WEBVTT 00:00:11.120 --> 00:00:16.879 lost in the fog of a cosmic storm floating on 00:00:16.879 --> 00:00:21.120 whimsical wavelengths is the norm dancing through 00:00:21.120 --> 00:00:31.219 the stars chasing spectrums of light It's that 00:00:31.219 --> 00:00:33.719 time again. It's time for Whimsical Wavelengths, 00:00:33.719 --> 00:00:36.079 where we uncover the best scientific pickup lines 00:00:36.079 --> 00:00:38.500 like this one for all those geophysicists out 00:00:38.500 --> 00:00:41.439 there. You must be a gravity anomaly because 00:00:41.439 --> 00:00:45.899 you're attracting all my attention. This time 00:00:45.899 --> 00:00:48.479 on Whimsical Wavelengths, we're going to explore 00:00:48.479 --> 00:00:50.759 machine learning and artificial intelligence 00:00:50.759 --> 00:00:55.539 applied to... Computer overlords rejoice. No, 00:00:55.560 --> 00:00:58.299 really, though. Artificial intelligence and machine 00:00:58.299 --> 00:01:01.700 learning are extremely useful in scientific applications. 00:01:02.119 --> 00:01:08.299 No copyright issues here. What? I think, actually, 00:01:08.359 --> 00:01:10.299 this is going to be the first time we've really 00:01:10.299 --> 00:01:14.239 applied or really using artificial intelligence 00:01:14.239 --> 00:01:16.459 or machine learning as part of the underlying 00:01:16.459 --> 00:01:18.739 science. Yes, here in Whimsical Wavelengths, 00:01:18.739 --> 00:01:20.560 it's come up a few times when talking about certain 00:01:20.560 --> 00:01:24.290 topics on... different aspects of science, but 00:01:24.290 --> 00:01:26.930 usually really kind of near the end of an episode 00:01:26.930 --> 00:01:29.010 where you're like, where is this going for the 00:01:29.010 --> 00:01:33.569 future? So where to start for today? I guess 00:01:33.569 --> 00:01:36.189 what is geophysics and inversions seems like 00:01:36.189 --> 00:01:39.250 a good place. Geophysics is basically combining 00:01:39.250 --> 00:01:42.170 my two favorite subjects, geology and physics, 00:01:42.310 --> 00:01:46.250 together, using physics to study the Earth, its 00:01:46.250 --> 00:01:49.709 structure, and its processes. And geophysical 00:01:49.709 --> 00:01:52.870 inversions? That one ends up being a little bit 00:01:52.870 --> 00:01:55.310 harder to define without using a bunch of jargon. 00:01:55.450 --> 00:01:59.109 Here is going to be our starting point. It's 00:01:59.109 --> 00:02:00.290 going to get a lot more technical, especially 00:02:00.290 --> 00:02:03.269 once our guest joins us. A geophysical inversion 00:02:03.269 --> 00:02:06.489 is a mathematical process of estimating the distribution 00:02:06.489 --> 00:02:09.210 of physical properties like conductivity, density, 00:02:09.469 --> 00:02:13.349 or perhaps even velocity, wave speed velocity, 00:02:13.430 --> 00:02:16.569 if we're talking about seismic data. We're looking 00:02:16.569 --> 00:02:18.349 at these properties beneath the Earth's surface 00:02:18.349 --> 00:02:21.719 from measurements... collected usually on or 00:02:21.719 --> 00:02:26.080 above the Earth's surface. So obviously when 00:02:26.080 --> 00:02:29.520 it comes to the Earth, we're not able to tunnel 00:02:29.520 --> 00:02:31.580 really easily. So when we try to measure something, 00:02:31.699 --> 00:02:33.080 we're measuring it at the Earth's surface. We 00:02:33.080 --> 00:02:36.340 can't measure these properties directly in situ. 00:02:36.740 --> 00:02:38.620 I'm not going to be there at a kilometer beneath 00:02:38.620 --> 00:02:42.000 the Earth's surface with an ohmmeter measuring 00:02:42.000 --> 00:02:44.400 the resistance of all these rocks everywhere 00:02:44.400 --> 00:02:46.280 underneath the Earth. It's just not feasible 00:02:46.280 --> 00:02:49.360 or possible. So inversions are special because 00:02:49.360 --> 00:02:51.620 they use the physical properties or the physical 00:02:51.620 --> 00:02:54.939 measurements that we take on the surface and 00:02:54.939 --> 00:02:57.580 then work backwards to infer what the Earth must 00:02:57.580 --> 00:02:59.620 be like to produce those observations through 00:02:59.620 --> 00:03:03.759 solving the physics through mathematics. The 00:03:03.759 --> 00:03:06.800 data itself, though, I've spent years of my life 00:03:06.800 --> 00:03:09.620 doing this and collecting it across the most 00:03:09.620 --> 00:03:13.099 remote places in North America. The data is typically 00:03:13.099 --> 00:03:15.979 collected via helicopter walking across the terrain. 00:03:16.159 --> 00:03:19.219 Me, I was doing this on the ground, meaning I 00:03:19.219 --> 00:03:21.740 was walking many, many, many, many, many, many 00:03:21.740 --> 00:03:25.000 kilometers or miles carrying wire and instruments 00:03:25.000 --> 00:03:26.900 across hills, swamps and mountains and everything 00:03:26.900 --> 00:03:29.240 else in between, pushing buttons and problem 00:03:29.240 --> 00:03:31.680 solving. And problem solving is a really big 00:03:31.680 --> 00:03:33.539 thing when you're hundreds of kilometers from 00:03:33.539 --> 00:03:36.800 the nearest road or town of anything. Meaning 00:03:36.800 --> 00:03:39.719 if something breaks, you have to kind of cobble 00:03:39.719 --> 00:03:42.599 it back together to continue to collect data 00:03:42.599 --> 00:03:46.879 or handling crew dynamics when somebody wants 00:03:46.879 --> 00:03:50.500 to quit or doesn't like somebody else. My point 00:03:50.500 --> 00:03:54.900 here is that the data can be really hard to collect. 00:03:55.219 --> 00:03:57.500 Some examples of data that you can collect are 00:03:57.500 --> 00:04:00.360 measuring changes to the Earth's gravitational 00:04:00.360 --> 00:04:03.500 or magnetic field over space. So walking across 00:04:03.500 --> 00:04:05.680 the Earth's terrain, pressing buttons, collecting 00:04:05.680 --> 00:04:09.020 how that field changes spatially. or looking 00:04:09.020 --> 00:04:12.639 at how magnetic fields or electric fields travel 00:04:12.639 --> 00:04:15.400 through the Earth. Once we have the measurements, 00:04:15.639 --> 00:04:17.920 the inversion, like I've already alluded to, 00:04:18.040 --> 00:04:21.120 is a mathematical way of taking those measurements 00:04:21.120 --> 00:04:23.579 and trying to get something or looking at how 00:04:23.579 --> 00:04:27.540 different physical properties change beneath 00:04:27.540 --> 00:04:30.899 the Earth's surface. To get a model, particularly 00:04:30.899 --> 00:04:34.079 a 3D model of the Earth, requires a lot of assumptions 00:04:34.079 --> 00:04:37.660 and simplifications. because there are way more 00:04:37.660 --> 00:04:40.819 unknowns than knowns. We're collecting data at 00:04:40.819 --> 00:04:44.040 the Earth's surface and trying to infer what's 00:04:44.040 --> 00:04:47.680 going on deep beneath the surface. Broadly speaking, 00:04:47.839 --> 00:04:49.319 that means we have an infinite number of solutions 00:04:49.319 --> 00:04:51.579 or models that can reproduce the data we've collected 00:04:51.579 --> 00:04:54.139 at the surface. Today we're going to explore 00:04:54.139 --> 00:04:56.839 tools that we can use to limit that number or 00:04:56.839 --> 00:04:59.519 infinite number and illuminate what we cannot 00:04:59.519 --> 00:05:03.050 directly observe. We're going to... illuminate 00:05:03.050 --> 00:05:05.550 the void of nothingness you know some people 00:05:05.550 --> 00:05:08.149 fear the void or the unknown but geophysicists 00:05:08.149 --> 00:05:12.629 just call it a zone of low sensitivity time to 00:05:12.629 --> 00:05:15.069 dive in and bring in today's guest like myself 00:05:15.069 --> 00:05:17.389 he spent time in rural areas of the world pushing 00:05:17.389 --> 00:05:19.310 buttons and pulling wire across the landscape 00:05:19.310 --> 00:05:21.310 differently though he's much more mathematically 00:05:21.310 --> 00:05:24.589 inclined to build the tools i'm privileged to 00:05:24.589 --> 00:05:27.680 leverage to probe the earth so Today, please 00:05:27.680 --> 00:05:30.860 welcome my co -worker, PhD candidate at University 00:05:30.860 --> 00:05:34.279 of British Columbia and geophysical guru, Jonathan 00:05:34.279 --> 00:05:38.360 Kataj. Thanks. Thanks for having me on the show 00:05:38.360 --> 00:05:40.980 and such a great introduction. That's great. 00:05:42.220 --> 00:05:44.920 Short and sweet, but we'll get into all of your 00:05:44.920 --> 00:05:48.439 backstory in just a second. First, PhD land, 00:05:48.699 --> 00:05:50.720 at least that's how I called it when I was there, 00:05:50.819 --> 00:05:53.939 using as little jargon as you can, which is... 00:05:54.139 --> 00:05:56.660 Always tough. Describe what you expect your thesis 00:05:56.660 --> 00:06:00.959 will aim to answer or investigate. Yeah, well, 00:06:01.040 --> 00:06:04.540 I guess as simple as I can. Yeah, so I'm using 00:06:04.540 --> 00:06:08.699 image segmentation methods, which are usually 00:06:08.699 --> 00:06:12.660 not typically used in geophysics. And I'm trying 00:06:12.660 --> 00:06:14.720 to use them to answer the questions of how we 00:06:14.720 --> 00:06:18.579 can resolve our subsurface images better, given 00:06:18.579 --> 00:06:23.399 our data. We tend to use really... kind of just 00:06:23.399 --> 00:06:25.620 regularizers and they're pretty general that 00:06:25.620 --> 00:06:27.819 we're just looking for smooth models and the 00:06:27.819 --> 00:06:30.300 images are they're not so much blurry they're 00:06:30.300 --> 00:06:31.860 a whole bunch of stuff they're a whole lot of 00:06:31.860 --> 00:06:34.439 everything but we can uh but what i'm trying 00:06:34.439 --> 00:06:37.279 to do is use segmentation methods um to help 00:06:37.279 --> 00:06:41.199 resolve that okay image segmentation methods 00:06:41.199 --> 00:06:46.379 what what what are those i mean i'm playing koi 00:06:46.379 --> 00:06:48.980 because i i I know more than your average person, 00:06:49.040 --> 00:06:51.620 but I don't feel like the full explanation, like 00:06:51.620 --> 00:06:54.199 what is, everyone knows what an image is. So 00:06:54.199 --> 00:06:57.680 what is image segmentation methods? Like give 00:06:57.680 --> 00:07:01.699 an example from, I don't know, using some AI 00:07:01.699 --> 00:07:03.779 thing, like recognizing streets or something. 