PubTalk 1/2018 — ShakeAlert: Path to West Coast EQ Early Warning

Video Transcript

Detailed Description

Title: ShakeAlert: The Path to West Coast Earthquake Early Warning ... how a few seconds can save lives and property

• The ShakeAlert earthquake early warning system will begin limited operations this year.
• Alerts could save lives and properties but several challenges remain.
• With millions at risk, why isn't full public alerting happening yet?

Details

Image Dimensions: 480 x 360

Date Taken:

Length: 01:14:34

Location Taken: Menlo Park, CA, US

Transcript

[ Silence ]

[ Silence ]

[background conversations]

Good evening, everyone.
Welcome to the evening lecture series.

My name is Robert de Groot.
I’m the national coordinator

for communication, education, and

And it’s my great pleasure tonight
to introduce you do Doug Given.

And I’ll tell you

But one thing, before we get started,
I wanted to spend a second to remind

Observatory talk that’s being given
on February 22nd, again at 7:00 p.m.

And there are fliers in the back.
The other piece I wanted to mention

as well is that there are

fact sheets in the back as well.
So please feel free to take one,

take two, take one home
to a friend, whatever you like.

But they’re available in the back,
and they’ll provide you with

program that you can learn later on.

and those sorts of things.

So I’d like to introduce Doug Given.
He is the USGS national

earthquake early warning coordinator.
And he coordinates USGS,

state organizations and universities,
and private companies to build the

system for the West Coast of the U.S.

system became operational

in California in 2012 and
in the Pacific Northwest in 2015.

In February 2016, the more
capable production prototype was

completed on a year – and a year later,
extended to the Pacific Northwest.

With this – with this version,
USGS began soliciting

real-world pilot
implementations.

USGS has the goal of regionally limited
public notifcations in 2018 – this year.

A little bit of

He joined the USGS in 1978, and he’s
on his 40th year, which is much longer.

I’ve only been with the USGS for about
two years, so he has a lot of years on me,

so I’m learning a lot
from him every single day.

He has pursued research into
the seismicity and tectonics of

southern California and
conducted field investigations

of several large earthquakes,
including Haiti in 2010.

He managed the Southern California
Seismic Network for 19 years.

Doug has been instrumental in the
development of several automated

earthquake processing systems and
was a prime architect of the

Quake Management System.

Doug is also an adjunct professor of

He’s been doing
this since 1991.

He is a lifetime resident of southern
California and lives with his wife

in La Crescenta, two blocks

And that’s something we tend to
do in southern California. [laughter]

So, yeah, I spent 17 years living on the
hanging wall of the Hollywood Fault.

So we typically discuss this geography.
[laughter]

He is an elder in the Presbyterian
Church USA and enjoys hiking,

So I’d like to welcome Doug

Thank you very much.

- Thank you.
- Enjoy.

[ Applause ]

- Yeah, with that intro, it sounds
almost like you should be considering

whether to date me or not.
[laughter]

I like long walks on the beach and …
[laughter]

Well, I want to thank you all for
coming out this evening to hear about

this really exciting technology that the
USGS is bringing to bear to reduce

the impacts of earthquakes in the
West Coast of the United States.

My goal tonight is to describe
the current status of the system,

a little bit of the history of how we
got here, and a lot about the challenges

that this project has, and to describe
what you can expect in this year,

calling a limited public rollout.

And I’ll explain what that means and
what some of those limitations are.

I want to start, though, with this slide.
The far right side shows a lot of logos,

and that is there to emphasize
the fact that this is an extremely

collaborative project.
It’s not just USGS, but many other

organizations, including –
well, I’m not going to read them all

because it would take too long.
But university partners, state agencies,

and others are all working
together to realize this.

And we actually need more partners
to consume – not to consume –

to consummate the
finalization of this.

It’s our goal to do earthquake
early warning in the three states

of the West Coast – Washington,
Oregon, and California.

And this slide illustrates why
we’re concentrating on the

West Coast of the United States.
This is an estimate of the impacts

and losses due to earthquakes in the
United States on an annualized basis.

And so you can see, those numbers end
with B’s – billions of dollars of losses.

And, again, this is an annualized idea.
And so that means that,

when the big one does occur,
even though it’s infrequent,

the losses are
going to be huge.

