WEBVTT Kind: captions Language: en 00:00:02.110 --> 00:00:03.860 All right, good morning, everyone. 00:00:03.860 --> 00:00:06.300 So next week, we’ll be hearing from our very own Roger, 00:00:06.300 --> 00:00:08.240 who will be talking about advances in anelastic and 00:00:08.240 --> 00:00:12.219 viscoelastic rate theory and the implications of those advancements. 00:00:12.220 --> 00:00:14.200 The following week, we’ll be doing something a little bit interesting. 00:00:14.200 --> 00:00:18.040 We’ll be doing an SSA preview. So we’ll be presenting – 00:00:18.050 --> 00:00:21.070 or, some of our scientists will be presenting work that 00:00:21.070 --> 00:00:23.100 they’ll be presenting the following week at SSA. 00:00:23.100 --> 00:00:26.550 So we currently have one open slot. So if anyone woud like to present, 00:00:26.550 --> 00:00:29.070 you don’t have to be going to SSA. But if you’d like to focus on 00:00:29.070 --> 00:00:34.960 some of your research and highlight that on May 9th, we do have a slot open. 00:00:34.960 --> 00:00:38.440 So this morning, I’ll let Andy introduce Biondo. 00:00:39.340 --> 00:00:42.500 - Thanks, Rob. It’s my pleasure to introduce Biondo, 00:00:42.500 --> 00:00:48.160 who is a professor at Stanford. He got his master’s in 00:00:48.160 --> 00:00:52.680 electrical engineering from the Politecnico di Milano in 1984. 00:00:52.680 --> 00:00:58.560 Got a master’s in geophysics at Stanford in 1987 and then a Ph.D. in 1990. 00:00:58.560 --> 00:01:02.280 And he’s been basically affiliated with Stanford ever since. 00:01:03.160 --> 00:01:06.900 In 1998 – there’s sort of a whole list of things on his CV, 00:01:06.909 --> 00:01:09.720 so I just picked a few, but [laughter] – 1998 … 00:01:09.720 --> 00:01:12.549 - That’s from being [inaudible]. [laughter] 00:01:12.549 --> 00:01:16.200 - Okay. Well, I’ll just point out that, since 1998, he’s been the co-director 00:01:16.200 --> 00:01:21.700 of the Stanford Exploration Project. He currently, and has for a while, 00:01:21.700 --> 00:01:25.770 taught 3D seismic imaging and reflection seismology. 00:01:25.770 --> 00:01:29.270 So really looking forward to his discussion today of some 00:01:29.270 --> 00:01:32.940 promising new fiber optic technology. So go ahead. 00:01:32.940 --> 00:01:37.260 - Thank you, Andy. Thanks, Robert. And thanks, all of you, for coming. 00:01:37.260 --> 00:01:41.189 It’s a pleasure finally to come to USGS after I have been at Stanford 00:01:41.189 --> 00:01:46.409 for 30 years. And actually, I do come here because I’m a seismologist. 00:01:46.409 --> 00:01:51.689 But, as you may know from my CV or whatever, I tend to be an 00:01:51.689 --> 00:01:57.290 active-source seismologist for – mostly for oil and gas exploration. 00:01:57.290 --> 00:02:02.540 However, a few years ago, I came across this technology 00:02:02.540 --> 00:02:06.360 of being able to record seismic data through fiber optics. 00:02:06.360 --> 00:02:11.520 And about six years ago, we built my home on campus, 00:02:11.530 --> 00:02:17.750 and Google Fiber was promising as to deliver 1 gigabit for free. 00:02:17.750 --> 00:02:21.950 And so took advantage of that, and I had to take care of, basically, 00:02:21.950 --> 00:02:27.620 routing of the fiber from the street to my home. 00:02:27.620 --> 00:02:32.010 And I saw how easy it is to put the fiber in a conduit. 00:02:32.010 --> 00:02:34.950 And when my friends – in particular, Steve Cole and 00:02:34.950 --> 00:02:40.500 Martine Karrenbach, that are SEP Stanford alumnis as well, 00:02:40.500 --> 00:02:43.260 they start to talk about fiber in seismology. 00:02:43.260 --> 00:02:45.879 I say, well, let’s – can be used something more than oil and gas. 00:02:45.879 --> 00:02:51.440 Let’s see if we can actually listen to the Earth – large-scale and major cities. 00:02:51.440 --> 00:02:54.770 And that is what is really inspiring me. 00:02:54.770 --> 00:02:59.880 As I would say, I call this project a billion-sensor array. 00:02:59.880 --> 00:03:03.580 Of course, a little over-ambitious, but I think that it sounds good. 00:03:03.580 --> 00:03:12.040 And I do hope that, before that I retire, we’ll actually have huge arrays based on 00:03:12.040 --> 00:03:15.560 fiber technology to listen in the Bay Area, Los Angeles, 00:03:15.560 --> 00:03:22.810 and basically other critical earthquake- prone areas and better understand them. 00:03:22.810 --> 00:03:28.080 So what is DAS? So this stands for distributed acoustic sensing, 00:03:28.080 --> 00:03:38.680 is based on measuring basically the seismic wave field using fiber optics. 00:03:38.680 --> 00:03:46.819 And you use what is called a laser interrogator that is 00:03:46.820 --> 00:03:52.680 a laser plus some electronics at one end of the – of the fiber. 00:03:52.680 --> 00:03:56.820 And you can interrogate up to something like 20, 30 kilometers. 00:03:56.830 --> 00:04:00.530 So depends very much on the parameters. 00:04:00.530 --> 00:04:04.099 And can be used also to measure temperature and strain. 00:04:04.100 --> 00:04:09.580 But the basic idea is that the fiber has impurity within the cable 00:04:09.580 --> 00:04:15.620 and is really a reflection seismology experiment, just performed with laser. 00:04:15.620 --> 00:04:19.939 You basically, by measuring the back- scattered energy and the time of the 00:04:19.939 --> 00:04:25.870 back-scattered energy, you can see if this impurity gets closer and farther apart. 00:04:25.870 --> 00:04:28.729 And you can do that at very high frequency, thanks to the 00:04:28.729 --> 00:04:32.760 speed of light compared to the speed of sound of the reflection seismology. 00:04:32.760 --> 00:04:38.599 And so you can basically measure the strain tensor 00:04:38.600 --> 00:04:43.169 component longitudinal to the fiber. That’s all of what it is. 00:04:43.169 --> 00:04:48.720 There is a lot of sophisticated electronic there, but the basic idea is pretty close. 00:04:48.720 --> 00:04:55.610 Simple. And also, things to keep in mind, and I will mention a few times, 00:04:55.610 --> 00:05:01.370 is this idea of a gauge length is – that is basically – is the stretch 00:05:01.370 --> 00:05:07.069 of the fiber over which you’re averaging this strain signal. 00:05:07.069 --> 00:05:11.009 And obviously, you would like, in many cases, to have shorter than 00:05:11.009 --> 00:05:14.540 the wavelengths. And it’s something that can be set up as one of your 00:05:14.540 --> 00:05:18.720 parameters so that you set up your experiment. We have the length from 00:05:18.720 --> 00:05:24.000 1-meter gauge length to 7 meters gauge length in our experiment. 00:05:24.000 --> 00:05:27.740 So, as a seismologist, you almost need to know all what – I told you 00:05:27.740 --> 00:05:32.899 all what you need to know, really, from a technology of the acquisition. 00:05:32.900 --> 00:05:37.260 Now, if are interested to large – to get large arrays, there is 00:05:37.260 --> 00:05:43.469 a very important part of the cost. And these are some kind of 00:05:43.469 --> 00:05:49.500 very back-of-the-envelope estimates of the cost of the – if you have 00:05:49.500 --> 00:05:56.440 a virtual sensor every 10 meters. A fiber cable is about $1 a meter. 00:05:56.440 --> 00:05:59.720 The interrogator, these days, are still quite expensive. 00:05:59.720 --> 00:06:05.669 But notwithstanding what my friends at OptaSense don’t like me here to say that, 00:06:05.669 --> 00:06:11.270 is a really basic laser and electronic technology. 00:06:11.270 --> 00:06:15.089 If we get a large number of this interrogator produced, 00:06:15.089 --> 00:06:20.590 the economy of scales will drive down to probably only a few thousand dollars. 00:06:20.590 --> 00:06:28.369 That is not, though, the price today. Even better, and that is what – 00:06:28.369 --> 00:06:33.320 really what this part here, as I told the story of the fiber in my home, 00:06:33.320 --> 00:06:39.980 is the fact that, if you convince the telecom companies or whoever 00:06:39.990 --> 00:06:47.649 owns the fibers under the ground, to give you some of those fibers, 00:06:47.649 --> 00:06:52.039 then the cost per meter goes to zero, as well the cost of installation 00:06:52.039 --> 00:06:57.669 goes to zero – almost zero. There is installation of the laser interrogator. 00:06:57.669 --> 00:07:03.140 So this can be extremely effective way of collecting large amount of data. 00:07:03.140 --> 00:07:11.140 And, as I said, that is what is really inspiring most of this project. 00:07:11.150 --> 00:07:18.990 Now, we were really the first one to put a fiber cable in telecom conduits. 00:07:18.990 --> 00:07:22.369 I will show you the more details in a few minutes. 