WEBVTT 00:00:00.000 --> 00:00:01.420 I'm gonna start the recording now. 00:00:02.520 --> 00:00:02.960 Perfect. 00:00:04.350 --> 00:00:10.100 Thank you all for attending the Earthquake Science Seminar Weekly Seminar Series. 00:00:10.210 --> 00:00:15.920 If you're new, welcome, and if you would like to be added to our email distribution groups, please send us an email. 00:00:16.410 --> 00:00:21.940 Seminars are recorded and mostly all talks are posted on the USGS Earthquake Science Center website. 00:00:22.290 --> 00:00:27.100 Closed captioning can be turned on by clicking the CC icon on the "more" tab. 00:00:27.450 --> 00:00:30.260 Sorry, "more" tab on the three dots at the top of the page. 00:00:30.690 --> 00:00:32.930 Attendees, please mute your mics. 00:00:32.940 --> 00:00:43.090 Turn off your cameras until the Q&A session at the end of the talk, submit your questions via the chat at any time, or wait to turn on your camera and ask your questions. 00:00:43.100 --> 00:00:48.060 during the Q&A session, uh specific announcements for today. 00:00:48.470 --> 00:00:50.360 Next, I think this is next week. 00:00:50.370 --> 00:00:50.720 Let's see. 00:00:50.730 --> 00:00:56.380 So that earthquake science seminar, all hands meeting is July 27th at 11:00 AM. 00:00:56.770 --> 00:01:03.020 See your inbox for the link and the additional information. If you're hoping to attend SCEC, this is a big one. 00:01:03.030 --> 00:01:10.890 The deadline to sign up internally is today, July 19th, so please check your email for logistics on that. 00:01:11.030 --> 00:01:13.210 Sign up if you plan to go to SCEC. 00:01:13.220 --> 00:01:16.710 Today's the deadline, and let's see. 00:01:16.720 --> 00:01:24.720 So, Thursday, as is usual this summer, the weekly intern meeting is at 1:00 PM and Not Ready for Prime time 00:01:24.730 --> 00:01:26.670 is at 3:00 PM in person only. 00:01:27.180 --> 00:01:41.850 We are still organizing location for that because we are not in the Yosemite Room this week, but we will keep you posted, so please stay tuned. With that, I will turn the introduction over to Shana and she will introduce our speaker for today. 00:01:43.830 --> 00:02:00.060 OK, so it's my pleasure to introduce today's Itzhak Lior got his bachelors in physics and geology from the Hebrew University of Jerusalem and his Masters and PhD in Seismology, both from Tel Aviv University working with Alan Zipf. During his PhD 00:02:00.070 --> 00:02:05.000 he was a visitor in the earthquake seismology group at Stanford, where I had the pleasure of making his acquaintance. 00:02:05.330 --> 00:02:19.530 Itzhak followed his PhD with the postdoctoral fellowship at Joe Azure in France, working on the use of ocean bottom fiber optic networks for earthquake source characterization with Pablo Ampuero, Anthony Slayden and Deon Rivey. 00:02:20.040 --> 00:02:37.450 Starting in 2021, he returned to the Hebrew University of Jerusalem as senior lecture, where he heads the seismology group. Itzhak, has worked broadly in the field of source size, modify and more recently, and leveraging distributed acoustic sensing and novel ways for a range of applications, including early warning, source studies and imaging. 00:02:37.600 --> 00:02:41.320 So without further ado, I'll let Itzhak take it away on this very exciting topic. 00:02:43.570 --> 00:02:43.850 Great. 00:02:43.860 --> 00:02:44.780 Thank you very much, Shanna. 00:02:47.390 --> 00:02:47.740 [noise] 00:02:49.350 --> 00:02:52.170 So alright. 00:02:51.830 --> 00:02:52.290 [noise] 00:02:55.220 --> 00:02:56.600 Can you see the screen and the pointer? 00:02:58.300 --> 00:02:58.430 Yes. 00:02:58.480 --> 00:02:59.100 Yes, we can. 00:02:59.980 --> 00:03:00.390 Thanks. 00:03:00.400 --> 00:03:00.650 Great. 00:03:01.560 --> 00:03:01.890 Yeah. 00:03:01.900 --> 00:03:09.890 So thank you for the opportunity to present my recent results here and I'm going to talk about how we can use distributed acoustic sensing for earthquake. 00:03:09.940 --> 00:03:19.510 early warning. I'm only focusing on this talk on 92 estimation and ground motion prediction. Leaving detection and location for future talks. 00:03:20.580 --> 00:03:22.970 This is the type of data that we're working with. 00:03:22.980 --> 00:03:27.540 These are 2D arrays or matrixes where we have distance along the fiber as a function of time. 00:03:28.540 --> 00:03:36.750 This specifically is an earthquake recorded near the Sea of Galilee in Israel using a 34 kilometer long fiber. 00:03:37.080 --> 00:03:43.990 And we can clearly see the P wave S wave, a very complex and vivid image of the seismic wave field. 00:03:44.250 --> 00:03:50.700 And these are the types of images that we can use to better our early warning system using distributed acoustic sensing. 00:03:51.770 --> 00:03:55.640 The results are already published in a paper in scientific reports earlier this year. 00:03:55.650 --> 00:04:06.960 if you're interested in learning more about the methods, I'm going to present. So, the basic procedures that we need to follow to produce earthquake early warnings are these four. 00:04:07.390 --> 00:04:12.840 First of all, we need to detect the earthquake and discriminate earthquakes from other local noise sources. 00:04:13.270 --> 00:04:20.110 we need to determine the earthquake location, epicenter or hypocenter, and then we need to quantify the size of the earthquake, 00:04:20.120 --> 00:04:23.300 the magnitude and possibly the stress job into one. 00:04:23.310 --> 00:04:43.790 Once we detected, and we are sure that it's an earthquake, we know the location of the earthquake and we know it's size, we can go ahead and predict P ground shaking intensities to further locations. Based on these P grand motion and predictions we can decide whether or not, and if and where we should issue early warning alerts. 00:04:44.180 --> 00:04:50.540 So, the end goal or the final product of an early warning system is whether an alert is issued or not. 00:04:50.820 --> 00:04:56.500 The end user of a warning system doesn't care about the first, doesn't care about any of these procedures. 00:04:56.510 --> 00:05:00.460 They only cares about whether he needs to take mitigation actions or not. 00:05:01.540 --> 00:05:07.460 So again, in this talk, I'm only going to present a method to do these two with distributed acoustic sensing. 00:05:09.090 --> 00:05:13.770 The classic approach to early warning relies on seismometers that are mostly located on land. 00:05:14.700 --> 00:05:18.230 There are few limitations to using seismometers for early warning. 00:05:18.660 --> 00:05:22.710 The first one is we require typically 4 detections to issue warning. 00:05:23.320 --> 00:05:26.400 We need to avoid false detections. 00:05:27.870 --> 00:05:41.050 This is especially problematic in urban environments where we have lots of anthropogenic noises and in the regions where the network is sparse and it takes a lot of time for the seismic waves to reach the fourth station. 00:05:41.750 --> 00:05:54.500 Another problem relating to sparse networks is that at the edge of the network we usually have to wait for the waves to reach the first station, and this is an example from the Tohoku- oki earthquake that occurred offshore Japan. 00:05:54.890 --> 00:06:07.720 It took the waves emitted from the earthquake roughly 20 seconds to reach the first stations on the Japanese coastline, and these are 20 seconds that could have been spared if we would have more censored near the epicenter. 00:06:08.930 --> 00:06:15.910 So again, if we would have offshore sensors here, the people of Japan would have had an additional 20 seconds to take mitigation actions. 00:06:16.580 --> 00:06:25.290 So this is another problem, and we have telemetry delays because there are seismic sensors that are deploying the inter-station spacings of 10s of kilometers. 00:06:25.620 --> 00:06:37.530 We need data to transfer from these stations to some local hub to analyze the data and we have another delay because of data telemetry and DAS can overcome most of these issues. 00:06:39.570 --> 00:06:42.440 This is an example of the solution that they came up with in Japan. 00:06:42.450 --> 00:07:00.040 They installed a very expensive, very large network of ocean bottom seismometers called S-net. This solution is very expensive to both install and maintain and it's not implemented in many regions around the world and thus can substitute these types of solutions at a low cost approach. 00:07:01.790 --> 00:07:03.340 So a few words about DAS. 00:07:03.490 --> 00:07:11.940 I guess that by now I guess all of you heard at least one talk about DAS and have some idea on what DAS is and what are the capabilities. 