And then, spread out over years, you can
see that there’s still $6 billion of losses. Now, most of these losses have to do with infrastructure – buildings and roads and so forth. And earthquake early warning is not going to be able to mitigate all of those losses, but it will mitigate some of them and reduce the impact and make our society more resilient to bounce back from damage caused by those earthquakes. [beeping sounds] Sorry. I can’t – that’s not an earthquake. [laughter] The earthquake hazards program of the U.S. has the mission of reducing death, injury, and property damage in earthquakes. And to do that, we do a number of different activities. We assess the seismic risks long-term. We conduct research into the risk. And the underlying part, we provide earthquake monitoring and notification. That notification is the category that ShakeAlert falls into. And we also build public awareness, and ShakeAlert is going to be instrumental in that as well. The pictures on this slide are the pre-event products – the things that help managers and insurance industry and others plan for earthquakes by characterizing how bad the risk is, where it is, and so forth. So these are all pre-event, pre-earthquake products. Now, when an earthquake occurs, there are a number of products that are produced by USGS to help with situational awareness for the understanding of what just happened. How do you allocate resources? How many resources do you need to bring to bear? Should you activate mutual aid agreements or not? And so this continuum here, you can see, starts on the left when the earthquake occurs. And then, moving to the right – I’m going to skip over the red – and we – traditionally, in the seismic network monitoring that we do, and have done for decades, produce a location and a magnitude to describe what happened within about a minute or two. That’s all automated. And, in a lot of ways, earthquake early warning is doing exactly that same job, but just compressing the timeframe into a few seconds instead of a few minutes. And then, following the earthquake, we can send out the information through our event notification system, or ENS. You can go on the website, sign up, and get an email or a text message for earthquakes in areas that you are concerned about. That’s a free service, and it has more than 400,000 subscribers. We also do a quick characterization of the impacts of the earthquake – ShakeMap is a product that shows the extent of ground shaking. And then another of our products is called PAGER. And that does an estimate of the dollar losses and the fatalities resulting from earthquakes worldwide. So there are a number of products that we provide, post-earthquake. But the one that we’re talking about tonight is ShakeAlert, the earthquake early warning system. So the basics of earthquake early warning. The fundamental idea is, it is not earthquake prediction. We are not guessing where an earthquake is about to occur. Instead, we are detecting an earthquake that has already begun and doing that very, very rapidly. We have to figure out where the earthquake is, how big the earthquake is, and ultimately, what impact it’s going to have. What is the resulting ground motion going to be? How hard is it going to shake things? Is it likely to do damage or not? And we want to do that in just a few seconds and get that information out to people who will be impacted. If we can do it fast enough, we can actually send warning information before the shaking arrives. And that is the essence of earthquake early warning. Information can move at the speed of light. Seismic waves, as you see here, move at a couple of miles per second through the rocks of the crust. Now, when we think about who can benefit from this information, they fall into two broad categories. First is people. You are people. And when you hear about earthquake early warning, most people’s reaction is, great, how can I get it? When do I get it? And how do I use it? And there are many things that people can do – drop, cover, and hold on is the advice we currently give when you feel an earthquake. And it’s the same advice that we would give you if you hear an earthquake early warning alert. You just get a little extra time to do that protective activity. But there are other contexts that you might find yourself in. If you’re driving. Might give you time to slow down and pull over to the side of the road. If you are in a work situation with hazardous materials, you might be able to protect yourself from harm by moving away from machinery or chemicals or some other hazardous area. And one that always resonates is, if you’re on the operating table [laughter], you would probably want your doctor to know that things are about to move around. [laughter] That’s the people part. The other category is things. Automatic processes – machinery, trains. So you can slow and stop trains. Here in the Bay Area, you’re all familiar with BART. BART has been slowing their trains based on earthquake early warning since 2012. And so they were a very early adopter because they saw the value of slowing down their trains. If a full train at rush hour, going at full speed, were to derail, it would be the largest mass casualty incident in U.S. history. So there are – it’s something you want to prevent. You can close valves. You can stop factories. You can do all sorts of things in a automatic environment. And so the options there are actually endless, and there are many things that we, I’m sure, have not even conceived of that are possible uses for earthquake early warning in that automatic mode. So why don’t we have earthquake early warning? No talk is complete without a meme, so there is our meme. I want it now. From Willy Wonka. I’d like to give it to you now, but there are some limitations and some challenges to why we cannot yet do early warning for the masses – general public alerting. And I’ll describe those. So those challenges include limitations of resources – in other words, funding. This project is not fully funded. And that’s a real restriction to our ability to complete the system and get full-scale alerts to everyone. But, in addition to that, we have some bureaucratic hurdles. It’s difficult to hire within the government. And more importantly, it’s difficult to get environmental permits to put in new stations. So we’ve found that, as we try to build out a large number of sensors, we run into this problem, and it increases the cost of putting in stations, and it slows down the process of putting in stations. Much of the West Coast is controlled by federal government agencies – national parks, national forests, BLM land – and those all require environmental impact reports. And the fact that we’re spending federal dollars to do that work means that we still have to meet those environmental criteria, even on private lands. Because it’s a federally funded project. And that, as I said, can be a challenge. There are some physical challenges due to just the physics of earthquakes. And I’ll be describing those in greater detail later. And then the system itself. We have to ensure that it’s reliable and accurate. And doing so has its own challenges, which I’ll also describe briefly. And then a big challenge is just the ability to get the alerts out to you all. You may think that that’s easy. You’ve got a cell phone. You get messages all the time. There is no current technology that can do mass communication fast enough for earthquake early warning to be effective. And we are working on that, and I’ll describe that in greater detail. But the silver lining is, despite all these challenges, the USGS is dedicated to doing limited public alerting this year. The funding part – way too much detail on this slide. The lower right-hand box shows our original estimate for what the cost of completing the early warning system would be. This was a report that we published in 2014. And, at that time, we estimated that, to build out the system for the three states of the West Coast would take another$38 million.

