1 00:00:04,040 --> 00:00:07,040 This is Geology Bites with Oliver Strumpel. 2 00:00:08,200 --> 00:00:11,800 In a couple of earlier podcast episodes, we've talked about 3 00:00:11,800 --> 00:00:15,400 using seismic waves to make images of the Earth's interior. 4 00:00:16,480 --> 00:00:20,600 Barbara Romanowitz introduced the topic, and Anna Ferreira 5 00:00:20,600 --> 00:00:24,440 talked about using anisotropy of seismic waves to see flows in 6 00:00:24,440 --> 00:00:27,920 the mantle. In both of these episodes, it 7 00:00:27,920 --> 00:00:30,880 was on the one hand, exciting to be able to image the lithosphere 8 00:00:30,880 --> 00:00:35,240 and mantle at all, and on the other hand, somewhat frustrating 9 00:00:35,280 --> 00:00:37,360 that our images were still so blurry. 10 00:00:38,520 --> 00:00:41,680 Clearly, having a sharper view of the mantle would be immensely 11 00:00:41,680 --> 00:00:45,880 desirable and help us answer all sorts of geological questions. 12 00:00:46,840 --> 00:00:50,400 In this episode, we explore the latest developments in the field 13 00:00:50,680 --> 00:00:53,800 and the obstacles that stand in the way of rapid progress. 14 00:00:54,960 --> 00:00:59,240 Andreas Fischner and his team develop and apply new techniques 15 00:00:59,240 --> 00:01:03,160 for converting seismic data into 3 dimensional pictures of the 16 00:01:03,160 --> 00:01:06,720 mantle. Recently, he has used fibre 17 00:01:06,720 --> 00:01:09,400 optic cables to provide information about Earth's 18 00:01:09,400 --> 00:01:13,400 structure and seismicity with unprecedented resolution. 19 00:01:13,920 --> 00:01:16,360 He is a professor in the Department of Earth and 20 00:01:16,360 --> 00:01:19,440 Planetary Sciences at the Federal Institute of Technology 21 00:01:19,440 --> 00:01:23,040 in Zurich. Andreas Fischner, welcome to 22 00:01:23,040 --> 00:01:25,120 Geology Bites. Thank you very much. 23 00:01:25,600 --> 00:01:29,120 Can you refresh listeners on the basic principles of seismic 24 00:01:29,120 --> 00:01:33,240 imaging or seismic tomography? The standard and still most 25 00:01:33,240 --> 00:01:37,240 useful way of explaining seismic tomography is with the analogy 26 00:01:37,240 --> 00:01:40,040 of medical ultrasound that many of us can relate to. 27 00:01:40,880 --> 00:01:45,400 In medical ultrasound, small loud speakers act as the sources 28 00:01:45,400 --> 00:01:48,240 of ultrasound waves. And those ultrasound waves, they 29 00:01:48,240 --> 00:01:52,640 travel through the human body. As they travel through different 30 00:01:52,640 --> 00:01:57,560 types of tissue, the tissue properties affect the properties 31 00:01:57,560 --> 00:02:00,080 of those waves. These properties include the 32 00:02:00,080 --> 00:02:02,880 speed of those waves, but also the amplitudes. 33 00:02:04,080 --> 00:02:08,560 And then we can record those ultrasound waves and knowing 34 00:02:08,840 --> 00:02:11,720 where they propagated, that knowing the physics of 35 00:02:11,720 --> 00:02:16,160 ultrasound wave propagation, we can reconstruct a 3D image 36 00:02:16,440 --> 00:02:18,480 provided that we have sufficient data. 37 00:02:19,000 --> 00:02:23,240 Now what is interesting to note here is that even in medical 38 00:02:23,240 --> 00:02:27,520 ultrasound, the positions of the transducers can essentially be 39 00:02:27,520 --> 00:02:31,000 arbitrary. Those medical ultrasound images 40 00:02:31,000 --> 00:02:35,760 have, compared to the size of the Earth, still a much poorer 41 00:02:35,760 --> 00:02:38,960 resolution than we can achieve for our planet. 42 00:02:39,120 --> 00:02:42,040 So actually our images are not that blurry. 43 00:02:42,040 --> 00:02:43,600 We are, in fact, doing pretty well. 44 00:02:44,360 --> 00:02:48,480 What is the best resolution the state-of-the-art methods can 45 00:02:48,480 --> 00:02:51,720 achieve today? This is difficult question and 46 00:02:51,720 --> 00:02:56,360 it depends a lot on the spatial scale of the experiment and on 47 00:02:56,360 --> 00:02:59,560 the data that we are collecting. So just to give you an example, 48 00:02:59,800 --> 00:03:03,360 with dedicated seismic experiments we can image the 49 00:03:03,360 --> 00:03:06,560 upper 200 meters of the earth, which are very relevant for 50 00:03:06,560 --> 00:03:10,080 seismic hazard, for example, with a resolution of about 10 51 00:03:10,080 --> 00:03:12,800 meters. Of course, in global tomography 52 00:03:12,800 --> 00:03:15,560 this is very different. I would say that order of 53 00:03:15,560 --> 00:03:19,960 magnitude, we can see details with spatial scales of about 10 54 00:03:19,960 --> 00:03:24,080 kilometres in the crust, about 100 kilometres in the upper 55 00:03:24,080 --> 00:03:27,320 mantle and maybe a few 100 kilometres in the lower most 56 00:03:27,320 --> 00:03:29,480 mantle close to the core mantle boundary. 