00:07:22.369 --> 00:07:27.129 And the big question is, if we want to use them, 00:07:27.129 --> 00:07:32.020 we need to see how well it works and understand the advantages. 00:07:32.020 --> 00:07:34.959 So the advantages are cost, as I just mentioned. 00:07:34.960 --> 00:07:38.480 Receiver density – so we can have a virtual receiver 00:07:38.480 --> 00:07:43.860 every 5, 10 meters – potentially even more densely. 00:07:44.760 --> 00:07:48.389 We can put the receivers under cities where there cannot be easy 00:07:48.389 --> 00:07:54.440 access to geophones, or even less, broadband sensors. 00:07:54.440 --> 00:07:56.629 We don’t need to power the individual receivers. 00:07:56.629 --> 00:08:00.759 So the only thing that we need to provide power is the interrogators. 00:08:00.759 --> 00:08:04.659 The interrogator must be secured because they are expensive piece of equipment, 00:08:04.659 --> 00:08:09.139 but the cable – the value is so spread along the cable itself, 00:08:09.139 --> 00:08:13.169 nobody would steal your fiber cable. And you don’t need to secure that. 00:08:13.169 --> 00:08:16.050 So that’s another important thing to know about environment. 00:08:16.050 --> 00:08:23.050 And, in some cases, like we were talking before about CO2 sequestration, really 00:08:23.050 --> 00:08:30.349 fiber may last for 20, 30, 40 years – anyway, they are really very durable. 00:08:30.349 --> 00:08:36.459 They are resilient in a harsh environment. Oil and gas company, they put down 00:08:36.460 --> 00:08:42.740 in deep water wells where we have high temperature and high pressure. 00:08:42.740 --> 00:08:47.420 So that is in our advantage compared to geophones. 00:08:47.420 --> 00:08:53.560 As a single sensor, they are clearly not up – not even – let’s forget even 00:08:53.560 --> 00:08:57.790 the broadband seismometer, which probably most of you work with. 00:08:57.790 --> 00:09:00.440 Not even the typical exploration geophones. 00:09:00.440 --> 00:09:05.500 So it’s clearly – the quality of the signal is not as good. 00:09:05.500 --> 00:09:08.920 It’s getting better and better in terms of signal-to-noise. 00:09:08.920 --> 00:09:14.279 But there is – a main limitation is that we’re measuring the 00:09:14.280 --> 00:09:19.100 strain tensor along the fiber. So there is very much an issue 00:09:19.100 --> 00:09:25.840 of directivity and sensitivity – kind of a [inaudible] cos-squared 00:09:25.840 --> 00:09:32.070 with respect to the angle that is parallel to the fiber itself. 00:09:32.070 --> 00:09:39.399 So it’s zero orthogonal to the fiber and 1 along the fiber. 00:09:39.399 --> 00:09:42.589 And then there is an issue of gauge length, as I mentioned. 00:09:42.589 --> 00:09:47.100 So the wavelength versus frequency because of that averaging, basically is 00:09:47.100 --> 00:09:51.440 the low-pass filter that you’re applying while that you’re acquiring. 00:09:51.440 --> 00:09:57.490 So I’ve said all of that. We went ahead in September of – 00:09:57.490 --> 00:10:01.940 now 19 months ago or so. We put a fiber under Stanford. 00:10:01.940 --> 00:10:07.699 And for the one of you that are familiar with Stanford, this is the map. 00:10:07.699 --> 00:10:13.319 That little node on the lower right corner is myself sitting in my office in Mitchell. 00:10:13.319 --> 00:10:18.420 And while that I am computing away, the fiber keeps collecting data. 00:10:18.420 --> 00:10:23.450 And we have done that for 19 months, as I said. 00:10:23.450 --> 00:10:28.510 We have built a database, of course, of the earthquake events, built on your 00:10:28.510 --> 00:10:36.860 [inaudible] database on the web, about 1,000 events that we have been looking. 00:10:36.860 --> 00:10:39.850 And I should say, at the beginning, most of the work is done, 00:10:39.850 --> 00:10:44.010 as usual, by students. Eileen Martin is my senior student 00:10:44.010 --> 00:10:49.149 that has supervised the project and, until now – she is graduating 00:10:49.149 --> 00:10:53.250 this quarter and moving on to a junior faculty at Virginia Tech. 00:10:53.250 --> 00:10:54.569 And she would be very happy to – 00:10:54.569 --> 00:10:59.009 collaborating as a junior faculty there if you are interested. 00:10:59.009 --> 00:11:01.060 And then I have a new student that – his name is 00:11:01.060 --> 00:11:05.380 Siyuan Yuan, but he is taking over Eileen’s work. 00:11:05.380 --> 00:11:09.160 And we kept recording. This is, of course, is just a simple 00:11:09.160 --> 00:11:13.620 display of a small fraction of the data that we have been recording. 00:11:13.620 --> 00:11:18.080 Our array has this squarish-8 shape. 00:11:18.080 --> 00:11:22.300 And I’m not going to tell you – or, need to tell you why we think 00:11:22.310 --> 00:11:25.490 that we are recording good data. Here is the Bay Area. 00:11:25.490 --> 00:11:30.710 We have [inaudible] ocean waves, so microseismic generation 00:11:30.710 --> 00:11:35.260 for interferometry, and we have a lot of faults. 00:11:35.260 --> 00:11:40.130 And I’m not necessarily tell you about that. 00:11:40.130 --> 00:11:44.579 As I said, this is really – I’m hoping that these are the first step 00:11:44.580 --> 00:11:49.800 towards a much more ambitious project, is to see if it is worthwhile and 00:11:49.800 --> 00:11:55.980 convincing people with the money to do it, to build a – call it a billion-sensor 00:11:55.980 --> 00:11:59.079 array under the Bay Area. Taking advantage of extensive 00:11:59.079 --> 00:12:03.860 fiber optic networks that is under the Bay Area for telecom. 00:12:03.860 --> 00:12:10.721 I should also have said that an important statistic is, typically, fiber 00:12:10.721 --> 00:12:17.939 optic networks owned by the telecom is busy only 30% of the cables are used. 00:12:17.939 --> 00:12:22.699 So 70% are there that are for redundancy, future growth. 00:12:22.699 --> 00:12:28.129 So it is really matter of convincing them versus really using valuable 00:12:28.129 --> 00:12:34.759 resources for the telecom. Because most of those fibers, they’re actually idle. 00:12:34.759 --> 00:12:38.290 So should be no real reason for not sharing that. 00:12:38.290 --> 00:12:42.810 So as I said, billion-sensor array, 100 kilometers by 100 kilometers. 00:12:42.810 --> 00:12:46.529 And then we start to listen to every little pops and crack 00:12:46.529 --> 00:12:49.660 of the Earth under the Bay Area. And hopefully some – making some 00:12:49.660 --> 00:12:55.040 progress in understanding the geology and earthquake mechanics. 00:12:55.040 --> 00:13:00.480 Also, shorter-term, and actually, we do have a proposal with 00:13:00.490 --> 00:13:07.689 Greg Beroza and Jack Baker that – start to use this for soil characterization 00:13:07.689 --> 00:13:15.170 to help civil engineers to basically build better, more efficient structure 00:13:15.170 --> 00:13:17.430 and more resilient to earthquake. 00:13:17.430 --> 00:13:22.649 And that’s, I think, can be done now. Doesn’t need to wait too long. 00:13:22.649 --> 00:13:27.089 So what is our experiment? It was basically using a fiber. 00:13:27.089 --> 00:13:32.839 We want to emulate what would be this billion-sensor array project. 00:13:32.839 --> 00:13:36.410 We put the fiber in the conduits under Stanford campus. 00:13:36.410 --> 00:13:38.860 Stanford campus has a lot of fibers around. 00:13:38.860 --> 00:13:42.259 Still remember the students, at the end of ’80s, when they were 00:13:42.259 --> 00:13:46.110 digging all around campus to start to put in the first fibers. 00:13:46.110 --> 00:13:48.250 And since then, they just kept going. 00:13:48.250 --> 00:13:50.910 So we have a very extensive fiber network. 00:13:50.910 --> 00:13:54.439 I went to talk with the IT service at Stanford and said, do you want to 00:13:54.439 --> 00:13:57.860 participate with scientific experiment? They were enthusiastic. 00:13:57.860 --> 00:14:02.120 And so that’s what we did. And this is a manhole. 00:14:02.120 --> 00:14:06.319 Eileen and another student and me in the background. 00:14:06.319 --> 00:14:09.910 This is the way that the cables goes in in the ground. 00:14:09.910 --> 00:14:11.790 So basically, it’s one wall in the manhole. 00:14:11.790 --> 00:14:15.680 You may see exactly – might be the one in the background. 00:14:15.680 --> 00:14:22.529 And the standard configuration is PVC conduits with cables sitting there. 00:14:22.529 --> 00:14:26.500 So the coupling is really only provided by gravity and friction. 00:14:26.500 --> 00:14:30.120 And before that we did that, people were very skeptical 00:14:30.120 --> 00:14:34.559 that we were getting any useful signal. And I say, well, you can do 00:14:34.559 --> 00:14:36.860 a lot of modeling, but the only way to 00:14:36.860 --> 00:14:40.319 prove it works or doesn’t work is that we’re trying. 