spatial gridding is uniform, earthquakes are not uniform. Then there's thousands of earthquakes next to a bin with 10 earthquakes. 00:57:07.000 --> 00:57:15.000 That's not always ideal from a stress inversion standpoint. So one of the motivations with this work is to address these sorts of problems also. 00:57:15.000 --> 00:57:24.000 So this is an example of an algorithm that I'm calling a quad root. So rather than looking at these uniform spatial cells. 00:57:24.000 --> 00:57:30.000 Rather than looking at uniform spatial cells, each one of these grids has the same number of earthquakes. 00:57:30.000 --> 00:57:35.000 I'm happy to talk more about that later. But this is built into the code. 00:57:35.000 --> 00:57:40.000 And whether you want to use that or the uniform spatial grid, it can handle both cases. 00:57:40.000 --> 00:57:50.000 So we've also included a number of built-in plotting tools. So on the left here is an example of a uniform dimensional stress inversion applied to the Ridgecrest sequence. 00:57:50.000 --> 00:57:58.000 So the background colors there are the SHmax measurements and on top of that we have I'm just staring us with the principal stress axis plotted. 00:57:58.000 --> 00:58:07.000 Then on the right is an example of that quad root example applied to Southern California. 00:58:07.000 --> 00:58:13.000 Actually, all the plots in this presentation come from the built-in plotting from this code. 00:58:13.000 --> 00:58:29.000 So we've revamped how uncertainties are added in the code, but it's still a large portion of this is still trying to convince yourself that the results and uncertainties are dependable. One way you can do that is with a checkerboard test. So using the exact same catalog that you used to produce your stress versions. 00:58:29.000 --> 00:58:40.000 We can automatically determine the sort of checkerboard tests. Here's a plot of results from one example showing the SHmax measurements and the corresponding errors. 00:58:40.000 --> 00:58:51.000 On the other side of things, you can also do a gradient test, right? So you have these kind of two end members of a synthetic test to give you a confidence. Results are dependable. 00:58:51.000 --> 00:58:57.000 We can also look at the impact of damping. So here's a similar example. Now damping is turned off. 00:58:57.000 --> 00:59:03.000 Chop out this square in the middle. So if you don't have earthquakes, you can't, you don't use stress inversions using them. 00:59:03.000 --> 00:59:12.000 But through the magic of damping, we can produce estimates for what those values are, which is a new feature that we've added to this code. 00:59:12.000 --> 00:59:17.000 So a large portion of this work has been designed to make this code as efficient as possible. 00:59:17.000 --> 00:59:22.000 So with pySATSI it's very trivial to run this in parallel. 00:59:22.000 --> 00:59:30.000 So here's some examples of relative runtimes applied to three different examples, kind of increasing order of complexity and size. 00:59:30.000 --> 00:59:36.000 Just running it on my laptop, that running PySATSI in parallel is about five times faster than running it in serial. 00:59:36.000 --> 00:59:40.000 Compared to the old SATC algorithm, it's about an order of magnitude faster. 00:59:40.000 --> 00:59:46.000 So our algorithm scales quite well. I mean, it's a large catalogs. 00:59:46.000 --> 00:59:52.000 Well, SATSI starts to explode when you start dealing with some pretty large problems. 00:59:52.000 --> 01:00:12.000 So the code is freely available online. It's all open source. If you ever need to compute stress inversions, I hope you'll give it a try. All right. Thank you. 01:00:12.000 --> 01:00:27.000 Hi, I'm John Rumble from UC Davis. Happy to be here talking about earthquake nowcasting and machine learning nowcasting is a commonly used in economics, finance, and letter. And so we've adapted it for earthquakes. 01:00:27.000 --> 01:00:42.000 So the basic method is shown here. This particular case in continuous time, the two parameter filter is applied to the monthly seismicity rate data, and then the parameters of the filter are optimized using a receiver operating characteristic. 01:00:42.000 --> 01:01:02.000 Signal is defined. That is a threshold value of the nowcast curve. An event is defined, the occurrence of a large earthquake magnitude 6.75 within the next three years in this case, and all possible cases are considered true positive, true negative, false positive, false negative. This is a complaint that 01:01:02.000 --> 01:01:12.000 Dave Jackson and Jan Kagan used to say about most earthquake prediction schemes, they didn't consider all four cases. The ROC curve is constructed. The skill is the area under the curve, 01:01:12.000 --> 01:01:17.000 and then the positive predictive value, which is the forecast of a large event within a future time window. 01:01:17.000 --> 01:01:22.000 Is found and then we apply that to Southern California and to the state of California. 01:01:22.000 --> 01:01:28.000 This is the workflow on the left. Upper left is the data plotted spatially. 