WEBVTT Kind: captions Language: en 00:00:01.660 --> 00:00:08.480 [ Silence ] 00:00:08.480 --> 00:00:13.120 [inaudible background conversations] 00:00:13.120 --> 00:00:16.260 Good morning, everyone. Welcome to seminar. 00:00:16.260 --> 00:00:20.260 Next week, our speaker will be Chris Milliner from UC-Berkeley. 00:00:20.260 --> 00:00:24.180 This week, Tom Hanks will introduce our speaker. 00:00:29.640 --> 00:00:34.640 - Yeah. There’s lots of things that can be said about Bill Lettis. 00:00:34.640 --> 00:00:39.790 But first and foremost, he is the 14th Joyner Lecturer. 00:00:39.790 --> 00:00:47.489 And the Joyner Lecture series honors our colleague, Bill Joyner, who spent 00:00:47.489 --> 00:00:52.059 most of his professional career working for the Geological Survey. 00:00:52.060 --> 00:00:57.260 And most of that was spent writing one Joyner-Boore or 00:00:57.260 --> 00:01:03.260 Boore-Joyner paper after another for 35 or 40 years, which, in a lot 00:01:03.260 --> 00:01:09.980 of ways, is kind of the bedrock of ground motion estimation in the country. 00:01:11.140 --> 00:01:16.960 But Bill also was quite unusual in that he was very much interested 00:01:16.960 --> 00:01:22.630 in bringing the very best seismology into earthquake engineering, 00:01:22.630 --> 00:01:28.211 structural engineering, and especially aseismic design 00:01:28.220 --> 00:01:30.800 and construction that was going to save lives. 00:01:30.800 --> 00:01:34.270 And he spent a lot of time in structural engineering meetings 00:01:34.270 --> 00:01:38.630 doing very tedious things with building codes and this and that 00:01:38.630 --> 00:01:42.760 and R factors and T factors and F factors and stuff. 00:01:44.100 --> 00:01:49.231 There never have been very many earthquake scientists who care 00:01:49.240 --> 00:01:52.840 very much about earthquake engineering or make a dent in it. 00:01:52.840 --> 00:01:54.180 And vice versa. 00:01:54.180 --> 00:01:58.060 There have not been very many earthquake engineers who really got into 00:01:58.060 --> 00:02:04.040 the hows and whys of the earthquake science that was available to them. 00:02:04.810 --> 00:02:12.000 Bill was really one of the first to do so. And he passed away in 2001. 00:02:12.000 --> 00:02:16.400 And we set up this program in its honor – in his honor. 00:02:17.610 --> 00:02:20.780 So over the years, there’s been 14. 00:02:20.780 --> 00:02:26.000 Lloyd Cluff, actually, who’s getting courtesy for the slide there, was the first. 00:02:27.440 --> 00:02:31.260 Bill graduated from Humboldt State University, which has this 00:02:31.260 --> 00:02:36.680 remarkable [chuckles] – remarkably talented group of young scientists 00:02:36.680 --> 00:02:39.830 that have come out of there after they get reprogrammed 00:02:39.830 --> 00:02:44.480 by Gary Carver and Bud Burke, to a certain extent, I guess. 00:02:46.160 --> 00:02:50.700 And he graduated from Humboldt State as an undergraduate then 00:02:50.710 --> 00:02:53.719 did his master’s and Ph.D. at Berkeley. 00:02:53.719 --> 00:02:58.050 He spent most of his life in the consulting business 00:02:58.050 --> 00:03:05.209 with one Lettis et al. company or another over the past 27 years. 00:03:05.209 --> 00:03:16.010 But one of the scientific products of the Lettis companies and employees is 00:03:16.010 --> 00:03:21.280 this book, Quaternary Geochronology Methods and Applications. 00:03:21.280 --> 00:03:26.569 It’s 20 years old now, but still is – still, in a lot of ways, very much 00:03:26.569 --> 00:03:33.080 the bible of Quaternary geochronology that was done – well, Janet Sowers, 00:03:33.080 --> 00:03:35.970 I think, did most of the hard work on that. Is that correct? 00:03:35.970 --> 00:03:38.800 A lot of the editorial work? [laughter] 00:03:38.800 --> 00:03:43.760 But Bill’s name is in here a lot as well. So at any rate, here’s Bill. 00:03:45.920 --> 00:03:48.100 You’ve got one. Okay. 00:03:48.660 --> 00:03:52.220 - Okay. Am I on? Okay. 00:03:54.360 --> 00:03:59.990 And I need to stand here. So I apologize for being so formal. 00:03:59.990 --> 00:04:03.420 But thanks, Tom. That was very nice words. 00:04:03.420 --> 00:04:11.000 It’s really an honor to be named the 2017 Joyner Lecturer. 00:04:11.000 --> 00:04:16.460 And I want to thank Tom Hanks and Jim Mori and Mary [Camarillo] for – 00:04:16.460 --> 00:04:22.580 they’re the nomination committee – for appointing me as the 2017 Lecturer. 00:04:22.580 --> 00:04:27.920 So I’m very happy and honored to be here. 00:04:30.080 --> 00:04:34.340 Before I get started, this is my first go-through with my Joyner Lecture. 00:04:34.340 --> 00:04:35.630 This is my trial run. 00:04:35.630 --> 00:04:40.410 I should have a big red “practice” written across here. 00:04:40.410 --> 00:04:43.190 But Tom was kind enough to ask me to come here. 00:04:43.190 --> 00:04:48.400 My first Joyner Lecture will be next week at EERI up in Portland, 00:04:48.400 --> 00:04:51.500 and then I’ll be giving the Joyner Lecture again in the 00:04:51.500 --> 00:04:55.050 SSA meeting in Denver later in April. 00:04:55.050 --> 00:05:00.260 Plus I’ll be giving the lecture a couple of other times. 00:05:01.460 --> 00:05:02.810 This is my introductory slide. 00:05:02.810 --> 00:05:08.660 I do want to acknowledge Lloyd Cluff, the first recipient of the Joyner Lecture. 00:05:08.660 --> 00:05:12.510 This was his original title slide in honor of Bill Joyner. 00:05:12.510 --> 00:05:14.800 And I thought, you know, that was pretty cool, 00:05:14.800 --> 00:05:19.380 so I duplicated it but put my own name and title on it. 00:05:21.920 --> 00:05:25.280 Because this is really a acknowledgement of Bill Joyner 00:05:25.280 --> 00:05:28.020 and his career, as Tom mentioned. 00:05:28.020 --> 00:05:33.740 And the intent of the Joyner Lecture, as I quoted here, is bringing 00:05:33.740 --> 00:05:37.560 earthquake seismology and earthquake engineering closer together. 00:05:37.560 --> 00:05:42.540 I thought about what I might talk about. I love – I’m an Earth scientist, 00:05:42.550 --> 00:05:47.620 I’m a geologist, paleoseismologist. There’s a number of great studies 00:05:47.620 --> 00:05:52.710 we’re doing now in Taiwan and in Korea, so a research-oriented talk. 00:05:52.710 --> 00:05:58.090 But that’s really an Earth science talk. In the spirit of this interface between 00:05:58.090 --> 00:06:01.570 Earth science and the earthquake engineering community, I thought, 00:06:01.570 --> 00:06:08.720 you know, for the past 20 years, really, my main focus has – in consulting – 00:06:08.720 --> 00:06:14.910 has been in this interface performing – taking information from the scientific 00:06:14.910 --> 00:06:19.310 community, performing a seismic hazard analysis that meets the 00:06:19.310 --> 00:06:22.110 objectives of the earthquake engineering community. 00:06:22.110 --> 00:06:26.410 And I’ve been through the transition from deterministic studies 00:06:26.410 --> 00:06:31.940 to probabilistic studies and into probabilistic risk analyses. 00:06:31.940 --> 00:06:37.100 And the real key to that transition has been a better recognition 00:06:37.100 --> 00:06:39.160 and understanding of uncertainty. 00:06:39.160 --> 00:06:45.740 Uncertainty is critical to the stability of seismic hazard analysis results. 00:06:46.480 --> 00:06:51.460 And so I thought, that’s what I’ll focus on in my Joyner Lecture. 00:06:51.460 --> 00:06:57.009 Some of the lessons learned I’ve had over the years, and the difficulties. 00:06:57.009 --> 00:07:01.940 And I’m going to go through some of those with you. 00:07:01.940 --> 00:07:06.960 First, there’s many, many people I’d like to thank for my career. 00:07:06.960 --> 00:07:11.940 Brian Atwater, for one, was a thesis adviser of mine – was great. 00:07:11.940 --> 00:07:15.569 You mentioned Bud Burke. He was really one of my mentors 00:07:15.569 --> 00:07:19.540 here at the USGS when I was here in the late ’70s, early ’80s. 00:07:19.540 --> 00:07:25.000 But really what started me off was meeting Gary Carver at Humboldt State. 00:07:25.000 --> 00:07:28.250 I was a forestry major. I got a forestry bachelor’s degree. 00:07:28.250 --> 00:07:32.099 And I had to hold off taking my last class because I met Gary, 00:07:32.099 --> 00:07:34.400 and I had to do a geology degree. 00:07:34.400 --> 00:07:40.910 And so Gary really taught me the love of geology. 00:07:40.910 --> 00:07:44.080 And then I worked with Clyde Wahrhaftig at UC-Berkeley. 00:07:44.080 --> 00:07:48.690 And he is a great man. 00:07:48.690 --> 00:07:52.110 He really taught me to think big and to be creative. 00:07:52.110 --> 00:07:56.580 Bill, don’t just look at the outcrop. You know, walk all the way around and, 00:07:56.580 --> 00:07:59.789 you know, see how big this landslide really is. [chuckles] 00:07:59.789 --> 00:08:02.910 Or, you know, you’re missing these glacial moraines. 00:08:02.910 --> 00:08:08.020 Anyway, Clyde was – really taught me how to make observations. 00:08:08.020 --> 00:08:10.360 And then I want to also mention Lloyd Cluff. 00:08:10.360 --> 00:08:13.400 I used his slide in the introduction. 00:08:13.400 --> 00:08:16.930 Lloyd was not only my mentor. 00:08:16.930 --> 00:08:21.129 He was who I wanted to emulate in terms of consulting. 00:08:21.129 --> 00:08:27.639 Lloyd really emphasized and repeatedly said, good science is good business, Bill. 00:08:27.639 --> 00:08:30.330 Make sure you have the best-quality people, you’re doing 00:08:30.330 --> 00:08:38.729 the best-quality work you can. Bring science into your analyses 00:08:38.729 --> 00:08:43.380 and interpretations and don’t let anything impede that. 00:08:43.380 --> 00:08:47.000 You know, make sure there’s good science behind your work 00:08:47.000 --> 00:08:51.500 as you pass that on to your client, which Lloyd was at PG&E. 00:08:51.500 --> 00:08:56.220 So I’ve really taken to that as I’ve – as you mentioned, 00:08:56.220 --> 00:08:59.720 I’ve had a few companies. 00:08:59.720 --> 00:09:05.310 The key to having a fun – I believe in having fun and being happy, and both 00:09:05.310 --> 00:09:10.000 of those involve being surrounded by good people and doing fun work. 00:09:10.000 --> 00:09:14.880 And so I’ve taken Lloyd’s message to heart in my life. 00:09:14.880 --> 00:09:19.310 So those are the three key people I’d like to acknowledge. 00:09:19.310 --> 00:09:25.260 So going through this, I have to work with you guys. 00:09:25.260 --> 00:09:29.230 The earthquake scientist. Brilliant people, usually. 00:09:29.230 --> 00:09:32.380 Great people. I love what you do. 00:09:32.380 --> 00:09:34.600 I feel like I’m part of your community – 00:09:34.600 --> 00:09:38.920 the Earth science community because I love research as well. 00:09:38.920 --> 00:09:43.380 And we have to – we have to translate that information 00:09:43.390 --> 00:09:48.310 so that people can build buildings to withstand earthquakes. 00:09:48.310 --> 00:09:51.980 And in this interface is often the consultant. 00:09:51.980 --> 00:09:57.730 And we have to bring Mother Nature and Father Engineering together. 00:09:57.730 --> 00:10:01.740 And in my life, it’s really evolved around capturing 00:10:01.740 --> 00:10:05.029 uncertainty for seismic hazard analysis. 00:10:05.029 --> 00:10:07.950 So this is the outline of my talk. 00:10:07.950 --> 00:10:12.339 I’d like to first describe the practical issues and challenges 00:10:12.339 --> 00:10:15.940 I’ve faced over the years. 00:10:15.940 --> 00:10:19.180 What is uncertainty, and why is it important? 00:10:19.180 --> 00:10:23.130 What process is currently evolving to capture uncertainty? 