57 00:03:30,520 --> 00:03:34,920 Our problem has always been and actually continues to be the mid 58 00:03:34,920 --> 00:03:38,680 mantle, but is also exciting because it means that there's 59 00:03:38,680 --> 00:03:41,600 actually still structure to discover in that part of the 60 00:03:41,600 --> 00:03:44,760 Earth. But let me mention another issue 61 00:03:44,760 --> 00:03:48,560 that seismic tomographers are struggling with, which is what 62 00:03:48,560 --> 00:03:52,040 exactly do geologists actually mean when they say resolution? 63 00:03:52,480 --> 00:03:56,600 And is that something that we can really quantify in a way 64 00:03:56,600 --> 00:03:59,880 that is useful for them? But first of all, I think it is 65 00:03:59,880 --> 00:04:03,000 important to find a more quantitative language so that we 66 00:04:03,000 --> 00:04:06,800 as tomographers understand which quantification of resolution a 67 00:04:06,800 --> 00:04:11,800 geologist would actually like. Is it more complicated than then 68 00:04:11,800 --> 00:04:14,720 when you talk about the resolution of an image, say 69 00:04:15,160 --> 00:04:19,079 where you talk about how much you can resolve or how many 70 00:04:19,079 --> 00:04:22,640 pixels you have in your X direction and Y direction? 71 00:04:22,840 --> 00:04:24,960 Yes, it is a lot more complicated. 72 00:04:25,120 --> 00:04:26,840 I will try to explain this intuitively. 73 00:04:27,760 --> 00:04:32,720 If your pixel or your voxel is very small, then resolution 74 00:04:32,720 --> 00:04:37,200 essentially means two things. One is the size of the voxel 75 00:04:37,200 --> 00:04:40,200 that you have. This is the the amount of detail 76 00:04:40,200 --> 00:04:44,240 you may see with your eye. But the other aspect of 77 00:04:44,240 --> 00:04:47,320 resolution that you're interested in is how well can 78 00:04:47,320 --> 00:04:50,360 you constrain A parameter inside that voxel. 79 00:04:50,440 --> 00:04:53,280 For example, how well do you know the P wave speed inside 80 00:04:53,280 --> 00:04:57,280 that voxel? And so the problem is that you 81 00:04:57,280 --> 00:05:02,320 can't have both 1 trades off against the other, meaning the 82 00:05:02,320 --> 00:05:06,600 larger you make the voxel, the more accurately you can say what 83 00:05:06,600 --> 00:05:08,480 P wave speed is inside the voxel. 84 00:05:09,360 --> 00:05:12,880 If you make your voxel smaller so that you see more detail, the 85 00:05:12,880 --> 00:05:16,680 consequence is that you see the properties of the voxel. 86 00:05:16,680 --> 00:05:19,600 For example, it's P wave speed with less accuracy. 87 00:05:20,080 --> 00:05:23,720 And part of the art of seismic tomography is to find a 88 00:05:23,720 --> 00:05:25,200 reasonable balance between the two. 89 00:05:25,960 --> 00:05:30,080 So at the end of the day, what does this mean in terms of 90 00:05:30,080 --> 00:05:33,960 structures we can say are revealed by seismic tomography 91 00:05:33,960 --> 00:05:36,280 today to a high level of confidence? 92 00:05:36,720 --> 00:05:39,440 There are several structural features that are being imaged 93 00:05:39,520 --> 00:05:42,680 very very consistently. Basically no matter how the 94 00:05:42,680 --> 00:05:45,640 tomography is set up from bottom to top. 95 00:05:45,760 --> 00:05:49,200 These are the large low shear velocity provinces and the ice 96 00:05:49,200 --> 00:05:53,240 cream plume, which is probably the only really plume like 97 00:05:53,240 --> 00:05:56,600 structure that is consistently being imaged from the core 98 00:05:56,600 --> 00:05:58,600 mantle boundary all the way to the surface. 99 00:05:59,360 --> 00:06:02,560 Then we very clearly see subtracting lithospheric slabs, 100 00:06:02,560 --> 00:06:07,240 some of those descending below the 660 into the lower mantle. 101 00:06:07,840 --> 00:06:10,160 By 660 you mean that's the mantle transition zone? 102 00:06:10,160 --> 00:06:14,280 Exactly at the 660 kilometer discontinuity inside the mantle 103 00:06:14,640 --> 00:06:18,760 then we very clearly see fast continental interiors, so old 104 00:06:18,760 --> 00:06:22,360 lithosphere that sometimes we image down to about 300 105 00:06:22,360 --> 00:06:25,200 kilometer depths. Good examples are Debaltic and 106 00:06:25,200 --> 00:06:28,800 the Canadian Shields. Also very clearly visible are 107 00:06:29,080 --> 00:06:32,520 the mid ocean ridges which are very pronounced low velocity 108 00:06:32,520 --> 00:06:35,800 features. And the good news is that as we 109 00:06:35,800 --> 00:06:40,520 collect more and more data, the agreement on the shape of these 110 00:06:40,520 --> 00:06:43,560 features in different tomographic images is becoming 111 00:06:43,560 --> 00:06:47,000 better and better. You and several other groups 112 00:06:47,000 --> 00:06:50,280 around the world are working on improving our imagery of the 113 00:06:50,280 --> 00:06:54,280 mantle, both in terms of resolution on a global scale and 114 00:06:54,280 --> 00:06:58,280 on various regional scales. Can you categorize the various 115 00:06:58,280 --> 00:07:01,480 obstacles that hinder our progress towards this end? 116 00:07:02,120 --> 00:07:04,720 I would say there are four different categories. 117 00:07:05,240 --> 00:07:08,600 The first one is inherent or physical non uniqueness. 