And then it would take
$16 million to run it every year. You can see our funding profile in the graph – in the table on the left-hand side. This year, we are funded at$10 million.

So it’s not even the
run-it-every-year number.

Certainly not even close to
the build-it-out number.

And so we are operating on a shoestring,
but making progress nevertheless.

Now, it’s not all
federal funding.

The state of California has recently
allocated $10 million for the project. About two-thirds of that is going to station build-out. The remaining third is going to be used for public education purposes. State of Oregon has funded some instrumentation. The Gordon and Betty Moore Foundation has given us$10 million

And some federal funding that

passed through, first the state,
and then through the City of

Los Angeles, was another $5.6 million for southern California build-out. So there are other sources of funding, but we still have not been able to complete the system with that available. To make things just a little more interesting, our original estimate of the cost of doing it is under review. We are getting really close to releasing revised numbers. And guess what direction the numbers are going to go? [laughter] Yeah. They’re going up. Not only because of inflation and the fact that we missed some things in the original estimate, but also because, as I said, some of the things, like putting stations in the ground and getting environmental approvals is more timely – or, more time-consuming and more expensive than we had originally thought. Well, so here’s a slide with lots of information. Over on the right-hand side is the map of the three states and an approximate map of station locations where some of them are already located and where others are anticipated to be located. This gives you a kind of a big view of what the system is like. The yellow stars are the alert centers, located in Seattle, the Bay Area, and southern California. The ShakeAlert earthquake early warning infrastructure is layered on top of previously existing seismic monitoring capabilities funded by the USGS and partners, sometimes, in the states. And these seismic networks, that have been in operation for years, are jointly run by the USGS and university partners. So in southern California, it’s Caltech. In the Bay Area, it’s Berkeley. In the Pacific Northwest, University of Oregon, University of Washington. The path is in the center bottom. We began seriously working on ShakeAlert in 2006. At that time, it was mostly research on the science of detecting earthquakes rapidly. We went through a demonstration phase, operationalized it, and turned it on in a serious way in 2012. At that time, we started to solicit beta users that could see the signals, but we asked them, please don’t take any actions based on the alerts because the system is not yet reliable enough. In 2017 – that’s last year, in April, we made a very important jump, and we said, we now believe it’s reliable enough that we are soliciting pilot users – users that could actually take actions and start to develop mostly those automatic types of processes. Now, we’re not soliciting for individuals to participate. We’re looking for organizations that will build capabilities and lead the way within their particular industries. And then, this year, we are doing the limited public rollout phase of the operation. The central part of the slide shows the five major components of an earthquake early warning system. Those five components, running from left to right, are the sensor networks – the field telemetry to bring those data back into the central processing sites. The box in the center is that central processing, which includes hardware infrastructure – networking, computers, and so forth. And then, the scientific software that does the job of detecting the earthquakes generating the alerts. So you see, those are dark blue boxes, and there’s an arrow below. That’s the alert generation section of the processing thread. But the alerts don’t do any good unless you send them to somebody. And so the next thing you have to do is distribute those alerts to the users, whether they be industrial users or individuals. And then, of course, they have to know what to do with them. You would need some kind of machinery in your factory to automatically respond. Or individuals need education about what the system is and how to use it when you hear an alert. So the light blue boxes are the alert delivery and use part. The USGS does the blue boxes – alert generation. But the light boxes – the distribution of the alerts and the use of the alerts – is outside of our capabilities. We depend on companies, on individuals, on end users, to fulfill that need of the system and that part of the system. Quick look at the stations. You’ve already seen a map similar to this one. This shows the current distribution of stations that are contributing. Our plan called for 1,675 stations to support the needs of this system. The reason for that number is we looked at the area of those three states, and we did some research about the optimal spacing for stations. Turns out that the optimal spacing is about 20 kilometers apart, which is about 12 miles. Now, in the urban areas, where there are lots of