00:14:40.319 --> 00:14:44.249 And that was really the main goal of experiment is prove if that works. 00:14:44.249 --> 00:14:46.779 And then you will see it does work pretty well. 00:14:46.779 --> 00:14:52.480 We’re also lucky that we have a broadband station at Jasper Ridge 00:14:52.480 --> 00:14:56.370 and that Greg and Bob Kovach installed 30 years ago. 00:14:56.370 --> 00:14:59.540 So we used that as a kind of a ground truth 00:14:59.540 --> 00:15:02.410 and tried to validate some of our results. 00:15:02.410 --> 00:15:07.319 How does this data look like? This is just one example. 00:15:07.319 --> 00:15:10.670 All the data that I’m going to show are going to have channel number 00:15:10.670 --> 00:15:17.990 vertical and time horizontal. That was an earthquake around Berkeley. 00:15:17.990 --> 00:15:20.839 And clear, you can see the P’s and the S arriving. 00:15:20.839 --> 00:15:25.350 Then we have traffic noise. This is typically a car or truck 00:15:25.350 --> 00:15:30.949 or a Marguerite bus going around Stanford that gets clear [inaudible]. 00:15:30.949 --> 00:15:37.740 This is pump noise out of the buildings. And then we do have just nature noise 00:15:37.800 --> 00:15:43.900 or freeway noise that we can use for interferometry. 00:15:44.980 --> 00:15:49.540 Our questions is, first, are we recording repeatable signal? 00:15:49.550 --> 00:15:52.970 So in this funky configuration that is [inaudible] scalable, 00:15:52.970 --> 00:15:57.329 are we recording useful signal? And in – first, repeatable. 00:15:57.329 --> 00:15:58.999 Are we recording P and S waves and 00:15:58.999 --> 00:16:03.889 surface waves for interferometry, for example? 00:16:03.889 --> 00:16:09.879 Is the limitation the quality of the laser interrogator, or is the coupling? 00:16:09.879 --> 00:16:12.660 I should also have said, the laser interrogator are 00:16:12.660 --> 00:16:15.889 relatively new technology. So they’re still very much 00:16:15.889 --> 00:16:20.470 on a steep part of a typical S curve of technologies. 00:16:20.470 --> 00:16:24.200 So every few years, they come up with new generation of interrogators 00:16:24.200 --> 00:16:31.519 and that are recording better data. So we have to try to record two 00:16:31.519 --> 00:16:37.459 interrogators at the same time and to look at comparing the two of them. 00:16:37.459 --> 00:16:42.200 Are we ready to interpret the data, and are we ready to process the data? 00:16:42.200 --> 00:16:45.680 Of course, the interpretation and the processing, that would be 00:16:45.680 --> 00:16:49.269 part of the research. You may know, I have 00:16:49.269 --> 00:16:54.389 a large research group – 15 grad students, and we are doing 00:16:54.389 --> 00:16:59.800 processing and imaging development. Data acquisition is new for us, and so – 00:16:59.800 --> 00:17:05.130 but I saw this as an opportunity to use our knowledge for different kind of 00:17:05.130 --> 00:17:10.010 array processing. So are we recording repeatable signal? The answer is yes. 00:17:10.010 --> 00:17:12.360 This is one of the many example that I can show. 00:17:12.360 --> 00:17:18.520 This is actually quarry blast down in San Antonio [inaudible]. 00:17:18.520 --> 00:17:22.880 And clearly, you can see that the waveforms are very recordable. 00:17:22.880 --> 00:17:25.630 And this is – as I said, is one of the many example. 00:17:25.630 --> 00:17:30.850 When I show a trace down here, is going to be a Jasper Ridge 00:17:30.850 --> 00:17:34.510 trace, and I’m going to say which component of Jasper Ridge – 00:17:34.510 --> 00:17:37.240 of the broadband Jasper Ridge in this particular case, 00:17:37.240 --> 00:17:42.659 was of the vertical component. Interesting enough, actually, in this case, 00:17:42.659 --> 00:17:46.070 the DAS data is more repeatable than the Jasper Ridge. 00:17:46.070 --> 00:17:48.870 This is really more an exception, but it just happened that, 00:17:48.870 --> 00:17:51.950 for the quarry blast, Jasper Ridge is not very good. 00:17:51.950 --> 00:17:55.480 Also, one thing that you will keep seeing are these arrows. 00:17:55.480 --> 00:18:01.120 The green points north, and the orange tells you the direction 00:18:01.120 --> 00:18:09.760 of arrival of a event. And the – and this is the map of channels. 00:18:09.760 --> 00:18:13.090 And if we had more time, I can spend time to see how, indeed, 00:18:13.090 --> 00:18:17.180 the kinemetics of these arrivals, they do correspond to the kinematics 00:18:17.180 --> 00:18:20.430 that we would expect for the arrival arriving from that direction. 00:18:20.430 --> 00:18:24.130 And these are surface waves in this particular case. 00:18:24.130 --> 00:18:27.230 So, can we record over different kind of events? 00:18:27.230 --> 00:18:29.980 Good question is, do we record P waves? 00:18:29.980 --> 00:18:33.520 We saw the directivity that is – basically says that the fiber should be 00:18:33.520 --> 00:18:40.160 blind to P waves arriving orthogonal, so do we actually record them? 00:18:40.160 --> 00:18:42.280 And actually, we do. 00:18:42.280 --> 00:18:46.040 And that was somewhat a surprise for which I don’t have a full answer. 00:18:46.040 --> 00:18:50.820 And we can get them very repeatable. We were very lucky that, 00:18:50.820 --> 00:18:54.820 this summer, there were two earthquakes happening in Ladera 00:18:54.820 --> 00:18:58.220 half an hour apart from each other that, actually, they were similar. 00:18:58.220 --> 00:19:02.000 And we will actually see the repeatability there. 00:19:02.010 --> 00:19:06.730 Here is just showing the P waves arrival that match well with the P waves on 00:19:06.730 --> 00:19:11.460 the Jasper Ridge. Jasper Ridge is closer to Ladera, so arrived earlier. 00:19:11.460 --> 00:19:19.620 And these are the S waves and surface waves that, as they arrive, 00:19:19.630 --> 00:19:23.820 the [inaudible] component on the north-south component 00:19:23.820 --> 00:19:26.360 at Jasper Ridge as well on the array. 00:19:26.360 --> 00:19:29.580 As I said, this actually was very repeatable. 00:19:29.580 --> 00:19:34.500 So this is – was one event, and this is the other event actually 00:19:34.500 --> 00:19:40.380 focused around the P waves arrival. And the other panel – the panel here 00:19:40.380 --> 00:19:45.300 in the middle shows the difference after that I scaled the two events. 00:19:45.300 --> 00:19:50.360 One was 1.6. The other was 1.8. Obviously, they happened 00:19:50.360 --> 00:19:56.120 in a very nearby part of the fault, if not exactly the same one. 00:19:56.120 --> 00:20:01.440 And you can see that we are really acquiring a very messy wave field. 00:20:01.440 --> 00:20:05.910 Not big surprise. We have a lot of basement. 00:20:05.910 --> 00:20:13.549 There is a five-floor subterranean parking lot here with – 00:20:13.549 --> 00:20:16.620 I had my mouse – here at Stanford. 00:20:16.620 --> 00:20:21.409 So there is a lot of – a lot of scattering going on, however. 00:20:21.409 --> 00:20:25.950 So we do have a complex wave field, but it’s all signal. 00:20:25.950 --> 00:20:29.250 This is not random noise or instrument noise. 00:20:29.250 --> 00:20:35.919 You see how repeatable these two events are in each – little small details. 00:20:35.919 --> 00:20:40.510 And only the pump noise and some traffic noise is not that repeatable, 00:20:40.510 --> 00:20:44.920 and that what comes up here in the difference. 00:20:44.920 --> 00:20:50.400 And so we have – yes, we have P and S, and they are very repeatable. 00:20:50.410 --> 00:20:54.649 And even the spectrum, as you see, is actually – is pretty similar. 00:20:54.649 --> 00:20:57.970 Other interest is looking at interferometry. 00:20:57.970 --> 00:21:03.780 And, as I said, one – I will say very low-hanging fruit 00:21:03.780 --> 00:21:09.730 of this technology is the one you producing very high-resolution maps 00:21:09.730 --> 00:21:15.740 at relatively low cost of soil property using interferometry. 00:21:15.740 --> 00:21:18.700 This is one of the – early example of interferometry. 00:21:18.700 --> 00:21:22.920 This is Rayleigh waves along this direction. 00:21:22.920 --> 00:21:29.400 And the thing that I want to point out, as you’d expect, we do see some 00:21:29.409 --> 00:21:33.790 frequency dispersion in the velocity as we expect. 00:21:33.790 --> 00:21:36.400 Lower frequency, they propagate faster because 00:21:36.400 --> 00:21:40.160 they interrogate a deeper part of the subsurface. 00:21:40.160 --> 00:21:44.760 It is not only for Rayleigh waves, also Love waves. 00:21:44.769 --> 00:21:47.350 So this is different kind of interferometry. 