00:07:11.950 --> 00:07:27.420 So, I'm just going to say a few words about the method and how the data looks like so distributed acoustic sensing allows us to turn any optical fiber into a larger and array of seismic sensors with measurements every roughly 10 meters along 10s of kilometers long fibers. 00:07:27.790 --> 00:07:48.220 So we can use fibers that we seismologists install for our purposes when we can use fibers that are deployed for other purposes. We can use those on land, offshore fibers deployed by telecommunication companies as long as we have one free strand, a dark fiber we can plug in this interrogated unit and get seismic measurements along the fiber. 00:07:48.730 --> 00:08:00.560 So the standard tangram interrogated unit consists of a laser source, data storage and processing computer. The way that it works, the interrogator sends laser pulses into the fiber. 00:08:00.800 --> 00:08:0. 7.470 and because the fiber is not homogeneous, we have many impurities heterogeneities basically scattering points inside the fiber. 00:08:08.090 --> 00:08:08.630 Umm. 00:08:09.010 --> 00:08:12.840 And the incoming photons interact with these trageneities. 00:08:13.170 --> 00:08:17.100 Some of the light is backscattered via Rayleigh backscattering mechanism. 00:08:17.570 --> 00:08:22.080 So the interrogator sends in laser pulses and receives the backscattered light. 00:08:22.710 --> 00:08:27.280 When a seismic wave front sweeps across the fiber, it causes it to slightly stretch and compress. 00:08:28.260 --> 00:08:38.000 This stretching and compressing all the fiber actually changes the physical location of these scattering points inside the fiber, and it also changes the phase of the backscattered light. 00:08:38.420 --> 00:08:49.480 The interrogated knows how to track the phase changes at specific locations along the fiber and translate these phase changes into strain or strain rate measurements every 10 meters along the fiber. 00:08:50.320 --> 00:08:54.380 Now, if everything I just said wasn't very clear, it doesn't really matter. 00:08:54.390 --> 00:08:56.790 The important thing is to understand how the data looks like. 00:08:57.860 --> 00:09:07.670 We have these matrices where we have distance along the fiber as a function of time and each line here is a seismogram recorded as a specific location along the fiber. 00:09:08.160 --> 00:09:12.230 You can see here an earthquake recorded at a sedimentary basin offshore Greece. 00:09:12.760 --> 00:09:14.650 These are the direct P waves. 00:09:14.960 --> 00:09:20.670 Direct S waves because we're inside a sedimentary basin, we see the reflections from the edges of the basin. 00:09:20.680 --> 00:09:22.210 We can measure their velocities. 00:09:22.680 --> 00:09:36.930 We see different interactions of waves, plain waves coming from both ends of the basin, and we have a very rich image of the seismic wave field inside the basin and this is something that we didn't have before when we used standard seismometers. 00:09:37.940 --> 00:09:46.130 So again, if before we had just one seismogram where we have some amplitude, in this case strain rate as a function of time, we see P&S waves. 00:09:46.500 --> 00:09:58.990 Now we have 10s of thousands of signals spaced at 10 meters along optical fibers, and we can use this very rich image of the seismic wave field to do many wonderful things 00:09:59.000 --> 00:10:12.060 in seismology, including earthquake early warning. Another example is we can use DAS to decipher signals that if we would have measured them using seismometers we would basically have no idea what they are. 00:10:12.370 --> 00:10:20.700 So this is an example of a seismic trace recorded at one location along the fiber, and if I would show this to you, you wouldn't know what's going on here. 00:10:20.710 --> 00:10:29.900 You would see something that you would probably classify as noise, you've seen on a seismometer, but once we add the spatial dimension we can measure the velocity of the waves. 00:10:30.250 --> 00:10:44.310 We can perform different analysis like 2D Fourier transform, FK transform to get the velocity of these very coherent plane waves and study phenomenons that up to now we wouldn't be able to study using standard seismometers. 00:10:44.740 --> 00:10:54.070 By the way, these are P waves that were generated by earthquake that occurred in North Africa and they were recorded by an optical fiber offshore South of France. 00:10:54.320 --> 00:11:01.470 So these P waves propagated all the way through the Mediterranean and recorded by an offshore optical fiber. 00:11:03.660 --> 00:11:07.440 We can use DAS with existing infrastructure with existing fibers. 00:11:08.240 --> 00:11:13.790 When you look at the deployment of optical fibers around the world, the image is very similar to what we have here. 00:11:14.240 --> 00:11:17.390 Many regions have abundance of optical fibers. 00:11:17.400 --> 00:11:31.970 For example, what we see here in Chile, many fibers with many landing points, different locations along the Chilean coastline, and we were able to use one of these fibers for measurements and I'm going to show results using these fiber and later in the talk. 00:11:33.030 --> 00:11:38.080 Another thing we can do is use existing infrastructure and existing interrogating units. 00:11:38.470 --> 00:11:49.340 Now this approach is used for seismology, but it's also used to monitor linear infrastructure like guest lines, pipelines, railroads and grid lines. 00:11:49.810 --> 00:12:01.040 And for example, in Israel the natural gas line company has optical fibers along most of its infrastructure and they also have several interrogator units already monitoring this infrastructure. 00:12:01.450 --> 00:12:06.610 So if we could collaborate with these types of companies, we would get seismic measurements using DAS 00:12:07.050 --> 00:12:12.140 basically free and do much more in seismology using this data. 00:12:14.530 --> 00:12:17.420 So how can DAS benefit earthquake early warning? 00:12:18.430 --> 00:12:22.940 As I said before, we can apply this to any optical fiber, including underwater optical fibers. 00:12:22.950 --> 00:12:24.240 We can also use boreholes. 00:12:24.550 --> 00:12:31.400 Basically getting us closer to earthquake hypocenters, we do know that most of the largest earthquakes on Earth occur underwater. 00:12:31.410 --> 00:12:34.820 I gave the example before from the Tohoku-oki earthquake. 00:12:35.150 --> 00:12:42.140 Many reasons around the world have their earthquakes offshore and they have to wait several seconds for this seismic waves to arrive. 00:12:42.450 --> 00:12:52.620 If we can harness these offshore fibers, we would improve the performance of early warning systems and basically give the population more time to prepare for intense shaking. 00:12:53.780 --> 00:12:59.670 We have continuous measurements in both time and space that can provide robust earthquake detection and location. 00:12:59.680 --> 00:13:08.130 These are still in progress by several research groups, and when we estimate the magnitude, we can average it over many stations, 00:13:08.170 --> 00:13:14.660 many virtual sensors along the fiber, and thus we minimize the effect of outliers and have a more reliable estimate of the magnitude. 00:13:15.460 --> 00:13:24.170 And because everything is obtained at the interrogator unit of the data, we don't have any telemetry delays, which is also a plus for cyber security. 00:13:24.180 --> 00:13:27.370 We don't need any information to be transferred wirelessly. 00:13:29.280 --> 00:13:31.650 We do have a few obstacles that we need to overcome. 00:13:32.160 --> 00:13:39.510 First of all, Das measures strains, while magnitude estimation, requires ground motions, displacement, velocities or accelerations. 00:13:39.520 --> 00:13:43.020 So we need to convert strains to ground motions in real time. 00:13:43.790 --> 00:13:54.760 Another problem is that DAS noise levels are basically mostly higher than those of seismometers, and they are frequency dependent, so this is something I'm going to present in a few slides. 00:13:55.370 --> 00:14:03.480 And what I think is the main difficulty at this point in time in the maturity of using DAS for seismology is that we don't have enough seismic data. 00:14:04.010 --> 00:14:06.470 We only have low to medium magnitude. 00:14:07.360 --> 00:14:11.830 We don't really have earthquakes that we would have wanted to issue alerts for. 00:14:12.380 --> 00:14:22.720 So having an empirical magnitude estimation approach is problematic in that sense, and here I'm going to show you a theoretical approach to magnitude estimation and ground motion prediction. 00:14:23.980 --> 00:14:28.610 So a few words about an empirical approach from a paper recently published by the Caltech Group. 