118 00:07:08,920 --> 00:07:14,040 This is an uncertainty that we cannot remove even if we had 119 00:07:14,040 --> 00:07:17,600 perfect data. The second class are the many 120 00:07:17,600 --> 00:07:21,600 subjective choices that we have to make when setting up a 121 00:07:22,080 --> 00:07:26,960 tomographic inversion. The third is data coverage, 122 00:07:27,000 --> 00:07:29,640 especially in the oceans, polar regions, the southern 123 00:07:29,640 --> 00:07:33,840 hemisphere. And the fourth top one is the 124 00:07:33,840 --> 00:07:37,520 actual size of the computational problem and the fact that our 125 00:07:37,520 --> 00:07:41,640 computers, although they are often portrayed as growing fast, 126 00:07:42,000 --> 00:07:45,080 they actually grow too slowly for the problem that we would 127 00:07:45,080 --> 00:07:47,960 like to solve. OK, there's a lot there. 128 00:07:48,080 --> 00:07:50,840 Let's talk about these challenges one at a time, 129 00:07:51,320 --> 00:07:54,280 starting with the non uniqueness of the solution. 130 00:07:55,000 --> 00:07:57,800 What does that mean and how can we address it? 131 00:07:58,040 --> 00:08:01,320 So what means non uniqueness? There's a hard mathematical 132 00:08:01,320 --> 00:08:06,120 definition here, but the intuitive 1 is that there simply 133 00:08:06,120 --> 00:08:09,880 infinitely many tomographic images, tomographic models that 134 00:08:09,880 --> 00:08:14,520 all explain our data to within the observational errors and 135 00:08:14,520 --> 00:08:17,520 there's non uniqueness essentially has two origins. 136 00:08:18,000 --> 00:08:19,640 The first one is just the nature of the data. 137 00:08:19,640 --> 00:08:22,840 Our data are noisy and they are sparse. 138 00:08:23,200 --> 00:08:26,440 This means that for some regions in the earth, we just very few 139 00:08:26,440 --> 00:08:30,320 data to actually constrain those regions, which means that in our 140 00:08:30,320 --> 00:08:33,600 tomographic models in those regions we can basically do 141 00:08:33,600 --> 00:08:37,120 whatever we want and our data don't really see that. 142 00:08:38,320 --> 00:08:43,640 The second reason or the second origin of non uniqueness is just 143 00:08:43,640 --> 00:08:49,120 the physics of the problem. One aspect of this is that some 144 00:08:49,120 --> 00:08:52,200 information that the waves in principle pick up when they're 145 00:08:52,200 --> 00:08:55,560 travelling through the earth is being eliminated by attenuation. 146 00:08:56,600 --> 00:09:00,880 The consequence is that we do not really record, for instance, 147 00:09:00,920 --> 00:09:04,360 high frequency S waves that do propagate through the earth but 148 00:09:04,360 --> 00:09:06,840 just don't make it all the way from the source to our 149 00:09:06,840 --> 00:09:10,800 receivers. And the absence of those high 150 00:09:10,840 --> 00:09:14,400 frequencies limits the resolution that we can achieve. 151 00:09:14,600 --> 00:09:17,040 And there's really nothing that can be done about it, because we 152 00:09:17,040 --> 00:09:18,920 can't switch off attenuation in the Earth. 153 00:09:19,840 --> 00:09:23,160 Just to butt in here, so when you say S wave, that means share 154 00:09:23,160 --> 00:09:24,520 waves. Correct. 155 00:09:25,000 --> 00:09:28,360 And the problem here is that in the earth shear waves are 156 00:09:28,360 --> 00:09:30,880 attenuated much more strongly than P waves. 157 00:09:31,280 --> 00:09:34,760 So for example for a large earthquake we can record AP wave 158 00:09:34,760 --> 00:09:39,280 at a frequency of 1 Hertz, but the one Hertz S wave is not 159 00:09:39,280 --> 00:09:41,960 there. So what can we do about this? 160 00:09:41,960 --> 00:09:45,920 There are basically 2 clean and one dirty way of addressing non 161 00:09:45,920 --> 00:09:48,520 uniqueness. The first clean 1 is just 162 00:09:48,520 --> 00:09:51,960 collect more information and that information could be more 163 00:09:52,000 --> 00:09:55,400 seismic information by putting more stations recording for a 164 00:09:55,400 --> 00:09:59,480 longer time, but it could also be filling the seismic 165 00:09:59,480 --> 00:10:02,920 information gap by independent knowledge, for example from 166 00:10:02,920 --> 00:10:07,840 mineral physics or geodynamics. Then the other cleaner way of 167 00:10:07,840 --> 00:10:12,520 addressing is would be to just be honest and instead of just 168 00:10:12,520 --> 00:10:16,080 producing one model that people look at, produce many models 169 00:10:16,080 --> 00:10:19,120 that are plausible and look at how different they are given the 170 00:10:19,120 --> 00:10:22,760 data that we use. This is easier said than done 171 00:10:23,000 --> 00:10:26,560 because producing many many models is computationally very 172 00:10:26,560 --> 00:10:30,160 very expensive. So then comes The Dirty 1, and 173 00:10:30,160 --> 00:10:32,720 The Dirty 1 is called regularisation. 174 00:10:33,600 --> 00:10:38,520 And regularisation, loosely speaking, means that we make up 175 00:10:38,640 --> 00:10:43,000 information that we do not actually have in order to make 176 00:10:43,000 --> 00:10:48,240 our inversion algorithms stable and well behaved, and to produce 177 00:10:48,440 --> 00:10:51,200 just one model in a computationally efficient way 178 00:10:51,600 --> 00:10:53,600 that then people can look at and interpret. 