00:10:23.130 --> 00:10:27.769 And I’m hoping it will become standard of practice as the years goes by. 00:10:27.769 --> 00:10:31.920 What problems have we encountered? There have been technical challenges, 00:10:31.920 --> 00:10:36.450 human challenges, and then there are all the influences that you have to be 00:10:36.450 --> 00:10:39.279 aware of – your client/owner who doesn’t have enough money to 00:10:39.279 --> 00:10:44.390 pay for something, or the regulatory body that’s usually a decade or 00:10:44.390 --> 00:10:48.920 several decades behind science in terms of regulations. 00:10:48.920 --> 00:10:51.520 And then there are all kinds of political and societal influences. 00:10:51.529 --> 00:10:55.690 And I’ll give some explicit examples of those influences 00:10:55.690 --> 00:10:57.010 that we’ve had to deal with. 00:10:57.010 --> 00:11:02.740 And then I’ll go through what I think is the solution or the path forward. 00:11:02.740 --> 00:11:04.500 So why is uncertainty important? 00:11:04.500 --> 00:11:09.779 I think I’m speaking to people who understand this. 00:11:09.779 --> 00:11:15.140 Many, many companies and regulators, especially for site-critical 00:11:15.140 --> 00:11:20.860 facilities, are moving toward formal risk-informed decision-making. 00:11:20.860 --> 00:11:25.700 And if not formal, then informally, they’re considering risk. 00:11:25.700 --> 00:11:29.010 Risk assessment, in turn, requires identifying and, 00:11:29.010 --> 00:11:31.160 where we can, reducing uncertainty. 00:11:31.160 --> 00:11:39.760 But at least capturing the range of uncertainty so that we can understand – 00:11:39.760 --> 00:11:42.190 when we say we’re going to use the mean hazard, well, just where 00:11:42.190 --> 00:11:45.500 does the mean hazard fall within the range of uncertainty? 00:11:45.500 --> 00:11:50.070 Because we typically design to mean hazard or ground motion 00:11:50.070 --> 00:11:52.580 developed from mean hazard curves. 00:11:53.840 --> 00:11:58.420 Capturing uncertainty correctly also leads to reproducible stable 00:11:58.420 --> 00:12:03.149 estimates of probabilistic hazard, both with contemporary information, 00:12:03.149 --> 00:12:04.730 but also over time. 00:12:04.730 --> 00:12:10.620 When new information is evolved, the hazard results shouldn’t be stable. 00:12:10.620 --> 00:12:15.519 And I’ll describe what I mean by “stable” in just a minute. 00:12:15.520 --> 00:12:18.620 And through all of this, we have to always remember that 00:12:18.620 --> 00:12:25.000 societal influence on us. The public is demanding that we be certain. 00:12:25.010 --> 00:12:28.180 And that’s a real conundrum that we have because we need to 00:12:28.180 --> 00:12:34.300 capture and inform the public of the – of the uncertainties 00:12:34.300 --> 00:12:35.990 that we have in terms of risk. 00:12:35.990 --> 00:12:40.480 We’re not in a risk-free world. But the public wants certainty. 00:12:40.480 --> 00:12:46.260 And that’s – it’s hard then to work for a company like PG&E, 00:12:46.260 --> 00:12:51.839 which reacts to the public, where they don’t want anything to happen. 00:12:51.839 --> 00:12:58.050 And yet, we can’t build to that standard. 00:12:58.050 --> 00:13:04.029 So what is uncertainty? In PSHA space, we’re attempting 00:13:04.029 --> 00:13:08.100 to characterize an element of nature – earthquakes. 00:13:08.100 --> 00:13:12.930 And models of nature, by definition, are not uniquely defined. 00:13:12.930 --> 00:13:15.920 We will never have a true model of nature. 00:13:15.920 --> 00:13:22.500 We can only develop models that attempt to capture nature, 00:13:22.500 --> 00:13:25.720 but there will always be uncertainty. 00:13:25.720 --> 00:13:31.660 And for every – in an earthquake assessment, for every model 00:13:31.660 --> 00:13:36.970 or parameter value, magnitude, slip rate, there is always 00:13:36.970 --> 00:13:39.800 a range of alternatives that warrant consideration. 00:13:39.800 --> 00:13:41.560 And they are all technically defensible, 00:13:41.560 --> 00:13:44.320 and we have to make sure we’re capturing that range. 00:13:44.320 --> 00:13:51.680 And then we also have, in PSHA calculations, scenarios that include 00:13:51.680 --> 00:13:57.529 events that aren’t represented in the data. And so we have to capture that. 00:13:57.529 --> 00:14:00.160 And this is, to me, the most important interface issue 00:14:00.160 --> 00:14:05.730 with the scientific community. How do we extract your opinions and 00:14:05.730 --> 00:14:13.550 your confidence or certainty versus uncertainty in your interpretations? 00:14:13.550 --> 00:14:18.850 So I’m also speaking, I think, to the informed, but just what is uncertainty? 00:14:18.850 --> 00:14:23.370 It comprises two different types – aleatory and epistemic. 00:14:23.370 --> 00:14:27.170 Aleatory’s uncertainty is just random variability that we 00:14:27.170 --> 00:14:32.480 cannot further constrain with data or further knowledge. 00:14:32.480 --> 00:14:36.520 Epistemic uncertainty, on the other hand, is based primarily on lack of knowledge. 00:14:36.529 --> 00:14:38.030 There is a true state. 00:14:38.030 --> 00:14:42.839 We just don’t know what it is, and we can collect more data to 00:14:42.839 --> 00:14:47.420 better reduce that uncertainty or better understand that particular feature. 00:14:47.420 --> 00:14:51.290 And in my talk, I’m going to focus primarily on capturing 00:14:51.290 --> 00:14:53.579 epistemic uncertainty. Because that’s where we’re 00:14:53.579 --> 00:14:58.520 really engaging the earthquake – Earth science community. 00:14:58.520 --> 00:15:02.260 And because I’m a geologist and I do seismic source modeling, 00:15:02.260 --> 00:15:06.120 as opposed to ground motion attenuation modeling, I’m going to 00:15:06.120 --> 00:15:14.800 use mostly examples from the seismic source characterization model. 00:15:16.200 --> 00:15:20.720 And in capturing epistemic uncertainty, we use logic trees. 00:15:20.730 --> 00:15:28.910 This is – since Ram Kulkarni in the maybe late ’60s, early ’70s, at the old 00:15:28.910 --> 00:15:33.209 Woodward-Clyde, began working with Lloyd Cluff, began using logic trees. 00:15:33.209 --> 00:15:35.050 That’s really become our standard of practice for 00:15:35.050 --> 00:15:41.529 how we capture uncertainty in seismic source models. 00:15:41.529 --> 00:15:45.940 And we assign weights that reflect our confidence in either the model 00:15:45.940 --> 00:15:50.930 or parameter being the correct state of nature. 00:15:50.930 --> 00:15:56.449 And at every node, any subsequent node of a – of the branching logic tree makes 00:15:56.449 --> 00:16:00.850 the assumption that all previous nodes are the correct state. 00:16:00.850 --> 00:16:08.480 You cannot continue to be thinking, as you develop the logic tree, that if you – 00:16:08.480 --> 00:16:14.240 if you gave a certain parameter a 50/50 weight over two alternatives that, 00:16:14.240 --> 00:16:17.470 in subsequent branches, you’re carrying forward that 50/50 weight. 00:16:17.470 --> 00:16:20.730 In subsequent branches, you have to assume that you’re 00:16:20.730 --> 00:16:24.440 on the correct state of knowledge on that branch. 00:16:25.420 --> 00:16:30.070 And just a comment. Aleatory variability influences 00:16:30.070 --> 00:16:35.570 the shape of the hazard curve, but epistemic uncertainty 00:16:35.570 --> 00:16:37.500 leads to multiple hazard curves. 00:16:37.500 --> 00:16:41.680 And it’s that epistemic uncertainty that we want to make sure that 00:16:41.680 --> 00:16:45.700 we understand the limits or the range of that uncertainty 00:16:45.700 --> 00:16:48.290 between the 5th and 95th percentile. 00:16:48.290 --> 00:16:56.580 So here’s just a simple logic tree from Rob McGuire’s monogram in 2004. 00:16:56.580 --> 00:17:05.120 And for every path through the logic tree, we develop a unique hazard curve. 00:17:05.120 --> 00:17:08.120 So this is a very simple logic tree. 00:17:08.120 --> 00:17:13.270 Many of our logic trees end up with a thousand or more hazard curves. 00:17:13.270 --> 00:17:17.839 And so we can look at the median of the – of the hazard. 00:17:17.840 --> 00:17:23.220 We can also look at the fractiles – in particular, the 5th to 95th range. 00:17:23.220 --> 00:17:28.400 That’s our total range of uncertainty that we want to carry forward into risk. 00:17:28.400 --> 00:17:32.900 Often the mean hazard, which we design to, at 10 to the minus 4, 00:17:32.910 --> 00:17:36.760 10 to the minus 5, 10 to the minus 6, and for nuclear plants, we look all the 00:17:36.760 --> 00:17:42.730 way through 10 to the minus 7, diverges strongly from the median curve. 00:17:42.730 --> 00:17:44.770 And for risk assessment, we really need to know, 00:17:44.770 --> 00:17:50.580 where is this mean hazard falling within the fractiles of the hazard? 00:17:53.320 --> 00:17:57.190 So what process is used to capture uncertainty? 00:17:57.190 --> 00:18:03.320 And there’s a broad range of approaches. 00:18:04.789 --> 00:18:13.580 In the early 1990s, for all nuclear plants in the United States, 00:18:13.590 --> 00:18:17.380 there were two fundamentally different studies performed – 00:18:17.380 --> 00:18:19.950 one by Lawrence Livermore Laboratory, 00:18:19.950 --> 00:18:23.890 and one by the Electric Power Research Institute – EPRI. 00:18:23.890 --> 00:18:27.700 And they came to very different conclusions about hazard at the 00:18:27.700 --> 00:18:29.840 nuclear plants in the United States. 00:18:29.840 --> 00:18:37.110 And so this led to a investigation of why these hazard results are different. 00:18:37.110 --> 00:18:42.710 Is it based on technical input? Is it based on the process that was used? 00:18:42.710 --> 00:18:47.900 Just why – or the codes that were used? Why were the hazard results different? 00:18:49.630 --> 00:18:53.720 And so they convened what they call the Senior Seismic 00:18:53.720 --> 00:18:55.210 Hazard Analysis Committee. 00:18:55.210 --> 00:19:00.680 That’s the acronym SSHAC that many of you, or all of you, have heard of. 00:19:00.680 --> 00:19:11.030 And the result of this publication in 1997 by Budnitz and others was that 00:19:11.030 --> 00:19:15.309 the primary reason for the difference in hazard results was the process 00:19:15.309 --> 00:19:21.620 that was used to capture uncertainty. And so that led to them developing 00:19:21.620 --> 00:19:24.820 a series of recommendations for how we should do that. 00:19:24.820 --> 00:19:28.310 Uncertainty, especially in seismic source parameters, 00:19:28.310 --> 00:19:31.770 is primarily expert judgment. In ground motion models, 00:19:31.770 --> 00:19:36.070 there’s a lot of data for statistical analyses and – 00:19:36.070 --> 00:19:39.120 that can support various models and interpretations. 00:19:39.120 --> 00:19:45.960 Geologists really have a few data points, and we make very confident 00:19:45.960 --> 00:19:49.470 interpretations of those very few data, often. 00:19:49.470 --> 00:19:54.460 And so the assessment uncertainty requires expert judgment. 00:19:54.460 --> 00:19:58.390 And we found that a single individual tends to be 00:19:58.390 --> 00:20:02.900 very overconfident in their interpretation of those few data. 00:20:02.900 --> 00:20:09.030 And so the best way to capture the range of uncertainty is to use multiple experts. 00:20:09.030 --> 00:20:16.160 And by substituting the range of uncertainty each expert has, 00:20:16.160 --> 00:20:18.510 we’re doing a better job of capturing the full range 00:20:18.510 --> 00:20:23.200 of uncertainty in the professional community. 00:20:23.200 --> 00:20:27.130 And so they developed a process to capture uncertainty from 00:20:27.130 --> 00:20:29.820 multiple experts or from the professional community. 00:20:29.820 --> 00:20:35.710 And this SSHAC process is emerging now as the domestic and international 00:20:35.710 --> 00:20:40.410 best practice for site-critical facilities, in particular nuclear plants. 