00:21:47.350 --> 00:21:51.560 So this is, again, Rayleigh waves, but you can see an opposite path 00:21:51.560 --> 00:21:58.200 of different – so this is basically measured here with yellow arrow 00:21:58.200 --> 00:22:05.000 and for two different virtual sources – channel 75 and 35. 00:22:05.010 --> 00:22:08.559 And you see that there are also some Love waves then being recorded. 00:22:08.559 --> 00:22:13.570 As we are going to see in about 10 minutes, Love waves recording 00:22:13.570 --> 00:22:17.799 is actually a little more challenge, and that has to do with the fact 00:22:17.799 --> 00:22:21.460 that we are measuring, as I said at the beginning, 00:22:21.460 --> 00:22:25.460 really just one component of the strain tensor. 00:22:25.460 --> 00:22:29.840 And so we are sensitive to Love waves and the different waves, but we know the 00:22:29.840 --> 00:22:34.240 sensitivity of the work to be done. Eileen is doing in her thesis. 00:22:34.240 --> 00:22:38.220 And you’re welcome to come and – to her defense. 00:22:38.220 --> 00:22:43.320 Her defense, if I remember right, is either 29th – I think it’s 29th of May 00:22:43.320 --> 00:22:47.059 is going to be on our website anyway. 00:22:47.059 --> 00:22:50.220 And you’re welcome to come to Stanford for her defense. 00:22:50.220 --> 00:22:54.880 So that is for Love waves. And also, since we measured 00:22:54.880 --> 00:22:59.669 the data for so long time, we start to look at time lapse. 00:22:59.669 --> 00:23:03.340 And can we use this technology for a cheap way 00:23:03.340 --> 00:23:07.289 of monitoring subsurface condition over time? 00:23:07.289 --> 00:23:13.860 And so this are different interferometric results at different months. 00:23:13.860 --> 00:23:18.060 No need to point out to you that this is the rainy season, 00:23:18.060 --> 00:23:24.740 in particular last year, and looks like the velocity has not really changed. 00:23:24.740 --> 00:23:27.760 Probably has to do – a lot to do with the near surface 00:23:27.769 --> 00:23:30.090 close to Stanford is – a lot of it is built up. 00:23:30.090 --> 00:23:37.260 It’s not just free soil in which you do have as much of saturation changes. 00:23:37.260 --> 00:23:41.820 But what has actually quite changed is the signal at [inaudible]. 00:23:41.820 --> 00:23:46.620 Dry season, we don’t get much [inaudible] signal. 00:23:46.620 --> 00:23:50.460 Wet season, we actually get more of a [inaudible] signal. 00:23:50.460 --> 00:23:55.080 So that is clearly visible. Probably in the – if there had been 00:23:55.080 --> 00:24:00.120 changes, we would be able to observe them over a fairly large frequency band. 00:24:00.120 --> 00:24:04.580 So these are dispersion curves. You may have seen similar curves 00:24:04.580 --> 00:24:09.980 for people that do surface waves analysis either for teleseismic 00:24:09.980 --> 00:24:16.610 or from interferometry like this in this case, using noise. 00:24:16.610 --> 00:24:20.850 So this is frequency. You may not see that is 5 hertz. 00:24:20.850 --> 00:24:26.820 This is 10 hertz. And this is apparent velocity. 00:24:26.820 --> 00:24:30.600 And you see typically that we do – at lower frequency, we do have 00:24:30.600 --> 00:24:38.000 increasing velocities. And it’s pretty resilient with – consistent with time. 00:24:38.000 --> 00:24:43.850 So we have September ‘16, March ‘17, September ‘17, and March of ‘18. 00:24:43.850 --> 00:24:49.360 So this is basically one year and a half. 00:24:49.360 --> 00:24:54.190 And one good question is, what’s happening here at the low frequency? 00:24:54.190 --> 00:24:57.830 So this is traffic noise. We [inaudible] are 00:24:57.830 --> 00:25:00.450 using a long [inaudible] array. We have shown that we can 00:25:00.450 --> 00:25:05.799 do interferometry using traffic noise reliably up to 10, 15 hertz. 00:25:05.799 --> 00:25:11.919 So this frequency band is traffic noise. Here, we will be [inaudible], 00:25:11.919 --> 00:25:18.570 the one naturally generated by ocean waves below 2 hertz. 00:25:18.570 --> 00:25:26.870 So a good question is, can we actually – oops – do a – reliably for below 1 hertz. 00:25:26.870 --> 00:25:30.909 And part of the answer – I don’t have it yet – we can for the next one, 00:25:30.909 --> 00:25:35.240 is when after we try a new interrogator. And it will turn out that the 00:25:35.240 --> 00:25:38.610 new interrogator has better response at lower frequencies. 00:25:38.610 --> 00:25:41.769 So I do expect – we have not done it yet – that we actually 00:25:41.769 --> 00:25:46.980 can get good interferometry reflection of hertz with the new interrogators. 00:25:46.980 --> 00:25:53.740 With what we have now is good – let’s say up from 1 hertz up. 00:25:53.740 --> 00:25:57.680 Still, pretty good in terms of civil engineering, in terms of 00:25:57.680 --> 00:26:03.740 wavelengths at 1 hertz, we still already have hundreds of meters of wavelength. 00:26:03.750 --> 00:26:06.610 So we go probably deeper than civil engineers want. 00:26:06.610 --> 00:26:11.860 Civil engineering really can benefit from around 5 hertz. 00:26:11.860 --> 00:26:17.120 That would be the ideal frequency in terms of getting 00:26:17.120 --> 00:26:24.490 velocity in the next – in the first 30, 50 meters in the subsurface. 00:26:24.490 --> 00:26:29.450 So this is the other big question. Is the data quality controlled 00:26:29.450 --> 00:26:35.100 by the fiber-to-rock coupling, given our open conduit configuration. 00:26:35.100 --> 00:26:37.210 It’s something that, if we want to use the 00:26:37.210 --> 00:26:41.169 telecom infrastructure, we cannot really change. 00:26:41.169 --> 00:26:46.630 Because the whole idea is using that. Of course, for other application in which 00:26:46.630 --> 00:26:54.269 we are more specialized, then in which you may cement the fiber to a well, 00:26:54.269 --> 00:26:58.809 or you may backfill a trench, then it’s a different story. 00:26:58.809 --> 00:27:02.580 But if we want to use the telecom infrastructure, that is – 00:27:02.580 --> 00:27:06.260 the coupling cannot be really played with. 00:27:06.260 --> 00:27:12.169 The question is, can we actually improve the signal-to-noise 00:27:12.169 --> 00:27:15.940 by changing interrogators? And we were lucky enough 00:27:15.940 --> 00:27:20.190 that OptaSense gave us – so what we have currently 00:27:20.190 --> 00:27:27.279 is the one that is continuously recording is – they call it 3.1. 00:27:27.279 --> 00:27:32.720 What they commercially rent for paying customers, 00:27:32.720 --> 00:27:38.720 that we are not, is the – generation 4. 00:27:38.720 --> 00:27:41.781 So we have been recording using 3.1, but they gave us 00:27:41.781 --> 00:27:45.260 a new-generation interrogator for a week. 00:27:45.260 --> 00:27:50.720 And so we were able to record data from both interrogator at the same time. 00:27:50.730 --> 00:27:56.860 Our fiber cable actually has six fibers within, so we used 00:27:56.860 --> 00:28:00.690 two of them at the same time. And we can compare the two of them. 00:28:00.690 --> 00:28:03.659 At the same time, actually, thanks to you, we had put 00:28:03.659 --> 00:28:09.269 three broadband geophones in the basements of three buildings at – 00:28:09.269 --> 00:28:12.310 very close to the array – at three corners of the array. 00:28:12.310 --> 00:28:19.659 So because of lack of manpower, so far, I have – we have start to look at the 00:28:19.659 --> 00:28:24.509 broadband sensor data now and have only relatively qualitative 00:28:24.509 --> 00:28:26.900 comparison between the two interrogators. 00:28:26.900 --> 00:28:31.649 But it does show that, indeed, the new interrogator looks like 00:28:31.649 --> 00:28:34.309 that has better response at lower frequency. 00:28:34.309 --> 00:28:42.210 So this is the spectrum at – red for the new interrogator. 00:28:42.210 --> 00:28:47.269 The blue is the older interrogator. We also were lucky that, in that week, 00:28:47.269 --> 00:28:52.100 there was a pretty large earthquake at Alum Rock. 00:28:52.100 --> 00:28:56.920 And so we actually have a nice earthquake that arrived 00:28:56.920 --> 00:29:00.180 on the array and recorded by the two interrogators. 00:29:00.180 --> 00:29:08.500 I am going to compare the P waves arrivals here in the green box and the – 00:29:08.520 --> 00:29:12.220 and the S wave surface waves arrival in the purple one. 00:29:12.220 --> 00:29:15.580 And the band pass using two different bands. 