people and lots of risk, we settled for a 10-kilometer spacing. That’s to make sure that the alerts were going to be fast, and we could lose a few stations – because not all equipment stays up all the time – and still have that optimal spacing. Outside of the urban areas, but where there were still earthquake sources and some people, we used the 20-kilometer spacing. And out in the boonies, where there aren’t very many people or earthquake sources, we used 40-kilometer spacing. So when we mapped that all out, we came up with that rather strange number – 1,675. We had some existing stations. We’ve been upgrading stations. We’ve been adding new stations. So our current station count is 859. Or at least the last time I counted. And we continue to work on that number. That means we’ve still got about 800 more to go. And even some of the stations that are currently contributing are not as fast as they could be and need to be upgraded. We’re also planning to add a different kind of sensor – not a seismic sensor that measures ground velocity or ground acceleration, but GPS stations – high-rate GPS that can measure displacements of the ground. And those are extremely useful for the biggest earthquakes – for characterizing the really huge earthquakes like the ones off the Cascadia subduction zone, off the Pacific Northwest coast, or huge San Andreas events. Our plans call for – well, we actually have funding for 250 additional stations in the next year and a half. And that will bring California up to about a 74% completion and the Pacific Northwest would be about halfway there. We knew that it was going to be a heavy lift to get in all the stations, and so we concentrated on the urban areas. And so the areas in the red circles are near or at target density. So major metropolitan areas – L.A., the bay, and Seattle/Tacoma. Well, this is a look at the interior of the system. This is a very complicated slide, but the test will be based on this one. [laughter] I don’t expect you to see this. This is sort of a shock-and-awe slide. Just to show you that there’s a lot of complicated stuff going on and a lot of redundancy built into the system. Data flows from the bottom toward the top. The ground motion data feeds into a variety of computers at the various station – or, the various network processing centers. And you can see them named across the bottom – Pasadena, Menlo Park, Berkeley, and Seattle. They examine the waveforms – the wiggles of the ground motions, derive information from those – what we call parameters. When did it start to move? How did it start to move? What’s the frequency content of that and other things like that? And that data is shared across all of the centers. So that means every computer in every center has access to all the data for the entire West Coast. And that’s a redundant feature so that we can lose any – or, any center and still stay up and generate our alerts. Once an event is declared, that information pops up to the top level. And that’s the decision-making level where alerts would be generated. Those are in a secure cybersecurity environment, and they would have public-facing servers. In this case, “public-facing” means the alerts are going to the end users, but not individuals, but to distribution mechanisms or to companies. Well, let’s talk about earthquake physics. You’ve probably heard somewhere in your academic career that earthquakes generate a variety of waves. There are more than just these two, but let’s talk about P waves – primary waves – and S waves – secondary waves. When an earthquake occurs, there is slip on a fault. And that slip on the fault causes elastic waves to radiate out from where that slip occurs. And the P wave moves fastest, but it’s fairly low in amplitude, so it doesn’t do, frequently, very much damage. The following S wave is where most of the damage occurs. And there’s a seismogram in green in the lower right-hand part of this slide. And you can see the arrival of the P wave, that happens first. Time is going from left to right. So the P wave arrives, and then some time passes, and then the S wave arrives later. And you see the S wave is much bigger. The peak ground acceleration – PGA – the heaviest shaking – may follow behind the S wave, or it may be at the same time as the S wave. It depends on the characteristics of the earthquake. So that S-minus-P interval varies according to your distance away from the earthquake. The closer you are, the smaller that interval will be. When we measure how much time you have before the strong shaking arrives, we’re talking about the arrival of the S wave. And it may be that you have a little bit more time if the strongest shaking actually lags a bit. The block diagram of the mountain is to show you that an earthquake does not happen at the surface. An earthquake happens at depth. It starts at depth, and so the epicenter is the place on the ground above where it started, but the hypocenter is the place at depth where the fracture actually starts and then starts to rip the fault like a zipper from that point either upwards, sideways, or both of those. And then those waves radiate out like the ripples in a pond when you throw a rock in. [coughs] Excuse me. We also have to consider the fact that