00:14:29.000 --> 00:14:42.970 What they did is develop an empirical approach using the first 2 seconds of the P wave and the S wave and estimated the magnitude using all available channels and you can see here magnitude is a function of time. 00:14:43.200 --> 00:14:52.060 They got a really excellent estimate of the magnitude 4.88 in a very short time of just under 2 seconds. 00:14:53.080 --> 00:14:59.580 When we look at the entire data set that they use, they have a very good agreement between the predicted magnitude and catalog magnitude. 00:15:00.170 --> 00:15:02.270 So this system performs very well. 00:15:02.280 --> 00:15:09.550 The problem is that when using empirical approach, we're not sure how it will extrapolate to different datasets. 00:15:10.700 --> 00:15:12.410 This is the data that they used mainly for 00:15:12.420 --> 00:15:27.090 a function of distance and I roughly outlined here the region or the earthquakes for which we would want to issue earthquake early warning and you can see that there are very few data points inside this triangle. 00:15:27.460 --> 00:15:34.210 So having an empirical approach is a bit problematic in that sense, and then the rest of the talk I'm going to present a theoretical approach. 00:15:35.140 --> 00:15:38.910 So let's start with the conversion of strains to grand motions. 00:15:40.820 --> 00:15:44.050 This conversion is commonly achieved using the apparent phase velocity. 00:15:44.920 --> 00:15:55.040 We can write strings as the spatial derivative of displacement, and we can use the chain rule to decompose the U to the X to the UDT, to the X. 00:15:55.350 --> 00:16:04.120 The first term is ground velocity and the second term is the slowness along the fiber or one over the apparent phase velocity. 00:16:05.340 --> 00:16:12.650 So strains can be converted to ground velocities and strain rates can be converted to ground accelerations. 00:16:13.460 --> 00:16:26.400 This approach requires or assumes that we are dealing with a plain wave that has a very well defined velocity, and it requires a straight fiber segment in order to reliably estimate the velocity of displaying wave. 00:16:27.030 --> 00:16:35.390 But in a recent paper, we show that even in the presence of various interactions between seismic waves, this method still performs well. 00:16:38.920 --> 00:16:42.870 And velocity there is as a function of both time and space. 00:16:42.940 --> 00:16:46.030 When we look at it as a function of time, we have different seismic velocities. 00:16:46.040 --> 00:16:49.410 We get the P wave S wave surface waves, scattered waves. 00:16:49.420 --> 00:17:00.080 Each phase has its own different velocity, and if you want to convert strains to ground motions reliably, we need to take that into account and to get the velocity as a function of time. 00:17:01.200 --> 00:17:05.310 The velocity also changes as a function of distance along the flyer as a function of space. 00:17:05.710 --> 00:17:10.770 We see here an example of an earthquake recorded on a fiber offshore grease. 00:17:10.780 --> 00:17:13.480 So this is fiber that's very well coupled. 00:17:13.490 --> 00:17:17.990 They buried it with an ROV all along its length, and we can see that here. 00:17:18.300 --> 00:17:37.030 It's deployed inside the setting entry basin and here it's deployed over Hard Rock and we clearly see the difference when it's deployed over low velocity settlements, the amplitudes are very high when it's deployed over high velocity sediments, the amplitudes are very low and in addition you can see that the velocity is. 00:17:37.210 --> 00:17:48.460 So the velocities of the waves themselves change abruptly along the fibers, so we need to have an approach that can obtain the velocities of the waves as a function of both time and space. 00:17:49.690 --> 00:17:55.220 Here I'm using the slant stack transform and I'm going to briefly go over the method because it's a bit technical. 00:17:55.650 --> 00:17:58.980 We use a short fiber segment we have here. 00:17:58.990 --> 00:18:13.600 The axes are flipped, so we have distance along the fiber as a function of time, and for this short fiber segment, at each time instant we check different slowness or different apparent velocities and calculate the semblance for these different velocities. 00:18:15.280 --> 00:18:23.990 What we see here is the slope or slowness as a function of time, and this actually gives us a time series of the velocity of these waves. 00:18:24.220 --> 00:18:27.930 So this is an example from a small earthquake recorded offshore frames. 00:18:28.320 --> 00:18:44.650 We see here that direct survivals immediately followed by scattered and surface waves, and we can reliably see that for the direct arrivals the velocity is quite higher two kilometers per second, and when we get these surface waves the velocity decreases. 00:18:44.660 --> 00:18:59.190 So this approach uh provides this time series of apparent velocity in using this simple expression, we can converge strains like we see here to ground velocities using the apparent velocity along the fiber. 00:19:00.600 --> 00:19:10.420 Now in real time we have a bit of a problem here because we can only see half of the wavefront, but not going too much into detail. 00:19:10.500 --> 00:19:22.480 Just looking at the bottom panel, we can see that even if you perform this procedure, only half of the wavefront the conversion quality is maintained because the ability to estimate the apparent velocity is the same. 00:19:22.840 --> 00:19:26.610 Here in black we see the slowness for the non real time approach. 00:19:26.620 --> 00:19:34.110 Looking at the full wavelength, sorry, the full wavefront and in red we see the real time approach where we only look at half of the wavefront. 00:19:34.960 --> 00:19:46.570 The slowness matches very well for most of the season, so we have an approach to convert in real time strains to ground motions, specifically strain rates to ground accelerations. 00:19:47.580 --> 00:19:49.670 Now let's talk about the high noise levels. 00:19:49.680 --> 00:19:55.990 But before, I just want to present the model that I used to derive the theoretical expression for the magnitude. 00:19:56.920 --> 00:20:00.190 So, like many studies before me, I'm using the Omega square model. 00:20:00.200 --> 00:20:02.060 We can see it here for ground displacements. 00:20:02.200 --> 00:20:06.060 It describes the far field radiation of either P or S waves. 00:20:06.500 --> 00:20:08.730 We have two parameters controlling the. 00:20:09.000 --> 00:20:21.980 Sorry when telling the model capital Omega Gamma that corresponds with the seismic momentum distance, and if not the corner frequency above which the Spectra falls off at the rate of Omega to the power of minus two. 00:20:23.360 --> 00:20:34.930 So using this model in a previous study together with alongside, we derived theoretical expression and approximations to understand how different source parameters. 00:20:35.110 --> 00:20:39.070 Specifically, these two fundamental parameters, this nice big moment and stress job. 00:20:39.410 --> 00:20:47.690 How they affect the different ground motion measures so we have PDP, GV and PGA ground displacements, velocities and accelerations. 00:20:47.920 --> 00:20:59.170 And how they are affected by these different parameters so we can see that for displacements, the seismic moment has a very high power of 5 / 6 while the stress drop has a very low power. 00:20:59.620 --> 00:21:05.060 And for accelerations, the seismic moment has a low power of 1/3 and the stress drop the power of 2/3. 00:21:06.320 --> 00:21:18.920 If you would want to estimate magnitudes in real time, it would be easier to use PGD because PGD or PD some form of displacement proxy is a better predictor of the seismic moment because it's better correlated with it. 00:21:19.970 --> 00:21:29.020 In contrast, if we would use accelerations, the correlation with seismic moment is not as good and the estimate would also be not as good. 00:21:30.950 --> 00:21:35.320 But we have a problem when we use does because noise levels are frequency dependent. 00:21:35.710 --> 00:21:41.150 Here I'm showing an earthquake recorded offshore Greece and magnitude 3.6 on the right. 00:21:41.160 --> 00:21:51.450 I'm showing the Spectra, the full bandwidth of the Spectra and in blue I'm showing the spectrum between bandpass between oh .06 and 10 Hertz. 00:21:51.880 --> 00:21:57.670 On the left, I'm showing the time series again, filtered between 1.00 point 06 and 10 Hertz. 00:21:58.830 --> 00:22:03.740 This panel here is for displacement which are proportional to the integral of strains. 00:22:04.350 --> 00:22:13.360 We can see that for low frequencies and they are dominated by instrumental noise, the noise decreases as F to the power of minus two. 00:22:13.670 --> 00:22:19.430 So if we would use this measure for magnitude estimation, it would be overestimated. 