179 00:10:54,520 --> 00:10:59,280 The classic example is that we fill in information, or 180 00:10:59,320 --> 00:11:03,080 artificial information, that the structure of the earth is 181 00:11:03,080 --> 00:11:05,640 smooth. There's not something that we 182 00:11:05,640 --> 00:11:07,960 know, but it's easy to implement. 183 00:11:08,240 --> 00:11:12,920 It makes the models easier to interpret and so we smooth and 184 00:11:12,920 --> 00:11:16,360 forcing our models to be smooth. This is the classic, most 185 00:11:16,360 --> 00:11:20,320 widely, and in fact always used method of regularization. 186 00:11:21,480 --> 00:11:25,720 OK, so you mentioned other subjective choices one has to 187 00:11:25,720 --> 00:11:29,040 make, in this case to do with selecting the physical 188 00:11:29,040 --> 00:11:33,840 properties that one would like to image and how they limit the 189 00:11:33,840 --> 00:11:37,600 quality of the results. So why does one have to select 190 00:11:37,960 --> 00:11:41,400 different data types and what are the implications of the 191 00:11:41,400 --> 00:11:44,760 choices one has to make? So the underlying problem is 192 00:11:44,760 --> 00:11:48,720 that seismic waves are sensitive to a very large number of 193 00:11:48,720 --> 00:11:53,040 physical properties in the Earth. 2 that we all know are 194 00:11:53,480 --> 00:11:55,400 the P and the S wave speed inside the Earth. 195 00:11:55,960 --> 00:11:59,080 But here's more. There's the attenuation of P 196 00:11:59,080 --> 00:12:02,280 waves and S waves. There's density, there is 197 00:12:02,280 --> 00:12:05,480 topography of the internal discontinuities, for example, 198 00:12:05,480 --> 00:12:08,760 the four hundred 10660 discontinuities and the core 199 00:12:08,760 --> 00:12:12,760 mental boundary, and there is a total of 19 additional 200 00:12:12,760 --> 00:12:16,080 parameters that describe seismic anisotropy in the Earth. 201 00:12:17,160 --> 00:12:21,520 Now it is intuitively clear that we can't constrain all of those 202 00:12:21,680 --> 00:12:26,600 physical properties at the same time, and as a consequence we 203 00:12:26,600 --> 00:12:29,240 need to make choices. For example, a scientist 204 00:12:29,240 --> 00:12:34,000 interested in attenuation in the Earth, which closely links to 205 00:12:34,000 --> 00:12:37,080 the distribution of temperature inside the Earth, may add 206 00:12:37,360 --> 00:12:41,200 attenuation to the standard P&S wave speed of a tomographic 207 00:12:41,200 --> 00:12:44,760 model. This is nice and good, but in 208 00:12:44,760 --> 00:12:49,000 doing so, the person also chooses to ignore all the other 209 00:12:49,000 --> 00:12:53,200 parameters. Now the issue here is that of 210 00:12:53,200 --> 00:12:58,200 course all of the other parameters have individually a 211 00:12:58,200 --> 00:13:02,720 small influence on the data, but there are many and they 212 00:13:02,720 --> 00:13:06,280 collaborate and their combined effect may actually be pretty 213 00:13:06,280 --> 00:13:10,200 large. And admittedly, so far we have 214 00:13:10,280 --> 00:13:12,400 no good technical solution for this problem. 215 00:13:12,720 --> 00:13:16,080 So I would say this is an open methodological challenge how to 216 00:13:16,080 --> 00:13:19,880 actually address this. And it is one of the reasons why 217 00:13:20,440 --> 00:13:24,160 tomographic images of attenuation are much more 218 00:13:24,160 --> 00:13:27,080 different than tomographic images of P and South wave 219 00:13:27,080 --> 00:13:29,600 speed, which are resolved much more robustly. 220 00:13:30,640 --> 00:13:35,080 There's also the question about how you parameterize the 221 00:13:35,080 --> 00:13:39,680 tomographic model. Can you explain that and why 222 00:13:39,760 --> 00:13:42,800 this presents a challenge to making better and more 223 00:13:42,800 --> 00:13:45,280 trustworthy imagery? So this is actually closely 224 00:13:45,280 --> 00:13:48,680 linked to the regularization issue that we just discussed, 225 00:13:49,000 --> 00:13:52,720 because the parameterization can be seen as a more subtle, more 226 00:13:52,720 --> 00:13:57,160 invisible way of regularizing A tomographic inversion. 227 00:13:57,640 --> 00:14:00,360 Now to explain this, it's a matter of fact that the Earth is 228 00:14:00,360 --> 00:14:04,400 a continuum, but all the methods that we are using and more 229 00:14:04,400 --> 00:14:07,880 fundamentally the computers that we use to run the inversion can 230 00:14:07,880 --> 00:14:12,080 only handle discrete things. And as a consequence, we somehow 231 00:14:12,080 --> 00:14:16,640 need to discretize the earth. And how this is done, again is a 232 00:14:16,640 --> 00:14:19,800 subjective choice. So some may partition the earth 233 00:14:19,800 --> 00:14:23,080 in two voxels that we already discussed, others may use 234 00:14:23,080 --> 00:14:26,800 polynomials, and again, others may use a combination of these 235 00:14:27,000 --> 00:14:30,960 or whatever basis functions they like, or they can plug into 236 00:14:30,960 --> 00:14:34,760 their numerical solver. And of course, in making 237 00:14:34,920 --> 00:14:39,160 different subjective choices on how to discretize the earth 238 00:14:39,680 --> 00:14:42,760 leads to slightly different tomographic images. 