00:07:04.269 --> 00:07:07.110 Yeah, well, we could even relate it to self -driving 00:07:07.110 --> 00:07:10.290 cars. They've got to take their images or take 00:07:10.290 --> 00:07:12.610 in images and be able to isolate structures. 00:07:13.129 --> 00:07:15.649 So I guess if you want to put it down to it, 00:07:15.689 --> 00:07:17.629 image segmentation is just trying to identify 00:07:17.629 --> 00:07:22.410 areas of commonality, and it kind of gives boundaries 00:07:22.410 --> 00:07:25.410 to objects and structures that you are able to 00:07:25.410 --> 00:07:30.189 then interpret. Okay, and then you're using this 00:07:30.189 --> 00:07:35.279 to look inside the Earth. i guess you're trying 00:07:35.279 --> 00:07:38.839 to use images of the 3d models that you can build 00:07:38.839 --> 00:07:41.339 from doing things like geophysics that we infer 00:07:41.339 --> 00:07:44.300 that they're fuzzy and then you're adding this 00:07:44.300 --> 00:07:47.500 process to try to infer structure and boundaries 00:07:47.500 --> 00:07:51.120 instead of something fuzzy Yeah, exactly. We're 00:07:51.120 --> 00:07:53.879 just trying to find these edges or just highly 00:07:53.879 --> 00:07:56.980 define these structures in the subsurface. And 00:07:56.980 --> 00:07:59.279 by using these image segmentation, we can kind 00:07:59.279 --> 00:08:02.319 of refine and improve our regularization to be 00:08:02.319 --> 00:08:05.060 able to pull out more like geologically real 00:08:05.060 --> 00:08:10.800 models. All right. Another deviation. Regularization. 00:08:12.860 --> 00:08:17.740 What is it? Yeah, it's pretty much the term in, 00:08:17.839 --> 00:08:19.819 well, I'm going to probably use some more jargon 00:08:19.819 --> 00:08:22.139 here, in the objective function. So some kind 00:08:22.139 --> 00:08:24.920 of function that we want to minimize. But yeah, 00:08:25.000 --> 00:08:27.199 to make it easier is the regularization holds 00:08:27.199 --> 00:08:29.600 all our prior assumptions, all the knowledge 00:08:29.600 --> 00:08:32.919 that we know or what we expect for an outcome 00:08:32.919 --> 00:08:36.200 of the model that we try to create. So you can 00:08:36.200 --> 00:08:39.399 use this word regularization on any problem that 00:08:39.399 --> 00:08:42.240 you're adding. prior information on, whether 00:08:42.240 --> 00:08:44.259 it's topography or anything that you're trying 00:08:44.259 --> 00:08:47.879 to... I keep wanting to use the word invert, 00:08:48.059 --> 00:08:51.279 but I'm also trying not to use jargon. So any 00:08:51.279 --> 00:08:53.379 problem you're trying to solve through mathematics, 00:08:53.659 --> 00:08:56.059 the regularization is providing some kind of 00:08:56.059 --> 00:08:58.899 structure to the solving of that problem. Is 00:08:58.899 --> 00:09:01.100 that fair to say? Exactly. We're just trying 00:09:01.100 --> 00:09:04.620 to give it as much information as we can on what 00:09:04.620 --> 00:09:08.019 we want to expect out of the model. Awesome. 00:09:08.059 --> 00:09:10.309 So now we're going to... back up before we go 00:09:10.309 --> 00:09:13.750 forward, because to get to PhD land, your journey 00:09:13.750 --> 00:09:17.990 wasn't exactly a straight line. So what was the 00:09:17.990 --> 00:09:20.470 winding path that brought you to this point of 00:09:20.470 --> 00:09:26.429 the world? Oof, that's a long story. Well, first, 00:09:26.629 --> 00:09:30.529 well, kind of similar. I really liked dinosaurs. 00:09:30.769 --> 00:09:33.710 I really liked volcanoes back when I was really 00:09:33.710 --> 00:09:37.580 young. But, you know, then I kind of just wandered 00:09:37.580 --> 00:09:40.740 into math and science land and thought I would 00:09:40.740 --> 00:09:43.539 become an engineer first so I tried that and 00:09:43.539 --> 00:09:47.539 then yeah then I found a geophysics class and 00:09:47.539 --> 00:09:50.580 it really hooked me so I decided well I'm gonna 00:09:50.580 --> 00:09:53.100 go and do my bachelor's in geophysics so I went 00:09:53.100 --> 00:09:56.879 and did that and after four years of yeah first 00:09:56.879 --> 00:09:59.600 doing engineering and then into geophysics I 00:09:59.600 --> 00:10:02.779 really wanted to get into industry so then industry 00:10:02.779 --> 00:10:05.799 I joined yeah an active data acquisition group 00:10:06.240 --> 00:10:09.360 and just toured the world collecting data all 00:10:09.360 --> 00:10:12.600 over the place. All sorts of data, mostly what 00:10:12.600 --> 00:10:14.980 we call direct current resistivity and induced 00:10:14.980 --> 00:10:17.559 polarization, where we're just ejecting lots 00:10:17.559 --> 00:10:19.600 of current in the ground and mapping electric 00:10:19.600 --> 00:10:22.960 fields. And then from there, I don't know, I 00:10:22.960 --> 00:10:24.740 guess I got a little bored. So then I got into 00:10:24.740 --> 00:10:27.840 programming and then I started writing software 00:10:27.840 --> 00:10:31.620 to do signal processing. And then I joined another 00:10:31.620 --> 00:10:35.200 group that had a lot of data. So at the time... 00:10:35.519 --> 00:10:37.100 the stuff that we use, like the software that 00:10:37.100 --> 00:10:39.360 we use to interpret that data was insufficient. 00:10:39.580 --> 00:10:43.399 So that made me get into Inversion, which I joined 00:10:43.399 --> 00:10:46.799 a group called Simpeg. It's an open source group 00:10:46.799 --> 00:10:49.620 out of the University of British Columbia. And 00:10:49.620 --> 00:10:51.620 I played around with them for a while, which 00:10:51.620 --> 00:10:54.399 then, you know, they kind of grew on me. And 00:10:54.399 --> 00:10:55.679 they're like, well, John, why don't you do a 00:10:55.679 --> 00:11:02.370 PhD? So here I am. Yeah, I guess that's one thing 00:11:02.370 --> 00:11:04.629 we definitely share in common. Although I got 00:11:04.629 --> 00:11:07.169 to say, I was only really within Canada when 00:11:07.169 --> 00:11:09.730 I was traveling about pushing buttons. You really 00:11:09.730 --> 00:11:13.070 got to go the whole world. So at some point I 00:11:13.070 --> 00:11:15.850 will circle back near the end and you're going 00:11:15.850 --> 00:11:18.190 to have to give us a story of the craziest place 00:11:18.190 --> 00:11:20.789 you've ever gone or the craziest event without 00:11:20.789 --> 00:11:23.149 giving too much away so you don't get in trouble. 00:11:26.539 --> 00:11:29.440 So there's always, as we've been kind of alluding, 00:11:29.559 --> 00:11:31.220 like you went and collected the data, then you 00:11:31.220 --> 00:11:33.980 got into inversions. So there's always two sides 00:11:33.980 --> 00:11:36.960 to this, two parts, which might be a whole side 00:11:36.960 --> 00:11:40.940 discussion, the data and what happens afterwards 00:11:40.940 --> 00:11:44.460 or in our case for our day jobs, the inversion. 00:11:44.639 --> 00:11:47.039 So let's dwell on the data just for a moment 00:11:47.039 --> 00:11:48.659 before we get into the more technical stuff. 00:11:49.299 --> 00:11:53.460 Like how is the data collected generally? Like 00:11:53.460 --> 00:11:57.490 what's involved? Well, it all depends on the 00:11:57.490 --> 00:12:00.389 survey itself, but the ones that I've been into, 00:12:00.549 --> 00:12:04.190 electromagnetic surveys, oh man, yes, it's very 00:12:04.190 --> 00:12:06.610 impressive the logistics that you have to do 00:12:06.610 --> 00:12:09.409 to carry out one of these surveys. Bringing in 00:12:09.409 --> 00:12:12.250 kilometers and kilometers of wire, dragging that 00:12:12.250 --> 00:12:15.570 through bush, you know, there's not always a 00:12:15.570 --> 00:12:19.789 path. So you're going to be in snowshoes, climbing 00:12:19.789 --> 00:12:23.980 hills, swinging from vines. Maybe not swinging 00:12:23.980 --> 00:12:26.879 from vines. Well, actually, there was a time 00:12:26.879 --> 00:12:30.139 in Mexico, we were in a pretty steep topography 00:12:30.139 --> 00:12:32.779 and the vines were helping me climb those mountains. 00:12:32.960 --> 00:12:35.679 So maybe I wasn't swinging from them, but I was 00:12:35.679 --> 00:12:40.200 holding on for dear life. Okay, so it's a lot 00:12:40.200 --> 00:12:41.980 of hard work is really what it comes down to 00:12:41.980 --> 00:12:45.179 in some really remote and special places sometimes. 00:12:46.440 --> 00:12:49.399 So inversions, building off what... you've already 00:12:49.399 --> 00:12:53.000 read in the intro, I set a basis for what geophysics 00:12:53.000 --> 00:12:57.740 is and what an inversion is. One term that I 00:12:57.740 --> 00:13:01.620 danced around on purpose was the term ill -posed 00:13:01.620 --> 00:13:05.220 problem. So in our context, in geophysics, what 00:13:05.220 --> 00:13:08.919 does that mean? Oh, boy. Well, the technical 00:13:08.919 --> 00:13:13.240 term would be it's the null space, the space 00:13:13.240 --> 00:13:16.639 of infinite kind of models that we can actually 00:13:16.639 --> 00:13:19.580 fit our data. simply what I could put it as. 00:13:19.879 --> 00:13:24.000 Sure. But I don't think anybody outside of taking 00:13:24.000 --> 00:13:27.360 university math has ever heard of a null space. 