00:29:15.580 --> 00:29:19.519 And so what I tried to do a poor man compensation of 00:29:19.520 --> 00:29:26.920 spectral balancing within each arrival. As I said, this is still a fairly qualitative 00:29:26.920 --> 00:29:32.360 comparison, but the high-frequency comparison, so is in – for the P waves, 00:29:32.360 --> 00:29:38.820 the green box, and the lower frequency is for the – 00:29:38.820 --> 00:29:41.980 S wave and surface waves in the purple box. 00:29:41.980 --> 00:29:46.399 So this is the older-generation interrogator, and this is the newer one. 00:29:46.399 --> 00:29:53.279 So you can see that, indeed, we have more coherent arrivals. 00:29:53.280 --> 00:29:58.120 And, as I said, this is really qualitative. This is a visual comparison. 00:29:58.120 --> 00:30:02.440 But both of the high-frequency – if you look at this arrival here, 00:30:02.440 --> 00:30:06.400 the new generation looks like to have more coherent arrival 00:30:06.400 --> 00:30:11.560 on the P waves arrival, as well also in – at the lower frequency. 00:30:11.560 --> 00:30:17.080 But it’s not a big surprise given the different spectra that we have. 00:30:17.080 --> 00:30:20.580 We have better signal-to-noise. 00:30:20.580 --> 00:30:24.830 More quantitative comparison really is – will follow. 00:30:24.830 --> 00:30:30.900 But I think that it shows that, so far, we are not limited only 00:30:30.900 --> 00:30:34.710 by the quality of the coupling. So improving the interrogators 00:30:34.710 --> 00:30:38.700 will actually improve the quality of the data that we will be recording, 00:30:38.700 --> 00:30:44.779 both in terms of active surveys, but as well, also, passive – 00:30:44.779 --> 00:30:48.880 for active survey – sorry. I should say earthquake events 00:30:48.880 --> 00:30:55.150 as well for noise like interferometry in particular around the 1 hertz 00:30:55.150 --> 00:31:02.200 or below 1 hertz that can be very useful using ocean waves. 00:31:02.840 --> 00:31:07.160 So that’s – is the data quality controlled by the 00:31:07.160 --> 00:31:11.400 fiber optic interrogator or the coupling? 00:31:11.400 --> 00:31:15.840 Need further analysis, but first indication, I think that are positive. 00:31:15.850 --> 00:31:22.110 So for the next 15 minutes, I’m going to talk now about how to – 00:31:22.110 --> 00:31:27.059 we ready to interpret the data and the processing of the data 00:31:27.059 --> 00:31:28.919 So first interpreting. 00:31:28.919 --> 00:31:33.790 Interpreting really means understanding what we are recording. 00:31:33.790 --> 00:31:38.740 We needed to be basically reminded by the data themselves 00:31:38.740 --> 00:31:41.379 that we are not recording geophones data. 00:31:41.379 --> 00:31:45.770 This is not particle velocity. It’s not acceleration. But is a strain thing. 00:31:45.770 --> 00:31:50.090 So it behaves in a different way in interesting ways. 00:31:50.090 --> 00:31:53.000 And here is where, actually, is one of the many example 00:31:53.000 --> 00:31:57.909 that the data showed that in our face. When we start looking at the data, 00:31:57.909 --> 00:32:03.679 we’re wondering why actually we see this – we called that polarity flip. 00:32:03.679 --> 00:32:10.409 You see that there is, indeed, a polarity flip across a different part of the array. 00:32:10.409 --> 00:32:15.100 And this actually corresponds to a corner of the array. 00:32:15.100 --> 00:32:19.169 And around each corner, in particular for surface waves, 00:32:19.169 --> 00:32:24.779 but if you look careful for shear waves as well, you observe this polarity flip. 00:32:24.779 --> 00:32:27.100 And we needed to understand it. 00:32:27.100 --> 00:32:32.320 And Eileen did a quite careful analysis, and I will show you only a few snippet 00:32:32.320 --> 00:32:37.480 of Eileen analysis that do explain this polarity flip or even more. 00:32:37.480 --> 00:32:40.250 And interesting enough, this part of – this particular corner 00:32:40.250 --> 00:32:43.990 is actually not a sharp corner. The other corners in the array, 00:32:43.990 --> 00:32:49.720 they are really manholes in which the cable cross 90 degrees. 00:32:49.720 --> 00:32:53.840 In this part of the array that I’m showing here some of the data, 00:32:53.840 --> 00:32:57.850 is recorded here in which the conduit turns more gently. 00:32:57.850 --> 00:33:01.230 And actually, we see that this polarity flips a little more gentle 00:33:01.230 --> 00:33:05.610 and there is an amplitude turn. Why that is the case? 00:33:05.610 --> 00:33:09.880 So Eileen went through the proper analytical analysis. 00:33:09.880 --> 00:33:15.149 This – I show this kind of a way of giving intuition. 00:33:15.149 --> 00:33:18.659 We are looking at strain tensor, so we really should look in these 00:33:18.659 --> 00:33:22.899 movies that I downloaded from the web how the distance 00:33:22.899 --> 00:33:29.360 between these dots changes with time. That what we really measuring. 00:33:29.360 --> 00:33:37.840 And we see that it’s quite different, the behavior, if we are measuring 00:33:37.840 --> 00:33:42.860 the seismic waves, like at P waves, or would be also Rayleigh waves, 00:33:42.860 --> 00:33:50.009 the particle motion is in the same direction as the direction 00:33:50.009 --> 00:33:54.290 of the phase velocity of the waves. But actually what we measure is 00:33:54.290 --> 00:33:59.710 quite different in the case of S waves, of Love waves, in which the particle 00:33:59.710 --> 00:34:04.490 motion is normal to the direction. So if you follow just those dots, 00:34:04.490 --> 00:34:11.380 you will see that actually the response is quite different and actually can be 00:34:11.380 --> 00:34:14.860 seen that analytically, in one case, in terms of amplitude, 00:34:14.860 --> 00:34:20.430 is a cos-squared theta, and that’s the reason why that is when we are 00:34:20.430 --> 00:34:25.990 moving 90 – we have 90-degree fibers, and then you see that, indeed, 00:34:25.990 --> 00:34:29.390 the vertical distance between the dot on the left panel 00:34:29.390 --> 00:34:32.070 doesn’t change as the waves propagate. 00:34:32.070 --> 00:34:38.070 And instead, it’s a sign of 2-theta for the S waves and Love waves. 00:34:38.070 --> 00:34:41.200 And that is where the polarity flips comes when we have 00:34:41.200 --> 00:34:49.580 a 90-degree rotation in theta. And that is where we do know that. 00:34:49.580 --> 00:34:56.700 As I said, Eileen made a quite more careful and in-depth analysis. 00:34:56.700 --> 00:35:02.470 That is actually posted on the web as part of some of the publication, 00:35:02.470 --> 00:35:06.270 of course, within her thesis. And that is one of the 00:35:06.270 --> 00:35:11.040 panel that she created. So this would be the sensitivity 00:35:11.040 --> 00:35:16.840 of a geophone that measured particle velocity at different 00:35:16.840 --> 00:35:22.770 wavelengths and as a function of a direction of arrival for 00:35:22.770 --> 00:35:28.640 Rayleigh waves is the green, and Love waves is in red. 00:35:28.640 --> 00:35:34.770 That in case that you have a point-wise strain tensor measurement. 00:35:34.770 --> 00:35:40.240 And the right column is in which you have a 10 meters gauge length. 00:35:40.240 --> 00:35:44.660 So that’s the gauge length average, of course, will make it different than 00:35:44.660 --> 00:35:48.930 the point-wise, in particular as the wavelengths decreases. 00:35:48.930 --> 00:35:53.720 So we go from a 44 meters wavelength on the top 00:35:53.720 --> 00:35:58.490 to a 10 meters wavelength at the bottom. 00:35:58.490 --> 00:36:02.820 What is interesting here that is, why the Rayleigh waves behaves 00:36:02.820 --> 00:36:11.050 not very different, at least up to where the gauge length is shorter 00:36:11.050 --> 00:36:20.660 than the wavelengths of the waves. The Love waves have actually this 00:36:20.660 --> 00:36:25.660 sensitivity pattern that is four lobes and is matching with the sensitivity 00:36:25.660 --> 00:36:32.660 at 45 degrees instead what you would expect, the 90 degrees of – 00:36:32.660 --> 00:36:36.620 if you measure acceleration of particle velocity. 00:36:36.620 --> 00:36:42.300 So that is where – what I was talking before that really doing Love waves 00:36:42.300 --> 00:36:46.420 analysis for teleseismic events or in general earthquakes event, 00:36:46.420 --> 00:36:52.030 as well for interferometry, it needs a little more understanding and 00:36:52.030 --> 00:37:00.600 really developing different interpretation and different processing tools. 00:37:00.600 --> 00:37:03.960 Here is doing interferometry for the Rayleigh waves, 00:37:03.960 --> 00:37:08.280 so in which we are looking at basically two segment of a cable, 00:37:08.280 --> 00:37:10.830 so that are aligned with each other. 00:37:10.830 --> 00:37:14.490 And we are looking at interferometry of the Rayleigh waves. 