00:22:20.650 --> 00:22:35.830 Same goes for velocities the the image is slightly better because noise levels decay at a rate of F to the power of minus one, but still a lot of low frequency noise that if we would use that to estimate the magnitude we would get magnitude over estimation. 00:22:37.170 --> 00:22:56.980 Uh, just to worried about magnitude saturation, if you would want to estimate the magnitudes of the all possible earthquakes, including the most the largest earthquakes there are, we would want to include as much of the low frequency signal as possible, and to do that we wouldn't want to apply any hypersphere to or apply a minimal hypersphere. 00:22:58.040 --> 00:23:03.110 And to do that, we're forced to use ground accelerations, which are proportional to strain rates. 00:23:03.460 --> 00:23:14.850 We can see here that for low frequencies the noise levels are flat, so we don't have to do anything with the low frequencies at the high frequencies, we see that they increase as frequency to the power of minus one. 00:23:14.900 --> 00:23:18.580 Sorry to the power of 1 so we have to apply some low pass filter. 00:23:19.810 --> 00:23:28.020 Umm, we do see that when we use accelerations, the correlation with the systemic moment is not as good as opposed to using displacement. 00:23:28.450 --> 00:23:31.190 But that is just what we have to do with. 00:23:32.020 --> 00:23:38.380 So as I said, you strain rates that are proportional to X relations and apply a low pass filter to remove the high frequency noise. 00:23:40.050 --> 00:23:42.120 So we dealt with the noise levels. 00:23:42.510 --> 00:23:44.720 Now let's talk about the derivation of the approach. 00:23:46.130 --> 00:23:53.810 So again, using the Omega Square model here I'm showing it for accelerations along with a decaying exponent that models high frequency attenuation. 00:23:54.700 --> 00:23:56.340 We calculate the RMS. 00:23:56.580 --> 00:24:01.990 This is definition for the Rams in the frequency domain, and now we own the integrate between 0. 00:24:02.260 --> 00:24:14.510 So keeping the low frequency portion intact and some maximum frequency that UM is used to get rid of the high frequency instrumental noise. 00:24:14.980 --> 00:24:21.710 In this case, the high frequency is set to be 5 Hertz, so the final model that we used is here. 00:24:21.940 --> 00:24:26.370 The solid curve that Annulation and a cut off at 5 Hertz. 00:24:27.750 --> 00:24:29.700 This integral has an exact solution. 00:24:29.710 --> 00:24:37.190 So what we did is solve the integral, derive an approximation in terms of the seismic moment and stress job, and we see it here again. 00:24:37.200 --> 00:24:47.530 The seismic moment has a power of 1/3 stress job of power of 2/3 and this relatively complicated equation has an analytic solution for the seismic moment and we can see it here. 00:24:48.080 --> 00:24:49.840 So it's relatively simple solution. 00:24:50.690 --> 00:24:53.840 It's analytics, so it's very fast to compute in real time. 00:24:54.250 --> 00:24:57.420 It's a function of XAB&C. 00:24:57.430 --> 00:25:03.160 We can see that functions are not really written here, but they're really easy to calculate. 00:25:03.410 --> 00:25:08.580 And what I'm showing here in red are the parameters that actually change in real time in real time. 00:25:08.590 --> 00:25:14.010 We only change the distance to the earthquake because the location estimate is updated. 00:25:15.040 --> 00:25:27.400 We change the data interval because we start with two seconds going forward to 345 so on and we change the acceleration RMS that we measure because we record more data. 00:25:28.590 --> 00:25:29.100 Umm. 00:25:29.110 --> 00:25:31.260 As you can see here, the stress drop is kept constant. 00:25:33.630 --> 00:25:51.270 So we have the theoretical approach to estimate the magnitude and now as I said, because the stress drop is kept constant, it would be wise to understand how the knowledge of stress drop, or more accurately, lack of knowledge regarding the actual stress job of the earthquake effects or magnitude prediction estimates. 00:25:51.710 --> 00:25:56.410 So going a bit back, because we can only use ground accelerations, we only have one proxy. 00:25:56.700 --> 00:26:04.480 We can only solve 1 parameter and we only solve for the moment, so we need to understand how the stress drop effects our performance. 00:26:04.970 --> 00:26:10.870 So we ran a few synthetic tests calculating RMS based on theoretical earthquake Spectra. 00:26:11.820 --> 00:26:17.210 Putting these arms into the theoretical estimation approach for the seismic moment. 00:26:17.760 --> 00:26:26.200 And we did it for variety of magnitudes you can see here on the vertical axis, the residuals between predicted and synthetic magnitude. 00:26:26.210 --> 00:26:30.880 So above 0 we are overestimating magnitude and below 0 we are underestimating it. 00:26:31.900 --> 00:26:41.480 When the stress drop is known, you see that the residuals are very close to zero and the only shape that you see here is because of the approximations made in deriving this equation. 00:26:42.270 --> 00:27:00.340 However, where the spiritual fees let's go with the overestimated, the magnitude is underestimated by in this case, if the threshold sorry is overestimated by one order of magnitude, the magnitude is underestimated by 1.33 units and we can see the cause for this number here. 00:27:01.740 --> 00:27:07.950 So we can see that this approach is a poor proxy for the main field. 00:27:07.960 --> 00:27:24.190 We have poor magnitude estimation when we don't know the actual value of the stress job, but as I said at the beginning of the presentation, we are not really interested in determining the magnitude the end user over an early warning system doesn't care about the magnitude or the location of the earthquake, he only cares about the ground motions. 00:27:24.200 --> 00:27:33.340 He only cares about whether he needs to take mitigation actions or not, so let's see how these bias affects our ability to predict P ground shaking. 00:27:34.530 --> 00:27:39.460 Here I'm using the uh GMPS that I derived together with Alonzo in 2018. 00:27:39.930 --> 00:27:43.380 They are derived from the same theoretical framework, the Omega square model. 00:27:43.570 --> 00:27:49.960 We can see the same proportions as here PGA with seismic moment to a power of 1/3 stress job power of 2/3. 00:27:50.380 --> 00:28:02.540 We can see various parameters that go in like the shear wave velocity, the distance OK from the Bruin or Madariaga models and these constants that gather various media parameters. 00:28:03.950 --> 00:28:11.630 So using this GMP we have we can input both seismic moment and stress job and get PGV and PGA estimates. 00:28:13.000 --> 00:28:26.940 So I'm showing you similar figure now with the vertical axis are PGA residual so predicted PGA as opposed to true PGA when we know the stress drop the PGA prediction is very good. 00:28:27.750 --> 00:28:33.450 However, if we use an overestimated stress job, let's go with this one first. 00:28:33.460 --> 00:28:44.930 If we use an overestimated stress job by 1 / 1 order of magnitude and use the true synthetic magnitude, we would significantly overestimate PGA. 00:28:45.240 --> 00:28:50.580 You see here that you stress drop is overestimated by this, but the seismic moment is true. 00:28:51.130 --> 00:28:54.220 PGA would go high because of the stress job. 00:28:54.650 --> 00:29:02.380 However, if stress job is overestimated and as we saw before mainly to these underestimated, so basically the stress drop is high. 00:29:02.970 --> 00:29:18.380 The seismic moment is low and PGA estimates are roughly OK you can see here that PGA residuals are bounded at half an order of magnitude, which is relatively good for an early warning system. 00:29:18.390 --> 00:29:23.840 And I'm going to show later on with the actual data that this difference is actually lower. 00:29:25.650 --> 00:29:50.300 So going over the algorithm for each fiber, we identified several well coupled fiber segments and for each of these fiber segments, we did the following for each does channel inside the fiber segment, we apply the low pass filter of five Hertz converted strain rates to grand accelerations using the slant stack transform applied an additional low pass filter and calculated the acceleration arms. 00:29:51.040 --> 00:29:55.080 Then we average the acceleration rematch for each short fibre segment. 00:29:55.390 --> 00:30:03.080 When we do this averaging, it mitigates the effect of outliers and then then we have one arms estimate per fiber segment. 00:30:03.870 --> 00:30:09.230 Then we estimate the seismic moment for each fiber segment and predict PGA and PGV to further locations. 00:30:10.880 --> 00:30:23.940 So these are the parameter tuning which are not very crucial, and this you already saw before the parameters that actually vary in real time and the results I'm going to show you, it didn't do the location in real time, it took it from the catalog. 