00:13:27.779 --> 00:13:33.320 So would it be fair to pose it as if there's 00:13:33.320 --> 00:13:37.519 more unknowns than knowns? You have less real 00:13:37.519 --> 00:13:40.120 world data and more things you're trying to solve 00:13:40.120 --> 00:13:42.860 out of an equation. So there's too many possible 00:13:42.860 --> 00:13:46.840 solutions to solve that set of equations. Exactly. 00:13:46.840 --> 00:13:49.340 It's just an underdetermined problem usually 00:13:49.340 --> 00:13:51.980 is what's the case where, yeah, again, you're 00:13:51.980 --> 00:13:55.419 solving for this model that's discretized up 00:13:55.419 --> 00:13:59.139 the subsurface and it's usually millions of parameters 00:13:59.139 --> 00:14:02.460 and you might only have, you know, dozens of 00:14:02.460 --> 00:14:05.639 data. So that is exactly what it is. It's just 00:14:05.639 --> 00:14:08.620 very underdetermined. You're trying to solve 00:14:08.620 --> 00:14:11.960 something with less knowledge than what parameters 00:14:11.960 --> 00:14:15.240 you're trying to solve for. Right. Wow. You've 00:14:15.240 --> 00:14:18.620 also just recently got your first paper published. 00:14:19.440 --> 00:14:22.659 It is published now, is it not? It's on its way. 00:14:23.220 --> 00:14:25.320 It's on its way, but it's been accepted. Is that 00:14:25.320 --> 00:14:27.500 where it's at? Yeah, it's just that they, yeah, 00:14:27.539 --> 00:14:29.679 long story short, they're changing the system. 00:14:29.840 --> 00:14:32.759 I have to submit to the new system and then it 00:14:32.759 --> 00:14:35.620 goes through. Okay, so it's accepted, but has 00:14:35.620 --> 00:14:38.429 more bureaucratic hoops to go through. Oh, you 00:14:38.429 --> 00:14:41.450 bet. Okay, so that paper, which is the first 00:14:41.450 --> 00:14:43.610 of your PhD, so congratulations on getting that 00:14:43.610 --> 00:14:45.350 through the system because that's always a pain 00:14:45.350 --> 00:14:47.730 in the butt, is learning structural information 00:14:47.730 --> 00:14:50.429 through inversion using image segmentation methods. 00:14:51.149 --> 00:14:54.070 So first, where did you submit it, which is a 00:14:54.070 --> 00:14:56.929 bit more of a question of the sausage of science, 00:14:57.090 --> 00:15:01.129 and why specifically there? Yeah, so I submitted 00:15:01.129 --> 00:15:03.250 to the Geophysics Journal. So that's kind of 00:15:03.250 --> 00:15:06.509 run out of the SEG in the, well, I guess it's 00:15:06.509 --> 00:15:09.490 very international, but it's the Society of Exploration 00:15:09.490 --> 00:15:12.470 Geophysicists. And yeah, I submitted to that 00:15:12.470 --> 00:15:14.929 one primarily because a lot of the work is based 00:15:14.929 --> 00:15:18.149 off of papers that were published in that journal. 00:15:19.389 --> 00:15:24.049 The original one was by Doug Oldenburg and Yago 00:15:24.049 --> 00:15:28.659 Lee. And they were looking at how do we form 00:15:28.659 --> 00:15:32.139 these regularizers to contain more of our information, 00:15:32.279 --> 00:15:36.080 our prior knowledge. And they found a way to 00:15:36.080 --> 00:15:40.340 be able to take its principal axes. And when 00:15:40.340 --> 00:15:42.480 I say principal axes, it's kind of like your 00:15:42.480 --> 00:15:45.960 model space. And we live in a 3D space, so we're 00:15:45.960 --> 00:15:50.120 going to have our x, y, and z axes. But in mathematics, 00:15:50.340 --> 00:15:53.190 we can change that. we can rotate that at coordinate 00:15:53.190 --> 00:15:55.809 axes so that it's kind of pointing a little bit 00:15:55.809 --> 00:15:59.350 down a little bit up and that's essentially the 00:15:59.350 --> 00:16:04.009 work i built on is upon theirs because now i 00:16:04.009 --> 00:16:06.230 have a way to be able to take you know spatial 00:16:06.230 --> 00:16:09.190 information and prior knowledge of dipping directions 00:16:09.190 --> 00:16:13.649 or like straight layered stratigraphy and i can 00:16:13.649 --> 00:16:16.110 program that into every cell in the model so 00:16:16.110 --> 00:16:19.779 that i can tell the regularization exactly I 00:16:19.779 --> 00:16:22.440 need to be able to, I want structures in this 00:16:22.440 --> 00:16:27.179 direction. And that's also another common, or 00:16:27.179 --> 00:16:29.840 maybe not a common theme in geophysics, but a 00:16:29.840 --> 00:16:31.639 lot of the geophysics papers are kind of the 00:16:31.639 --> 00:16:34.860 application style. It's not so much the theoretical, 00:16:35.080 --> 00:16:37.240 but how we can take that theoretical and then 00:16:37.240 --> 00:16:43.200 add it to our toolbox. OK, so I guess that answers 00:16:43.200 --> 00:16:45.480 the question of why you put it in geophysics, 00:16:45.559 --> 00:16:49.440 because the people that or that journal really 00:16:49.440 --> 00:16:54.940 specializes in these sorts of methods based papers. 00:16:55.600 --> 00:16:59.259 Yeah, it's an application of most of them. They're 00:16:59.259 --> 00:17:01.860 like it does do theoretical, but a lot of that 00:17:01.860 --> 00:17:04.000 is like the papers that I followed up on all 00:17:04.000 --> 00:17:06.539 structurally built. All the structure that I 00:17:06.539 --> 00:17:08.660 built from is from this geophysics journal because 00:17:08.660 --> 00:17:12.470 it's an applied journal. That's what I'm doing 00:17:12.470 --> 00:17:15.049 is I was trying to find a way to apply some of 00:17:15.049 --> 00:17:18.349 this APRO knowledge to an application. Okay, 00:17:18.390 --> 00:17:19.930 so we're going to do one of my favorite things 00:17:19.930 --> 00:17:22.490 to do as part of the show and take something 00:17:22.490 --> 00:17:25.569 I had no idea what it was when I first read it. 00:17:25.930 --> 00:17:27.930 Spent a lot of time trying to figure out what 00:17:27.930 --> 00:17:32.230 it actually meant and then still not sure. But 00:17:32.230 --> 00:17:35.190 better off than when I started, to be fair. So 00:17:35.190 --> 00:17:39.150 we're going to do some fun sounding jargon. Applying 00:17:39.150 --> 00:17:44.609 a Gaussian mixture Markov random field. I bet 00:17:44.609 --> 00:17:46.369 most people here have at least started listening, 00:17:46.470 --> 00:17:48.910 have heard of a Gaussian distribution or normal 00:17:48.910 --> 00:17:51.049 distribution. It's probably the distribution 00:17:51.049 --> 00:17:53.769 characterized by, you know, it's the symmetrical 00:17:53.769 --> 00:17:57.230 bell shaped curve. Fun fact, this is how some 00:17:57.230 --> 00:18:01.130 professors curve grades. You can totally. dump 00:18:01.130 --> 00:18:04.930 on Gaussian there. Grade's too terrible? Yeah, 00:18:05.089 --> 00:18:07.329 just adjust it so it fits the Gaussian distribution. 00:18:07.750 --> 00:18:10.490 In my own experience, in actual fact, it's rarely 00:18:10.490 --> 00:18:12.970 done. I think I've only had it done three times 00:18:12.970 --> 00:18:15.009 out of every single course and test that I ever 00:18:15.009 --> 00:18:20.470 took. But aside over, what is this thing? A Gaussian 00:18:20.470 --> 00:18:26.160 mixture Markov random field. Yeah, I really just 00:18:26.160 --> 00:18:29.519 go with the GMMRF. Even that one's hard to say. 00:18:29.680 --> 00:18:36.200 It's just, yeah, it's a mouthful. But yeah, essentially 00:18:36.200 --> 00:18:40.059 it is, yeah, it's a special kind of Gaussian, 00:18:40.279 --> 00:18:43.759 I guess, Gaussian distribution. It contains multiple. 00:18:44.119 --> 00:18:46.579 So it's going, you're going to have many means 00:18:46.579 --> 00:18:51.309 within this Gaussian mixture model. So it's, 00:18:51.309 --> 00:18:53.190 yeah, it's pretty much what the name says. It's 00:18:53.190 --> 00:18:56.630 a mixture of individual Gaussians. However, what 00:18:56.630 --> 00:18:59.789 makes it a Markov random field is that you can 00:18:59.789 --> 00:19:02.450 encode spatial information into it. So those 00:19:02.450 --> 00:19:04.690 Gaussians know how to talk to each other in space. 00:19:05.250 --> 00:19:09.730 So you have a bunch of bell curves that are distributed 00:19:09.730 --> 00:19:14.240 in space that can modify each other. Yeah, depending 00:19:14.240 --> 00:19:17.440 on how close they are spatially, not just necessarily 00:19:17.440 --> 00:19:19.720 how close their means are. They could have close 00:19:19.720 --> 00:19:22.099 means, but they could be very far apart. And 00:19:22.099 --> 00:19:25.859 these Markov random fields help us be able to 00:19:25.859 --> 00:19:27.720 encode that into our Gaussian mixture model. 00:19:29.180 --> 00:19:32.720 Okay, so the idea is that your mathematical algorithms 00:19:32.720 --> 00:19:37.339 or formulas or equations, the inversion in other 00:19:37.339 --> 00:19:41.660 words, has a series of these Gaussian distributions 00:19:41.660 --> 00:19:45.319 within it due to structure a prior information. 00:19:45.980 --> 00:19:49.000 Yeah, typically it'll be like a physical property. 