00:37:14.490 --> 00:37:17.760 I don’t know how many of you are familiar with this, 00:37:17.760 --> 00:37:22.260 that was a way simple – intuitive way of [inaudible] 00:37:22.260 --> 00:37:24.250 explaining what’s happening interferometry. 00:37:24.250 --> 00:37:29.800 So we have sources all over. We record correlations 00:37:29.800 --> 00:37:34.000 for all different sources. And that, what, really when we 00:37:34.000 --> 00:37:38.970 average over sources, we are going to look at the stationary point 00:37:38.970 --> 00:37:43.310 as a function of the azimuth. The black here is the geophones. 00:37:43.310 --> 00:37:45.860 The red is the DAS array. 00:37:45.860 --> 00:37:49.680 And what is important here, these two traces. 00:37:49.680 --> 00:37:53.460 Again, the red is the DAS. The black is the geophones. 00:37:53.460 --> 00:37:56.980 And for Rayleigh waves, the kinematic information and the waveform that 00:37:56.980 --> 00:38:02.750 you will get from those geophones and from the DAS is pretty comparable. 00:38:02.750 --> 00:38:08.120 So the interpretation of Rayleigh waves interferometry is relatively straightforward. 00:38:08.120 --> 00:38:12.180 For Love waves, it’s quite different because of a different sensitivity. 00:38:12.190 --> 00:38:17.780 So instead of having the maximum of amplitudes at the stationary point, 00:38:17.780 --> 00:38:23.190 that would be here and here, like we do have for the geophones, 00:38:23.190 --> 00:38:26.600 we do actually have around 45 degrees. 00:38:26.600 --> 00:38:32.910 So when we do actually [inaudible], assuming that we have only direction 00:38:32.910 --> 00:38:37.830 of sources – nature of sources, then we will get a signal 00:38:37.830 --> 00:38:42.680 that is much more complicated and doesn’t have a clear kinematic 00:38:42.680 --> 00:38:46.590 information as the geophones but is quite more complicated. 00:38:46.590 --> 00:38:50.750 Now, if we have arrays, we can filter it out, do bin forming, 00:38:50.750 --> 00:38:55.130 so there are many ideas of how to compensate and extract 00:38:55.130 --> 00:38:58.690 useful information for Love waves interferometry. 00:38:58.690 --> 00:39:02.560 The main method here, though, that is not as straightforward as 00:39:02.560 --> 00:39:06.210 Rayleigh waves and needs a way to understand what we are recording 00:39:06.210 --> 00:39:11.890 and what is the result of processing. And you need to do probably more 00:39:11.890 --> 00:39:17.820 source analysis and source directivity analysis than using geophones. 00:39:18.840 --> 00:39:21.580 So we are making progress in particular in the understanding 00:39:21.590 --> 00:39:28.600 how to interpret this data. However, definitely I think I don’t – 00:39:28.600 --> 00:39:31.850 I have a new Ph.D. working – students working on this. 00:39:31.850 --> 00:39:37.100 I do expect that there are going to be several first order progress, 00:39:37.100 --> 00:39:40.540 both in the understanding and in the processing. 00:39:40.540 --> 00:39:46.240 So this is – I’m using this data, actually, as – not only for 00:39:46.240 --> 00:39:49.600 seismologist, but also for computer scientist. 00:39:49.600 --> 00:39:55.780 Because this, in some way, is your typical big data analysis. 00:39:55.780 --> 00:39:59.620 Actually, we call that the big data stream. It’s not just big data. 00:39:59.620 --> 00:40:05.280 Is you do have a huge flow of data that keeps coming at you. 00:40:05.290 --> 00:40:12.380 Even in our teeny-tiny array at Stanford, we record several gigabytes per day, 00:40:12.380 --> 00:40:17.860 and that’s because we are band-passing at 25 hertz because we would not 00:40:17.860 --> 00:40:23.820 know what to do if we were recording at hundreds of hertz, of kilohertz. 00:40:23.820 --> 00:40:31.960 So that is a really – a challenge to – for continuous monitoring to deal with this 00:40:31.960 --> 00:40:36.150 huge – potentially huge amount of data. Let’s not talk about if we build 00:40:36.150 --> 00:40:40.130 the billion-sensor array or even a million-sensor array. 00:40:40.130 --> 00:40:46.270 So that is clearly an area where new ideas, new technology, 00:40:46.270 --> 00:40:51.180 is going to be crucial. And that is where getting our colleagues 00:40:51.180 --> 00:40:55.430 in computer science with a look at data analytics with big data can be important. 00:40:55.430 --> 00:41:01.070 So I actually gave this data – so I don’t know if you have seen it, but – not last 00:41:01.070 --> 00:41:06.830 weekend, but the weekend before, we had a Big Earth Hackathon at Stanford. 00:41:06.830 --> 00:41:13.620 And that’s – in which we had 130 students working on different 00:41:13.620 --> 00:41:18.890 Earth science data – large data set. And actually, this data was the 00:41:18.890 --> 00:41:22.840 only known satellite data that was part of the hackathon. 00:41:22.840 --> 00:41:28.850 So there were five satellite data and one seismic data that was – one week of our 00:41:28.850 --> 00:41:34.120 DAS data, and several students got interested and came up new ideas. 00:41:34.120 --> 00:41:39.570 So that is another interesting area that – and I totally control this data. 00:41:39.570 --> 00:41:44.400 So I don’t – I can give it away without any problem of asking permission. 00:41:44.400 --> 00:41:48.840 So we need to invent new algorithms to compensate for 00:41:48.840 --> 00:41:53.380 unusual sensitivity of DAS. I hinted a few minutes ago 00:41:53.380 --> 00:41:58.260 the fact that we are not really measuring acceleration and velocity. 00:41:59.580 --> 00:42:05.980 We need to estimate the location of the virtual receivers when – 00:42:05.990 --> 00:42:10.070 in telecom cables. So that is quite important. 00:42:10.070 --> 00:42:14.150 If we build a thousand- or million-sensor array, we need to 00:42:14.150 --> 00:42:18.260 know where those receivers are, where the cables are. 00:42:18.260 --> 00:42:22.460 So we need to have smart algorithms that give us an estimates where 00:42:22.460 --> 00:42:26.020 actually they are, just based on the data that we’re recording. 00:42:26.020 --> 00:42:29.410 We cannot rely on the maps of the telecom companies. 00:42:29.410 --> 00:42:31.690 That would be probably foolish. 00:42:31.690 --> 00:42:36.830 We need to – you saw pump noise. You saw car noise. 00:42:36.830 --> 00:42:40.700 We have very non-stationary – keep changing source of noise, 00:42:40.700 --> 00:42:42.620 in particular in urban environment. 00:42:42.620 --> 00:42:46.940 Good news that we have arrays. So doing noise separation 00:42:46.940 --> 00:42:52.460 and attenuation, identification, with arrays, that is something that seismology 00:42:52.460 --> 00:42:56.870 have looked – at least my kind of seismology have looked for long time. 00:42:56.870 --> 00:42:59.660 And, but again, here we have the stream of data. 00:42:59.660 --> 00:43:03.440 It’s not just one particular data set. 00:43:03.440 --> 00:43:06.680 We would like to detect, classify, and analyze the signal. 00:43:06.680 --> 00:43:09.140 We would like to find smaller earthquakes. 00:43:09.140 --> 00:43:12.640 That is one of the next thing that I’m going to do with my students. 00:43:12.640 --> 00:43:19.680 Can we actually, using our array, detect earthquakes that are not in your catalog? 00:43:19.680 --> 00:43:22.560 So far, we have used the one in your catalogs. 00:43:22.560 --> 00:43:28.980 If we want to use to – able to – for CO2 sequestration, or for just environmental 00:43:28.980 --> 00:43:35.280 monitoring, we need to be able to detect those events – cave-ins, and so on. 00:43:35.280 --> 00:43:42.180 And then, if we look at, for monitoring, the temperature has an effect on the 00:43:42.180 --> 00:43:47.510 signal as well. The fiber can be used as a temperature measurement as well. 00:43:47.510 --> 00:43:51.470 So we need to make sure, for example, that we identify – 00:43:51.470 --> 00:43:55.890 estimate the temperature effect and remove that from the seismic data. 00:43:55.890 --> 00:43:58.230 That should be not that difficult because probably 00:43:58.230 --> 00:44:04.430 have much longer time scales than the seismic themselves. 00:44:04.430 --> 00:44:11.350 And, as I said, that is – really we just start to scratch the surface. 00:44:11.350 --> 00:44:17.580 And one of my students – another student – her name is Fantine Huot. 00:44:17.580 --> 00:44:21.700 She is working with machine learning, and she has been playing around 00:44:21.700 --> 00:44:25.330 with this data quite a bit. We had actually a paper coming out 00:44:25.330 --> 00:44:30.700 on the – to policing the processing months ago, in which we were told some 00:44:30.700 --> 00:44:37.550 of the first things that we start to do in terms of machine learning – this data. 00:44:37.550 --> 00:44:41.460 So we are ready to process the data that we record. 