00:30:23.950 --> 00:30:27.590 So the distance is constant and taken from the catalog. 00:30:27.900 --> 00:30:33.190 T is the data interval and it changes starting from 2 seconds going forward in time and the acceleration. 00:30:33.200 --> 00:30:35.430 RMS is calculated in the time domain. 00:30:35.740 --> 00:30:42.330 We derived the model in the frequency domain, but we calculate the observed rooms in the time domain using this simple equation. 00:30:44.330 --> 00:30:59.160 We used 53 earthquakes magnitude 2 to 5.7 at distances of 17 to 365 kilometers using four different ocean bottom fibers 2IN Greece, one in Chile and one in France. 00:30:59.910 --> 00:31:03.860 Each of these fibers were measured using different interrogator unit. 00:31:03.870 --> 00:31:13.670 This one we used fabulous in this one we used Alcatel and here we used arrogant photonics and I'm going to show you the results are equivalent for these three interrogators. 00:31:14.470 --> 00:31:23.460 Again, just plotting the triangle that I showed you in the previous slide for the contact study, and again you see that the image is very similar. 00:31:23.470 --> 00:31:32.940 We hardly have any data points for which we would want to issue an alert, but again, I just want to emphasize that the approach I'm showing was derived theoretically. 00:31:32.950 --> 00:31:47.930 I didn't use any data to derive the approach and I'm only using data now to validate the approach and I'm using data from different regions, so this is summary of a workflow for one well coupled segment. 00:31:47.940 --> 00:31:50.270 So this is an earthquake in Chile. 00:31:50.760 --> 00:31:54.570 We see that the fiber segment is here very continuous. 00:31:54.900 --> 00:32:02.010 First arrival of the S wave and all the subsequent phases, we can estimate the slowness as a function of time. 00:32:02.280 --> 00:32:04.200 You can see it for the P wave and S wave. 00:32:04.210 --> 00:32:09.190 Then we can use this slowness to convert strain rates that you see them here in black. 00:32:09.560 --> 00:32:11.890 Two ground accelerations that are seen here in blue. 00:32:12.840 --> 00:32:14.440 You can see the difference here. 00:32:15.610 --> 00:32:33.780 Discarded the OR surface waves and then when we have ground accelerations, we can estimate the systemic moment you can see here the magnitude is a function of time starting two seconds after the P wave arrival, and we can see the different stress drop values gives significantly different magnitude estimates. 00:32:35.200 --> 00:32:43.620 But I'm going to show you in a few slides that like we show theoretically, magnitude estimation is not as important for grand motion prediction. 00:32:45.420 --> 00:32:53.790 So I'm going to show a few of these slides for several time instances, so this is 2 seconds from the first P wave arrival at the fiber. 00:32:54.360 --> 00:33:05.590 Here I'm showing real time magnitude as a function of catalog magnitude with a one to one line and here in the middle panel I'm showing PGV residuals at the bottom PGA residuals. 00:33:06.020 --> 00:33:27.750 On the left they are a function of catalog magnitude and on the right function of hyperspectral distance we have different markers for the different Richard Research regions or Chile, Greece and France and the way to read this legend is that we have 45 earthquakes in Chile, 742 PGA and PGV observations. 00:33:28.340 --> 00:33:35.810 The average residual 0, - 0.76 and the standard deviation is 1.14 and you see that these numbers improve when we go forward in time. 00:33:36.360 --> 00:33:46.780 So we estimated the seismic moment in the magnitude using does, but in the panels below, PGA and PGV were measured from standard seismometers on length. 00:33:47.660 --> 00:33:57.210 So in this validation, we estimated the main 2 dizzy offshore fibers and validated PGA and PGV predictions using small letters and accelerometers on length. 00:33:57.400 --> 00:34:00.530 So no conversion required for these bottom panels. 00:34:01.500 --> 00:34:02.960 Now I'm going to go forward. 00:34:02.970 --> 00:34:05.460 This is a 2 seconds going forward to four seconds. 00:34:05.470 --> 00:34:07.630 We've seen improvement in the magnitude. 00:34:07.640 --> 00:34:11.530 We also see improvements in the residuals for PGP and PGA forward. 00:34:11.540 --> 00:34:23.210 Again, for six seconds 10 seconds, you already see that the magnitudes are very well constrained for almost the entire data set, and the residuals are also already very close to zero. 00:34:23.250 --> 00:34:24.190 You can see them here, minus. 00:34:24.200 --> 00:34:26.450 Oh, point 2 for PGV minus. 00:34:26.460 --> 00:34:35.090 0.06 for PGA when you walk forward to 15 seconds, we still have a slight improvement minus 1.15 for Chile, for PGV and minus. 00:34:35.100 --> 00:34:38.410 0.02 for PGA Umm. 00:34:38.530 --> 00:34:42.180 So it's stopping hearing 15 seconds and showing this summary plot. 00:34:42.190 --> 00:34:44.240 So these are real time magnitudes. 00:34:44.250 --> 00:34:51.360 The function of catalog magnitude for 4:10 and 15 seconds and these are the panels you just saw for PGV and PGA. 00:34:51.950 --> 00:35:01.250 These were obtained using a stress drop of 10 megapascal and on the left we see what happens when we use one order of magnitude lower one megapascal. 00:35:01.760 --> 00:35:08.360 You can see that the larger earthquake does have an overestimation of the magnitude compared to what we see here. 00:35:09.680 --> 00:35:14.250 Let's look at it as a function of magnitude focusing on this earthquake. 00:35:14.260 --> 00:35:21.990 Here we hardly see a difference in PGV in PGA estimates when we use one megapascal or 10 megapascal. 00:35:22.680 --> 00:35:30.060 I'm going to show that in another plot, so here I'm only showing the largest earthquake we hand, which is the main code 5.7. 00:35:30.940 --> 00:35:36.290 This is the theoretical analysis we saw that the stress job has a significant effect on the main thread estimation. 00:35:36.600 --> 00:35:47.870 We also see it in real data when we use one megapascal and 10 megapascal, we get significantly different magnitude estimates and here are PGA and PGV estimates. 00:35:47.880 --> 00:36:01.370 The theoretical synthetic test that we did and actual observations you can see here that PGP residuals and PGA residuals show very little sensitivity to the value of the stress job, and there are virtually identical. 00:36:01.660 --> 00:36:10.150 You can see here for different colors correspond to the standard deviation of the estimates, and they are virtually overlapping. 00:36:10.920 --> 00:36:15.510 So we can say that mainly to estimates exhibit high sensitivity to the value of the stress. 00:36:15.680 --> 00:36:28.560 Yep, we saw it theoretically and empirically, but PGA and PGV exhibit low sensitivity to the stress job because the magnitude estimate and gramma motion predictions are derived here from the same framework. 00:36:29.850 --> 00:36:31.560 Umm, so I went a bit fast. 00:36:31.570 --> 00:36:35.920 This is actually the last uh figure I'm showing here on the left. 00:36:36.250 --> 00:36:42.030 An example from the fiber offshore Chile, a few earthquakes that occurred near the fiber. 00:36:42.540 --> 00:36:56.210 The color code here corresponds to the amount of time we had for each earthquake until the South waves reached the shoreline, and at this time point we measured the magnitude or provided real time estimate of the magnitude. 00:36:56.660 --> 00:37:06.110 So for each data point here you see the catalog magnitude and the real time magnitude and we have a very good agreement between the two, lower than half a magnitude unit. 00:37:06.480 --> 00:37:14.710 So by the time that S waves reach the Chilean shoreline, we already have a very reliable estimate of the magnitude to issue an early warning. 00:37:15.120 --> 00:37:20.500 Now we know that these are relatively small earthquakes, and for larger earthquakes it may. 00:37:20.510 --> 00:37:22.700 The magnitude may take a bit longer to evolve. 00:37:23.680 --> 00:37:37.850 So this issue again needs to be tested when we when we have a more observations on the right, we did another synthetic test to show how much additional warning time we would add by using this offshore fiber. 00:37:38.260 --> 00:37:53.040 So we tested different potential earthquake locations on this map, and for each earthquake location, we measured the theoretical P wave arrival time at the fiber compared to the P wave arrival time at the 4th closest station. 