00:20:40.410 --> 00:20:46.039 IAEA, the International Atomic Energy Association, the NRC, many regulators 00:20:46.039 --> 00:20:51.820 across the world, are now adopting SSHAC as their process. 00:20:51.820 --> 00:20:56.650 Many dam owners now are following through with SSHAC-type processes 00:20:56.650 --> 00:21:00.370 or something comparable, where we’re trying to engage 00:21:00.370 --> 00:21:02.210 the professional community. 00:21:02.210 --> 00:21:06.710 And not just in the old days, 30 years ago – 20 years ago, 00:21:06.710 --> 00:21:09.700 we’d have a small group. Three guys go, and we’d read the 00:21:09.700 --> 00:21:17.580 literature, and we’d make up our model. And that’s no longer the process. 00:21:17.580 --> 00:21:21.250 So just what is SSHAC? It’s a structured framework 00:21:21.250 --> 00:21:25.600 for facilitating the multiple-expert interactions. 00:21:25.600 --> 00:21:29.590 You know, we – and it’s not a poll of experts, but it’s a collection 00:21:29.590 --> 00:21:35.150 of experts informing a team of their opinions. 00:21:35.150 --> 00:21:40.250 And then that team is charged with capturing that full range 00:21:40.250 --> 00:21:42.480 of uncertainty from those experts. 00:21:42.480 --> 00:21:47.250 And our learning occurs through workshops, working meetings, 00:21:47.250 --> 00:21:51.700 where we actually meet with the expert community. 00:21:51.700 --> 00:21:56.880 I mentioned the first report in 1997, a subsequent report in 2012. 00:21:56.880 --> 00:22:01.750 Tom Hanks was heavily involved in some of these earlier studies. 00:22:01.750 --> 00:22:08.280 And Tom, I didn’t put yours up there, but Tom has published some guidelines also. 00:22:08.280 --> 00:22:13.190 But in 2017, Coppersmith, Julian Bommer, and a couple of others 00:22:13.190 --> 00:22:18.150 published sort of lessons learned in a practical implementation 00:22:18.150 --> 00:22:22.309 guideline for how to – how to perform a SSHAC study, 00:22:22.309 --> 00:22:26.049 with the primary objective being two-fold. 00:22:26.049 --> 00:22:32.980 One is, develop a process that will produce reproducible stable estimates 00:22:32.980 --> 00:22:37.290 of probabilistic hazard and to capture the center, body, and range 00:22:37.290 --> 00:22:40.240 of technically defensible interpretations. 00:22:40.240 --> 00:22:43.860 That’s the – the acronym for that is CBR of the TDI – 00:22:43.860 --> 00:22:47.560 of the available data, methods, and models. 00:22:48.510 --> 00:22:52.820 What do we mean by reproducible stable hazard results? 00:22:52.830 --> 00:22:56.340 This is my interpretation. It’s been informed by discussions 00:22:56.340 --> 00:23:01.500 with Norm Abrahamson and Rob McGuire and Julian Bommer and others. 00:23:01.500 --> 00:23:08.860 But what is – what does “stable” mean? Well, you can view stability either with 00:23:08.870 --> 00:23:14.580 what’s in the current environment – if we were to develop hazard in the 00:23:14.580 --> 00:23:21.940 Sierra Nevada for dams for PG&E – a regional study – and the Bureau of 00:23:21.940 --> 00:23:28.180 Reclamation or Army Corps of Engineers did a similar study, 00:23:28.180 --> 00:23:33.680 if they – if the second team were to evaluate the same data, methods, and 00:23:33.680 --> 00:23:39.580 models, stability means that they would develop a similar CBR of the TDI. 00:23:40.660 --> 00:23:44.580 Thus, the hazard fractiles would be approximately the same. 00:23:44.580 --> 00:23:48.400 I’m showing the mean hazard there. Let me bring that slide back up 00:23:48.400 --> 00:23:51.409 with the mean hazard, but also the fractiles. 00:23:51.409 --> 00:23:54.870 So the fractiles will be approximately the same, but the mean hazard result 00:23:54.870 --> 00:23:59.799 in ground motion can vary a little bit, but should be roughly 10%. 00:23:59.799 --> 00:24:02.860 When I say mean hazard in ground motion, that’s on the X axis. 00:24:02.860 --> 00:24:07.610 On the Y axis, the difference might be up to 30%. 00:24:07.610 --> 00:24:16.090 Rob McGuire published a chapter in the CEUS-SSC volume, or study, 00:24:16.090 --> 00:24:22.210 where he looked at a range of studies done by individual stability. 00:24:22.210 --> 00:24:31.080 We can’t be more precise than roughly 30%, just based on the science. 00:24:31.080 --> 00:24:35.830 And with respect to new information over time – so you – if similar people 00:24:35.830 --> 00:24:41.669 were to do the same study with the same information currently, that’s stability. 00:24:41.669 --> 00:24:46.840 But how stable is the result looking forward over the next five or 10 years? 00:24:46.840 --> 00:24:50.180 Stability would mean that any new hazard results based on 00:24:50.190 --> 00:24:57.000 new information will fall within the existing fractiles over here. 00:24:57.000 --> 00:25:01.900 So if anything, the range from 5 to – whoops. 00:25:02.960 --> 00:25:10.040 The range from 5 to 95% should reduce, never grow larger. 00:25:10.040 --> 00:25:13.640 And the only time where it might grow larger, if what we were 00:25:13.640 --> 00:25:17.220 currently considering to be aleatory variability, as we learn more, 00:25:17.220 --> 00:25:20.700 sometimes something we thought was totally random becomes 00:25:20.700 --> 00:25:24.750 something that we know about and can better define. 00:25:24.750 --> 00:25:29.299 And so when you shift a variable from aleatory into epistemic, 00:25:29.299 --> 00:25:31.690 that can expand the uncertainty range. 00:25:31.690 --> 00:25:36.000 But hopefully, going forward, we will always reduce uncertainty. 00:25:36.000 --> 00:25:39.539 And the mean hazard, with this new information, 00:25:39.539 --> 00:25:43.140 almost always will move around, and potentially more than 10%. 00:25:43.140 --> 00:25:46.950 But as long as the mean hazard is staying within the fractile range, 00:25:46.950 --> 00:25:48.799 we would consider that stable. 00:25:48.799 --> 00:25:53.110 An example would be, let’s say, in the eastern United States, 00:25:53.110 --> 00:25:58.539 almost all of the hazard is generated from aerial source zones, 00:25:58.539 --> 00:26:01.080 not from active faults or what we call repeating 00:26:01.080 --> 00:26:05.169 large-magnitude earthquake sources like Charleston or New Madrid. 00:26:05.169 --> 00:26:11.360 Let’s say, for a nuclear power plant, an active fault is now identified 00:26:11.360 --> 00:26:14.320 within that aerial source zone near that power plant. 00:26:14.320 --> 00:26:18.820 Well, for that site-specific study – whoops. 00:26:21.929 --> 00:26:27.840 For that site-specific study, the mean hazard might go up 00:26:27.840 --> 00:26:32.060 considerably with new information. 00:26:36.250 --> 00:26:41.300 So what do we mean by capturing the center, body, and range of uncertainty? 00:26:41.309 --> 00:26:48.539 The center is based on the best data we have – direct evidence, observations, 00:26:48.539 --> 00:26:52.250 our preferred interpretations of those data, lead to a best estimate. 00:26:52.250 --> 00:26:55.140 And that’s what we would consider the center. 00:26:55.140 --> 00:26:58.740 The range would be bringing in indirect evidence. 00:26:58.740 --> 00:27:01.210 In ground motion studies, it would be numerical modeling 00:27:01.210 --> 00:27:03.840 and some simulations. 00:27:05.700 --> 00:27:09.330 In our business, we also have – and I’m going to describe this more – we have 00:27:09.330 --> 00:27:16.450 certain beliefs, or paradigms, like segmentation is very polarizing issue. 00:27:16.450 --> 00:27:19.630 We don’t know if it’s true or not true. 00:27:19.630 --> 00:27:23.780 And are there testable data to test segmentation? 00:27:23.780 --> 00:27:30.270 But, so we – so we have sort of indirect lines of evidence for 00:27:30.270 --> 00:27:36.090 capturing the range of what might happen in a – in a seismic source model. 00:27:36.090 --> 00:27:40.080 And that would be sort of the body. So that’s the center and body. 00:27:40.080 --> 00:27:47.419 The range would be, what are the known unknowns, or the unknown unknowns? 00:27:47.419 --> 00:27:51.990 As long as something is physically plausible – you know, 00:27:51.990 --> 00:27:56.240 this is – this is the title of my talk. How far – how uncertain should we be? 00:27:56.240 --> 00:28:04.730 How far out should this range go? Because the mean hazard is really 00:28:04.730 --> 00:28:10.220 driven by the tails – the upper tail of that uncertainty range. 00:28:10.220 --> 00:28:14.909 And so where we clip that range off is critical 00:28:14.909 --> 00:28:17.480 to what the final hazard results will be. 00:28:17.480 --> 00:28:25.419 And, you know, should we incorporate something in the range that can’t 00:28:25.419 --> 00:28:31.280 be tested and is not observed? Like a magnitude 9 earthquake 00:28:31.280 --> 00:28:37.450 on a continental strike-slip fault. Is that within the range or not? 00:28:37.450 --> 00:28:42.460 So these are the kinds of issues that are – that we face in 00:28:42.460 --> 00:28:45.700 developing a PSHA model. So that’s just what we mean by 00:28:45.700 --> 00:28:50.600 the center, body, and range. And I’ll go into some more of that. 00:28:50.600 --> 00:28:56.120 So the experts involved in this process – there’s the evaluator expert. 00:28:56.120 --> 00:28:59.470 That’s – in SSHAC, they’re the members of what we call 00:28:59.470 --> 00:29:02.830 the Technical Integration Team. They have the final intellectual 00:29:02.830 --> 00:29:05.529 ownership – not the professional community. 00:29:05.529 --> 00:29:09.789 But their job is to evaluate all the available data, methods, 00:29:09.789 --> 00:29:13.520 and models in a very objective manner. 00:29:14.300 --> 00:29:18.460 Without the biased perspective of, I like this model. 00:29:18.460 --> 00:29:25.190 You know, we’re trying to capture what the professional community likes. 00:29:25.190 --> 00:29:31.140 And then, after that evaluation, integrate that into a model 00:29:31.140 --> 00:29:33.930 that represents the full center, body, and range. 00:29:33.930 --> 00:29:39.210 And when I say must think differently, we’re all trained as scientists. 00:29:39.210 --> 00:29:41.649 We follow the scientific method. 00:29:41.649 --> 00:29:47.140 We collect data, we analyze data, we develop an interpretation of those data. 00:29:47.140 --> 00:29:50.100 And often, when we write a publication, we’re trying to convince others 00:29:50.100 --> 00:29:52.860 that my interpretation of these data is the correct one. 00:29:52.860 --> 00:29:56.630 And so we write it in a very confident way. 00:29:56.630 --> 00:30:00.490 And then science progresses by someone going, no, I’m going to collect more data 00:30:00.490 --> 00:30:03.920 and show that you’re wrong, and I have a better model for those data. 00:30:03.920 --> 00:30:07.880 That’s the common scientific approach. 00:30:07.880 --> 00:30:11.809 As an evaluator expert, you cannot think like a scientist. 00:30:11.809 --> 00:30:16.809 You have to objectively look at each model from the perspective of the 00:30:16.809 --> 00:30:19.770 person who developed that model as if it’s the correct one and 00:30:19.770 --> 00:30:25.580 make sure that we’re representing those range of interpretations 00:30:25.580 --> 00:30:28.350 in the center, body, and range. 00:30:28.350 --> 00:30:31.800 And we rely on resource experts and proponent experts. 00:30:31.800 --> 00:30:33.640 That is the Earth science community. 00:30:33.640 --> 00:30:39.140 They’re people who understand the data, the uncertainty in that data set. 00:30:39.140 --> 00:30:43.370 Then there are people who have made interpretations or models from that 00:30:43.370 --> 00:30:47.430 data and who are proponents of that particular interpretation. 00:30:47.430 --> 00:30:51.690 And so we want to – as the evaluator expert, these are the individuals we want 00:30:51.690 --> 00:30:58.640 to talk to and really understand the range of uncertainty from those individuals. 