00:19:49.299 --> 00:19:53.319 We'll have an idea of how many kind of rock units 00:19:53.319 --> 00:19:55.759 that we're going to expect, and those will be 00:19:55.759 --> 00:19:59.640 assigned a physical property value, which stands 00:19:59.640 --> 00:20:03.799 for a mean in our Gaussian mixture model. Okay, 00:20:03.940 --> 00:20:08.890 so you can imagine... Audience, correct me if 00:20:08.890 --> 00:20:12.549 I'm wrong, if I overstep my skis here. I've been 00:20:12.549 --> 00:20:14.650 known to do that. But you can imagine if we just 00:20:14.650 --> 00:20:17.730 take a layered earth, which is rarely the case, 00:20:17.769 --> 00:20:20.549 but if each individual layer had a different 00:20:20.549 --> 00:20:23.730 mean, each one of those represent a different 00:20:23.730 --> 00:20:27.150 Gaussian, and each one of those is also representing 00:20:27.150 --> 00:20:29.670 a structure. And so you have these individual 00:20:29.670 --> 00:20:32.549 Gaussians or layers. that can affect one another 00:20:32.549 --> 00:20:35.630 within the inversion or the ill -posed problem 00:20:35.630 --> 00:20:39.250 we're starting with. Is that kind of right? That's 00:20:39.250 --> 00:20:43.609 pretty bang on. Yeah. Perfect. Okay. So when 00:20:43.609 --> 00:20:45.450 we start talking about structural information, 00:20:45.829 --> 00:20:48.410 which is almost a jargon term in itself, because 00:20:48.410 --> 00:20:51.349 to those that are just jumping in somehow into 00:20:51.349 --> 00:20:53.869 the halfway through the podcast, you might be 00:20:53.869 --> 00:20:57.170 thinking like we're building something. Is it? 00:20:57.630 --> 00:21:00.069 It's definitely not like creating a city under 00:21:00.069 --> 00:21:04.569 the ground, though, is it? No, I wouldn't say 00:21:04.569 --> 00:21:07.569 creating anything underground. It's just trying 00:21:07.569 --> 00:21:13.029 to interpret from the data what this subsurface 00:21:13.029 --> 00:21:17.609 image would be. Okay. Maybe to bring this together 00:21:17.609 --> 00:21:20.589 and make it make sense of why this is kind of 00:21:20.589 --> 00:21:23.890 a big thing, like it is actually a huge step 00:21:23.890 --> 00:21:26.809 forward for... people in our sphere for inversions 00:21:26.809 --> 00:21:29.450 is let's go back to what a classical geophysical 00:21:29.450 --> 00:21:33.569 inversion is also known as the unconstrained 00:21:33.569 --> 00:21:36.450 inversion when it comes to imaging geology i 00:21:36.450 --> 00:21:39.569 know we can image or illuminate it really well 00:21:39.569 --> 00:21:44.849 but generally where do they fail uh well you 00:21:44.849 --> 00:21:47.210 know they fail a lot mostly because of their 00:21:47.210 --> 00:21:49.309 sensitivity, depending on the data set that you're 00:21:49.309 --> 00:21:52.630 using and what you're trying to detect. So we're 00:21:52.630 --> 00:21:54.950 going to have lots, like, again, this comes into 00:21:54.950 --> 00:21:57.890 that ill -posed and this null space. There's 00:21:57.890 --> 00:22:01.049 so many models. So we've got models that can 00:22:01.049 --> 00:22:06.329 have very thick, say, conductive, so meaning, 00:22:06.470 --> 00:22:09.230 you know, currents travel really easy in these 00:22:09.230 --> 00:22:13.210 layers. And if that's over top of anything that 00:22:13.210 --> 00:22:15.269 we're actually trying to target, we're not going 00:22:15.269 --> 00:22:17.609 to be able to see it. It's going to suck up all 00:22:17.609 --> 00:22:20.849 that kind of imaging power, all that current, 00:22:20.910 --> 00:22:23.210 and we're only going to see some kind of conductive 00:22:23.210 --> 00:22:26.630 layer. So these really break down because in 00:22:26.630 --> 00:22:28.250 a standard L2, you've got no other way. You're 00:22:28.250 --> 00:22:30.170 just like, I want a smooth model. So you're just 00:22:30.170 --> 00:22:32.990 going to get this smooth conductor on top. Whereas 00:22:32.990 --> 00:22:36.819 when we can start modifying... that L2 and putting 00:22:36.819 --> 00:22:38.700 some more information into that regularization, 00:22:38.960 --> 00:22:41.380 we can start kind of see like, okay, we expect 00:22:41.380 --> 00:22:43.759 something to be shallower, but we want stuff 00:22:43.759 --> 00:22:46.279 to grow at depth. That's where we're going to 00:22:46.279 --> 00:22:49.059 get our advantage over our standard L2, as I 00:22:49.059 --> 00:22:53.380 call it, or the unconstrained inversion. Yeah, 00:22:53.480 --> 00:22:57.039 maybe this is like, again, I'm very much a user 00:22:57.039 --> 00:23:00.359 audience when it comes to inversions. I am much 00:23:00.359 --> 00:23:02.960 more of a closer to the geology side of geophysics 00:23:02.960 --> 00:23:05.819 than what John is. much within the mathematics 00:23:05.819 --> 00:23:09.160 and able to program some of this so um i'm gonna 00:23:09.160 --> 00:23:11.920 take a stab at this a little bit to illuminate 00:23:11.920 --> 00:23:15.920 the fact that we've used the term l2 and uh that's 00:23:15.920 --> 00:23:18.539 just if my understanding is correct and hopefully 00:23:18.539 --> 00:23:20.640 it is it's just another way of putting in uh 00:23:20.640 --> 00:23:25.079 an um Our prior information, we're making an 00:23:25.079 --> 00:23:27.359 assumption to be able to solve the ill -posed 00:23:27.359 --> 00:23:30.279 problem. And L2 is a way of bringing in that 00:23:30.279 --> 00:23:32.519 assumption, which assumes kind of a continuous, 00:23:32.539 --> 00:23:38.500 smooth property distribution. So it's going to 00:23:38.500 --> 00:23:40.660 try to smooth it out a little bit. It's not going 00:23:40.660 --> 00:23:43.319 to give you true layers. It's not going to give 00:23:43.319 --> 00:23:46.319 you hard boundaries until you do something to 00:23:46.319 --> 00:23:48.200 it, in which case it's no longer constrained. 00:23:48.660 --> 00:23:52.349 Is that about right? More or less? Yeah, that's 00:23:52.349 --> 00:23:54.890 completely correct. Because as soon as you start 00:23:54.890 --> 00:23:57.170 wanting more out of the inversion for details 00:23:57.170 --> 00:23:59.950 and boundaries, you're starting to put in, you 00:23:59.950 --> 00:24:02.589 want to put weights or hard constraints at certain 00:24:02.589 --> 00:24:05.890 places and certain spatial areas in the model. 00:24:06.569 --> 00:24:11.039 So in other words, the L2 is... It is fantastic 00:24:11.039 --> 00:24:13.700 when we have no other information. It provides 00:24:13.700 --> 00:24:15.779 us something that can track geology fairly well, 00:24:15.900 --> 00:24:18.559 but it's always a really kind of fuzzy image 00:24:18.559 --> 00:24:21.980 of what the real thing is. It's like, for lack 00:24:21.980 --> 00:24:23.579 of a better way of saying, it's like looking 00:24:23.579 --> 00:24:26.039 through a number of different glasses and each 00:24:26.039 --> 00:24:28.480 one is distorting the image a little bit more 00:24:28.480 --> 00:24:31.799 and a little bit more. So it kind of blurs it 00:24:31.799 --> 00:24:34.589 out. But in geology... it's often quite a stark 00:24:34.589 --> 00:24:37.309 contrast when you hit a lithology boundary you're 00:24:37.309 --> 00:24:40.029 really going from one property to the other so 00:24:40.029 --> 00:24:43.049 this smoothness doesn't really break down so 00:24:43.049 --> 00:24:47.829 perhaps it would be useful to state why we don't 00:24:47.829 --> 00:24:52.309 typically force that onto your standard unconstrained 00:24:52.309 --> 00:24:56.279 inversion Yeah. Well, the way to explain it mathematically 00:24:56.279 --> 00:24:59.519 is that the L2 is just a beautiful function. 00:24:59.640 --> 00:25:02.019 It's differentiable everywhere. It's smooth. 00:25:02.140 --> 00:25:05.259 It just works mathematically nice. When we want 00:25:05.259 --> 00:25:08.519 to start imposing in some of these prior information, 00:25:08.519 --> 00:25:12.259 or say we want to use norms that aren't smooth, 00:25:12.500 --> 00:25:16.960 that gets mathematically hard. And it doesn't 00:25:16.960 --> 00:25:20.079 always want to converge. You don't always get... 00:25:20.559 --> 00:25:24.759 to an actual model that satisfies the data. Well, 00:25:24.880 --> 00:25:28.039 let's take a tact back towards your paper. There's 00:25:28.039 --> 00:25:30.059 lots more we can get into, but I think for understanding 00:25:30.059 --> 00:25:32.440 the paper, we need one more component before 00:25:32.440 --> 00:25:36.799 trying the results. Neural networks, the AI or 00:25:36.799 --> 00:25:39.819 machine learning component of the paper, which 00:25:39.819 --> 00:25:42.259 if you're going down that road, I really hope 00:25:42.259 --> 00:25:45.799 you're not working at replacing me. No, no, no. 00:25:45.859 --> 00:25:48.500 No, definitely not. Most of the work that I do 00:25:48.500 --> 00:25:52.480 is we're pulling it to be a tool of ours. So 00:25:52.480 --> 00:25:55.400 that's the thing. What my paper was built off 00:25:55.400 --> 00:25:58.140 of is rotating that regularization, being able 00:25:58.140 --> 00:26:00.819 to encode the information into it. So technically, 00:26:00.940 --> 00:26:04.220 you don't need the segmentation or the AI. If 00:26:04.220 --> 00:26:06.619 you have an idea, you can put it in yourself. 