00:37:53.960 --> 00:37:55.940 And here we see this abstraction of the two. 00:37:56.830 --> 00:38:03.530 Uh, the regions marked here by red color or regions where we would have a positive time game when we use dust. 00:38:03.990 --> 00:38:05.450 So when we use the fiber you can see. 00:38:06.330 --> 00:38:08.580 A time gain of up to 25 seconds. 00:38:09.050 --> 00:38:22.290 So many additional seconds that could be used to provide warning for the people of Chile, we see that even on shore where the seismic network is sparse, we have a positive time gain of a few seconds. 00:38:23.400 --> 00:38:32.480 So this image really shows the benefit of using dust for offshore early warning, especially in regions where they don't have offshore observations. 00:38:34.920 --> 00:38:35.330 OK. 00:38:35.340 --> 00:38:42.440 So to conclude, I showed the theoretical approach for magnitude estimation and ground motion prediction using the Omega square model. 00:38:43.540 --> 00:38:50.100 The real strength of this approach is the self consistency between magnitude estimation and ground motion prediction. 00:38:50.180 --> 00:39:04.470 The fact that this case 2 wrongs actually make a right if we get the stress job wrong and we get as a consequence the magnitude wrong, we will still get big round checking relatively OK because this is a theoretical approach. 00:39:04.480 --> 00:39:06.210 It's geographically independent. 00:39:06.340 --> 00:39:14.180 I didn't use any data to derive the approach, just to validate the approach and I showed results from Chile, Greece and France. 00:39:14.310 --> 00:39:17.060 I also have results from Israel and it works equally well. 00:39:18.170 --> 00:39:25.080 We don't have, we don't expect to have magnitude situation because we include the low frequency portion as much as possible. 00:39:25.710 --> 00:39:30.700 This approach also allows for continuous magnitude and ground motion updates in real time. 00:39:32.020 --> 00:39:36.710 This is especially beneficial for large earthquakes with ruptured takes a few seconds to evolve. 00:39:36.980 --> 00:39:42.140 We're not stopping at 2 or 4 seconds and we can keep going as long as the earthquake keeps growing. 00:39:42.930 --> 00:39:49.580 You also see significant time gains for offshore earthquakes when we use an offshore fiber and when we use an existing fiber. 00:39:49.590 --> 00:39:55.580 This approach is low cost and ideal for developing countries that don't have a very well developed seismic network. 00:39:56.640 --> 00:40:09.380 The computational cost, they didn't really mention we talked about, but there are a few data reduction strategies that help us analyze these large volumes of data in real time and we can also harness parallel computing. 00:40:10.540 --> 00:40:12.370 Umm yeah, thank you. 00:40:22.910 --> 00:40:25.380 OK, that was a fantastic talk. 00:40:25.390 --> 00:40:28.460 With that, I'd like to open it up to questions from the audience. 00:40:29.840 --> 00:40:32.910 Either raise your hand or type it into the chat and we can call on you. 00:40:33.880 --> 00:40:34.810 Let's see. 00:40:34.820 --> 00:40:38.070 Somebody I saw a hand raised. 00:40:38.120 --> 00:40:41.100 Uh, Jeff McGuire, do you want to meet yourself in? 00:40:41.540 --> 00:40:42.230 Ask a question. 00:40:44.500 --> 00:40:45.090 Sure. 00:40:45.440 --> 00:40:45.990 That's great. 00:40:46.000 --> 00:40:46.350 Thanks. 00:40:46.360 --> 00:40:46.710 Attack. 00:40:46.720 --> 00:40:49.030 That's a amazing amount of work and great. 00:40:49.040 --> 00:40:50.900 How many offshore cables you've tried it on? 00:40:50.910 --> 00:40:55.400 That's really a unique and special umm, I was wondering. 00:40:55.410 --> 00:41:02.010 I think everybody can agree that if we can get systematic access to offshore cables, it'll be a huge thing for early warning and subduction zones. 00:41:02.680 --> 00:41:04.490 Am I? 00:41:09.980 --> 00:41:10.170 Right. 00:41:05.110 --> 00:41:18.070 You didn't talk much about saturation of the phase shift signal, and I'm wondering, particularly for cables in marine sediments with really low shear wave velocities and hence really low phase velocities. 00:41:19.090 --> 00:41:28.190 What level of ground velocity you think you can really reliably record, and how that will affect the magnitude estimation algorithm that you showed? 00:41:29.230 --> 00:41:30.960 Yeah, that's that's really an issue. 00:41:30.970 --> 00:41:33.520 And it really depends on the interrogator that you use. 00:41:33.530 --> 00:41:34.230 For example the. 00:41:36.330 --> 00:41:40.660 SN integrator, we did experience a few situation effects. 00:41:41.150 --> 00:41:49.410 I'm actually using a different unit here in Israel with Prisma Photonics, where it didn't see any saturation or any earthquake or quality blast. 00:41:49.420 --> 00:42:00.210 So that's, I mean you can actually put a number depending on the interrogator, the system that they use and and you know various parameters like the gauge length and sampling. 00:42:01.300 --> 00:42:06.590 So it we can quantify it, but it varies between different interrogators. 00:42:06.600 --> 00:42:27.600 And I think that when we apply it for early warning and we would expect to measure high amplitudes, we also need to take that into account, like taking small gauge length to avoid phase keeping and having high sampling rates to lower the actually having high sampling rates is not crucial and that's that stage. 00:42:29.410 --> 00:42:51.190 But what I'm what I'm trying to say is that when we want you actually use it for early morning, we would have to work with some and interrogate unit manufacturer to have everything working properly and make sure that we are able to measure the grand amplitude that we expect and it is it is a problem for soft sediments with low velocities. 00:42:52.610 --> 00:43:02.660 I can say that for these offshore cables that are horizontal, the P wave is not expected to saturate because it's a horizontal fiber and P was registered at low amplitude. 00:43:02.970 --> 00:43:12.190 So that's a plus that keeps me optimistic that even if sweave saturated, we can still use the P waves and somehow calibrate them to get the real magnitude. 00:43:16.170 --> 00:43:17.130 Hope that the answer is. 00:43:17.560 --> 00:43:18.280 OK. Thanks. 00:43:25.530 --> 00:43:26.170 Other questions? 00:43:31.650 --> 00:43:35.290 It looks like Armand has a question in the chat. 00:43:35.370 --> 00:43:38.210 If you'd like to read that out, you're welcome to or I can. 00:43:42.730 --> 00:43:43.190 I'll just. 00:43:43.200 --> 00:43:43.350 I'll. 00:43:43.360 --> 00:43:43.940 I'll go for it. 00:43:43.950 --> 00:43:46.380 So it says, uh, thank you, Doctor. 00:43:46.390 --> 00:43:52.230 Learn what is the projected lifespan and maintenance requirement for a DAS relative to traditional sensors. 00:43:54.090 --> 00:44:00.710 It's hard to answer and I I frankly don't really know because systems get updated very fast in these days. 00:44:01.490 --> 00:44:07.700 Umm, they get upgraded so I'm I'm not sure I I don't really know, sorry. 00:44:11.160 --> 00:44:12.380 Anne Marie, do you want to ask your question? 00:44:18.010 --> 00:44:18.440 Yeah. 00:44:18.450 --> 00:44:19.420 Hey, thanks so much. 00:44:19.430 --> 00:44:22.710 It's truck, and as a really nice talk, I have a question. 00:44:22.720 --> 00:44:37.340 Maybe this was obvious and I just missed it in the beginning, but when you're proposing this application for early warning and your are you using the magnitudes that you've estimated to then predict the pgas and PVS? 00:44:39.940 --> 00:44:41.520 I'm getting the magnitude from. 00:44:47.700 --> 00:44:48.030 Right. 00:44:51.970 --> 00:44:55.410 OK, right. 00:44:44.470 --> 00:44:58.410 These from this relation and I'm using it here to get PGA and PGB, so I'm having the same stress job that I assumed here, yeah. 00:44:56.830 --> 00:44:59.420 Yeah, sure is. 00:44:59.430 --> 00:45:03.760 So is that your proposed like method for the early warning application? 00:45:05.120 --> 00:45:05.660 Uh, yeah. 00:45:06.390 --> 00:45:09.740 And then so I guess I have a a devil's advocate question then why not? 00:45:09.830 --> 00:45:16.270 You've already measured this, the strain and and hence the PGV, so why not just use that? 00:45:16.840 --> 00:45:20.320 But it would be more accurate as far as. 00:45:19.870 --> 00:45:20.410 What do you mean? 00:45:20.460 --> 00:45:27.710 I measured the strain and the the PGV here is measured from online sites, monitors, PGB that I showed. 00:45:28.300 --> 00:45:29.280 Sorry here. 00:45:29.970 --> 00:45:30.170 Yeah. 00:45:31.240 --> 00:45:34.060 So what do you mean use the strains? 00:45:36.570 --> 00:45:37.920 Uh, yeah, I guess maybe I'm. 00:45:46.830 --> 00:45:47.030 Yeah. 00:45:37.930 --> 00:45:47.200 I'm just confused about the sort of work flow to get to the final right the the goal for the early warning is a is a ground motion prediction. Umm. 00:45:52.860 --> 00:45:53.060 Yeah. 00:45:48.020 --> 00:45:57.120 So I'm kind of saying that it's hard to get the magnitude right, but it doesn't matter because we get the ground motion prediction, right? 