00:31:01.549 --> 00:31:04.179 This is what I just described earlier. 00:31:04.180 --> 00:31:09.240 A scientist interprets data, and it’s the classical way that science progresses. 00:31:09.240 --> 00:31:13.200 You know, they develop an interpretation of their data. 00:31:13.200 --> 00:31:16.140 An evaluator expert needs to be impartial 00:31:16.159 --> 00:31:18.350 and perform an objective assessment. 00:31:18.350 --> 00:31:25.649 And also, the value of SSHAC – and one reason I adhere to SSHAC – 00:31:25.649 --> 00:31:31.440 I really accept it as the path forward is because of the transparent nature 00:31:31.440 --> 00:31:35.540 of the SSHAC process and the documentation that’s required in the 00:31:35.540 --> 00:31:39.799 SSHAC process so that, 10 years from now, people will know what you did. 00:31:39.799 --> 00:31:45.120 Who you talked to, what you learned, how you went through your assessment. 00:31:45.120 --> 00:31:49.000 Right now, when I pick up a study done five years, 10 years, 15 years ago 00:31:49.000 --> 00:31:53.880 by someone not using the SSHAC documentation process, I have no idea 00:31:53.880 --> 00:32:01.020 why they assigned a slip rate of X to a certain fault or a b value to a certain – 00:32:01.020 --> 00:32:05.789 you know, or how did they smooth the b value over a certain aerial source zone? 00:32:05.789 --> 00:32:09.070 So it’s the documentation and transparent process 00:32:09.070 --> 00:32:15.539 that I really advocate in the SSHAC process. 00:32:15.539 --> 00:32:16.940 And if you don’t follow the SSHAC process, 00:32:16.940 --> 00:32:18.669 at least do something comparable. 00:32:18.669 --> 00:32:21.620 Make sure you’ve documented – and one way we document 00:32:21.620 --> 00:32:24.570 is we prepare a hazard input document, which defines 00:32:24.570 --> 00:32:29.400 exactly how to input all the parameters into the PSHA. 00:32:30.410 --> 00:32:34.100 So implementing this SSHAC process, it sounds very easy, right? 00:32:34.100 --> 00:32:39.380 I just described it. And it is an easy process. 00:32:39.380 --> 00:32:44.040 But in practice, it really can be complex. As I mentioned, there is technical 00:32:44.040 --> 00:32:48.480 challenges, human challenges, client/owner challenges, and it really 00:32:48.490 --> 00:32:52.000 gets to be exhausting, especially when I have to argue with people 00:32:52.000 --> 00:32:56.010 like Dave Schwartz over segmentation or something like that. 00:32:56.010 --> 00:33:00.460 So I want to go through some of these now with some concrete examples. 00:33:00.460 --> 00:33:03.120 So technical challenges. 00:33:03.120 --> 00:33:08.180 You know, our goal is to capture all the uncertainty we can, and the data, the 00:33:08.180 --> 00:33:14.559 methods, the analyses that people use, the interpretations, their final models. 00:33:16.720 --> 00:33:19.840 I’ve already mentioned, just how uncertain should that range be? 00:33:19.840 --> 00:33:24.520 How far out should we go? What kind of guidelines or rules should we follow? 00:33:26.400 --> 00:33:29.120 We have what I call true believers. 00:33:29.780 --> 00:33:34.240 Or truthiness – you know, there’s a variety of articles written on this – 00:33:34.250 --> 00:33:37.210 which is entrenched belief in paradigms. 00:33:37.210 --> 00:33:43.210 That’s very difficult, when we talk to proponent experts, to shake that. 00:33:43.210 --> 00:33:49.090 And we, as evaluator experts, easily succumb to that. 00:33:49.090 --> 00:33:53.370 In particular, for myself. I’m an evaluator expert. 00:33:53.370 --> 00:33:59.659 Personally, I believe in segmentation. I’ve got to shake myself of that belief 00:33:59.659 --> 00:34:06.110 and that, what I call, a paradigm going forward. 00:34:06.110 --> 00:34:09.919 And extracting this uncertainty from the 00:34:09.919 --> 00:34:13.330 scientific community is a real technical challenge. 00:34:13.330 --> 00:34:16.780 So there are the human behavior or instincts that get in the way. 00:34:16.780 --> 00:34:19.320 There’s the regulatory uncertainty I mentioned. 00:34:19.320 --> 00:34:22.980 They’re typically one step behind science, always. 00:34:22.980 --> 00:34:30.450 They’re always in a reactive mode. Regulators want consistency. 00:34:30.450 --> 00:34:34.960 In their mind, that equates to stability. They can’t keep changing their 00:34:34.960 --> 00:34:40.290 rules because that leads to unstable results in their mind. 00:34:40.290 --> 00:34:45.490 So they’ll often adhere to guidelines or approaches that were 00:34:45.490 --> 00:34:51.139 used 10, 20, 30 years ago still. And science has passed them by. 00:34:51.140 --> 00:34:57.920 And so dealing with that and trying to develop a model that satisfies a 00:34:57.920 --> 00:35:04.220 regulator impacts our ability to capture uncertainty. It’s a real issue. 00:35:04.230 --> 00:35:08.420 Because the regulator, in reality, is who defines standard of practice 00:35:08.420 --> 00:35:11.750 in the earthquake engineering community. 00:35:11.750 --> 00:35:16.470 The client/owner uncertainty – they tell you how much budget and – you know, 00:35:16.470 --> 00:35:20.050 we want to do a SSHAC study. It’s going to take 30 months to complete. 00:35:20.050 --> 00:35:22.400 No, I need something in six months. 00:35:22.400 --> 00:35:28.830 And, you know, you can’t do your job in a complete manner in six months. 00:35:28.830 --> 00:35:34.579 And so we’re always dealing with the client/owner uncertainty and influences. 00:35:34.579 --> 00:35:39.750 Clients have vested interests, and that is really – it’s an unspoken 00:35:39.750 --> 00:35:44.640 pressure that we have to deal with when we’re capturing uncertainty. Oops. 00:35:45.240 --> 00:35:49.320 And then I mentioned political societal influence. 00:35:49.339 --> 00:35:54.130 You know, what’s been called the post-true society. 00:35:54.130 --> 00:35:58.730 When did this really begin in our business? 00:35:58.730 --> 00:36:01.980 And I’m going to give a specific example on this. 00:36:01.980 --> 00:36:06.839 You know, the alternate facts, fake facts – we’ve all been hearing this. 00:36:06.839 --> 00:36:10.940 Is this beginning to infiltrate us? Whether in the actual scientific 00:36:10.940 --> 00:36:15.040 community – it certainly has in the, say, global warming community. 00:36:15.040 --> 00:36:22.240 You know, there are scientists who have been – who have 00:36:22.240 --> 00:36:25.720 concluded global warming is not happening. 00:36:27.020 --> 00:36:31.580 So we have these issues, and I don’t want that to influence our earthquake science 00:36:31.589 --> 00:36:37.850 community or the consulting community trying to interpret your data. 00:36:37.850 --> 00:36:41.760 And then there are also interveners. This is a critical role in our society. 00:36:41.760 --> 00:36:45.050 Interveners keep us honest, especially in the nuclear industry 00:36:45.050 --> 00:36:50.440 and also in the dam industry, oil and gas industry. 00:36:51.600 --> 00:36:55.100 But we can’t be unduly influenced by them. 00:36:56.740 --> 00:36:58.960 So very quickly, you know, the scientific method. 00:36:58.960 --> 00:37:01.710 We acquire data. We analyze the data. 00:37:01.710 --> 00:37:05.390 We interpret the data and develop a proponent model. 00:37:05.390 --> 00:37:10.230 This is the resource expert with the data, the proponent expert. 00:37:10.230 --> 00:37:13.120 What we have over time is a whole bunch of these models. 00:37:13.120 --> 00:37:17.960 I’ve represented four of them here – data analysis proponent model. 00:37:17.960 --> 00:37:22.710 At what time does one succeed another one and so we can abandon it? 00:37:22.710 --> 00:37:26.750 That is highly dependent on the country you’re working in. 00:37:26.750 --> 00:37:31.290 Sometimes the individuals you’re working with. 00:37:31.290 --> 00:37:36.180 In the state of Israel, for example, no model is ever superseded 00:37:36.180 --> 00:37:40.500 by a subsequent model. Everybody has to have their say, 00:37:40.510 --> 00:37:42.930 and you’ve got to capture everything in uncertainty. 00:37:42.930 --> 00:37:46.360 In the U.S., we’re much more likely to supersede one model 00:37:46.360 --> 00:37:48.960 with a subsequent model. 00:37:50.180 --> 00:37:52.980 But all these models have to be brought into the SSHAC process. 00:37:52.980 --> 00:37:56.160 Sometimes there is new data which inform an existing model. 00:37:56.160 --> 00:37:59.760 Sometimes there’s new data that no one’s interpreted yet. 00:37:59.760 --> 00:38:01.400 Same thing with methods. 00:38:01.400 --> 00:38:06.220 And that represents the whole available knowledge, let’s say. 00:38:06.230 --> 00:38:09.720 Our job as an evaluator expert is to consider all this and then develop 00:38:09.720 --> 00:38:14.060 a model that represents all of those perspectives. 00:38:16.180 --> 00:38:19.040 The technical – how uncertain should we be? 00:38:19.040 --> 00:38:25.160 Underestimating uncertainty – epistemic uncertainty is due to lack of data. 00:38:25.160 --> 00:38:32.660 We can’t truly know a state of – and less data should imply larger uncertainty. 00:38:32.660 --> 00:38:36.441 But in past studies, and even today – this is something the evaluator expert 00:38:36.441 --> 00:38:44.680 has to be cognizant of – we often have few data and we’re very confident. 00:38:44.680 --> 00:38:47.520 Few available studies lead to small uncertainty. 00:38:47.520 --> 00:38:52.650 That’s been the norm in the Earth science community, believe it or not. 00:38:52.650 --> 00:38:56.650 Many available studies lead to larger uncertainty. 00:38:56.650 --> 00:39:00.040 And that’s counter-intuitive. The more we know about something, 00:39:00.040 --> 00:39:03.880 our uncertainties should reduce. It shouldn’t grow. 00:39:05.230 --> 00:39:09.880 Thus uncertainty often increases the more work we do. 00:39:09.890 --> 00:39:12.890 To me, what that means is, we underestimated uncertainty 00:39:12.890 --> 00:39:14.730 at the beginning. 00:39:14.730 --> 00:39:19.859 And we have to do a better job of truly representing the range. 00:39:19.859 --> 00:39:24.280 And this is one of Norm Abrahamson’s points. 00:39:24.280 --> 00:39:30.290 This means we may need to create models or interpretations that go beyond 00:39:30.290 --> 00:39:36.260 our current data or proponent models to make sure that we’re representing 00:39:36.260 --> 00:39:42.400 possible physical states that may be valid. 00:39:42.400 --> 00:39:45.210 And you never know what’s enough unless you know 00:39:45.210 --> 00:39:49.140 what is more than enough. That’s one of my favorite quotes. 00:39:49.140 --> 00:39:52.390 So questions the evaluator expert always has to consider. 00:39:52.390 --> 00:39:54.910 How do we constrain models that are outside the range? 00:39:54.910 --> 00:39:57.780 They’re well-constrained by empirical data, either the 00:39:57.780 --> 00:40:01.420 ergodic assumption or direct observations of something. 00:40:01.420 --> 00:40:04.300 How do extrapolate models beyond the data? 00:40:04.300 --> 00:40:07.600 How do we identify and then reject ideas or models 00:40:07.600 --> 00:40:12.550 that aren’t grounded in fact or have an underlying physical basis? 00:40:12.550 --> 00:40:17.530 And how do we incorporate the known unknowns or the unknown unknowns? 00:40:17.530 --> 00:40:20.850 And even should we? That’s a – I don’t know the answer to that. 00:40:20.850 --> 00:40:24.120 This is something I’m always faced with. 00:40:24.120 --> 00:40:29.690 This is the great Donald Rumsfeld quote about weapons of 00:40:29.690 --> 00:40:32.170 mass destruction in Iraq. You know, they’re the 00:40:32.170 --> 00:40:37.400 unknown unknowns, which is how he based his conclusion. 