00:26:07.000 --> 00:26:09.759 But what I'm working at here is being able to... 00:26:09.950 --> 00:26:12.750 Let's make the computer do the work for us. Why 00:26:12.750 --> 00:26:15.430 do we want to go out and measure dips and look 00:26:15.430 --> 00:26:18.490 at orientations and spend the time assigning 00:26:18.490 --> 00:26:20.710 these ourselves when we could actually have just 00:26:20.710 --> 00:26:24.670 a model that can take that information or take 00:26:24.670 --> 00:26:27.710 in the information that we have or the images 00:26:27.710 --> 00:26:30.630 that we have, geology, and then it can spit out 00:26:30.630 --> 00:26:34.369 information about the orientations and the structure 00:26:34.369 --> 00:26:38.369 directions. Okay, so I guess there's two questions 00:26:38.369 --> 00:26:41.569 overlapping here. First, so you're getting the 00:26:41.569 --> 00:26:45.869 AI or machine learning to pick out an appropriate 00:26:45.869 --> 00:26:49.529 structure to try to seed or use as prior information 00:26:49.529 --> 00:26:53.890 to the inversion. But like any machine learning 00:26:53.890 --> 00:26:57.569 or AI, you've had to train it on something. And 00:26:57.569 --> 00:27:00.549 I assume you haven't stolen a bunch of inversion 00:27:00.549 --> 00:27:03.289 models from across the internet and trained it 00:27:03.289 --> 00:27:07.039 on it. Maybe just explain that part a little 00:27:07.039 --> 00:27:10.500 bit for the audience. Yeah, well, at first, that's 00:27:10.500 --> 00:27:12.880 the avenue I started to look at is, well, you 00:27:12.880 --> 00:27:14.759 got these models, you have to train it on something 00:27:14.759 --> 00:27:17.319 that you're looking at. It's not very generalizable 00:27:17.319 --> 00:27:21.180 when I first started. But then trying to make 00:27:21.180 --> 00:27:24.099 examples, it takes a long time to run inversions. 00:27:24.099 --> 00:27:26.619 So getting a huge sample base was not really 00:27:26.619 --> 00:27:30.740 a thing. It was just not tractable. So, no, I 00:27:30.740 --> 00:27:33.660 didn't steal anything. But what I did was, is 00:27:33.660 --> 00:27:37.779 I've Actually, it was the spring, right before 00:27:37.779 --> 00:27:40.680 I started getting into this, the SAM model, Segment 00:27:40.680 --> 00:27:43.539 Anything model from Google, or from Meta, was 00:27:43.539 --> 00:27:47.980 released. And it looked really great. They were 00:27:47.980 --> 00:27:50.019 trying it on all sorts of images, all different 00:27:50.019 --> 00:27:52.000 sorts of things. I'm like, well, let's take a 00:27:52.000 --> 00:27:54.119 crack at geology. Let's see if it can look at 00:27:54.119 --> 00:27:56.859 a geophysical inversion and segment objects. 00:27:58.079 --> 00:28:00.759 And lo and behold. It actually worked right, 00:28:00.819 --> 00:28:03.660 really well, right out of the box. And with these 00:28:03.660 --> 00:28:07.279 models now, they train them on the whole internet 00:28:07.279 --> 00:28:11.299 or whatever have you, what have you. And yeah, 00:28:11.740 --> 00:28:13.779 they put these models, these checkpoints up and 00:28:13.779 --> 00:28:15.420 you can download them and you can start from 00:28:15.420 --> 00:28:18.599 there. So even if I wanted to, or I could, if 00:28:18.599 --> 00:28:22.240 I wanted to, is then use some of the... geophysical 00:28:22.240 --> 00:28:24.599 inversion kind of examples i had before and i 00:28:24.599 --> 00:28:26.779 can fine tune it but i found that i actually 00:28:26.779 --> 00:28:30.259 don't have to the sam model is quite great and 00:28:30.259 --> 00:28:34.160 it it's a great segmentation tool whoa wait a 00:28:34.160 --> 00:28:37.539 minute so you just took something that is trained 00:28:37.539 --> 00:28:40.019 for something else completely something i mean 00:28:40.019 --> 00:28:42.779 it's similar you're finding structure and identity 00:28:42.779 --> 00:28:46.000 and shapes and things yeah but you didn't you 00:28:46.000 --> 00:28:49.009 could just take that and apply it directly into 00:28:49.009 --> 00:28:51.369 this branch of science and you didn't really 00:28:51.369 --> 00:28:54.549 need to tinker with it? Yeah. So they call it 00:28:54.549 --> 00:28:58.369 foundational models. So I've found a foundational 00:28:58.369 --> 00:29:01.910 model that's applicable to the geophysical inversion 00:29:01.910 --> 00:29:06.309 and tried it out and it works. Something that 00:29:06.309 --> 00:29:08.710 was trained on ice cream trucks can actually 00:29:08.710 --> 00:29:13.250 also find segmentation in geophysical models. 00:29:13.750 --> 00:29:17.720 Wow, that's awesome. Okay, so now the results. 00:29:18.059 --> 00:29:20.880 Because we've got the ill -posed problem, how 00:29:20.880 --> 00:29:23.640 we're going to add information to that ill -posed 00:29:23.640 --> 00:29:26.380 problem through image segmentation. So identifying 00:29:26.380 --> 00:29:31.160 things that are created in the model by solving 00:29:31.160 --> 00:29:33.279 that ill -posed problem, things that have structure, 00:29:33.420 --> 00:29:36.519 and applying that structure and essentially rerunning 00:29:36.519 --> 00:29:40.130 the inversion, I guess. Yeah, okay, I saw the 00:29:40.130 --> 00:29:43.509 head nod. So you're going to go through that 00:29:43.509 --> 00:29:45.829 process. So what is the case study or proof of 00:29:45.829 --> 00:29:48.769 concept for this? Set the scene for us. Where 00:29:48.769 --> 00:29:53.289 was the data collected? So the data is collected 00:29:53.289 --> 00:29:57.630 in the wonderful Athabasca Basin. Oh, I know 00:29:57.630 --> 00:30:02.450 it well. We both do. Nice cold winters and lots 00:30:02.450 --> 00:30:06.309 of skidooing and fun. And a lot of bugs in the 00:30:06.309 --> 00:30:10.740 summer. Holy smokes. For those listening that 00:30:10.740 --> 00:30:12.059 don't know where we're talking about, the Athaka 00:30:12.059 --> 00:30:15.799 Basin sits in northern Alberta, and it's actually 00:30:15.799 --> 00:30:18.119 mostly in Saskatchewan, but it straddles the 00:30:18.119 --> 00:30:21.559 border there. It's got a big lake. It's a sedimentary 00:30:21.559 --> 00:30:26.640 basin, thick sandstone layer above it. Sandstone 00:30:26.640 --> 00:30:30.119 high in quartz. Quartz is a resistor. And then 00:30:30.119 --> 00:30:33.880 beneath it is the old, old rocks of the Craton. 00:30:33.880 --> 00:30:36.160 So the oldest rocks of the world actually sit 00:30:36.160 --> 00:30:39.099 in the North American Craton. I don't know about 00:30:39.099 --> 00:30:41.319 the area of Craton here, but it's certainly more 00:30:41.319 --> 00:30:43.220 than 2 billion years old. I just can't remember 00:30:43.220 --> 00:30:45.579 how old the Craton is beneath the Athabasca Basin. 00:30:46.160 --> 00:30:51.539 But back to setting the scene. So in this setting, 00:30:51.640 --> 00:30:55.539 we're going to have lots of structural faulting 00:30:55.539 --> 00:30:58.500 happening across that interface between the sandstone 00:30:58.500 --> 00:31:01.240 and the basement. And this allows hydrothermal 00:31:01.240 --> 00:31:06.279 fluids to drain upwards. And they carry a lot 00:31:06.279 --> 00:31:12.059 of hydrothermal fluids that are going to mineralize 00:31:12.059 --> 00:31:16.529 and deposit uranium. these boundaries or near 00:31:16.529 --> 00:31:20.650 this unconformity. And these conductors are really, 00:31:20.650 --> 00:31:22.589 these faults are really huge. So they form these 00:31:22.589 --> 00:31:26.569 long dipping conductors or yeah, they can, they, 00:31:26.630 --> 00:31:28.349 they form along the faults. So they're, they're 00:31:28.349 --> 00:31:30.230 not quite sheet like, but they're, they're very 00:31:30.230 --> 00:31:34.289 dipping like objects. And that is essentially 00:31:34.289 --> 00:31:37.690 what a nice, a great kind of object for image 00:31:37.690 --> 00:31:40.849 segmentation. And the idea is, is we want to 00:31:40.849 --> 00:31:45.029 be able to tell. where these conductors are faulting 00:31:45.029 --> 00:31:47.309 to or where they're dipping to so that we can 00:31:47.309 --> 00:31:50.029 plan the drill holes accordingly. We kind of 00:31:50.029 --> 00:31:52.390 know which direction to step out as we kind of 00:31:52.390 --> 00:31:57.829 probe these conductors. And yeah, then we also 00:31:57.829 --> 00:32:01.210 have these horizontal kind of... unconformity 00:32:01.210 --> 00:32:03.849 over top of them so if we can start looking at 00:32:03.849 --> 00:32:07.089 we can start segmenting structures that are not 00:32:07.089 --> 00:32:09.829 flat lying and we can get kind of a dip approximation 00:32:09.829 --> 00:32:11.990 out of those and then we can also get kind of 00:32:11.990 --> 00:32:14.930 the the thickness of that resistor on top as 00:32:14.930 --> 00:32:17.150 kind of a segment outing like flat layered blocks 00:32:17.150 --> 00:32:20.809 all right before i i try to summarize the whole 00:32:20.809 --> 00:32:22.750 picture here because i think you've done a good 00:32:22.750 --> 00:32:26.869 job for me um how thick is the uh the overburden 00:32:26.869 --> 00:32:31.569 here It varies a lot, but generally, yeah, generally 00:32:31.569 --> 00:32:33.930 about 500 meters in the areas that I'm looking 00:32:33.930 --> 00:32:37.710 up for. So yeah, it's quite deep. So for all 00:32:37.710 --> 00:32:39.750 the listeners, you can imagine you have, because 00:32:39.750 --> 00:32:43.130 it's Saskatchewan, it's pretty flat. So you have 00:32:43.130 --> 00:32:47.349 flat plains, lots of lakes and trees and swamps. 