00:45:57.870 --> 00:46:02.170 Right, but you're measuring strain, right? 00:46:01.710 --> 00:46:03.030 Right, yeah. 00:46:02.630 --> 00:46:03.970 And and then you're. 00:46:04.020 --> 00:46:05.610 So maybe maybe this is our misunderstanding. 00:46:12.260 --> 00:46:12.710 Right. 00:46:05.620 --> 00:46:13.550 You're measuring strain and then you have some conversion to magnitude and then you're putting it back into some ground motion prediction, right? 00:46:13.040 --> 00:46:15.000 Because if we use offshore fibers. 00:46:13.980 --> 00:46:16.240 But if you're already measuring strain. 00:46:18.630 --> 00:46:19.400 You maybe. 00:46:20.800 --> 00:46:21.030 Yeah. 00:46:19.410 --> 00:46:21.250 I mean, maybe this is essentially what you're doing. 00:46:21.040 --> 00:46:22.010 So maybe this? 00:46:21.260 --> 00:46:23.610 Just go from the strain straight to the ground motion. 00:46:26.070 --> 00:46:30.010 With this figure can better emphasize, I think, escaped it. 00:46:37.580 --> 00:46:37.880 Right. 00:46:31.040 --> 00:46:39.010 So we see that strains very abruptly along the fiber because of the subsurface velocities and ground motions don't. 00:46:42.500 --> 00:46:43.260 Right, right. 00:46:39.200 --> 00:46:49.810 So ground motions are a more stable predictor of check and in any way in this application I'm using offshore fibers to predict what's going on online. 00:46:50.340 --> 00:46:50.540 Yeah. 00:46:50.260 --> 00:47:00.850 So you have predict the strain, the peak strain on the fiber, it doesn't really help me quantify how the PGA or PGB or pig strain would look like online. 00:47:01.220 --> 00:47:02.860 Hmm, OK. 00:47:03.490 --> 00:47:05.090 I'm not sure if that's answered. 00:47:05.140 --> 00:47:08.910 Ah, no, but I think that helps understand the the workflow. 00:47:10.120 --> 00:47:10.430 Yeah. 00:47:10.250 --> 00:47:10.430 Yeah. 00:47:10.440 --> 00:47:11.020 What goes on? 00:47:11.340 --> 00:47:12.120 Alright, thanks so much. 00:47:12.990 --> 00:47:13.220 OK. 00:47:13.230 --> 00:47:13.510 Thank you. 00:47:19.640 --> 00:47:20.820 I had a question for you. 00:47:20.830 --> 00:47:28.930 I'm not particularly well versed with DAS, but I feel like you've shown quite nicely how you know. 00:47:28.940 --> 00:47:31.540 Important some of these fiber deployments could be. 00:47:33.000 --> 00:47:37.050 What's the order of magnitude in terms of cost for a fiber deployment? 00:47:37.060 --> 00:47:38.550 How much is that like? 00:47:38.600 --> 00:47:39.290 Like, what's the main? 00:47:40.970 --> 00:47:47.140 Hold up or hesitation for these deployments, other than the fact that they just must cost a decent amount of money for to to deploy. 00:47:49.280 --> 00:47:49.580 That. 00:47:48.960 --> 00:47:49.900 I guess that's two questions. 00:47:50.450 --> 00:47:51.720 That that's the main problem. 00:47:53.800 --> 00:47:55.770 Yeah, sure. 00:47:51.730 --> 00:48:09.040 They cost a lot of money to deploy, especially when we deploy offshore because you have to have the fiber either buried or somehow deployed that, you know, fishing activities or other currents, other effects and destroy it. 00:48:09.050 --> 00:48:11.410 And that can happen if you don't do it properly. 00:48:12.080 --> 00:48:12.320 Right. 00:48:11.890 --> 00:48:19.880 And it also has to withstand the test of time because fiber is offshore do get corroded if water gets in and then it's hard to maintain them. 00:48:21.110 --> 00:48:22.340 Uh, so it's very expensive. 00:48:22.350 --> 00:48:26.740 You know that if we have an existing fiber, you only need one strand. 00:48:26.750 --> 00:48:32.580 So if a telecommunication company deploys its own fiber and it has doesn't strands, you only need one. 00:48:33.180 --> 00:48:37.860 So that's a plus with using existing fibers and the cost of deployment. 00:48:38.600 --> 00:48:44.290 Umm, but yeah, that's the the main issue costs a lot of money. 00:48:44.560 --> 00:48:46.840 That's also the reason I'm not deploying a fiber. 00:48:48.020 --> 00:48:51.440 Umm, because a lot of money and the money can go to waste if you don't do it properly. 00:48:54.030 --> 00:48:54.190 So. 00:48:53.740 --> 00:48:59.700 Of course, yeah, I've heard a decent amount about, you know, turbidity, currents and things severing telecommunications fibers. 00:48:59.710 --> 00:49:03.290 So I understand you know there would be some challenges associated with that. 00:49:03.680 --> 00:49:05.670 You have a few more questions in the chat. 00:49:05.790 --> 00:49:08.320 Shannon, do you wanna do you wanna ask your question? 00:49:09.890 --> 00:49:11.240 Umm yeah. 00:49:11.590 --> 00:49:22.700 Sometimes when I go to like the more site seismology, types of sessions, they're always talking about how to extract the nonlinear site response and so on. 00:49:23.490 --> 00:49:34.090 And from that example that you showed with the difference between the soft and the hard sediment amplitudes, if you'd considered trying to use desk for doing something like that. 00:49:35.800 --> 00:49:41.630 Umm, I don't really know how they do the nonlinear extraction of the site response. 00:49:42.810 --> 00:49:44.390 Do they need different components? 00:49:44.400 --> 00:49:46.280 Because that's only measures one. 00:49:46.830 --> 00:49:49.870 So for example, doing H / V ratio is problematic here. 00:49:51.650 --> 00:49:52.030 Uh. 00:49:51.560 --> 00:49:55.860 So we depends on how they do it, because the problem is you only have 1 component. 00:49:57.020 --> 00:49:57.440 Umm. 00:49:58.500 --> 00:50:16.710 But I did do some work with ambient noise where you see the change in the sediment velocities and many other many other people did similar works comparing like VS 30 derived from a subsurface imaging with ground motion applications from earthquakes. 00:50:16.720 --> 00:50:18.660 There, there is a paper by the Caltech group. 00:50:19.550 --> 00:50:22.130 Umm, so you can do these types of things. 00:50:23.410 --> 00:50:24.390 So OK. 00:50:28.970 --> 00:50:32.170 Uh, you've got another question in the chat from Ola. 00:50:32.180 --> 00:50:34.790 Do you wanna unmute and ask where I can read it aloud either way? 00:50:40.510 --> 00:50:41.640 Uh, yeah, I can do that. 00:50:41.650 --> 00:50:42.900 We had a lively discussion here. 00:50:42.910 --> 00:50:44.120 Thanks for a great talk. 00:50:44.290 --> 00:50:54.740 I'm just wondering, obviously there's a lot of data to be converted from, you know, light poles and then interferometry to actual strain rate for these kilometer long with cables. 00:50:55.250 --> 00:51:01.480 What is the time delay in actually getting to strain rate before you can predict ground motions or magnitudes? 00:51:02.390 --> 00:51:02.700 Yes. 00:51:02.710 --> 00:51:12.190 So I didn't really talk about it, but here I downsampled the data to 20 Hertz and it didn't show the maps. 00:51:15.260 --> 00:51:16.870 Don't use the entire fiber. 00:51:16.880 --> 00:51:20.730 I use just specific fiber segments, so it's kind of hard to see it here. 00:51:22.040 --> 00:51:22.720 See better. 00:51:25.670 --> 00:51:28.340 They're here, so I'm using these fiber segments. 00:51:28.590 --> 00:51:34.090 So these are relatively short, so that also reduces computational cost. 00:51:35.190 --> 00:51:49.110 There's a difference between all the processing done at the interrogated level by the manufacturer, like the conversion from optical data to phase rate to strain rate, and then all the processing that we do. 00:51:49.820 --> 00:51:58.390 So the processing done at the interrogated to get strain rate doesn't really have a delay because the data is also written at the speed at which it's collected. 00:51:58.800 --> 00:52:15.400 So it's not an issue and with this application going down to 20 Hertz, umm, I saw that even with my poor coding skills on Python, it's still runs faster than then it needs to be to to actually work in real time. 00:52:16.840 --> 00:52:18.880 Umm, so it's not a big issue. 00:52:20.230 --> 00:52:20.480 I don't. 00:52:20.630 --> 00:52:21.200 OK. 00:52:21.270 --> 00:52:21.710 Thank you. 00:52:27.110 --> 00:52:27.480 Umm. 00:52:28.300 --> 00:52:29.280 Let's see. 00:52:29.330 --> 00:52:31.370 I saw a hand raise, but I can't. 00:52:33.290 --> 00:52:34.870 Maybe it was put ohh. 00:52:34.910 --> 00:52:36.810 Dean, do you wanna ask your question? 00:52:42.210 --> 00:52:42.580 OK. 00:52:42.590 --> 00:52:42.800 Yeah. 00:52:42.810 --> 00:52:43.320 Can you hear me? 00:52:43.930 --> 00:52:44.110 Yeah. 00:52:45.200 --> 00:52:45.590 Yeah. 00:52:45.600 --> 00:52:53.290 So my question is, you know, what's the effect of the cable orientation to the wavefront? 