239 00:14:43,480 --> 00:14:46,800 And these differences just between the parameterization, 240 00:14:46,800 --> 00:14:50,200 they may be small, but the important thing here is that 241 00:14:50,200 --> 00:14:54,440 they add to the differences that we already get from explicit 242 00:14:54,440 --> 00:14:57,720 regularization, from data selection and many other things. 243 00:14:58,720 --> 00:15:03,040 The other kind of problem that you mentioned is data coverage. 244 00:15:03,480 --> 00:15:07,560 Does this apply both to the sources of seismic waves, IE 245 00:15:07,560 --> 00:15:11,760 where the earthquakes occur, as well as to the distribution of 246 00:15:11,760 --> 00:15:14,360 seismometers that detect the seismic waves from the 247 00:15:14,360 --> 00:15:16,000 earthquakes? Yes, absolutely. 248 00:15:16,160 --> 00:15:19,320 It is both. Obviously, we cannot move to 249 00:15:19,320 --> 00:15:21,400 earthquakes. There's nothing to be done. 250 00:15:21,520 --> 00:15:25,440 They are where they are. But in the past decades, we have 251 00:15:25,440 --> 00:15:30,240 seen really fast progress in improving coverage of seismic 252 00:15:30,240 --> 00:15:35,160 stations, for example, in the oceans, in the polar regions and 253 00:15:35,160 --> 00:15:38,000 in the Southern Hemisphere, although there are still far 254 00:15:38,000 --> 00:15:41,280 less seismic stations than in Europe, North America and China. 255 00:15:42,400 --> 00:15:46,520 Interestingly, currently due to its size, Russia is probably the 256 00:15:46,520 --> 00:15:50,640 biggest problem here as there are very few openly available 257 00:15:50,640 --> 00:15:54,440 data from there. Is that because the data has 258 00:15:54,440 --> 00:15:57,560 potential military implications? To some extent, yes. 259 00:15:58,840 --> 00:16:03,600 Lastly, you mentioned the computational problem, and I 260 00:16:03,600 --> 00:16:07,440 assume that's to do with the scaling of the numerical methods 261 00:16:07,440 --> 00:16:09,600 as you try and improve the imagery. 262 00:16:10,080 --> 00:16:13,440 So I suppose the amount of computation does not scale 263 00:16:13,440 --> 00:16:16,280 linearly with the quality of the imagery. 264 00:16:16,400 --> 00:16:18,720 This is absolutely true, and this is a tough one. 265 00:16:19,640 --> 00:16:24,760 The first part to the answer here is that resolution scales 266 00:16:24,760 --> 00:16:29,080 approximately with frequency. So this means, intuitively 267 00:16:29,080 --> 00:16:34,320 speaking, if I want to see a feature that is half as small 268 00:16:34,640 --> 00:16:38,360 inside the earth or inside the human body, I need to use 269 00:16:38,480 --> 00:16:41,640 frequencies that are twice as high. 270 00:16:43,000 --> 00:16:45,960 And so for example, if I have a 10 kilometer feature in the 271 00:16:45,960 --> 00:16:49,240 earth, I have a resolution of 10 kilometers somewhere and I would 272 00:16:49,240 --> 00:16:52,800 like to see a resolution of five kilometers, so a sharper image. 273 00:16:53,120 --> 00:16:58,240 Then I need to use frequencies that are twice as high as the 274 00:16:58,240 --> 00:17:02,960 ones that I have used before. Now the issue here is that 275 00:17:02,960 --> 00:17:06,839 computational cost does not scale linearly with frequency. 276 00:17:07,240 --> 00:17:10,680 In fact, the computational cost of the imaging problem. 277 00:17:10,839 --> 00:17:13,800 So if the seismic tomography it scales with frequency to the 278 00:17:13,800 --> 00:17:19,000 power of at least 5, not 5, at least five if all the 279 00:17:19,000 --> 00:17:22,200 computations run fine. Now what does this mean 280 00:17:22,440 --> 00:17:25,440 practically? In global scale waveform 281 00:17:25,440 --> 00:17:29,000 tomography we still miss a factor of 10 in frequency. 282 00:17:29,840 --> 00:17:34,560 So loosely speaking, as already said, we can observe seismic P 283 00:17:34,560 --> 00:17:39,600 waves at a frequency of 1 Hertz, but what we can simulate and 284 00:17:39,600 --> 00:17:44,440 actually bring into our seismic tomography is AP wave with the 285 00:17:44,440 --> 00:17:47,200 frequency of something like .1 Hertz. 286 00:17:47,320 --> 00:17:50,200 So there's this factor of 10 in frequency that we still miss 287 00:17:50,440 --> 00:17:53,840 between what we can simulate and what we actually have in the 288 00:17:53,840 --> 00:17:56,360 data. So now let's go back to that 289 00:17:56,360 --> 00:18:00,120 scaling frequency to the power of 5 at least. 290 00:18:00,560 --> 00:18:05,040 So this means that we miss a factor of 10 to the five S 291 00:18:05,280 --> 00:18:09,480 100,000 in, say, the power of our computers. 