00:32:47.730 --> 00:32:52.769 And then 500 meters, the top 500 meters is basically 00:32:52.769 --> 00:32:57.809 all pure quartz sand. Beneath that 500 meters 00:32:57.809 --> 00:33:01.549 is a surface that we call an unconformity in 00:33:01.549 --> 00:33:04.329 geology. That's basically where we've had erosion 00:33:04.329 --> 00:33:08.369 cut and just remove material. So you have a gap 00:33:08.369 --> 00:33:10.849 in age between the sediments above and below, 00:33:11.009 --> 00:33:14.029 or in this case, above and the rocks below of 00:33:14.029 --> 00:33:17.549 the craton, the metamorphic rocks. In that metamorphic 00:33:17.549 --> 00:33:20.069 rocks, there's a series of faults. And those 00:33:20.069 --> 00:33:22.369 faults have carried hydrothermal fluids, which 00:33:22.369 --> 00:33:25.680 have deposited nice... conductive minerals and 00:33:25.680 --> 00:33:28.980 which is what we're targeting here right exactly 00:33:28.980 --> 00:33:31.900 it's these conduct we're not targeting the uranium 00:33:31.900 --> 00:33:34.819 or the uranite exactly but it's the associated 00:33:34.819 --> 00:33:37.859 mineralization components which is these big 00:33:37.859 --> 00:33:41.140 graphitic conductors yeah and graphitic just 00:33:41.140 --> 00:33:43.319 means as it sounds like it has graphite in it 00:33:43.319 --> 00:33:47.589 it's got lots of carbon in it um and these because 00:33:47.589 --> 00:33:50.829 they are bound by these faults and the faults 00:33:50.829 --> 00:33:53.789 are roughly linear they don't do any weird geometry 00:33:53.789 --> 00:33:59.730 they provide good structural basis they should 00:33:59.730 --> 00:34:02.009 be somewhat simple for the image segmentation 00:34:02.009 --> 00:34:05.079 is that correct Yeah, and again, maybe depending 00:34:05.079 --> 00:34:09.159 on the inversion, or sorry, the cervic type that 00:34:09.159 --> 00:34:12.480 you've used. So far, I've experimented with magnetotolerics 00:34:12.480 --> 00:34:17.179 and DC, or I guess DC resistivity. The DC resistivity, 00:34:17.199 --> 00:34:20.280 you know, if you can't quite image all the way 00:34:20.280 --> 00:34:24.360 down to there, down to the basement, it's going 00:34:24.360 --> 00:34:26.579 to be hard to get something that's going to segment 00:34:26.579 --> 00:34:30.510 out properly. However, in my... case uh it has 00:34:30.510 --> 00:34:34.150 been working out not too bad and that's uh but 00:34:34.150 --> 00:34:36.570 that since given that limitation because with 00:34:36.570 --> 00:34:39.369 the sensitivity just decreases with depth for 00:34:39.369 --> 00:34:42.110 that kind of survey the magnetotelerics though 00:34:42.110 --> 00:34:45.130 gives me that definition i can get that resolution 00:34:45.130 --> 00:34:48.349 at depth so that's so far for the basin kind 00:34:48.349 --> 00:34:51.170 of dipping conductors it works really well unfortunately 00:34:51.170 --> 00:34:55.070 that's not the most economical survey or easily 00:34:55.070 --> 00:34:58.719 or easy survey to do in the basin All right. 00:34:58.739 --> 00:35:01.059 And you use two different surveys. They're direct 00:35:01.059 --> 00:35:02.940 current, where you stick a bunch of electrodes 00:35:02.940 --> 00:35:05.559 and pump a bunch of current into the ground. 00:35:05.599 --> 00:35:08.539 You measure the voltage between points. And then 00:35:08.539 --> 00:35:14.699 magnetotellurics, which are... It's a passive 00:35:14.699 --> 00:35:18.519 method. So we're using sources that are, you 00:35:18.519 --> 00:35:21.619 know, lightning storm down in the Amazon or solar 00:35:21.619 --> 00:35:25.420 kind of... solar influence it's kind of the reverberation 00:35:25.420 --> 00:35:28.179 of the magnetosphere and the signals kind of 00:35:28.179 --> 00:35:31.400 trap in with from there and then those travel 00:35:31.400 --> 00:35:34.739 in plane waves and those plane waves hit our 00:35:34.739 --> 00:35:38.159 ground and we can measure the different frequencies 00:35:38.159 --> 00:35:41.739 and each frequency corresponds to a different 00:35:41.739 --> 00:35:46.400 depth given a conductivity so i assume then because 00:35:46.400 --> 00:35:49.840 of the differences in frequencies you can look 00:35:49.840 --> 00:35:54.039 at really long wavelengths or to look deeper 00:35:54.039 --> 00:35:56.239 than what the direct current can, and that's 00:35:56.239 --> 00:35:59.519 why you can see these areas better? Exactly. 00:35:59.780 --> 00:36:02.280 So if you record long enough for these MT and 00:36:02.280 --> 00:36:05.039 these low enough frequencies, you can see down 00:36:05.039 --> 00:36:08.159 to kilometers, even tens of kilometers. Not here, 00:36:08.280 --> 00:36:11.500 but the general surveys around here are low enough 00:36:11.500 --> 00:36:13.500 frequency that we get a good couple kilometers 00:36:13.500 --> 00:36:17.679 easily out of them. Okay, so you were able to 00:36:17.679 --> 00:36:22.400 segment and get these structures. And how did 00:36:22.400 --> 00:36:25.940 it line up with what has been actually ground 00:36:25.940 --> 00:36:28.860 truth, which basically means drilled 500 meters 00:36:28.860 --> 00:36:31.519 down to intersect these things? How did it compare? 00:36:32.190 --> 00:36:36.329 Yeah, I'm pretty fortunate. The survey that I'm 00:36:36.329 --> 00:36:39.630 going over and using the data from has been extensively 00:36:39.630 --> 00:36:44.250 drilled. It's a mine already. So I had a lot 00:36:44.250 --> 00:36:47.630 of good ground truth. And lo and behold, it worked 00:36:47.630 --> 00:36:51.289 out very well. Compared to the standard L2, as 00:36:51.289 --> 00:36:53.230 we were talking about before, that smooth assumption, 00:36:53.730 --> 00:36:56.409 you really didn't... Like, you couldn't really 00:36:56.409 --> 00:36:58.630 tell. You can see that there's a conductor. There's 00:36:58.630 --> 00:37:00.809 maybe a break in the unconformity. But you'd 00:37:00.809 --> 00:37:02.929 have no idea if it was dipping to the east, dipping 00:37:02.929 --> 00:37:05.409 to the west. Where do you plan your drill? Do 00:37:05.409 --> 00:37:07.329 you just drill straight down on top of what you 00:37:07.329 --> 00:37:09.570 have? You're not getting a lot of information. 00:37:10.269 --> 00:37:13.349 Again, this is 500 meters here, so this is expensive 00:37:13.349 --> 00:37:16.889 drilling. It's not just cheap just to send a 00:37:16.889 --> 00:37:19.389 drill down. So we really need this information. 00:37:19.929 --> 00:37:23.230 And also, we've had seismic in this area, which... 00:37:23.449 --> 00:37:25.869 lights up really well as kind of like another 00:37:25.869 --> 00:37:28.750 ground truth. And with the image segmentation, 00:37:28.889 --> 00:37:32.250 being able to just kind of, it can see a lot 00:37:32.250 --> 00:37:34.550 of things. Like it looks at data, not just like 00:37:34.550 --> 00:37:37.869 a visual eye. So it can kind of find those boundaries 00:37:37.869 --> 00:37:40.130 a lot better, you know, given maybe tweaking 00:37:40.130 --> 00:37:42.789 some parameters. But with that little bit of 00:37:42.789 --> 00:37:45.449 influence, the regularization can be like, okay, 00:37:45.510 --> 00:37:48.389 I have something here and it actually drives 00:37:48.389 --> 00:37:51.849 it home. It doesn't, it can kind of, for lack 00:37:51.849 --> 00:37:53.610 of a better term, latches on to the solution 00:37:53.610 --> 00:37:57.789 and keeps it pushing from there. And that was 00:37:57.789 --> 00:38:02.329 able to recover. Maybe the dip was about pretty 00:38:02.329 --> 00:38:05.510 shallow, about 36 degrees. My estimation was 00:38:05.510 --> 00:38:09.570 about 40, 39. I'd say it was like 39 .8, but 00:38:09.570 --> 00:38:13.789 I'll maybe have to call it 40. But yeah, in hindsight, 00:38:14.489 --> 00:38:17.570 yeah, significant figures, exactly. But in hindsight, 00:38:17.710 --> 00:38:21.679 it worked really well in this application. Yeah, 00:38:21.719 --> 00:38:24.139 I mean, to be fair, that's close enough for what 00:38:24.139 --> 00:38:26.360 they needed to do. They need to know rough direction 00:38:26.360 --> 00:38:29.480 and, you know, plus or minus 15 degrees. If you 00:38:29.480 --> 00:38:32.099 get within that, to be honest, at 500 meters 00:38:32.099 --> 00:38:34.719 depth, you're doing amazing. It's a bit like 00:38:34.719 --> 00:38:38.039 trying to throw a dart blindfolded, right? 500 00:38:38.039 --> 00:38:40.780 meters away, and now you're giving them, you're 00:38:40.780 --> 00:38:43.039 taking off the blindfold. It's still 500 meters 00:38:43.039 --> 00:38:44.980 away. You still got to be able to hit the spot, 00:38:45.119 --> 00:38:49.559 but you're increasing your chances for sure. 00:38:50.360 --> 00:38:53.300 That's pretty good. I mean, I think this is new. 00:38:53.360 --> 00:38:56.300 And in our opinions, we've opined on this over 00:38:56.300 --> 00:38:59.260 beers. The industry working is so slow in adopting 00:38:59.260 --> 00:39:02.559 new technologies and techniques. Will we see 00:39:02.559 --> 00:39:10.260 this any time soon? Ah, well, I'd like to say 00:39:10.260 --> 00:39:13.940 yes, but I have moved on in making the... Well, 00:39:14.000 --> 00:39:16.019 I've kind of rounded out some of the edges that 00:39:16.019 --> 00:39:19.360 kind of... came out of this this study as it 00:39:19.360 --> 00:39:22.820 is but yeah it's it's kind of funny because the 00:39:22.820 --> 00:39:25.900 industry actually i don't think realizes how 00:39:25.900 --> 00:39:28.460 close they are to being able to do something 00:39:28.460 --> 00:39:32.139 like this the this code like was back the paper 00:39:32.139 --> 00:39:34.900 that i got the original idea from was from 99 00:39:34.900 --> 00:39:38.539 and it's actually buried in some of the software 00:39:38.539 --> 00:39:42.150 that some of these industry leaders use but nobody 00:39:42.150 --> 00:39:45.250 really understands or knows how to use them. 00:39:45.269 --> 00:39:48.349 Or maybe it's a lack of willingness to do it. 00:39:49.449 --> 00:39:51.909 These tools are all here, but they're more, I 00:39:51.909 --> 00:39:53.889 guess, like Lego blocks. You've got to put them 00:39:53.889 --> 00:39:56.670 together yourself and design it. But all these 00:39:56.670 --> 00:40:00.409 tools are available. But having someone push 00:40:00.409 --> 00:40:03.349 it and put it to a point where it's like, oh, 00:40:03.369 --> 00:40:06.750 I want to do image segmentation with structural 00:40:06.750 --> 00:40:11.119 information incorporated button. That takes a 00:40:11.119 --> 00:40:12.820 little bit more than just the science to make 00:40:12.820 --> 00:40:16.960 that stuff happen. So, in other words, we have 00:40:16.960 --> 00:40:21.679 the tools, but we don't have a user interface 00:40:21.679 --> 00:40:25.519 to play. That's essentially it, unless you're 00:40:25.519 --> 00:40:27.619 willing to kind of dig into the blocks yourself 00:40:27.619 --> 00:40:31.039 and start building it. That's the unfortunate 00:40:31.039 --> 00:40:33.679 bridge where I guess there's the academia and 00:40:33.679 --> 00:40:36.800 then there's industry. So you're going to build 00:40:36.800 --> 00:40:40.909 it for me? Well, actually, I think we do have 00:40:40.909 --> 00:40:46.650 it in our codes. Well, at a level that I can 00:40:46.650 --> 00:40:50.230 use. I mean, that's a pretty high bar, I know, 00:40:50.369 --> 00:40:54.070 but because around work for the listeners, I 00:40:54.070 --> 00:40:56.230 tend to break all the codes. It's just a skill. 