01:01:28.000 --> 01:01:37.000 In the middle, here is the monthly rate of small earthquakes. This blue line here is an exponential moving average. 01:01:37.000 --> 01:01:56.000 And you can see that activation here is at the top and quiescence is at the bottom. This curve in the middle, in the middle of this is the time varying mean seismicity rate so we have to correct for that. That's the other parameter in the two parameter filter. When we take this blue curve and invert it, you get 01:01:56.000 --> 01:02:01.000 turn it upside down, you got this curve, which is the now cast curve. Pliesen is at the top. 01:02:01.000 --> 01:02:09.000 Activations at the bottom. This is the ROC curve, the red curve here and this is the optimization diagram. 01:02:09.000 --> 01:02:16.000 So now we can, let's see, what I'm going to show you, 01:02:16.000 --> 01:02:26.000 this is the backwards here this is the backwards details of the constructing the ROC curve. I won't go into any details. You assume a threshold. 01:02:26.000 --> 01:02:33.000 And you classify all the points, whether they're above or below this curve, and whether a large earthquake occurs in this future time window. 01:02:33.000 --> 01:02:39.000 So there's four possible cases there. Again, I'm not going to go into that. You can look at the slides in detail. 01:02:39.000 --> 01:02:43.000 So I'm going to show you a movie, actually potentially two movies if we have time. 01:02:43.000 --> 01:02:51.000 The red dot is the current state. These vertical lines here are the major earthquakes that occurred in the region of Southern California here. 01:02:51.000 --> 01:02:57.000 The red lines are the magnitude seven earthquakes. The current date is shown here. Middle curve, this hockey stick curve. 01:02:57.000 --> 01:03:03.000 Is the probability of a major earthquake. See, it goes rapidly upward when you get to this point here. 01:03:03.000 --> 01:03:10.000 On the right hand is a traffic light alerting system. On the right hand is a plot of where the earthquakes are likely to occur. 01:03:10.000 --> 01:03:24.000 Based upon agreeing algorithm. There are the circles here, they appear three years before the earthquake actually occur and so you can see where they will occur during the next three years or so. 01:03:24.000 --> 01:03:30.000 [You can play this movie, please. There we go.] 01:03:30.000 --> 01:03:42.000 So it starts off in 1970. And we're heading up towards the earthquake in 1979, which will occur over here somewhere. 01:03:42.000 --> 01:04:02.000 And then that occurs, a big drop through the aftershocks Here, and then we're heading up towards the Loma Prieta earthquake. So you see a red circle appear up here, the big earthquake Excuse me. And there is a slight brightening of the region. And then the red circle appears there. 01:04:02.000 --> 01:04:09.000 And then we get the aftershock bloom, so to speak. And here's the Landers earthquake coming up just occurred. 01:04:09.000 --> 01:04:13.000 And then we're going to have the Hector Mine earthquake over here. 01:04:13.000 --> 01:04:26.000 In this region, you'll see the circle appear. Things get a little bit brighter And then the earthquake occurs. And then we keep going until we get to the Miracucapah earthquake in 2010. 01:04:26.000 --> 01:04:40.000 And that's coming up shortly. So this is obviously meant to be like a stress accumulation and release curve that you see here. So you can actually see this earthquake process happen so that occurred and then we're going towards the Ridgecrest earthquake 01:04:40.000 --> 01:04:52.000 and the interesting on these traffic light indicator lights show that what's actually going on sort of alert for predictive purposes for communication purposes. 01:04:52.000 --> 01:05:03.000 And then we are up to the current state and we're just now at the point where we're starting to move up the hockey stick curve in the rapidly accumulating phase. 01:05:03.000 --> 01:05:15.000 The thing that's a little worrisome is down here at the intersection of the Garlock fault and the San Andreas fault, this tongue of high probability appears to be reaching down from the Ridgecrest area and that will worries me a little bit. 01:05:15.000 --> 01:05:22.000 Next curve. Let's see, I want to go to the California region. 01:05:22.000 --> 01:05:30.000 And let's see, can we play that one. 01:05:30.000 --> 01:05:38.000 Actually, if you can skip ahead towards the end, I just want to show the Mendocino earthquake. 01:05:38.000 --> 01:05:48.000 Because the Mendocene earthquake here will appear obviously up in this region and you can see some cursory activation going on there, which is sort of interesting. 01:05:48.000 --> 01:06:00.000 So that's the Ridgecrest earthquake and then that's The Mendocino earthquake is going to occur right there. 01:06:00.000 --> 01:06:16.000 So that's the kind of thing we're doing here. So basically to summarize, this is a way of looking at in real-time the increase in probability of major earthquakes occurring in California. 01:06:16.000 --> 01:06:34.000 And the next thing we're doing, by the way, is building something called QuakeG PT, which is a science transform based neural network based method for using training a model based on ETAS type beta stochastic simulations data and then applying it to these kind of curves and predicting beyond the end of the now cast curve. 