00:40:37.400 --> 00:40:42.760 Considerations for extending uncertainty are that we have to go – 00:40:42.760 --> 00:40:47.690 this is – these are the things I consider about, how far out should I go 00:40:47.690 --> 00:40:52.099 with my range, knowing that the farther I push the range out, 00:40:52.100 --> 00:40:55.940 we often are directly impacting mean hazard. 00:40:55.940 --> 00:41:00.100 What’s the future stability of results? You know, I want to make sure 00:41:00.119 --> 00:41:04.060 that future work is always going to reduce uncertainty. 00:41:04.060 --> 00:41:07.940 Is something grounded in physical concepts or reality? 00:41:07.940 --> 00:41:10.150 What’s the acceptance credibility? 00:41:10.150 --> 00:41:14.849 If I don’t include it, is the professional community going to believe the results? 00:41:14.849 --> 00:41:16.940 Is the regulator going to believe the results? 00:41:16.940 --> 00:41:21.329 You know, it’s something – the credibility, to me, 00:41:21.329 --> 00:41:27.220 of the work we do is something that highly influences my judgments. 00:41:27.220 --> 00:41:32.320 Underestimating or overestimating hazard, and thus risk, 00:41:32.329 --> 00:41:36.910 is an assessment we need to – you know, we run a sensitivity analysis. 00:41:36.910 --> 00:41:42.460 If I include this, just how impactful is that to the final hazard result? 00:41:46.160 --> 00:41:49.500 You know, setting a precedent. 00:41:49.500 --> 00:41:54.260 If we include it, is this setting a dangerous precedent for future studies? 00:41:54.260 --> 00:41:59.100 You know, we need to really be concerned with that. 00:41:59.100 --> 00:42:03.460 And then considerations also are that it can help us 00:42:03.460 --> 00:42:07.540 focus future work on what matters. 00:42:07.540 --> 00:42:11.720 This brings me back to, when I graduated with my Ph.D. in 1982, 00:42:11.720 --> 00:42:15.900 I was Clyde Wahrhaftig’s last student in 1982. 00:42:15.910 --> 00:42:17.330 And he gave the commencement address, 00:42:17.330 --> 00:42:21.060 and his whole address was about global warming. 00:42:21.060 --> 00:42:22.420 And this was in 1982. 00:42:22.420 --> 00:42:24.260 You know, nobody was thinking about global warming. 00:42:24.270 --> 00:42:27.040 And I was, like, afterwards, Clyde, how could you talk about global warming? 00:42:27.040 --> 00:42:28.640 Why didn’t you talk about earthquakes? 00:42:28.640 --> 00:42:31.300 You know, this is – that’s what we’ve been working on. 00:42:31.319 --> 00:42:35.900 And, at that time, he’s an outlier. I’m not even sure it would have been 00:42:35.900 --> 00:42:41.250 in a model – would have even been accepted as within the range. 00:42:41.250 --> 00:42:46.650 But it just goes to show you that what we think may not be true today, maybe – 00:42:46.650 --> 00:42:48.840 as long as there’s a physical basis. 00:42:48.840 --> 00:42:51.810 You know, and Clyde’s observations were the glaciers were melting. 00:42:51.810 --> 00:42:54.000 There’s changes in fish migration patterns. 00:42:54.000 --> 00:42:57.589 You know, all these reasons that he came up with that there’s 00:42:57.589 --> 00:43:01.900 global warming. And then he had all these reasons why it was man-induced. 00:43:03.120 --> 00:43:09.660 So the primary issue is significance to hazard and how far out we go. 00:43:09.660 --> 00:43:14.079 Do we focus more work – many of you have seen tornado diagrams 00:43:14.080 --> 00:43:17.800 where we’ll look at every parameter of the logic tree. 00:43:17.800 --> 00:43:21.839 You know, the number one there is the base case ground motion model. 00:43:21.839 --> 00:43:26.319 Just how, as we move the parameter to alternative models, 00:43:26.319 --> 00:43:28.050 what influences the hazard? 00:43:28.050 --> 00:43:32.590 In this case, M-max doesn’t even influence hazard. 00:43:32.590 --> 00:43:35.650 So spending more work trying to reduce uncertainty in M-max 00:43:35.650 --> 00:43:40.820 is not really useful in this particular example. 00:43:42.700 --> 00:43:47.660 And we always have to remember there’s a penalty to pay if we 00:43:47.670 --> 00:43:53.339 don’t correctly understand uncertainty. It’s been known from biblical time 00:43:53.340 --> 00:43:58.000 all the way through to the present. So this isn’t a new concept. 00:44:01.190 --> 00:44:03.580 And this is an example that I’m guilty of – 00:44:03.589 --> 00:44:06.870 overconfidence and ignoring uncertainty. 00:44:06.870 --> 00:44:12.070 In 1986, I was responsible, at Diablo Canyon – Diablo Canyon is located 00:44:12.070 --> 00:44:20.260 along the coast right here in this picture and just over here in this picture. 00:44:20.260 --> 00:44:23.280 The data I had to work with in 1986 is on the left. 00:44:23.280 --> 00:44:26.970 That’s the offshore bathymetry. And PG&E asked, Bill, 00:44:26.970 --> 00:44:32.069 is there a fault along the coastline? And I said with 100% certainty, no. 00:44:32.069 --> 00:44:38.180 I collected hundreds of offshore geologic samples. 00:44:39.160 --> 00:44:43.980 I made offshore/onshore geologic maps to show continuity of stratigraphy. 00:44:43.990 --> 00:44:48.540 I looked at the bathymetry, and I didn’t identify any type of lineament, other 00:44:48.540 --> 00:44:54.580 than, as you can see, that there is a linear coastline along the Diablo Canyon coast. 00:44:54.580 --> 00:44:58.260 I attributed that to possibly oblique slip and uplift 00:44:58.260 --> 00:45:01.660 along the Hosgri Fault, which is directly offshore. 00:45:01.660 --> 00:45:06.010 As it turns out, Jeanne Hardebeck, as you know, re-interpreted 00:45:06.010 --> 00:45:10.670 seismicity data and identified a seismicity alignment. 00:45:10.670 --> 00:45:15.650 We went out. We collected detailed multibeam bathymetry. 00:45:15.650 --> 00:45:20.980 And there’s clearly an obvious fault directly offshore of Diablo Canyon. 00:45:20.980 --> 00:45:26.280 And so I was – given limited data, I was very overconfident in my result. 00:45:26.280 --> 00:45:28.160 And it wasn’t that nobody thought about it. 00:45:28.160 --> 00:45:31.589 Burt Slemmons, who was the peer reviewer of the – of the – 00:45:31.589 --> 00:45:34.510 for the NRC of the Diablo Canyon study asked me, Bill, 00:45:34.510 --> 00:45:38.000 don’t you think there should be a fault right offshore? 00:45:38.000 --> 00:45:40.950 And I said, yeah, you would think so 00:45:40.950 --> 00:45:43.050 because this whole range is being uplifted. 00:45:43.050 --> 00:45:44.950 We have the Los Osos Fault on the far side. 00:45:44.950 --> 00:45:50.880 Shouldn’t there be a fault on the offshore side uplifting the coast? 00:45:50.880 --> 00:45:54.060 And that’s why we spent a lot of effort looking for that fault. 00:45:54.060 --> 00:45:56.860 And we concluded there wasn’t one. 00:46:00.620 --> 00:46:05.140 And so the result of that – here’s that offshore seismicity lineament. 00:46:05.140 --> 00:46:08.290 And now the new mapping based on bathymetry and 00:46:08.290 --> 00:46:12.690 how all of these faults link up. We developed new models and 00:46:12.690 --> 00:46:16.079 new methods now to better represent the range of uncertainty. 00:46:16.079 --> 00:46:18.520 You know, we incorporated the new fault. 00:46:18.520 --> 00:46:23.660 We now have revised fault geometries. We revised all the fault slip rates. 00:46:23.660 --> 00:46:27.630 The new methods – we allow – as UCERF3 does, we allow for 00:46:27.630 --> 00:46:32.430 linked fault ruptures. We revised the magnitude frequency 00:46:32.430 --> 00:46:38.880 distribution to allow a tail to incorporate magnitudes up to 8-1/2 or larger. 00:46:40.040 --> 00:46:43.660 Anyway, there’s a whole number of new methods and new models 00:46:43.660 --> 00:46:48.180 that came out of the more recent work at Diablo Canyon. 00:46:50.640 --> 00:46:53.810 Trying to better represent this center, body, and range. 00:46:53.810 --> 00:46:58.200 Another recent example that’s currently happening right now is Lake Oroville. 00:46:58.200 --> 00:47:03.160 This is the spillway that’s – as you all know, has been damaged and 00:47:03.160 --> 00:47:07.170 just basically gone from about here all the way down now. 00:47:07.170 --> 00:47:11.300 Yesterday they closed the spillway, and we had a number of 00:47:11.300 --> 00:47:13.730 geologists in there mapping this. 00:47:13.730 --> 00:47:18.030 It’s just a big hole right here, potentially eating headward. 00:47:18.030 --> 00:47:20.930 So when this first happened, then they opened the emergency spillway. 00:47:20.930 --> 00:47:27.350 And the emergency spillway is unlined. It’s just on rock. 00:47:27.350 --> 00:47:31.400 And the conclusion at that time – this emergency spillway is a safe 00:47:31.400 --> 00:47:36.210 and stable structure founded on solid rock that will not erode. 00:47:36.210 --> 00:47:38.960 With 100% certainty. 00:47:38.960 --> 00:47:43.460 And they opened that emergency spillway, and immediately, within hours, 00:47:43.460 --> 00:47:47.740 began eroding headward, encroaching up toward the dam. 00:47:47.740 --> 00:47:51.490 Which, if it breached the dam, it would have just – as you know, 00:47:51.490 --> 00:47:55.690 the erosion would have just slotted down through that emergency spillway. 00:47:55.690 --> 00:47:58.609 So they closed it. This is the hole that had eroded. 00:47:58.609 --> 00:48:03.260 There’s many more current YouTube videos and things on this. 00:48:03.260 --> 00:48:06.930 And, you know, the governor’s response is, we live in a world of risk. 00:48:06.930 --> 00:48:09.160 Stuff happens, and we respond. 00:48:09.160 --> 00:48:12.240 You know, I’d like to think, yes, we live in a world of risk, 00:48:12.240 --> 00:48:16.920 but let’s really understand that risk and not just be reactive to that risk. 00:48:16.920 --> 00:48:24.280 Let’s try to design to best – you know, best design for that risk. 00:48:26.770 --> 00:48:30.040 So technical uncertainty. You know, I mentioned the 00:48:30.040 --> 00:48:32.660 true believers – entrenched belief in paradigms. 00:48:32.660 --> 00:48:36.819 A paradigm – this is the definition – it’s a set of assumptions, concepts, 00:48:36.820 --> 00:48:40.700 values, practices, that constitute a way of viewing reality. 00:48:40.700 --> 00:48:47.240 It’s not based – there’s no data in there, or facts, or truth. 00:48:47.250 --> 00:48:51.710 It’s a paradigm. It’s a belief we have based on our judgments. 00:48:51.710 --> 00:48:54.440 That once they become entrenched in our mind, 00:48:54.440 --> 00:48:59.940 they’re very difficult to dislodge. And why do we do that? 00:48:59.940 --> 00:49:02.420 You know, why do we continue with these paradigms? 00:49:02.430 --> 00:49:07.170 And there’s many of them in our earthquake hazard community. 00:49:07.170 --> 00:49:10.180 Because it conforms to the way we’ve done things before. 00:49:10.180 --> 00:49:15.119 And the benefits of that is it’s a – it’s stable to the regulator. 00:49:15.120 --> 00:49:19.560 They – regulators like stable, consistent approaches. 00:49:19.560 --> 00:49:22.800 And, you know, do we have past or present paradigms 00:49:22.820 --> 00:49:24.480 in our earthquake science? 00:49:24.480 --> 00:49:30.900 There’s a whole bunch that you might consider paradigms. 00:49:30.910 --> 00:49:34.570 All may be true, and so maybe we need to carry them all forward. 00:49:34.570 --> 00:49:37.110 But we need to really challenge these 00:49:37.110 --> 00:49:41.620 and really critically examine the underlying data now. 00:49:41.620 --> 00:49:45.740 They may – some of these paradigms were developed in the very infancy of 00:49:45.740 --> 00:49:50.250 probabilistic hazard analysis, like the assumption of Poisson fault behavior. 00:49:50.250 --> 00:49:53.720 That’s in every PSHA code. That’s the default. 00:49:53.720 --> 00:49:57.190 Everything’s Poisson. Well, as a geologist, when you 00:49:57.190 --> 00:50:01.609 look at a fault, would you say there’s time-dependent behavior on that fault? 