00:40:56.710 --> 00:40:59.050 I'm a bug tester by accident. Yeah, it's a good 00:40:59.050 --> 00:41:02.449 thing. It keeps our codes pretty top -notch. 00:41:04.009 --> 00:41:06.469 OK, so what about AI and machine learning more 00:41:06.469 --> 00:41:08.349 broadly? I mean, we're talking about toys. We're 00:41:08.349 --> 00:41:12.030 talking about different how our industry is moving 00:41:12.030 --> 00:41:15.769 slow forward in a slow space. But it seems like 00:41:15.769 --> 00:41:19.730 AI, the possibilities are broad for use. And 00:41:19.730 --> 00:41:22.869 I just can't imagine somebody not using it. So 00:41:22.869 --> 00:41:25.849 is it going to be used more broadly or is it 00:41:25.849 --> 00:41:27.510 already being used more broadly? Where is it 00:41:27.510 --> 00:41:31.769 going? I think it is and it isn't. Like you said, 00:41:31.789 --> 00:41:34.369 there's... The industry seems like it's slow 00:41:34.369 --> 00:41:36.469 to adopt, but there's also people within that 00:41:36.469 --> 00:41:38.730 industry that want this stuff that, you know, 00:41:38.750 --> 00:41:40.949 they're paying money for the academia to research 00:41:40.949 --> 00:41:43.869 it and to make it happen. And yeah, I guess that's 00:41:43.869 --> 00:41:48.010 kind of where things, well, hopefully they're 00:41:48.010 --> 00:41:52.389 going to go is that we're like, from my experience 00:41:52.389 --> 00:41:55.570 on being part of this impact group and being 00:41:55.570 --> 00:41:57.849 at the university of British Columbia, we're 00:41:57.849 --> 00:42:00.940 trying to take. take all that stuff and make 00:42:00.940 --> 00:42:04.719 it more simpler so like even my next work after 00:42:04.719 --> 00:42:08.800 this paper is incorporating it more into being 00:42:08.800 --> 00:42:10.579 like more like a geophysical inversion where 00:42:10.579 --> 00:42:13.480 you can be like okay here's my data here's the 00:42:13.480 --> 00:42:16.139 regularizations I want to use here or here's 00:42:16.139 --> 00:42:19.690 my assumptions and then invert and then get our 00:42:19.690 --> 00:42:23.110 outputs. So we're, I guess they, it's pushed 00:42:23.110 --> 00:42:24.909 academia enough that we're starting to like, 00:42:24.929 --> 00:42:26.969 look at the problems that way so that how do 00:42:26.969 --> 00:42:29.389 we, how do we transfer that knowledge a lot better? 00:42:29.449 --> 00:42:32.269 So I think from the academia side, we're definitely 00:42:32.269 --> 00:42:37.469 pushing that way and industry, it's kind of up 00:42:37.469 --> 00:42:40.090 to them to hopefully keep up with what we're 00:42:40.090 --> 00:42:43.750 doing and pull it in. But you've, yeah, the industry 00:42:43.750 --> 00:42:47.789 is so, so variable. Some people like what they 00:42:47.789 --> 00:42:51.150 have and just going to stick with it or yeah 00:42:51.150 --> 00:42:52.690 there's the people who are out there who want 00:42:52.690 --> 00:42:55.150 to adopt it and are kind of keeping up with what 00:42:55.150 --> 00:42:58.590 we're doing a lot of say these newer companies 00:42:58.590 --> 00:43:00.670 that are coming out that are more like tech mining 00:43:00.670 --> 00:43:04.889 companies they are full in both that dove in 00:43:04.889 --> 00:43:07.949 with both feet and they're even on to the next 00:43:07.949 --> 00:43:10.150 level of doing things like prospectivity mapping 00:43:10.889 --> 00:43:14.070 And that's incorporating more data than just 00:43:14.070 --> 00:43:17.650 geophysical data. So seeing the adoption of that, 00:43:17.730 --> 00:43:20.750 but until we start seeing, I think, a lot of 00:43:20.750 --> 00:43:25.550 outcome from that, because I don't know, I can't 00:43:25.550 --> 00:43:28.610 really, some people are saying they found deposits 00:43:28.610 --> 00:43:31.670 and they're finding things with AI, but I don't 00:43:31.670 --> 00:43:33.590 think that there has been a mine that's going 00:43:33.590 --> 00:43:37.170 to start anytime soon that was found by pure 00:43:37.170 --> 00:43:41.659 AI. But until I think on that step, until they 00:43:41.659 --> 00:43:44.500 can start proving results. Yeah, well, it's because 00:43:44.500 --> 00:43:46.699 they're also looking for something. They use 00:43:46.699 --> 00:43:49.940 a mine as a reason to, oh, I found the mine. 00:43:50.119 --> 00:43:52.019 Well, yeah, you found the mine. We knew it existed 00:43:52.019 --> 00:43:54.860 30 years ago. Like it's already mined out. You're 00:43:54.860 --> 00:43:56.820 just finding a signature. It's like I've seen 00:43:56.820 --> 00:43:59.400 geophysicists do the same thing. And they're 00:43:59.400 --> 00:44:01.559 like, oh, we need to invert and drill that. And 00:44:01.559 --> 00:44:04.690 you're like, no, that's the mine. There's already 00:44:04.690 --> 00:44:07.090 buildings there. It's this little spot over here 00:44:07.090 --> 00:44:10.389 that we're interested in So yeah, I wonder where 00:44:10.389 --> 00:44:12.469 that's gonna go and whether or not our techniques 00:44:12.469 --> 00:44:16.429 are sophisticated for it I also should really 00:44:16.429 --> 00:44:18.210 at one of the times go into like a historical 00:44:18.210 --> 00:44:20.570 look at my own industry and inversions because 00:44:20.570 --> 00:44:25.059 what we used to do back in like the 60s is wild 00:44:25.059 --> 00:44:27.920 all the inversions were done in people's heads 00:44:27.920 --> 00:44:32.179 and it really came down to who like who you got 00:44:32.179 --> 00:44:34.820 for a particular area because the inversion was 00:44:34.820 --> 00:44:38.159 all in their brain matter it was not a repeatable 00:44:38.159 --> 00:44:41.260 solving of physics unlike it is today today it's 00:44:41.260 --> 00:44:43.940 a real science back then it was a dark art holy 00:44:43.940 --> 00:44:47.420 smokes but um before we move on and i get into 00:44:47.420 --> 00:44:49.760 the last couple of fun questions is there anything 00:44:49.760 --> 00:44:52.719 you want to call out from uh Anything from your 00:44:52.719 --> 00:44:56.000 project, your ongoing PhD work, or maybe the 00:44:56.000 --> 00:44:58.579 Simpeg group itself? I don't know. Anything you 00:44:58.579 --> 00:45:00.579 want to call out and bring up at this point? 00:45:01.019 --> 00:45:03.480 Yeah, just I guess to promote a little bit of 00:45:03.480 --> 00:45:05.980 my future work. So the first kind of work here 00:45:05.980 --> 00:45:08.980 is building on, you know, what is image segmentation 00:45:08.980 --> 00:45:12.820 and how can it be applicable to geophysics? And 00:45:12.820 --> 00:45:16.030 so I find a way to be able to apply that. and 00:45:16.030 --> 00:45:17.969 you know we can get some results we can use that 00:45:17.969 --> 00:45:20.389 information but it's kind of something that's 00:45:20.389 --> 00:45:23.070 abstract outside of the inversion it's not really 00:45:23.070 --> 00:45:25.710 incorporated into it where my new work is now 00:45:26.760 --> 00:45:30.440 using actual like this kind of the same equations 00:45:30.440 --> 00:45:33.239 that they do to do image segmentation problems 00:45:33.239 --> 00:45:36.440 well with images say like deblurring so if you 00:45:36.440 --> 00:45:39.380 take a picture of a fast moving car you get a 00:45:39.380 --> 00:45:41.139 picture of the car but everything behind it is 00:45:41.139 --> 00:45:43.340 a little bit blurred so there's techniques that 00:45:43.340 --> 00:45:45.940 you can use that use image segmentation to correct 00:45:45.940 --> 00:45:47.900 that and then you can actually kind of get your 00:45:47.900 --> 00:45:51.860 background resolved better and upon reading those 00:45:51.860 --> 00:45:55.400 papers they've been applying it to medical and 00:45:55.400 --> 00:45:58.619 all sorts of other applications. And now that 00:45:58.619 --> 00:46:00.960 I've now taken that and adopted that to the geophysical 00:46:00.960 --> 00:46:02.599 inversion. And then now it's kind of getting 00:46:02.599 --> 00:46:05.119 to that where you were saying, like getting it 00:46:05.119 --> 00:46:07.860 into the hands of someone, maybe an industry, 00:46:08.019 --> 00:46:10.360 they can use it a little bit easier. Maybe they 00:46:10.360 --> 00:46:12.679 don't need to understand all the crazy math behind 00:46:12.679 --> 00:46:15.599 it, but it still functions exactly like the tools 00:46:15.599 --> 00:46:19.000 that they're kind of used to. And yeah, and that's 00:46:19.000 --> 00:46:22.380 all done through the Sympeg framework where I'm 00:46:22.380 --> 00:46:24.099 actually part of the new initiative where we're 00:46:24.099 --> 00:46:27.030 kind of. looking back at the optimization tools 00:46:27.030 --> 00:46:29.829 within there. And we're redesigning that so that 00:46:29.829 --> 00:46:33.869 we can plug in my ideas or even any others, like 00:46:33.869 --> 00:46:37.449 even using neural networks as regularizations 00:46:37.449 --> 00:46:40.309 themselves being put into the SIMPEG framework 00:46:40.309 --> 00:46:46.019 there. Cool. Okay, so I want to ask... One more 00:46:46.019 --> 00:46:48.079 joke before, or one more joke. I want to ask 00:46:48.079 --> 00:46:50.760 one more question. Before we get to the joke, 00:46:50.780 --> 00:46:52.780 you can see how excited I am for his science 00:46:52.780 --> 00:46:55.880 joke. I am too. Okay, so first things first. 