01:06:34.000 --> 01:06:43.000 So thank you. 01:06:43.000 --> 01:06:57.000 Hello. So my name is Lei Li. Today I'm going to share our recent observations on the dynamic triggering the Geysers Geothermal Field by two recent larger earthquakes in California. 01:06:57.000 --> 01:07:09.000 So as you can see the Mendocino and Ridgecrest events are located in the northwest and the southeast parts of the Geysers Geothermal Field respectively. 01:07:09.000 --> 01:07:30.000 And they're the local network named BG for monitoring [indiscernible] seismicity monitoring So this is the outline and first I will give a brief introduction of the guidance. So it has been under production for over 60 years and left the figure shows monthly injection and production data. 01:07:30.000 --> 01:07:36.000 And on the right you choose the corresponding production and injection world distributions. 01:07:36.000 --> 01:07:42.000 And we have obtained an enhanced catalog aided by deep learning algorithms. 01:07:42.000 --> 01:08:06.000 And here shows the distribution of numbers and injection and production data volumes and we can find there the positive correlation between the even numbers the injection volumes. And we have also obtained more than two times even numbers compared with the official Northern California catalog. 01:08:06.000 --> 01:08:13.000 So let's see what happened during the occurrence of the two large earthquakes in 2024 and 2019. 01:08:13.000 --> 01:08:30.000 So here we focus on 10 day long micro-seismicity sequence and we can see a clearly increase in both the event numbers and magnitudes and especially for the Mendocino scenario we can find the dynamic trigger event with local magnitude 4.3 01:08:30.000 --> 01:08:41.000 And we also estimated the beta-value and d-value it shows large value which is larger than the threshold of dynamic triggering. 01:08:41.000 --> 01:08:57.000 And so this is a spatial distribution of the events during the 10 days. And we also calculate the seismic source mechanism of the three largest events in Mendocino case, it shows different forwarding types and orientations. 01:08:57.000 --> 01:09:12.000 And we also find that the most events in the northwest part just occurred around the boundaries of the velocity anomalies and this is also consistent with previously signal studies. 01:09:12.000 --> 01:09:17.000 So we also checked the waveforms to further validate the dynamic triggering. 01:09:17.000 --> 01:09:32.000 The left figure shows one hour reference waveform data so from top to bottom are the raw waveforms and the fitted waveforms and also the spectrogram those raw waveforms. 01:09:32.000 --> 01:09:54.000 So the red figure shows the one-hour waveforms for the Ridgecrest foreshock, so we did not observe dynamic trigger here since this atmosphere rate is normal and they are around 10 events within the hour within an hour. So during the occurrence of the two largest weeks, we can clearly see very obvious dynamic triggering and 01:09:54.000 --> 01:10:01.000 the event numbers during the last hours increased to more than 300 in the 200 respectively. 01:10:01.000 --> 01:10:06.000 So this is a very evidence for dynamic triggering. As mentioned before 01:10:06.000 --> 01:10:28.000 the Mendocino earthquake also triggered a magnitude 4.3 event within three minutes and from the rescaled RT plot, we can find the cluster over here, left the bottom corner, it corresponds to the aftershock sequence of the magnitude 4.3 event 01:10:28.000 --> 01:10:50.000 And for the Ridgecrest event we did not observe any [indiscernible[ cluttering. So we also estimated the estimated dynamic stress and we can find they are around 40-30 keeper several keeper dynamic stress can indeed trigger earthquakes. 01:10:50.000 --> 01:11:15.000 So the concluding remarks that both the 2024 and 2019 large earthquakes dynamic triggered earthquakes in the Geysers Geothermal Field, and the dynamic stress yet instant and short impact but can trigger moderate earthquakes preferentially oriented a force. And just have a second micro-seismicity is still dominated by local GFMO operations and the stress perturbations 01:11:15.000 --> 01:11:28.000 And next step may be to further investigate the potential trigger mechanisms to investigate the impact of the dynamic stress on shear stress and or pore pressures. 01:11:28.000 --> 01:11:47.000 Okay, that's the end of my talk. Thank you. Thank you for your attention. 01:11:47.000 --> 01:12:02.000 [Just lots of slides. Here we go.] Hi, everyone. I'm here on behalf of colleagues across the Bay Area to invite scientists and others here today to participate in Earthquake Country Alliance Bay Area(ECA). 01:12:02.000 --> 01:12:11.000 The ECA is statewide. With regions in Southern California, Central Coast, Bay Area, and also the Redwood Coast Tsunami Work Group. 01:12:11.000 --> 01:12:17.000 Big shout out to a lot of the leadership provided by Laurie Dangler and others there. 