292 00:18:09,720 --> 00:18:14,320 So if we wanted to exploit the data that we actually have, we 293 00:18:14,320 --> 00:18:17,840 would need computers that are at least 100,000 times more 294 00:18:17,840 --> 00:18:21,240 powerful than the most powerful ones that we have right now. 295 00:18:21,840 --> 00:18:25,200 And I assume you're using GPU's and all the latest? 296 00:18:25,200 --> 00:18:27,280 That is already accounted for, yes. 297 00:18:28,240 --> 00:18:31,800 In the introduction, I mentioned that you've developed methods 298 00:18:31,800 --> 00:18:36,120 that use fiber optic cables. Are these existing 299 00:18:36,120 --> 00:18:39,800 telecommunications cables or are these specially installed ones 300 00:18:39,800 --> 00:18:43,680 for this purpose? And can you explain how fibre 301 00:18:43,680 --> 00:18:46,040 optic cables can give us information about earth 302 00:18:46,040 --> 00:18:46,920 structure? Yes. 303 00:18:46,920 --> 00:18:49,600 So concerning the first question, we can actually use 304 00:18:49,600 --> 00:18:52,960 both, for example in cities or in the oceans. 305 00:18:53,080 --> 00:18:56,720 The Co use of telecommunication fibres is just very, very 306 00:18:56,720 --> 00:18:59,120 convenient. We don't need to deploy them, 307 00:18:59,360 --> 00:19:01,400 they're already there. So this is pretty nice. 308 00:19:01,400 --> 00:19:05,000 This makes the cost of our experiments very low, provided 309 00:19:05,000 --> 00:19:07,320 that we actually get access to those fibers. 310 00:19:08,320 --> 00:19:12,320 In challenging terrain, which often includes glaciers or 311 00:19:12,320 --> 00:19:16,800 volcanoes, it can actually be much easier to deploy a fiber 312 00:19:16,800 --> 00:19:19,960 optic cable than to install many seismometers. 313 00:19:20,520 --> 00:19:24,040 So people are doing both Now. How does this work? 314 00:19:24,040 --> 00:19:27,160 Why does it work? The underlying idea is that 315 00:19:27,160 --> 00:19:31,640 optical fibers can be transformed or can be used as 316 00:19:31,760 --> 00:19:34,760 chains of densely spaced seismometers. 317 00:19:35,440 --> 00:19:39,040 Densely spaced here meaning spaced at about 1m. 318 00:19:39,440 --> 00:19:43,840 Which means that if one has a 10 kilometre long optical fibre, 319 00:19:44,040 --> 00:19:47,480 this can act as a chain of 10,000 seismometers, which is a 320 00:19:47,480 --> 00:19:50,160 lot compared to what seismologists are traditionally 321 00:19:50,160 --> 00:19:52,560 used to. Now why does this work? 322 00:19:52,840 --> 00:19:58,320 It works because those optical fibres, they contain teeny tiny 323 00:19:58,360 --> 00:20:00,680 heterogeneities from the fabrication. 324 00:20:01,760 --> 00:20:04,680 The telecommunication people, of course they don't like it 325 00:20:04,680 --> 00:20:07,040 because it attenuates the waves as they propagate. 326 00:20:07,400 --> 00:20:11,280 But for us this is nice because each time a laser pulse hits one 327 00:20:11,280 --> 00:20:16,520 of those small heterogeneity, it backscatters a tiny amount of 328 00:20:16,520 --> 00:20:19,160 the light to the beginning of the fiber. 329 00:20:19,680 --> 00:20:23,080 Now when the fiber deforms, for example when a seismic wave 330 00:20:23,080 --> 00:20:26,720 passes by, the scatter moves a tiny bit. 331 00:20:27,560 --> 00:20:31,920 And in response to the movement of the scatter, the travel time 332 00:20:31,920 --> 00:20:35,840 of the back scattered laser pulse gets modified a tiny bit. 333 00:20:36,760 --> 00:20:39,520 And that small modification of the travel time of the back 334 00:20:39,520 --> 00:20:42,520 scattered laser pulse in response to deformation of the 335 00:20:42,520 --> 00:20:45,320 fibre, this is actually something that can be measured. 336 00:20:46,080 --> 00:20:48,720 And that measurement of travel time shift of the back scattered 337 00:20:48,720 --> 00:20:53,360 light can be translated back into a distributed measurement 338 00:20:53,520 --> 00:20:56,440 of deformation along the fibre. And so we have those 339 00:20:56,440 --> 00:21:01,040 seismometers that produce seismograms, and we can use them 340 00:21:01,040 --> 00:21:04,560 just as we use regular seismograms in our tomographic 341 00:21:04,560 --> 00:21:07,400 inversions. Well that's really incredible. 342 00:21:07,560 --> 00:21:13,240 Can you even use these super long transoceanic fiber optic 343 00:21:13,240 --> 00:21:15,400 cables? Yes and no. 344 00:21:15,720 --> 00:21:19,280 There is a little bit of a trade off here, and the trade off is 345 00:21:19,280 --> 00:21:25,040 between the length of the fiber that one can interrogate and the 346 00:21:25,040 --> 00:21:27,760 spacing of the seismometers that I can achieve. 