00:46:56.780 --> 00:46:59.940 Just, I know this is going to put you on the 00:46:59.940 --> 00:47:01.760 spot. And I know sometimes, because I've been 00:47:01.760 --> 00:47:04.139 in the field myself, you don't necessarily want 00:47:04.139 --> 00:47:06.039 to give out too much information to get somebody 00:47:06.039 --> 00:47:09.780 in trouble. But collecting this geophysical data 00:47:09.780 --> 00:47:12.920 is... special in its own right like where you 00:47:12.920 --> 00:47:15.860 are the people you're dealing with the locations 00:47:15.860 --> 00:47:19.519 the logistics it can all be incredibly special 00:47:19.519 --> 00:47:23.000 sometimes it's not but quite often it is so perhaps 00:47:23.000 --> 00:47:28.239 a short story or an example from your own time 00:47:28.239 --> 00:47:34.500 in the field Oh, boy. Well, let's do a fun one, 00:47:34.559 --> 00:47:38.719 but also a devastating one. Oh, it's both a tragedy 00:47:38.719 --> 00:47:41.980 and a comedy at the same time. Shakespeare would 00:47:41.980 --> 00:47:45.599 be proud. So I did actually a lot of work in 00:47:45.599 --> 00:47:48.940 China. We were one of the geophysical contractors 00:47:48.940 --> 00:47:52.159 that actually would do work and people were able 00:47:52.159 --> 00:47:54.139 to get us in. We had an entity there as well. 00:47:54.820 --> 00:48:00.500 And yeah, it was in northern China. Kind of an 00:48:00.500 --> 00:48:02.960 old gold mine. So, you know, it's kind of more 00:48:02.960 --> 00:48:05.039 like maybe the foothills of Alberta. But again, 00:48:05.199 --> 00:48:07.500 everything has thorns on it. So everything wants 00:48:07.500 --> 00:48:10.260 to scrape you up and do all sorts of things. 00:48:10.340 --> 00:48:12.460 But, you know, it was really fun because there 00:48:12.460 --> 00:48:14.960 was it was a. It was a really large scale survey. 00:48:15.059 --> 00:48:18.079 We covered a lot of little small towns. And, 00:48:18.219 --> 00:48:21.360 you know, the helpers that came and helped us, 00:48:21.400 --> 00:48:23.340 they would invite us over to their house and 00:48:23.340 --> 00:48:27.420 have meals. And, you know, it'd be like they'd 00:48:27.420 --> 00:48:29.719 be giving us apples, all these fresh fruits in 00:48:29.719 --> 00:48:31.519 their fields that we're going through with our 00:48:31.519 --> 00:48:35.219 wires. And yeah, it was just generally a really 00:48:35.219 --> 00:48:39.340 good experience, except for when the mine site 00:48:39.340 --> 00:48:44.110 owner would come. And this guy, someone would 00:48:44.110 --> 00:48:46.250 hear his name and they're like, oh boy, we have 00:48:46.250 --> 00:48:49.789 to run. But it wasn't because he was a scary 00:48:49.789 --> 00:48:53.769 guy. It's because this guy really liked to party. 00:48:54.050 --> 00:48:56.590 And he didn't care if you had to work the next 00:48:56.590 --> 00:48:59.750 day. He didn't care if you had a deadline. He 00:48:59.750 --> 00:49:02.090 would take pretty much everybody from the mine 00:49:02.090 --> 00:49:06.059 out and you'd go have hot pot, karaoke. And a 00:49:06.059 --> 00:49:13.139 lot of Baishal. So much Baishal. And I'm not 00:49:13.139 --> 00:49:16.840 going to name names because that's just to keep 00:49:16.840 --> 00:49:20.800 things a little general. But yeah, it was always 00:49:20.800 --> 00:49:22.960 scary because when you heard his name, you didn't 00:49:22.960 --> 00:49:24.079 want to go out that night because you're like, 00:49:24.139 --> 00:49:28.159 I have to go hike 30 ,000 steps minimum tomorrow. 00:49:28.780 --> 00:49:31.300 And I don't want to. And, you know, when you're 00:49:31.300 --> 00:49:34.739 in different countries, you know, you... Sometimes 00:49:34.739 --> 00:49:38.179 saying no is not culturally accepted. So you 00:49:38.179 --> 00:49:40.599 kind of have to say yes. So it was funny because 00:49:40.599 --> 00:49:43.619 people would run and hide and they would try 00:49:43.619 --> 00:49:46.039 to avoid these situations. It was like, oh, sounds 00:49:46.039 --> 00:49:49.960 fun. It does. But after, you know, once every 00:49:49.960 --> 00:49:52.019 couple of weeks, this guy comes around, you're 00:49:52.019 --> 00:49:55.239 just like, oh, I got to hide under my bed. Yeah, 00:49:55.360 --> 00:50:00.289 for sure. Like, it's not it's no joke. You say, 00:50:00.369 --> 00:50:02.210 yeah, you know, you're young, you're fit. You 00:50:02.210 --> 00:50:05.190 could do that one time. Yes, but you have to 00:50:05.190 --> 00:50:08.989 be doing that 30 ,000 steps or depending on the 00:50:08.989 --> 00:50:11.869 terrain, it could be, you know, 20 kilometers 00:50:11.869 --> 00:50:13.869 of walking or it could be like 10 kilometers 00:50:13.869 --> 00:50:17.809 of walking, pulling wire, bushwhacking, whatever 00:50:17.809 --> 00:50:20.880 it is, but you're doing it day in, day out. Day 00:50:20.880 --> 00:50:25.000 in, day out, for weeks at a time. Sure, any one 00:50:25.000 --> 00:50:28.119 of those days. If you're relatively fit, not 00:50:28.119 --> 00:50:29.840 a big deal, but try doing that for six weeks 00:50:29.840 --> 00:50:36.280 straight. It's special. Oh, yes. So it sounds 00:50:36.280 --> 00:50:39.539 like that was an adventure, both fun and it wasn't 00:50:39.539 --> 00:50:42.539 very tragic. Slightly terrifying. Unless you 00:50:42.539 --> 00:50:44.360 skipped over something tragic, in which case 00:50:44.360 --> 00:50:45.860 we'll leave it there and we'll go straight to 00:50:45.860 --> 00:50:47.659 the final question because we're going to keep 00:50:47.659 --> 00:50:50.619 on the comedy instead of the tragedy. And your 00:50:50.619 --> 00:50:56.199 favorite science joke. It might not be my favorite, 00:50:56.300 --> 00:50:58.440 but it's a segment, my favorite segmentation 00:50:58.440 --> 00:51:01.420 joke. Oh, even better. The nerdier, the better. 00:51:01.880 --> 00:51:05.739 Come on, John, talk nerdy to me. Okay. All right. 00:51:05.840 --> 00:51:09.699 So why don't segmentation models ever fight? 00:51:10.510 --> 00:51:13.829 I don't know. They always find clear boundary 00:51:13.829 --> 00:51:19.309 first. I tried to make it as terrible as possible. 00:51:19.429 --> 00:51:21.750 I don't know. I don't know. I love it that it's 00:51:21.750 --> 00:51:24.409 thematic. It's even better. All right. Well, 00:51:24.489 --> 00:51:26.610 thanks so much for being on, John. It means the 00:51:26.610 --> 00:51:28.690 world to me. And for all those listening, he 00:51:28.690 --> 00:51:31.010 is one of the few people that have been listening 00:51:31.010 --> 00:51:33.750 to this podcast from the beginning. I'm so glad 00:51:33.750 --> 00:51:36.869 I finally got him on the podcast because he's 00:51:36.869 --> 00:51:41.079 awesome. Thank you very much. Not too bad yourself. 00:51:41.900 --> 00:51:44.280 Thanks so much for being here and enjoying Whimsical 00:51:44.280 --> 00:51:47.980 Wavelengths with me. Of course, this is the time 00:51:47.980 --> 00:51:50.139 of the show where I ask you to subscribe, tell 00:51:50.139 --> 00:51:53.719 a friend, reach out to me on social media. You 00:51:53.719 --> 00:51:56.699 know, I actually had somebody suggest on Blue 00:51:56.699 --> 00:51:59.599 Sky that I should do a show all about why mammals 00:51:59.599 --> 00:52:01.840 weren't, wasn't able to take over the skies. 00:52:03.210 --> 00:52:06.710 We managed to take over and become apex predators 00:52:06.710 --> 00:52:09.650 in the marine sphere, as well as everywhere on 00:52:09.650 --> 00:52:12.230 land. Why not the skies? It's a pretty good, 00:52:12.250 --> 00:52:13.949 interesting idea, getting into the evolution 00:52:13.949 --> 00:52:16.570 of flight and just how evolutionary pressures 00:52:16.570 --> 00:52:21.329 really didn't favor mammals in that way. I don't 00:52:21.329 --> 00:52:22.650 know if I'll get around to doing that, but that's 00:52:22.650 --> 00:52:24.750 a really cool idea. I love it. Keep it coming. 00:52:25.349 --> 00:52:28.190 I'm serious. So yeah, subscribe, tell a friend, 00:52:28.250 --> 00:52:35.010 and I'll see you in two weeks. So chaotic, so 00:52:35.010 --> 00:52:39.769 misbehaved. Echoes of melodies remind us it's 00:52:39.769 --> 00:52:46.710 always. Colors weave stories painting the sky. 00:52:49.730 --> 00:53:18.510 Swaying to rhythms as the galaxies fly by. Chasing 00:53:18.510 --> 00:53:50.769 down wonders, always seeking. Huh?