01:12:17.000 --> 01:12:22.000 Especially on tsunami matters that we've been extending across the state. 01:12:22.000 --> 01:12:37.000 The ECA-Bay Area along with all the regions, coordinate statewide activities that, of course, ShakeOut in October tsunami preparedness week, along with the state in March. 01:12:37.000 --> 01:12:40.000 Coming up the last week of March, the 22nd to the 30th. 01:12:40.000 --> 01:12:50.000 And a lot of other activities, which you can see on our calendar. We also have statewide committees that you could be a part of or if you know someone interested in these particular sectors. 01:12:50.000 --> 01:12:55.000 They want to develop resources and activities, be at workshops or conferences. 01:12:55.000 --> 01:13:01.000 And share earthquake and tsunami resources. These are groups you could be a part of. 01:13:01.000 --> 01:13:11.000 And in our regional alliances have quarterly workshops have people who go out to a lot of local events, a lot of other coordination. 01:13:11.000 --> 01:13:20.000 And ECA Bay Area is led by Jeff Airth from San Francisco International Airport, Ray Bonilla from Elevance Health in the city of Hayward. 01:13:20.000 --> 01:13:38.000 Is on their city council and Dena Gunning from San Mateo County and Susan Garcia involved and many others. And there are some up in positions so If you're interested in getting involved with local leadership and an earthquake and tsunami resilience. This is one way you could be a part of the coordinating committee. 01:13:38.000 --> 01:13:53.000 We've also been having presenters at recent workshops. Some of which may be on right now. And we thank you for your presentations and partnerships, Mike Oskin, Austin Elliott. 01:13:53.000 --> 01:14:04.000 Grace Parker, Roland Bergman, Savas Marco, Richard Allen, Anne Rosinski over the last year. And we are looking for you to join us and present in our upcoming workshops. 01:14:04.000 --> 01:14:07.000 Next is going to be on February 25th in Marin County. 01:14:07.000 --> 01:14:18.000 And it's going to be focused on lessons learned from the December 5th tsunami in terms of the messaging we're going to have presentations from CGS and Cal OES there. 01:14:18.000 --> 01:14:30.000 But future workshops coming up, we kind of move around the region. So we'll be looking for you if you're interested and sharing information with the emergency managers, general public, this is a way of doing that. 01:14:30.000 --> 01:14:43.000 Also, a lot of materials available for your use at earthquakecountry.org/languages and 16 languages. So a lot of good guidance here and free to everyone to download. 01:14:43.000 --> 01:14:57.000 We're soon going to have a way that you can order printed copies of these. So if you're looking for materials to give to you know, your children in school or to other groups, this is where you can look, the earthquakecountry.org website. 01:14:57.000 --> 01:15:10.000 If you're doing a presentation or have a request to do a presentation that includes preparedness guidance. We can provide slides to you. We can provide a handout that people can fill out during that presentation. 01:15:10.000 --> 01:15:20.000 Reach out to us as well. We do have an events bureau that goes to local events and does presentations. So if you're interested in being involved 01:15:20.000 --> 01:15:27.000 and doing more public presentations. We get a lot of requests. And so we're looking for your involvement there too. 01:15:27.000 --> 01:15:42.000 This year's shakeouts on October 16th. And each year we have a big media event in 2024, it was at UC Berkeley. But we are looking for a Bay Area location in 2025. 01:15:42.000 --> 01:15:52.000 Really for a great drill activity that we can highlight and bring the media to and have early morning live broadcasts with the earthquake simulator and the whole thing. 01:15:52.000 --> 01:15:58.000 So if you're aware of any plans for doing something special for Shakeout this year 01:15:58.000 --> 01:16:05.000 you can also let us know. And also, we've been supporting the HayWired Scenario Exercise Toolkit; 01:16:05.000 --> 01:16:26.000 30 themes that have been identified; 45 ideas that have led to now 20, soon 20 facilatator tool sets that we're training local governments and businesses on how to use and we are looking for some host locations or groups for 01:16:26.000 --> 01:16:34.000 trainings for which we have funding for this year. So all in all, you can email me at benthian@usc.edu. 01:16:34.000 --> 01:16:57.000 Or go to info@earthquakecountry.org to get involved with us in any way that you might see it is interesting to you. Thank you. 01:16:57.000 --> 01:17:08.000 Thanks, everyone, for sticking around for our wonderful thunder talks. Lunch is next and I'll hand it off to Sarah for any additional announcements. 01:17:08.000 --> 01:17:14.000 I just want to say we return at 1:15 unless, of course, you are speaking in the next session. So Brian, Allison, Jessie. 01:17:14.000 --> 01:17:19.000 Oz and Belle, if you could join us at 1PM to check on your mic and your video, that would be great. Otherwise, let's everyone get to lunch while we still can. 01:17:19.000 --> 01:17:29.000 Thanks.