347 00:21:28,240 --> 00:21:34,280 Loosely speaking, the longer the fiber, the coarser the spatial 348 00:21:34,280 --> 00:21:37,480 resolution along the fiber, which means that if I have a 349 00:21:37,480 --> 00:21:40,760 fiber that is relatively short, a few kilometers or a few tenths 350 00:21:40,760 --> 00:21:44,200 of kilometers, I can have a very high spatial resolution meter 351 00:21:44,200 --> 00:21:47,520 scale. If I have a fiber that extends 352 00:21:47,880 --> 00:21:51,720 through the North Atlantic, then that spatial resolution along 353 00:21:51,720 --> 00:21:55,000 the fibre is much, much lower. So I can't have both. 354 00:21:55,520 --> 00:21:59,480 And how to deal with those ultra long fibres in the tomographic 355 00:21:59,480 --> 00:22:02,440 inversion or how to use them? This is still very much work in 356 00:22:02,440 --> 00:22:04,640 progress, still something that we need to figure out. 357 00:22:05,240 --> 00:22:08,880 Is this something to do with distinguishing between the many 358 00:22:09,000 --> 00:22:12,040 backscattered pulses that come from the impurities that if you 359 00:22:12,040 --> 00:22:14,520 have too many of them it's hard to distinguish them? 360 00:22:14,960 --> 00:22:18,880 This is 1 component. The more fundamental 1 is that 361 00:22:18,880 --> 00:22:22,560 the longer those backscatter pulses travel, the more they 362 00:22:22,560 --> 00:22:27,640 attenuate, and that attenuation eliminates some of the spatial 363 00:22:27,640 --> 00:22:30,520 information that you have when the fiber is shorter. 364 00:22:30,960 --> 00:22:33,360 So it's incredible. You're getting really orders of 365 00:22:33,360 --> 00:22:37,080 magnitude more effective seismometers than you would by 366 00:22:37,080 --> 00:22:39,800 traditional methods. Absolutely yes. 367 00:22:40,480 --> 00:22:44,320 And on the episode web page, we illustrate how you've used fibre 368 00:22:44,320 --> 00:22:49,000 optics to study Iceland's most active volcano and the Northeast 369 00:22:49,120 --> 00:22:53,160 Greenland Ice stream. We also show an example where 370 00:22:53,160 --> 00:22:56,240 fibre optic sensing was used to make a high resolution 371 00:22:56,240 --> 00:22:59,960 tomographic image of a seismically active caldera in 372 00:22:59,960 --> 00:23:04,000 California. In addition to the lithosphere 373 00:23:04,000 --> 00:23:08,680 and mantle, are we also able to use seismic methods to reveal 374 00:23:08,680 --> 00:23:12,080 structures within the core? I would also say that this is 375 00:23:12,080 --> 00:23:14,600 work in progress. People have been interested in 376 00:23:14,600 --> 00:23:18,440 the structure of the inner core for a very long time, and what 377 00:23:18,440 --> 00:23:22,560 is known pretty well is the seismic anisotropy of the inner 378 00:23:22,560 --> 00:23:27,160 core, which we can infer quite directly from the arrival times 379 00:23:27,160 --> 00:23:30,040 of P waves travelling through the core in different 380 00:23:30,040 --> 00:23:33,200 directions. Now, why is it difficult to make 381 00:23:33,280 --> 00:23:36,360 a 3D image of the core just as we do for the mantle? 382 00:23:37,040 --> 00:23:40,320 Well, essentially the problem is that the inner core occupies 383 00:23:40,480 --> 00:23:44,080 just about 1% of the Earth's volume, actually less. 384 00:23:44,360 --> 00:23:48,640 So this means we're looking at a teeny tiny volume of the Earth 385 00:23:49,520 --> 00:23:53,600 that is very far away from the surface, so very far away from 386 00:23:53,600 --> 00:23:55,720 where where we actually measure the information. 387 00:23:56,640 --> 00:24:01,000 And as a consequence, 3D images of the inner core are really 388 00:24:01,000 --> 00:24:03,360 just emerging. So the first ones have really 389 00:24:03,360 --> 00:24:08,200 appeared in the past few years and they're based on very 390 00:24:08,400 --> 00:24:11,560 careful data selection procedures. 391 00:24:11,800 --> 00:24:15,400 And I would say that even though this goes in a very promising 392 00:24:15,400 --> 00:24:19,080 direction, how robust these models are against the many 393 00:24:19,080 --> 00:24:22,560 subjective choices that we already discussed, I think this 394 00:24:22,560 --> 00:24:26,680 is still a little bit open. So there's something moving, but 395 00:24:26,680 --> 00:24:30,320 those images are not yet as robust as those that we have put 396 00:24:30,320 --> 00:24:34,720 across and the matter. So given all the numerous 397 00:24:34,720 --> 00:24:39,360 hurdles we've been discussing, what sort of progress do you 398 00:24:39,360 --> 00:24:43,480 think we'll be able to make, say, over the next 10 years? 399 00:24:44,040 --> 00:24:47,920 And would it be mainly one of the categories you mentioned of 400 00:24:47,920 --> 00:24:51,720 the challenges that would yield to our efforts to make 401 00:24:51,720 --> 00:24:53,920 significant improvements in seismic imagery? 402 00:24:54,440 --> 00:24:58,120 I'm actually quite hopeful and my guess is that we will see 403 00:24:58,120 --> 00:25:02,280 improvements mainly in two aspects, that's data coverage 404 00:25:02,520 --> 00:25:06,800 and computational methods. So for example, I'm pretty sure 405 00:25:06,800 --> 00:25:10,920 that fibre optic sensing will most likely allow us to expand 406 00:25:10,920 --> 00:25:14,200 coverage significantly into the oceans and to fill one of our 407 00:25:14,200 --> 00:25:17,280 major data gaps there. And hopefully also some 408 00:25:17,280 --> 00:25:19,840 political changes in certain countries will make a 409 00:25:19,840 --> 00:25:21,800 contribution. Now, when it comes to 410 00:25:21,800 --> 00:25:25,360 computation, my hope would be that the clever combination of 411 00:25:25,360 --> 00:25:28,800 the classical numerical methods that we're already using and 412 00:25:28,800 --> 00:25:32,160 evolving machine learning tools will allow us to overcome some 413 00:25:32,160 --> 00:25:33,960 of the current computational challenges. 