00:44:41.460 --> 00:44:45.560 I think that really the bottom line, that we’re only getting started. 00:44:45.560 --> 00:44:49.580 We are really – there is a lot of space for young people 00:44:49.580 --> 00:44:55.080 to do first order contribution. And if, indeed, the tool becomes 00:44:55.080 --> 00:45:00.580 important and useful for permanent monitoring of, let’s say, 00:45:00.580 --> 00:45:05.680 CO2 sequestration or, as I mentioned, for really large-scale monitoring 00:45:05.680 --> 00:45:12.430 in urban environment, then those first order improvements 00:45:12.430 --> 00:45:16.720 really can get young people career started. 00:45:16.720 --> 00:45:21.030 So conclusion. We can continuously listen to the 00:45:21.030 --> 00:45:24.980 Earth and hear it well – not perfectly – by using 00:45:24.980 --> 00:45:30.870 cost-effective distributed acoustic sensors with fibers 00:45:30.870 --> 00:45:36.230 laid in PVC conduits, like the telecom, I repeated many times. 00:45:36.230 --> 00:45:42.430 Once that we take into account that we record the axial strain, 00:45:42.430 --> 00:45:45.570 we can extract the kinematics of waveforms pretty easily from the 00:45:45.570 --> 00:45:52.410 recorded data, both from discrete events as well using interferometry. 00:45:52.410 --> 00:45:55.600 We need to develop new signal processing and customize the 00:45:55.600 --> 00:45:59.960 algorithms for both the processing and the interpretation. 00:45:59.960 --> 00:46:04.380 And I would say that the experiment showed that there are potential. 00:46:04.390 --> 00:46:08.030 There are really promises in this technology for all the application 00:46:08.030 --> 00:46:13.770 that have been talking about. And this can open very much 00:46:13.770 --> 00:46:18.920 new opportunities for widespread seismic monitoring. 00:46:18.920 --> 00:46:24.190 And I was lucky that Andy invited me to come here because I do think 00:46:24.190 --> 00:46:27.530 that this is an area that a partnership between Stanford 00:46:27.530 --> 00:46:35.020 and USGS can be really proficient. We do have different strengths, 00:46:35.020 --> 00:46:40.160 but in the seismic monitoring and permanent seismic monitoring 00:46:40.160 --> 00:46:45.180 in a cost-effective way of an urban area, I think that we do have a lot of 00:46:45.180 --> 00:46:52.740 overlapping interest. And the good news is that I – this data, as I said, I own it. 00:46:52.740 --> 00:46:58.840 So if anybody is interest to work on the data, I will be happy to share. 00:46:58.840 --> 00:47:02.710 As well, as I mentioned earlier, Greg Beroza, Jack Baker, and I, 00:47:02.710 --> 00:47:05.740 we have a proposal, and part of the proposal is 00:47:05.740 --> 00:47:11.600 actually doing this in the city of San Francisco. 00:47:11.600 --> 00:47:15.710 And that if we record the data can be, again, 00:47:15.710 --> 00:47:18.530 quite a bit of collaborative projects as well. 00:47:18.530 --> 00:47:26.000 So I would like to end on this note. This is a long list of acknowledgement. 00:47:26.000 --> 00:47:29.480 And, in particular, I would like to acknowledge, again, the students 00:47:29.480 --> 00:47:34.360 that worked on this that are Eileen Martin and Siyuan Yuan. 00:47:34.360 --> 00:47:36.350 And I would like to thank you for your attention. 00:47:36.350 --> 00:47:40.200 I would be happy to answer all the question that I know how to answer. 00:47:40.200 --> 00:47:41.300 Thank you. 00:47:41.300 --> 00:47:47.040 [Applause] 00:47:49.080 --> 00:47:54.340 - So I’m a little – [inaudible] clarifying [inaudible]. 00:47:54.340 --> 00:47:57.860 So, like, in 100 channels [inaudible] Stanford array, 00:47:57.860 --> 00:48:01.740 do you have 100 interrogators? - No. One. 00:48:01.740 --> 00:48:04.720 Sorry. I might have not – gone too fast. 00:48:04.720 --> 00:48:08.500 So let’s go back to the map. 00:48:08.500 --> 00:48:14.710 Well, actually, let’s do the map just to – so just to give you a little more details. 00:48:14.710 --> 00:48:16.770 We have an interrogator in the Green’s building. 00:48:16.770 --> 00:48:20.650 That is more or less where my pointer is in the computer room 00:48:20.650 --> 00:48:26.400 of the School of Earth Sciences. And we connect there to the fiber. 00:48:26.400 --> 00:48:30.750 And the fiber goes around. And you see this is the channel count. 00:48:30.750 --> 00:48:34.990 And actually, we then spliced – as I said before, the cable itself 00:48:34.990 --> 00:48:39.990 has six fiber strands inside. 00:48:39.990 --> 00:48:46.420 We actually fused – at the same end, inside the computer room, two of them. 00:48:46.420 --> 00:48:50.380 So we – actually, we have 600 channels. 00:48:50.380 --> 00:48:55.150 We interrogate basically going in one direction and then coming back. 00:48:55.150 --> 00:49:01.080 But we have one interrogator. And altogether, that is about 00:49:01.080 --> 00:49:06.660 4.8 kilometers of fiber that we are interrogating with one interrogator. 00:49:06.660 --> 00:49:11.300 And in reality, you can go farther. As I said, 10, 20 kilometers, 00:49:11.300 --> 00:49:14.700 30 kilometers, that depends on many parameters. 00:49:14.700 --> 00:49:17.180 - So then – and the interrogator is a Michelson interferometer, 00:49:17.180 --> 00:49:21.260 so you’re getting – it’s got – it’s following fringe patterns? 00:49:21.260 --> 00:49:23.120 - That’s correct. 00:49:23.120 --> 00:49:27.980 And it has quite sophisticated signal processing machinery inside, 00:49:27.980 --> 00:49:32.600 but it’s basically doing interferometry between 00:49:32.600 --> 00:49:37.170 the laser signal going in and the laser signal that comes back. 00:49:37.170 --> 00:49:42.520 - So, in this case, it’s actually getting a reflection from an endpoint. 00:49:42.520 --> 00:49:44.340 Is that right? - Oh … 00:49:44.340 --> 00:49:46.490 - Not just little imperfections along the line. 00:49:46.490 --> 00:49:50.840 - Yeah. You can do the reflection. And actually, maybe some of you 00:49:50.840 --> 00:49:54.960 are familiar with the SAFOD experiment. 00:49:54.960 --> 00:50:01.100 Bill Ellsworth got OptaSense to record some data at SAFOD. 00:50:01.100 --> 00:50:06.700 And, which you did have a fiber there, mostly for strain measurements. 00:50:06.700 --> 00:50:09.680 And it has a – put the reflector there, and you can see 00:50:09.680 --> 00:50:14.730 the clear and strong reflection from there. But so the endpoint is not that 00:50:14.730 --> 00:50:18.660 interesting, or what the – my point of view is, they really were distributed that 00:50:18.660 --> 00:50:23.800 way, and we have a virtual sensor, as I said, every 7, 8 meters. 00:50:23.800 --> 00:50:27.020 And that is a parameter depends how much averaging 00:50:27.020 --> 00:50:30.700 you do over the gauge length. So there is a frequency. 00:50:30.700 --> 00:50:34.460 Depends at which frequency you want to record, which signal-to-noise, 00:50:34.460 --> 00:50:38.140 then you can decide basically the spacing of the virtual interrogators – 00:50:38.140 --> 00:50:39.570 of the virtual sensors, sorry. 00:50:39.570 --> 00:50:42.320 - And how much does an interrogator cost now? 00:50:42.780 --> 00:50:45.780 - I cannot really say that. 00:50:46.560 --> 00:50:53.260 As it is now, I think is in the $7,000 to $10,000. 00:50:53.260 --> 00:51:00.000 We are paying zero, so – and, as I said, is sophisticated 00:51:00.000 --> 00:51:02.710 signal processing and is still very expensive. 00:51:02.710 --> 00:51:06.810 But probably order of tens of these machines around the world. 00:51:06.810 --> 00:51:13.040 But if you find – if you can scale up and build billion-sensor arrays, 00:51:13.040 --> 00:51:18.000 then there is no reason why that cost should go down to only a few thousand. 00:51:18.010 --> 00:51:21.260 But that is my personal comment. 00:51:21.260 --> 00:51:25.620 Is not an official statement from interrogator vendors. 00:51:26.380 --> 00:51:30.000 [Silence] 00:51:30.860 --> 00:51:35.280 - Have you looked to see where the coupling – at what amplitude 00:51:35.290 --> 00:51:40.210 of shaking the coupling is lost or the signal starts to really degrade? 00:51:40.210 --> 00:51:44.750 - Wonderful question. Is one of the many things that I would like to do. 00:51:44.750 --> 00:51:47.540 And do that as a function of frequency as well. 00:51:47.540 --> 00:51:54.300 And part of that, I’m hoping to use both the fact that we record two different – 00:51:54.300 --> 00:51:58.380 interrogator with a different theoretical response. 00:51:58.390 --> 00:52:01.910 And by the way, interrogator company are very sensitive. 