00:50:01.609 --> 00:50:03.859 Or is purely Poisson? 00:50:03.859 --> 00:50:08.570 So I think, you know, we need to challenge the Poisson fault behavior. 00:50:08.570 --> 00:50:15.050 This is a term I love. Belief echoes, by Emily Thorson. 00:50:15.050 --> 00:50:18.800 It’s a continued belief in a paradigm despite new facts. 00:50:18.800 --> 00:50:22.819 And – you know, and she concluded, once a paradigm gains traction, 00:50:22.819 --> 00:50:28.170 confronting true believers with facts is simply not even useful. 00:50:28.170 --> 00:50:33.540 It challenges – it challenges them. 00:50:33.540 --> 00:50:37.180 So if paradigms are wrong, you know, what do we do? 00:50:37.180 --> 00:50:39.190 Do we tear everything down and rebuild it? 00:50:39.190 --> 00:50:41.200 Which, basically, that’s what UCERF 3 did. 00:50:41.200 --> 00:50:43.050 They just stated over and said, we’re going to develop 00:50:43.050 --> 00:50:47.819 a whole new way of looking at hazard in California. 00:50:47.819 --> 00:50:50.809 Or do we simply expand uncertainties and accommodate 00:50:50.809 --> 00:50:55.700 all these paradigms with expanded range of uncertainty? 00:50:55.700 --> 00:50:59.130 Or do we ignore it for convenience or regulatory stability? 00:50:59.130 --> 00:51:02.329 And believe me, there’s enormous pressure from regulators 00:51:02.329 --> 00:51:08.290 to not change anything. Because it’s unstable. 00:51:08.290 --> 00:51:12.370 Getting the NRC to recognize time-dependent behavior at 00:51:12.370 --> 00:51:17.079 Diablo Canyon was a real challenge. They did not want to recognize 00:51:17.080 --> 00:51:23.720 time-dependent behavior on faults. Even within the range of uncertainty. 00:51:27.740 --> 00:51:31.300 So at Diablo Canyon, we had some of these what I’ll call old paradigms – 00:51:31.309 --> 00:51:34.970 fault segmentation, Poisson, accepted fault geometries – 00:51:34.970 --> 00:51:38.140 and we changed it all at Diablo Canyon. 00:51:38.140 --> 00:51:40.980 We now, as I mentioned, have linked fault behavior. 00:51:40.980 --> 00:51:46.660 We developed an equivalent Poisson rate to model time-dependent recurrence, 00:51:46.660 --> 00:51:49.910 and we have a range of alternate fault geometries. 00:51:49.910 --> 00:51:53.030 So just using that as an example from my own – so human behaviors 00:51:53.030 --> 00:52:01.240 and instincts that get in the way. Man is man because they can reason. 00:52:01.240 --> 00:52:04.720 And so we believe that we’re all doing rational reasoning. 00:52:04.720 --> 00:52:07.480 Facts and careful weighing of evidence determines which model we 00:52:07.480 --> 00:52:13.200 accept and which we reject. The reality is, many of us have cognitive bias. 00:52:13.200 --> 00:52:17.190 We have what’s called motivated reasoning. 00:52:17.190 --> 00:52:23.839 We start with, we think this is the right answer, and we develop 00:52:23.840 --> 00:52:29.070 facts and models to prove that answer. 00:52:32.360 --> 00:52:35.080 And as I mentioned, psychologists call this cognitive bias. 00:52:35.089 --> 00:52:38.940 So what does cognitive bias or motivated reasoning lead to? 00:52:38.940 --> 00:52:44.620 It leads to scientific overconfidence, overestimating what’s known. 00:52:46.100 --> 00:52:48.680 And this actually leads to greater uncertainty over time 00:52:48.680 --> 00:52:53.460 as we do more studies. It leads to anchoring. 00:52:53.460 --> 00:52:59.020 Proponent models become accepted belief, which then become paradigms. 00:52:59.030 --> 00:53:02.410 And it also leads to polarized views on factual questions. 00:53:02.410 --> 00:53:05.740 We can have the same set of facts, and different scientists with 00:53:05.740 --> 00:53:13.910 different experiences or biases or whatever can lead to 00:53:13.910 --> 00:53:17.980 very different interpretations of the same body of facts. 00:53:19.470 --> 00:53:23.400 One of the first things – when I joined the USGS back in the late 1970s, 00:53:23.400 --> 00:53:29.520 Al Lindh pulled me aside, and he goes, Bill, what are you geologists doing? 00:53:29.520 --> 00:53:33.690 Every time you dig a new trench on a fault, your uncertainty gets greater. 00:53:33.690 --> 00:53:37.589 You know, you – every time a new trench, there’s a new segment on the 00:53:37.589 --> 00:53:43.119 San Andreas Fault. And he goes, that’s – the reverse should be true. 00:53:43.119 --> 00:53:45.851 The more trenches you dig, you should be more confident in the 00:53:45.851 --> 00:53:49.609 behavior of the San Andreas Fault. And I – that’s always stuck with me 00:53:49.609 --> 00:53:54.099 since the late 1970s when Al pulled me aside. 00:53:54.099 --> 00:54:01.339 And so anyway, so scientific overconfidence – in a way, 00:54:01.339 --> 00:54:03.510 it’s bred into our scientific method. 00:54:03.510 --> 00:54:07.080 Because we’re trying to develop a model and then defend it. 00:54:07.080 --> 00:54:14.540 And it’s human nature. I’m going to defend what I believe or what I interpret. 00:54:16.960 --> 00:54:26.000 And it – you know, it leads to a number of issues. 00:54:26.000 --> 00:54:32.660 Anchoring is past belief – in past beliefs or paradigms, your – by human nature, 00:54:32.660 --> 00:54:38.910 once again, we’re inclined to like models we’re familiar with. 00:54:38.910 --> 00:54:42.790 And especially if we’ve generated them ourselves. 00:54:42.790 --> 00:54:46.490 That’s where self-esteem comes in. It took a lot for me to finally 00:54:46.490 --> 00:54:50.410 accept that there’s a fault offshore of Diablo Canyon. 00:54:50.410 --> 00:54:52.869 You know, Jeanne Hardebeck is showing seismicity lineaments, 00:54:52.869 --> 00:54:55.960 and I’m going, no, there’s no fault out there. 00:54:55.960 --> 00:54:58.099 And then it blew me away when it turned out to be 00:54:58.100 --> 00:55:01.860 a strike-slip fault instead of a reverse fault. 00:55:01.860 --> 00:55:06.700 You know, so it was my self-esteem initially rejecting 00:55:06.700 --> 00:55:10.500 or reacting to this new fault offshore. 00:55:10.500 --> 00:55:13.780 Conditioning events – if something dramatic has happened, 00:55:13.790 --> 00:55:16.859 like Lake Oroville, that’s a conditioning event. 00:55:16.860 --> 00:55:21.360 Now every dam owner wants to go evaluate their spillways. 00:55:24.609 --> 00:55:30.760 So cognitive bias – we all have it. And the only solution in terms of 00:55:30.760 --> 00:55:35.809 capturing the CBR of the TDI – and this is a quote from the original 00:55:35.809 --> 00:55:39.160 NUREG volume – you can only counter this cognitive bias or 00:55:39.160 --> 00:55:42.599 motivated reasoning by making the evaluator experts 00:55:42.600 --> 00:55:46.400 aware of it and to encourage them to avoid that. 00:55:46.400 --> 00:55:50.280 You know, we have to – it’s really – once again, evaluator experts 00:55:50.280 --> 00:55:54.869 have to think differently. They have to be aware of this. 00:55:54.869 --> 00:56:01.839 It’s very susceptible to fall into when you’re assessing uncertainty 00:56:01.840 --> 00:56:08.480 to have your own bias or preference for a particular model. 00:56:09.640 --> 00:56:13.580 I promised an example. So political uncertainty. 00:56:13.580 --> 00:56:20.299 I’m going to use Panama as an example. I worked down at the Panama Canal. 00:56:20.299 --> 00:56:25.290 This is for the new Panama Canal for the Panamax container ships. 00:56:25.290 --> 00:56:29.630 But why was there a Panama Canal in the first place? 00:56:29.630 --> 00:56:33.500 Back in the late 1800s, early 1900s, there were 00:56:33.500 --> 00:56:39.359 two alternative routes selected – Panama versus Nicaragua. 00:56:39.359 --> 00:56:44.130 Initially, our Congress – House of Representatives voted 00:56:44.130 --> 00:56:53.410 strongly to build a canal in Nicaragua. But then Panama went on a – 00:56:53.410 --> 00:56:59.599 the government of Panama went on a mission to convince people that Panama, 00:56:59.599 --> 00:57:03.059 as opposed to Nicaragua or their other Central American neighbors, 00:57:03.059 --> 00:57:04.460 we have no volcanoes. 00:57:04.460 --> 00:57:07.220 We have no hurricanes. We have no earthquakes. 00:57:07.220 --> 00:57:11.829 And, you know, when you look at the distribution of volcanoes, 00:57:11.829 --> 00:57:15.450 yeah, okay, that’s true. A volcano-free zone in Panama. 00:57:15.450 --> 00:57:18.800 When you look at all instrumental seismicity in Panama, 00:57:18.800 --> 00:57:21.880 well, okay, there’s very few earthquakes. 00:57:25.700 --> 00:57:32.540 And so what Panama did is, three days prior to the Senate final vote, they sent – 00:57:32.540 --> 00:57:36.240 every Senator received a postcard with a postage stamp from Nicaragua 00:57:36.240 --> 00:57:38.560 showing volcanoes. [laughter] 00:57:38.560 --> 00:57:44.180 And, you know, rational reasoning? 00:57:44.190 --> 00:57:48.820 The Senate voted strongly to support the Panama Canal. 00:57:48.820 --> 00:57:51.970 Even today, this is the travel guide you would pick up 00:57:51.970 --> 00:57:54.790 if you go visit Panama today. 00:57:54.790 --> 00:57:59.480 It has the conclusion over there, which I’ll highlight. 00:57:59.480 --> 00:58:04.040 Panama has no major earthquakes. Panama also has none of the destructive 00:58:04.040 --> 00:58:08.309 earthquakes that plague its Central American neighbors. 00:58:08.309 --> 00:58:11.040 Then along came Eldon Gath and Tom Rockwell. 00:58:11.040 --> 00:58:14.790 And they did a study for the Panama Canal and identified several – 00:58:14.790 --> 00:58:18.760 and it’s shown in yellow – several significant faults, 00:58:18.760 --> 00:58:23.420 each of which – these are multiple- millimeter-per-year slip rate faults. 00:58:25.180 --> 00:58:30.040 Panama’s initial response was, we don’t believe you, Eldon and Tom. 00:58:30.040 --> 00:58:36.100 In fact, at one point, they even refused to allow Eldon Gath to re-enter the country. 00:58:37.000 --> 00:58:39.420 We’re going to hire Bill Lettis to come down and give us an 00:58:39.420 --> 00:58:44.650 independent assessment of these faults. And, you know, the work Eldon 00:58:44.650 --> 00:58:51.610 and Tom did was very good. And obviously, there are active faults. 00:58:51.610 --> 00:58:55.510 And still, the Panamanian government would not believe it. 00:58:55.510 --> 00:58:58.030 And we were working with URS at that time. 00:58:58.030 --> 00:59:02.410 Lelio Mejia was guiding it, and Lelio was, how do we – 00:59:02.410 --> 00:59:05.340 we want to get these fault assessments into the ground motion, 00:59:05.340 --> 00:59:07.710 and fault – in particular, fault rupture. 00:59:07.710 --> 00:59:13.710 This Pedro Miguel Fault right here crosses the new 00:59:13.710 --> 00:59:17.740 Panamax Canal at one of the concrete locks. 00:59:17.740 --> 00:59:21.619 And if there’s fault rupture through the concrete locks, it will drain 00:59:21.619 --> 00:59:24.569 Lake Gatun right here, and it will take years for them 00:59:24.569 --> 00:59:29.880 to repair and then refill Lake Gatun to restore the Panama Canal. 00:59:33.360 --> 00:59:37.480 And so Eldon and Tom pointed out, in 1621, there’s well-documented 00:59:37.490 --> 00:59:43.020 destruction of the original Panama City. And that coincides with the function 00:59:43.020 --> 00:59:50.010 of a road built by the Spanish across Panama, and which crosses – 00:59:50.010 --> 00:59:54.770 out in the jungle, crosses the Pedro Miguel Fault where the 00:59:54.770 --> 00:59:58.610 Spanish would get gold from Peru and carry it across Panama 00:59:58.610 --> 01:00:04.510 on mule and put it on galleons and ship it over to Spain. 01:00:04.510 --> 01:00:10.640 And so I thought, Tom and Eldon, let’s get machetes – it’s all buried 01:00:10.640 --> 01:00:15.090 in jungle – and follow this road to where it crosses the fault. 