00:52:53.380 --> 00:52:59.780 You know, I imagine that the what if it is parallel to the wavefront, your sensitivity would be much less. 00:53:00.540 --> 00:53:00.680 Yeah. 00:53:01.100 --> 00:53:06.810 And what are the main impediments for getting into use commercial cables? 00:53:06.820 --> 00:53:07.390 Like what? 00:53:07.400 --> 00:53:10.950 The companies, what is their main concern? 00:53:10.960 --> 00:53:12.990 You say, hey, I'd like to interrogate your cable. 00:53:14.090 --> 00:53:14.390 OK. 00:53:14.300 --> 00:53:16.590 Is it more like security or is it just this? 00:53:16.700 --> 00:53:29.280 You know you're gonna foul up the communication system and then that really calls back to that issue about changing hardware, which is, you know, technology moves along so fast. 00:53:35.820 --> 00:53:36.050 Umm. 00:53:29.940 --> 00:53:40.490 You might be able to get an interrogator in there, and then one year later they changed their hardware and it doesn't work, so it's kind of like a just go kind of a general question three part. 00:53:41.330 --> 00:53:41.740 OK. 00:53:41.810 --> 00:53:46.070 So I'll start with the beginning and then maybe I'll need to remind with the rest. 00:53:46.780 --> 00:54:05.790 Umm so as you can see here we have the direct as wave and for the direct swave and P wave the back as a message to the earthquake does matter because it affects the apparent velocity which effects the amplitude and our ability to reconstruct the signal. 00:54:06.120 --> 00:54:22.020 But the way that we do it here by actually measuring the apparent phase velocity, umm, we are able to get the ground motions more accurately and you can see here for example, uh, the S waves you can see here the accelerations in blue and strains in black. 00:54:22.470 --> 00:54:23.670 And there are normalized. 00:54:23.680 --> 00:54:29.510 So for the direct swave they match, but for later phases which are slower, they don't match anymore. 00:54:30.540 --> 00:54:32.170 Ground accelerations, they are lower. 00:54:32.780 --> 00:54:46.580 So by using this approach we are able to discriminate between the different phases, the velocities, and because the velocity is also a function of the orientation of the fiber and the back azimuth, we also account for that or later phases. 00:54:46.970 --> 00:54:47.880 Surface waves. 00:54:47.890 --> 00:54:48.760 Scattered waves. 00:54:49.210 --> 00:54:55.990 Uh, in a recent paper, I showed that for four different earthquakes recorded in in Greece. 00:54:57.050 --> 00:55:12.850 Umm, when they did the FK over the S wave the FK image looked very similar, meaning the velocity is looked very similar because or offshore fibers the wave field is dominated by these scattered waves. 00:55:13.810 --> 00:55:16.380 And when you think of these headed waves, it doesn't matter. 00:55:16.390 --> 00:55:22.620 After a few seconds from which direction they initial wave came from, so it doesn't matter for the first arrival. 00:55:22.630 --> 00:55:24.300 For later arrivals? 00:55:24.380 --> 00:55:24.770 Not so much. 00:55:25.810 --> 00:55:30.300 Umm, so that's for the first part or the second. 00:55:30.310 --> 00:55:39.230 It was about difficulties in getting access to fibers, so I can say that I had a lot of difficulty. 00:55:40.830 --> 00:55:57.820 I do have access to the one fiber in Israel, but that also took a long time because it's a gas company and they were very suspicious and you know, there's an NDA and then go over the data and then go over manuscripts before they are submitted and stuff like that. 00:55:57.830 --> 00:56:02.830 So it does, uh, detour some people from, from working with them. 00:56:03.870 --> 00:56:16.140 There are also some offshore fibers in Israel, and I couldn't get access because they are guests and oil companies that I really scared with the cybersecurity and we don't want them. 00:56:16.510 --> 00:56:21.740 They don't want us to touch their equipment, so I couldn't get access there. 00:56:21.750 --> 00:56:27.200 The fiber in Chile is owned by a GTT, which is the National Telecommunication Company. 00:56:27.590 --> 00:56:29.840 And they did provide us with access. 00:56:29.850 --> 00:56:34.200 It's actually part of the end of the events ELC, Umm. 00:56:34.750 --> 00:56:39.180 And you know, it's like a commercial process. 00:56:39.190 --> 00:56:42.550 You pay for the bandwidth that you use, so you pay for the fiber. 00:56:43.370 --> 00:56:51.780 Uh, you got a similar proposal for a fiber in Israel going from North Israel to Cypress, and the price was astronomical. 00:56:51.850 --> 00:56:52.640 So I couldn't do that. 00:56:53.710 --> 00:57:13.060 Umm, I hope that answered the second part and for the third part it doesn't really matter if you get a new interrogated unit, you can still use the old one and if they upgrade the fiber, for example in South of France they repaired part of the fiber, they replaced it, they changed the geometry. 00:57:13.430 --> 00:57:21.130 You can still use it without a problem if you need to convert the like to to go from the interrogating unity used to another one. 00:57:21.900 --> 00:57:27.070 Umm, in terms of connecting to the fiber, it's it's not an issue. 00:57:27.380 --> 00:57:43.170 The only issue is if you have to fine tune some of the parameters of the measurements to correspond with the algorithm that estimates the magnitude and predicts ground motion, but it's not really a problem like it did I answer everything. 00:57:46.270 --> 00:57:46.510 Yeah. 00:57:47.710 --> 00:57:53.330 Well, so then you're actually, you're renting out a a section of of of the cable. 00:57:54.220 --> 00:57:57.480 So there's no other data that's being transmitted on that section. 00:57:58.730 --> 00:57:59.090 Ohh. 00:57:57.490 --> 00:58:00.050 That fiber during the time that you're eating it. 00:58:00.060 --> 00:58:00.650 Is that correct? 00:58:01.910 --> 00:58:03.640 So there are two models. 00:58:03.650 --> 00:58:22.260 That's one model where you pay them, and then you get the fiber and the agreement that I got with this company, they are actually monitoring that the fiber for security purposes to the guest pipe to understand if there are leakages or is somebody's trying to steal gas. 00:58:22.270 --> 00:58:27.360 So they are monitoring the fiber with an operational system anyway. 00:58:27.850 --> 00:58:30.240 So in that case I didn't pay anything. 00:58:31.480 --> 00:58:35.170 But you know like, like every other agreement, you have to give something back. 00:58:35.180 --> 00:58:44.640 So there are like uh IP issues and security issues, but in that specific case I don't pay them anything. 00:58:45.360 --> 00:58:50.680 So you can also get these types of agreements with infrastructure that's already been monitored. 00:58:53.650 --> 00:58:54.030 Thank you. 00:58:53.460 --> 00:58:55.570 There are actually 2 fibers in Israel. 00:58:55.580 --> 00:59:06.260 Another one deployed somewhere here that I'm going to get access soon, and in that case the the company is actually paying me to use the fiber and conduct some research for them. 00:59:07.090 --> 00:59:09.960 So I get the data and it gets funding from them. 00:59:10.030 --> 00:59:14.280 So there are also pluses of using this this approach. 00:59:18.640 --> 00:59:41.550 Well, I can add in a fourth question what's how about the cost comparison, you know generally like in a really general answer between laying out a few OS or other sensors on that version floor, I mean you get a nicer view of the wave field, but how much, how much of that view of the wave field do you actually need to make these calculations or to early warning? 00:59:42.330 --> 00:59:42.600 Yeah. 00:59:42.610 --> 00:59:46.090 So for real time you need the OS to be real time connected. 00:59:47.050 --> 00:59:53.010 Umm, so you need telemetry and everything so it's more expensive because you you have to deploy a fiber anyway. 00:59:54.810 --> 01:00:09.020 Uhm, if you use OS for you know basic research like people used to do until a few years ago, you would just and you know, shoved the obvious from a ship and collected a few weeks. 01:00:09.030 --> 01:00:13.890 Months later, if you wanted to use it for your time, you have to have a fiber connected to it anyway. 01:00:15.020 --> 01:00:21.640 Umm, so you don't really gain much by using OBS for eight time at least. 01:00:23.070 --> 01:00:23.430 Thank you. 01:00:29.110 --> 01:00:31.710 Do we have any last questions? 01:00:35.720 --> 01:00:37.080 Going once, going twice. 01:00:39.460 --> 01:00:39.880 OK. 01:00:39.890 --> 01:00:42.770 With that, uh, let's thank our speaker one last time. 01:00:42.780 --> 01:00:43.160 It's OK. 01:00:43.170 --> 01:00:50.950 Thank you for a great talk and for staying up far past your, you know, typical work day hours. 01:00:50.960 --> 01:00:51.740 We appreciate it. 01:00:52.980 --> 01:00:53.780 So thanks a lot. 01:00:55.400 --> 01:00:55.760 Thank you.