00:52:01.910 --> 00:52:06.640 So I’ve been given some information and some 00:52:06.640 --> 00:52:10.670 kind of nondisclosure because – but I would like to make the comparison. 00:52:10.670 --> 00:52:17.500 As well as the broadband sensors that we installed thanks to your help. 00:52:17.500 --> 00:52:20.580 So comparing the data from the broadband sensors that are really 00:52:20.580 --> 00:52:26.080 close to the array and the data recorded by the two interrogators 00:52:26.080 --> 00:52:29.860 that was contemporaneous. So we had the broadband sensors 00:52:29.860 --> 00:52:35.020 in the building basements and the two interrogators at the same time. 00:52:35.020 --> 00:52:38.200 So we have a data set there that is waiting for analysis 00:52:38.200 --> 00:52:42.050 in which I will try to get some answers to that question. 00:52:42.050 --> 00:52:45.220 At the moment, I have to say I really don’t know. 00:52:45.220 --> 00:52:50.600 - And completely different question is, yes, it’s nice to make use of 00:52:50.600 --> 00:52:54.500 telecom cables that already exist, but have you thought about 00:52:54.500 --> 00:52:59.240 what would be sort of the ideal arrangement or pattern 00:52:59.240 --> 00:53:04.200 of fiber optic cable to really try and get more than one component? 00:53:04.200 --> 00:53:07.250 Like, a grid-type pattern or something like that? 00:53:07.250 --> 00:53:12.920 - So I think that is an excellent question. In particular, the oil industry has 00:53:12.920 --> 00:53:18.560 looked at that because they can afford to put specialized fibers. 00:53:18.560 --> 00:53:23.950 So Shell, I think, may have some patents, but there is open publication. 00:53:23.950 --> 00:53:31.070 The basic idea is to have the fiber going on a helical trajectory. 00:53:31.070 --> 00:53:34.760 And then you can – then you can even elaborate on that. 00:53:34.760 --> 00:53:38.010 You can a helix within the helix. 00:53:38.010 --> 00:53:42.540 And so Shell has some publication on that. 00:53:42.540 --> 00:53:47.020 [inaudible] at University of Calgary has some publication on that, 00:53:47.020 --> 00:53:52.480 as well as Colorado School of Mines. So you can then, analytically, 00:53:52.480 --> 00:53:56.240 get different sensitivity at two different components. 00:53:56.240 --> 00:53:58.020 In the case that you put the one particular fiber, 00:53:58.020 --> 00:54:01.000 but you – so you deploy a specialized cable. 00:54:01.000 --> 00:54:04.650 The problem there, that building those cable is quite expensive. 00:54:04.650 --> 00:54:07.600 And then you need to have a specialized cable deployed. 00:54:07.600 --> 00:54:09.680 The other question, which maybe that’s what you 00:54:09.680 --> 00:54:19.930 were hinting, is, if we can choose in a open system in which we may 00:54:19.930 --> 00:54:27.820 have fibers all over the direction, what is the array geometry 00:54:27.820 --> 00:54:35.290 using the convention of just normal fiber that is measuring along the strain tensor 00:54:35.290 --> 00:54:41.600 along the fiber itself, what will be an optimal acquisition geometry? 00:54:41.600 --> 00:54:46.120 And that is another question that we would like to answer, but to be 00:54:46.120 --> 00:54:50.690 able to answer, we need to do that – so basically groundwork of 00:54:50.690 --> 00:54:55.790 understanding what is the sensitivity of each individual sensor – 00:54:55.790 --> 00:54:59.820 basically what the – some of the results that I showed you from Eileen’s 00:54:59.820 --> 00:55:05.780 thesis that I showed some of the slides. So it was an excellent question. 00:55:05.780 --> 00:55:09.580 And again, I’m sorry to say, yes, we are going to do that, 00:55:09.580 --> 00:55:14.100 but we are not there too precisely yet. 00:55:15.240 --> 00:55:19.030 - Yeah, fascinating talk. What’s the smallest seismic event 00:55:19.030 --> 00:55:23.470 that you’ve detected, and how far was that from the array? 00:55:23.470 --> 00:55:30.430 - Oh, that is – we have detected some on Stanford campus 00:55:30.430 --> 00:55:32.920 and the Felt Lake. I don’t know if you know where it is. 00:55:32.920 --> 00:55:36.270 It’s close to Arastradero Preserve. 00:55:36.270 --> 00:55:40.470 And that was something like 0.8, 0.9, I think. 00:55:40.470 --> 00:55:47.640 It was still flagged by your website. So that’s the way that we saw it. 00:55:47.640 --> 00:55:52.220 And if you want, I have some slides on this computer I would 00:55:52.220 --> 00:55:58.100 be happy to share to show to you. However, what we really would like, 00:55:58.100 --> 00:56:02.680 and what we don’t have an answer yet, is can we detect something that is 00:56:02.690 --> 00:56:08.530 not flagged in your – in your database on the web. 00:56:08.530 --> 00:56:15.070 And we have – hoping to use both pattern-matching as well 00:56:15.070 --> 00:56:18.660 some machine learning based on the database. 00:56:18.660 --> 00:56:24.030 Now that we have the database with 1,000 events, we may have 00:56:24.030 --> 00:56:28.310 enough of a database to, for example, train a new network and to be able to 00:56:28.310 --> 00:56:33.700 see, maybe, weaker ones. But it’s excellent question. 00:56:33.700 --> 00:56:36.210 As I said, the only one that I can think about is Felt Lake. 00:56:36.210 --> 00:56:39.000 But it was, like, 3 kilometers from the array. 00:56:39.000 --> 00:56:43.200 I have the exact numbers in the computer if you’re interested. 00:56:45.100 --> 00:56:54.040 [Silence] 00:56:54.720 --> 00:56:59.260 - So, you know, it’s measuring uniaxial strain along the fiber. 00:56:59.260 --> 00:57:03.010 But in your array here, you’ve got crossings. 00:57:03.010 --> 00:57:07.850 So presumably, you’re measuring uniaxial strain in orthogonal directions. 00:57:07.850 --> 00:57:13.330 So if I understand this right, that means that, with clever design 00:57:13.330 --> 00:57:17.050 of the layout or selection of the segments that you analyze, 00:57:17.050 --> 00:57:20.800 that you could get the full horizontal strain tensor. Is that correct? 00:57:20.800 --> 00:57:24.000 - That’s a part of the array geometry design. 00:57:24.000 --> 00:57:26.880 - Okay. - In our case, we – 00:57:26.880 --> 00:57:30.280 depends on the wavelengths. Because, unfortunately, 00:57:30.280 --> 00:57:34.920 for both branches of the array, the actually – the corner here, 00:57:34.920 --> 00:57:37.480 there are manholes. So they are not well-coupled, 00:57:37.480 --> 00:57:39.660 neither of them, at the manhole. 00:57:39.660 --> 00:57:42.760 But if the wavelength is sufficiently long, you can assume 00:57:42.760 --> 00:57:47.670 that whatever is outside of manhole is, indeed – is a kind of approximation 00:57:47.670 --> 00:57:49.670 of the point measurement. - Okay. 00:57:49.670 --> 00:57:53.360 And then I had another question. If you were to, for example, 00:57:53.360 --> 00:57:58.600 couple two ends of, you know, a section, is there any possibility of recording to 00:57:58.600 --> 00:58:04.130 DC basically static changes in strain? Or is that totally beyond? 00:58:04.130 --> 00:58:06.960 - In theory, yes. As you know, or you can – 00:58:06.960 --> 00:58:12.790 I think that this is really back-scattering. When we do strain meters, they do – 00:58:12.790 --> 00:58:17.580 it’s a different kind of back-scattering. Don’t remember the name, frankly. 00:58:17.580 --> 00:58:22.040 So that’s because it’s easier to – really, to do strain meter. 00:58:22.040 --> 00:58:24.440 And so we can use the different physical phenomena. 00:58:24.440 --> 00:58:30.030 Now, for Rayleigh back-scattering, I think that limitation is not physical, 00:58:30.030 --> 00:58:38.200 but is in the signal processing of the – of the interrogator but is quite promising, 00:58:38.200 --> 00:58:40.780 just looking at this spectrum. 00:58:40.780 --> 00:58:46.640 And the comparison of the – of the signal between the different – the two. 00:58:46.650 --> 00:58:49.900 I do think that already the newer generation has a 00:58:49.900 --> 00:58:54.830 better response at the low frequency. So far, the comparison, as I said, 00:58:54.830 --> 00:59:00.000 is qualitative and visual, but I think that they’re a good indication of that. 00:59:00.000 --> 00:59:06.420 What is the endpoint in terms of how close can get to DC? 00:59:06.960 --> 00:59:08.480 I don’t know. 00:59:11.240 --> 00:59:12.880 - Any other questions? 00:59:13.880 --> 00:59:16.580 Okay, so if you’d like to join us for lunch, probably just hang out 00:59:16.580 --> 00:59:21.180 here for a couple minutes. And let’s thank Biondo. Thanks. 00:59:21.180 --> 00:59:22.180 - Thank you. 00:59:22.200 --> 00:59:26.140 [Applause] 00:59:26.140 --> 00:59:35.040 [Silence]