01:00:15.090 --> 01:00:18.990 And see if it’s been offset by the 1621 earthquake, 01:00:18.990 --> 01:00:22.620 which should offset the road. And it’s out there in the jungle. 01:00:22.620 --> 01:00:28.490 This is the – this is the Pedro Miguel Fault. This is a blow-up of it. 01:00:28.490 --> 01:00:33.280 Pedro Miguel Fault, and the road crosses it out here. 01:00:33.280 --> 01:00:36.990 Here is that road crossing. It just – we got lucky. 01:00:36.990 --> 01:00:41.140 At that point, the road happened to drop down into a – near a creek bed 01:00:41.140 --> 01:00:44.020 without very much vegetation. And I was holding the GPS. 01:00:44.020 --> 01:00:47.530 You know, we should – through the canopy, we were finally getting covered. 01:00:47.530 --> 01:00:49.609 We should be near this fault. 01:00:49.609 --> 01:00:54.130 And right here the road comes up, hits about a half-meter-high scarp, 01:00:54.130 --> 01:00:56.490 and is offset several meters to the right. 01:00:56.490 --> 01:00:59.119 And this is a right-lateral strike-slip fault. 01:00:59.119 --> 01:01:03.990 And so Tom went back, and with Pat Williams and others, 01:01:03.990 --> 01:01:06.650 and did detailed surveying of this road. 01:01:06.650 --> 01:01:09.599 And the road was beautifully marked by culverts, even. 01:01:09.599 --> 01:01:12.950 They really built this road very nice. So it’s very easy, with piercing 01:01:12.950 --> 01:01:17.240 points along the margin of the road, to identify it. 01:01:17.240 --> 01:01:22.559 Tom restored the offset, and there was about 2.8 meters 01:01:22.559 --> 01:01:28.310 of offset during an earthquake, which we attribute to the 1621. 01:01:28.310 --> 01:01:37.609 Still, Panama did not design the lock to withstand fault rupture. 01:01:37.609 --> 01:01:40.829 So they – at the end of the day, they incorporated this fault 01:01:40.829 --> 01:01:45.819 into the ground motion analysis for the levees or the Borinquen Dam 01:01:45.819 --> 01:01:50.690 and other structures to – but where it’s entrenched, and they built 01:01:50.690 --> 01:01:57.600 a concrete lock, the concrete lock, it’s not designed for fault rupture. 01:01:58.400 --> 01:02:02.320 They just won’t believe that there is active faulting. 01:02:03.440 --> 01:02:06.619 So that’s an example of political alternate facts 01:02:06.620 --> 01:02:14.860 impacting our assessment and ultimately rejecting our uncertainty. 01:02:15.760 --> 01:02:17.500 And so the solution going forward, 01:02:17.500 --> 01:02:22.060 it’s just rediscovering the same conclusion reached in 1997. 01:02:23.130 --> 01:02:26.480 The primary pitfall in executing a successful PSHA, 01:02:26.480 --> 01:02:29.480 capturing uncertainty, is procedural. 01:02:30.600 --> 01:02:38.710 And it emphasizes why we need a good transparent structured process 01:02:38.710 --> 01:02:46.760 for capturing uncertainty that hopefully, even governments will believe in, 01:02:46.760 --> 01:02:51.760 but certainly has the credibility of the technical community and the regulator. 01:02:52.840 --> 01:02:56.740 Second, we really need to train evaluator experts to think 01:02:56.740 --> 01:03:02.170 as evaluator experts and not as scientists, which all evaluator experts are. 01:03:02.170 --> 01:03:08.700 We’ll all trained using the scientific method. 01:03:08.700 --> 01:03:13.130 And we can’t think that way as evaluator experts. 01:03:17.300 --> 01:03:25.559 As I mentioned, we need to train ourselves. 01:03:25.559 --> 01:03:28.670 We need to apply the concept of hazard informed 01:03:28.670 --> 01:03:30.630 when we are capturing uncertainty. 01:03:30.630 --> 01:03:33.089 This is something that Norm Abrahamson in particular 01:03:33.089 --> 01:03:40.160 really harps on. He said this to me over a hundred times. 01:03:40.160 --> 01:03:43.579 Focus on what matters, not on what is interesting, Bill. 01:03:43.579 --> 01:03:48.020 Once again, as scientists, you want to understand everything about a model. 01:03:48.020 --> 01:03:54.240 But in limited-scope budget, schedule, you really need to focus on 01:03:54.240 --> 01:03:58.300 what is significant to hazard. And so going through a full sensitivity 01:03:58.300 --> 01:04:04.940 analysis, and in an iterative manner, during the entire process is critical. 01:04:04.950 --> 01:04:10.820 Once again, in a transparent and explicit manner with full documentation 01:04:10.820 --> 01:04:14.590 of all these hazard iterations that you’ve done. 01:04:14.590 --> 01:04:19.280 And then finally, quality people. Going back to Lloyd Cluff. 01:04:19.280 --> 01:04:21.849 Quality people equates to quality work. 01:04:21.849 --> 01:04:28.150 Your evaluator experts really need to be selected carefully. 01:04:28.150 --> 01:04:33.579 We need senior experienced people. We also need young people on our – 01:04:33.580 --> 01:04:38.530 on our evaluation teams to train the next generation. 01:04:39.440 --> 01:04:43.920 But this goes all the way back to way back when, when David 01:04:43.920 --> 01:04:48.620 was just a young man – David Schwartz is in here somewhere. 01:04:48.620 --> 01:04:53.160 But Kerry Sieh – here’s Lloyd Cluff, Kerry Sieh, Dave Schwartz, 01:04:53.170 --> 01:04:57.480 Gary Carver, Burt Slemmons. George Brogan is in here. 01:04:57.480 --> 01:05:00.900 [Nancy Bigger], who just passed away, is here. 01:05:00.900 --> 01:05:05.460 But really a great team was put together, and this is way before SSHAC, 01:05:05.460 --> 01:05:10.000 to evaluate the TAPS Pipeline, or the Alyeska Pipeline. 01:05:10.000 --> 01:05:15.220 And, you know, they found the location of the fault. 01:05:15.230 --> 01:05:17.380 They had uncertainty where the fault might be. 01:05:17.380 --> 01:05:21.099 The maximum displacement. They had the average displacement 01:05:21.099 --> 01:05:24.680 then added uncertainty to that average displacement. 01:05:24.680 --> 01:05:29.490 And the TAPS performance during the 2002 Denali earthquake was beautiful. 01:05:29.490 --> 01:05:31.039 It performed as designed. 01:05:31.039 --> 01:05:34.090 There’s not very much remaining tolerance, but that’s okay. 01:05:34.090 --> 01:05:37.260 It performed as designed. There was no spills. 01:05:37.260 --> 01:05:39.790 And so they were right. And was it because 01:05:39.790 --> 01:05:42.410 of the wrong reasons? Because they were lucky? 01:05:42.410 --> 01:05:45.560 In my opinion, it’s because they had good judgment by 01:05:45.560 --> 01:05:49.400 highly skilled people understanding uncertainty 01:05:49.400 --> 01:05:56.120 and designing for that uncertainty in this fault rupture assessment. 01:05:57.740 --> 01:06:01.920 So just in conclusion, uncertainty is critical for hazard assessment. 01:06:01.920 --> 01:06:07.340 It’s essential for risk-informed decision-making, which, as I mentioned, 01:06:07.350 --> 01:06:12.170 almost all of our clients are moving more and more toward. 01:06:12.170 --> 01:06:16.430 It does help with stability of results over time. 01:06:16.430 --> 01:06:19.990 I hate it when a new hazard study is done and hazard suddenly 01:06:19.990 --> 01:06:23.829 has increased by a factor of 2 in a certain location. 01:06:23.829 --> 01:06:29.120 It’s just – that’s not – that means we underestimated uncertainty the first time. 01:06:30.460 --> 01:06:34.840 Characterizing uncertainty challenges our human instincts. 01:06:34.840 --> 01:06:39.560 We need to think beyond available data and methods and models to capture 01:06:39.560 --> 01:06:44.520 the full range because we simply don’t understand the state of nature. 01:06:45.580 --> 01:06:50.680 SSHAC provides a process, or some comparable type of process. 01:06:50.680 --> 01:06:53.540 We need a transparent process. 01:06:55.480 --> 01:07:00.420 And it must be hazard-informed to help focus us and to really 01:07:00.420 --> 01:07:04.090 capture the uncertainty on those issues that matter. 01:07:04.090 --> 01:07:07.760 And one interface suggestion I would have, as a consultant 01:07:07.760 --> 01:07:14.319 looking at my Earth science community is, if there is a way, 01:07:14.319 --> 01:07:18.460 in our future publications, after you’ve developed your 01:07:18.460 --> 01:07:25.460 interpretation and your model, we need some – almost a section by itself. 01:07:25.460 --> 01:07:29.000 What’s your range of uncertainty in that assessment? 01:07:29.000 --> 01:07:33.230 It’s not meant to challenge your interpretation or anything. 01:07:33.230 --> 01:07:38.400 It’s just to help future users of your information understand, 01:07:38.400 --> 01:07:40.980 is there uncertainty in your data? How good are the data? 01:07:40.980 --> 01:07:44.400 Is there uncertainty in your analytical approach? 01:07:44.400 --> 01:07:46.349 Is there uncertainty in your interpretation? 01:07:46.349 --> 01:07:49.380 Is there uncertainty in your final model? 01:07:49.380 --> 01:07:53.340 Understanding that from your perspective makes it easy for us. 01:07:53.349 --> 01:07:59.549 Because then I don’t need to read what’s in your mind or make something up. 01:07:59.549 --> 01:08:03.260 So that’s just a thought going forward. 01:08:03.260 --> 01:08:07.160 So thank you very much. I appreciate it. 01:08:07.160 --> 01:08:09.140 Any questions or … 01:08:09.140 --> 01:08:15.840 [ Applause ] 01:08:15.840 --> 01:08:20.240 That was good practice. I went long, so … 01:08:24.360 --> 01:08:26.940 - Questions for Bill? 01:08:29.100 --> 01:08:31.940 - Or if you have any comments and would just email them. 01:08:31.940 --> 01:08:38.360 I really want to modify this presentation for next week and future presentations. 01:08:39.700 --> 01:08:43.860 [ Silence ] 01:08:45.000 --> 01:08:49.440 - Thanks very much for a really entertaining talk. 01:08:49.440 --> 01:08:54.339 I wanted to ask you a question about the notion of decreasing uncertainty – 01:08:54.339 --> 01:08:59.089 or it seemed to be a statement that there was a goal to decrease uncertainty 01:08:59.089 --> 01:09:02.640 as you went along in studies. But then you also mentioned 01:09:02.640 --> 01:09:07.380 that sometimes the scientific method leads us to a new understanding, 01:09:07.380 --> 01:09:10.940 and probably the experience of many people in the room is that 01:09:10.940 --> 01:09:16.909 then you realize that your understanding is worse, so your uncertainty increases 01:09:16.909 --> 01:09:21.009 as you come up with a better physical understanding of a process. 01:09:21.009 --> 01:09:25.799 So I guess my question is, which is it? Is it a goal to – whose goal 01:09:25.799 --> 01:09:29.920 is it to decrease uncertainty? Is it the regulator or the client? 01:09:29.920 --> 01:09:34.859 - Well, it’s – the evaluator expert. In the PSHA, we have to 01:09:34.859 --> 01:09:40.699 understand that, as you – as a scientist, if you do a new study, and it – 01:09:40.699 --> 01:09:47.429 and opinions change or interpretations changes, our initial assessment 01:09:47.429 --> 01:09:51.069 has to make sure we’re bracketing within the range, that possibility that, if you did 01:09:51.069 --> 01:09:56.650 another study or had more information, that your interpretation might change. 01:09:56.650 --> 01:10:01.150 So the goal of the evaluator expert is to be broad enough in our 01:10:01.150 --> 01:10:03.730 initial assessment of the center, body, and range 01:10:03.730 --> 01:10:08.590 of uncertainty that any future studies are going to fall somewhere within that. 01:10:08.590 --> 01:10:12.400 We don’t want it to go beyond the range. Certainly beyond the range. 01:10:12.400 --> 01:10:18.300 Hopefully it’s within the body or even our best estimate. 01:10:18.300 --> 01:10:22.250 But I fully appreciate, you know, new knowledge will come along. 01:10:22.250 --> 01:10:25.210 I use the example, we might discover an active fault 01:10:25.210 --> 01:10:28.600 in the central eastern United States. 