414 00:25:34,600 --> 00:25:36,400 And here I want to stress that I think it's really the 415 00:25:36,400 --> 00:25:38,960 combination of the two. I think brute force machine 416 00:25:38,960 --> 00:25:40,840 learning will not bring us further. 417 00:25:41,120 --> 00:25:44,800 The classical numerical methods probably won't do so either. 418 00:25:45,160 --> 00:25:47,680 We need to think about how to combine them in a temper way, 419 00:25:48,040 --> 00:25:49,320 and that is really just beginning. 420 00:25:50,280 --> 00:25:53,680 But do you think there's some fundamental physical or 421 00:25:53,680 --> 00:25:57,680 mathematical barriers that will ultimately limit the quality of 422 00:25:57,680 --> 00:26:01,240 the seismic imagery we can get? Ultimately, I think 423 00:26:01,480 --> 00:26:06,080 mathematical, No, I think there it is a matter of being clever 424 00:26:06,640 --> 00:26:10,840 and having bigger computers. But there certainly are 425 00:26:10,840 --> 00:26:13,920 barriers. No matter how densely we cover 426 00:26:13,920 --> 00:26:16,960 the earth with centers at the surface or how big our super 427 00:26:16,960 --> 00:26:21,040 computers become, we will not be able to place earthquakes where 428 00:26:21,040 --> 00:26:23,840 we need them to be. And that would actually be deep 429 00:26:23,840 --> 00:26:26,280 inside the earth. If we wanted to improve 430 00:26:26,280 --> 00:26:31,080 resolution drastically, we would need to have earthquakes near 431 00:26:31,080 --> 00:26:33,360 the core mental boundary, so distribute it through the 432 00:26:33,360 --> 00:26:34,960 volume. But that's not going to happen. 433 00:26:35,280 --> 00:26:39,160 And also we will just not be able to switch off attenuation. 434 00:26:39,160 --> 00:26:43,240 As I already explained, we will not have one Hertz as waves, so 435 00:26:43,240 --> 00:26:45,360 that information that is contained in those high 436 00:26:45,360 --> 00:26:47,560 frequency waves, we will just never have it. 437 00:26:48,480 --> 00:26:52,320 And also seismic noise will always be present and that will 438 00:26:52,320 --> 00:26:55,560 always inherently the certainty that we can achieve. 439 00:26:56,520 --> 00:26:58,040 What are you working on at the moment? 440 00:26:58,480 --> 00:27:00,640 When it comes to seismic tomography, I would say we have 441 00:27:00,640 --> 00:27:03,440 three main topics. The first one already discussed 442 00:27:03,440 --> 00:27:06,640 is the incorporation of fiber optic sensing data into seismic 443 00:27:06,640 --> 00:27:10,640 waveform tomography. This is a tough one because in 444 00:27:10,640 --> 00:27:14,240 contrast to traditional seismometers, we don't know as 445 00:27:14,240 --> 00:27:18,000 well what the physical quantity is that those optical fibers are 446 00:27:18,000 --> 00:27:22,480 really recording. Then we are working a lot on the 447 00:27:22,480 --> 00:27:25,560 quantification of image quality, and this actually has two 448 00:27:25,560 --> 00:27:27,400 components. It has a technological 449 00:27:27,400 --> 00:27:31,040 component, a computational one, but it also has the other 450 00:27:31,040 --> 00:27:34,960 component that we discussed earlier, what are measures of 451 00:27:34,960 --> 00:27:38,960 resolution that geologists actually find useful that enter 452 00:27:39,520 --> 00:27:42,560 different types of geological interpretation? 453 00:27:42,840 --> 00:27:44,840 And there this is not technological. 454 00:27:44,840 --> 00:27:46,960 This is a matter of having closer interaction with 455 00:27:46,960 --> 00:27:50,960 geologists who tell us more precisely what exactly they want 456 00:27:50,960 --> 00:27:54,440 to know, what is resolution for them, because in many cases 457 00:27:54,680 --> 00:27:58,360 resolution for us applied mathematicians is something 458 00:27:58,360 --> 00:28:01,320 else, and this needs to be homogenized well. 459 00:28:01,320 --> 00:28:03,600 And then finally, the third big topic is overcoming 460 00:28:03,600 --> 00:28:06,880 computational challenges that allow us to include more of the 461 00:28:06,880 --> 00:28:11,000 data that we already have. Andreas Fischner, thank you very 462 00:28:11,000 --> 00:28:13,480 much. Many thanks to you, it was 463 00:28:13,480 --> 00:28:16,480 really a pleasure. To see pictures and 464 00:28:16,480 --> 00:28:21,880 illustrations that support this podcast, go to geologybytes.com, 465 00:28:22,000 --> 00:28:24,760 where you'll also find a subject matter index of all the 466 00:28:24,760 --> 00:28:27,120 episodes. There you can also give me 467 00:28:27,120 --> 00:28:31,080 feedback which I welcome, as well as sign up to get my emails 468 00:28:31,080 --> 00:28:32,280 about new episodes.