01:10:28.600 --> 01:10:35.860 If we’ve done our study correctly, in general, the aerial source zone 01:10:35.860 --> 01:10:40.960 should capture the possibility that there is that earthquake – 01:10:40.960 --> 01:10:45.160 that active fault within that zone. 01:10:45.170 --> 01:10:51.330 You might localize now the hazard on that fault, which is why I said 01:10:51.330 --> 01:10:56.530 the mean could vary widely, but if we captured that maximum 01:10:56.530 --> 01:10:59.630 magnitude and the rate of that earthquake occurring in the 01:10:59.630 --> 01:11:05.810 aerial source zone within our 95th percentile, the hazard 01:11:05.810 --> 01:11:09.960 shouldn’t go beyond that 95th percentile. I hope. 01:11:11.260 --> 01:11:14.100 That’s – but that’s the way of thinking is, I want to make sure 01:11:14.100 --> 01:11:19.200 my range is broad enough that any new information – new models – 01:11:19.200 --> 01:11:21.420 that’s why I said we need to go beyond the data, 01:11:21.420 --> 01:11:25.980 beyond what we know now, in our capturing of uncertainty. 01:11:25.980 --> 01:11:30.520 If there’s a physical basis or possibility for that. 01:11:30.520 --> 01:11:35.970 As Norm pointed out, for some faults or some earthquake sources, 01:11:35.970 --> 01:11:39.400 we simply don’t have any information. 01:11:39.400 --> 01:11:43.820 We need to create an interpretation of that fault with a broad enough 01:11:43.820 --> 01:11:48.700 uncertainty that, when somebody finally works on it, it falls within that range. 01:11:48.700 --> 01:11:52.460 - So it’s like adding something – you might maybe call it, like, 01:11:52.469 --> 01:11:58.130 the Rumsfeld Distribution, that you add to every analysis, sort of. 01:11:58.130 --> 01:11:59.440 - Sure. - The possibility that there 01:11:59.440 --> 01:12:02.180 might be this fault – this central United States … 01:12:02.180 --> 01:12:09.420 - I fully expect new research by the science community will change and 01:12:09.420 --> 01:12:14.260 develop new models and – I mean, things will change. 01:12:14.880 --> 01:12:19.360 And in the past, when we’ve had – you know, 20 years ago, 01:12:19.360 --> 01:12:26.270 we would say this fault can produce a magnitude 7-1/4 plus or minus 1/4. 01:12:27.500 --> 01:12:31.691 You know, we were very confident that any future maximum magnitude 01:12:31.700 --> 01:12:36.540 would fall between 7 and 7-1/2 with a mean of 7-1/4. 01:12:36.540 --> 01:12:43.840 And that’s simply – future work might show – like the UCERF3 01:12:43.840 --> 01:12:45.970 model now where we’re linking faults together. 01:12:45.970 --> 01:12:48.130 Well, maybe it could grow into a magnitude 8. 01:12:48.130 --> 01:12:54.850 Well, that falls way outside our range and will influence hazard. 01:12:54.850 --> 01:12:59.500 So given limited data, we – that original assessment 01:12:59.500 --> 01:13:04.219 of 7-1/4 plus or minus 1/4 was way too narrow a range of uncertainty. 01:13:04.220 --> 01:13:09.880 When we do work, we have to make sure we’re capturing a full range. 01:13:11.880 --> 01:13:16.860 Until data rejects that or brings that range down. 01:13:16.860 --> 01:13:19.140 That’s why I’m hoping future studies will only 01:13:19.140 --> 01:13:23.000 narrow that range and not grow that range. 01:13:24.200 --> 01:13:35.780 [ Silence ] 01:13:36.780 --> 01:13:40.160 - Yeah, so when they dug the foundation for the locks – 01:13:40.170 --> 01:13:45.670 for the new locks on the Panama Canal, that the fault went right through, 01:13:45.670 --> 01:13:48.760 did you see the fault in the foundation excavation? 01:13:48.760 --> 01:13:55.510 - Okay, so a couple things happened. And I don’t mean to belittle Panama. 01:13:55.510 --> 01:13:59.940 I mean, the people working on the Panama Canal, the technical people, 01:13:59.940 --> 01:14:02.610 really wanted to do a good job. 01:14:02.610 --> 01:14:08.590 And there was other pressures and cost pressures by the contractor and things. 01:14:08.590 --> 01:14:11.320 Because they did not want to have to move this lock. 01:14:11.320 --> 01:14:18.190 So when they made this giant excavation – when Tom Rockwell – 01:14:18.190 --> 01:14:22.890 Tom worked on one side of the canal. I worked on the other side. 01:14:22.890 --> 01:14:25.900 They wouldn’t let us work at the same location. 01:14:25.900 --> 01:14:28.260 Because they wanted independent assessment. 01:14:28.260 --> 01:14:31.050 But the whole place was jungle. 01:14:31.050 --> 01:14:36.110 When they – and when they made the large excavation, 01:14:36.110 --> 01:14:40.940 they did not invite Tom Rockwell or Eldon Gath back to look at it 01:14:40.940 --> 01:14:43.960 and help them find it – find the fault. 01:14:43.960 --> 01:14:50.160 And remember, this is a jungle, and the saprolite thickness is 20 meters thick. 01:14:50.160 --> 01:14:54.880 I mean, it’s really not an easy chore to find this fault. 01:14:57.780 --> 01:15:00.460 They invited me back. 01:15:00.460 --> 01:15:05.840 And I walked the length of the canal, this excavation, with them. 01:15:05.840 --> 01:15:10.440 And their technical people said, see, Bill, there’s no fault. 01:15:10.440 --> 01:15:15.660 I said, well, this canal is, in places, entrenched. 01:15:15.660 --> 01:15:20.980 And in other places, where it goes through a saddle, it’s got embankments. 01:15:20.980 --> 01:15:24.180 So it’s not always entrenched. 01:15:24.190 --> 01:15:33.090 And this fault has a dip to it, and it’s very simple three-dimensional geometry. 01:15:33.090 --> 01:15:36.330 The Panamanians were projecting this fault as a straight line. 01:15:36.330 --> 01:15:41.429 And here’s where it should be. I said, no, no, it’s dipping. 01:15:41.429 --> 01:15:44.770 And where it’s dipping, it goes right through this saddle 01:15:44.770 --> 01:15:49.989 where you don’t have an excavation, in the topographic saddle. 01:15:49.989 --> 01:15:52.300 And the reason this topographic saddle is here 01:15:52.300 --> 01:15:59.840 is that’s because that’s where the fault goes. And – but we can’t see it. 01:15:59.840 --> 01:16:03.310 And I said, but you don’t have an excavation here. 01:16:03.310 --> 01:16:08.100 You’re constructing the landfill over it, or, you know, 01:16:08.100 --> 01:16:13.250 these embankments over it. And then going to put the lock inside. 01:16:13.250 --> 01:16:18.949 So it – but they could not go to their superior or 01:16:18.949 --> 01:16:23.440 to the government and say, here is the fault – or to the contractor. 01:16:23.440 --> 01:16:28.280 And so they just didn’t include it in the fault rupture. 01:16:28.280 --> 01:16:31.660 To me, it was a very frustrating experience. 01:16:32.840 --> 01:16:37.980 And I hope, in my lifetime, they don’t have that earthquake. [chuckles] 01:16:40.260 --> 01:16:45.500 - Bill, you mentioned a couple times that the regulatory guidelines 01:16:45.500 --> 01:16:50.580 were decades behind. And I see that for critical facilities. 01:16:50.580 --> 01:16:54.440 And so I’d like to hear a comment on what you suggest 01:16:54.449 --> 01:16:57.690 could be improved there. But I also see it the other way around 01:16:57.690 --> 01:17:02.949 where we’re defining the regulatory – the state of the practice. 01:17:02.949 --> 01:17:09.290 And so, for more common consulting practices, we’re ahead 01:17:09.290 --> 01:17:16.040 of the consulting practice. - The Earth science community? 01:17:16.040 --> 01:17:17.489 - Yeah. - The Earth science community 01:17:17.489 --> 01:17:22.440 is always – in terms of knowledge, you’re always ahead of the consultant. 01:17:22.440 --> 01:17:24.920 - No, no, the regulatory. 01:17:24.920 --> 01:17:27.940 - You’re always ahead of the regulator also. 01:17:28.940 --> 01:17:33.860 - See, that’s – I see it both ways. I see it sort of – so, for critical facilities, 01:17:33.860 --> 01:17:38.710 I see that the regulatory state of the practice might be behind 01:17:38.710 --> 01:17:42.780 what the Earth science community is doing, or the consultants. 01:17:42.780 --> 01:17:47.380 But for general consulting, it seems like it’s the opposite. 01:17:47.380 --> 01:17:52.660 - Well, the building code is typically based on the National Hazard Map. 01:17:52.660 --> 01:17:54.340 - That’s [inaudible]. Right. - That, in some cases, 01:17:54.349 --> 01:17:58.070 is five or eight years old. And we already know that there are 01:17:58.070 --> 01:18:03.030 changes that should be made, but they’re not in the building code yet. 01:18:03.030 --> 01:18:08.380 So in that – the building code does the best job of sort of staying current 01:18:08.380 --> 01:18:15.560 with hazard assessments. But places like – there are some – 01:18:15.560 --> 01:18:20.730 like Division of Safety of Dams still requires a deterministic 01:18:20.730 --> 01:18:23.950 analysis of dams. - Right. 01:18:23.950 --> 01:18:26.530 - The technical staff acknowledge, and they want to know what 01:18:26.530 --> 01:18:31.150 the probabilistic hazard is, so we typically do both. 01:18:31.150 --> 01:18:37.260 But they have not yet moved from a deterministic to a probabilistic approach. 01:18:37.260 --> 01:18:39.949 The Bureau of Reclamation has moved to a probabilistic approach, 01:18:39.949 --> 01:18:41.960 but that took a while. The Corps of Engineers with 01:18:41.960 --> 01:18:46.060 [Ellis Kraninski] in the old days weren’t going to budge off of a 01:18:46.060 --> 01:18:49.180 deterministic approach for dams. 01:18:49.180 --> 01:18:56.360 The oil and gas industry now is moving and requiring a SSHAC study. 01:18:57.800 --> 01:19:01.460 Or the way they phrase it is a complete, explicit quantitative 01:19:01.460 --> 01:19:08.780 assessment of uncertainty in the – in your PSHA. 01:19:08.780 --> 01:19:11.500 And the oil and gas industry has interpreted that as 01:19:11.500 --> 01:19:14.860 doing a SSHAC-like study. 01:19:18.400 --> 01:19:22.600 We’re still working hard on probabilistic fault rupture. 01:19:22.600 --> 01:19:25.880 You know, we had that workshop not long ago. 01:19:25.880 --> 01:19:29.040 I mean, the science is there for us to do probabilistic fault displacement 01:19:29.040 --> 01:19:37.680 hazard analyses, but many firms still want to have a deterministic 01:19:37.680 --> 01:19:40.540 fault displacement analysis. 01:19:41.640 --> 01:19:44.380 Anyway, I don’t know if I’m answering your question, but … 01:19:44.380 --> 01:19:47.340 - Well, it helps. I just – I don’t know what … 01:19:47.340 --> 01:19:48.870 - I don’t mean – I’m not condemning a regulator. 01:19:48.870 --> 01:19:51.050 - No, no, no. - Regulators are in a tough spot. 01:19:51.050 --> 01:19:53.550 - Right. - The technical staff are keeping up 01:19:53.550 --> 01:19:59.010 with current approaches and techniques. But the regulation is designed 01:19:59.010 --> 01:20:03.570 for what they believe to be stability. They don’t want to impose on one 01:20:03.570 --> 01:20:09.000 utility or one owner what they didn’t impose on another owner. 01:20:09.000 --> 01:20:12.340 And they want stability through time. 01:20:12.340 --> 01:20:17.320 And so they’ll make changes, but it’s a slow evolutionary process. 01:20:17.320 --> 01:20:21.140 - Right. So I guess it’s a challenge we just have to live with and maybe … 01:20:21.150 --> 01:20:23.480 - Yeah. - Maybe more frequent updates 01:20:23.480 --> 01:20:29.500 could help, but it’s not – right. Okay. - Yeah. 01:20:30.860 --> 01:20:36.440 [ Silence ] 01:20:37.400 --> 01:20:41.820 - Okay. If there’s no more questions, we’re going to take Bill to lunch 01:20:41.820 --> 01:20:50.320 at the patio. Let’s meet in, let’s say, less than 10 minutes at the flagpole. 01:20:50.320 --> 01:20:52.460 And let’s thank our speaker again. 01:20:52.460 --> 01:20:56.960 [ Applause ] 01:20:56.960 --> 01:20:58.480 - Thank you. And please, this is my trial …