there are different kinds of earthquakes in different tectonic situations. So if you look at the picture on the right, there are three red boxes – so these are the three primary types of earthquakes that we have to detect. There are shallow crustal earthquakes, like the diagram we showed in the last slide. But in addition to those, if you are in a subduction zone environment, like the Cascadia subduction zone in the Pacific Northwest, you can also have subduction zone earthquakes at the interface between the North American continent and an oceanic plate that’s being shoved down underneath the continent. So slip on that interface between those two tectonic plates can cause the largest earthquakes that we know of. And in addition to that, the oceanic plate that’s being shoved back down into the Earth’s asthenosphere will twist and bend and crack and cause very large earthquakes as well. The two big earthquakes that just happened in Mexico were of that last type – those deep slab earthquakes. And if anybody remembers the Nisqually earthquake in Seattle, that was also a deep slab earthquake. Another thing that you need to understand about earthquakes is they don’t happen at a point. Epicenter, hypocenter – those are just where the earthquake starts. But large earthquakes rupture long faults. The bigger the earthquake is, the longer the rupture is. This diagram is a animation running 12 times real time of a hypothetical rupture on the southern San Andreas Fault, generating a magnitude 7.8 earthquake. So big earthquakes are not points. The magnitude of the earthquake is proportional to the fault rupture. That fault rupture takes time, which means that, when an earthquake happens, the bigger it is, the longer it takes to happen. And so it’s a challenge to estimate its magnitude. It’s not done until it’s done. You don’t know what the magnitude is until it has fully developed and basically stopped. So you can’t predict what the rupture length is going to be. So the system must map the rupture in real time, as it occurs, and continue to update the estimate of magnitude, and therefore, the estimate of effects, as the earthquake continues to evolve. And, in this particular scenario, that earthquake takes a minute and a half to happen. You also have to consider that it’s not the distance to the epicenter that will get you. It’s the distance to where the fault is rupturing. So, in this case, if we were worried about L.A. – and I know this is northern California, you’re not worried about L.A. [laughter] In fact, you’re hoping for this earthquake to happen. [laughter] But, in this scenario, if you try to estimate the impacts by measuring the distance to the epicenter, you’re going to way underestimate the impacts. Because that rupture is headed toward L.A., and you have to measure the distance to the fault. Now, the detection of earthquakes is a very touchy thing. It’s difficult to do. And there’s a tradeoff between speed and accuracy. How much data are you going to wait around for to get a better answer? Well, it’s early warning. We need to get answers out quickly. So we need to use the minimum amount of information to make those alerts as fast as possible. And this illustrates some of the challenges of doing that. These are ground motion records coming from real seismograms in the field. The one in the center top is just background noise. There’s no earthquake in there. The Earth is noisy. There’s traffic. There’s water flowing. There’s wind blowing through the trees. There’s the waves beating on the coastline. There’s all sorts of sources of noise that could fool algorithms – that is, scientific ideas of how to detect earthquakes. They can fool them into thinking there’s an earthquake when there really isn’t one. That’s one of the hardest parts and why so much science goes into the detection of these events. Also, the instruments themselves do weird things. So, over on the right-hand side, you see a calibration pulse. That’s an instrument going through a regular test and slamming its weight against the stops, causing a seismogram that looks, at first blush, like a huge earthquake. And so we have to control for that. And the diagram in the center bottom is a teleseism. Teleseism is a large distant earthquake. So, for example, there was just a magnitude 7.9 in Alaska. Well, those waves swept across our network from north to south. And our system has trouble discriminating between teleseisms and local earthquakes. But the good news is, just today, we propagated forward software that has a new filter in it that does an excellent job of discriminating between teleseisms and local earthquakes. And it’s really going to reduce that problem significantly. Okay, so now it’s your turn to be an earthquake early warning system. Here’s the beginning of an earthquake in seismograms at the bottom of the screen. That’s what you get. We trigger on the first four stations in order to make sure it’s fast. So that’s the information you get to decide, is this noise, or is this an earthquake? And if it’s an earthquake, where is it, and how big is it? So now’s your time. Take your guess. Earthquake or noise? [laughter] Earthquake. This is the south Napa earthquake. You can see the P waves kind of going at a slant up and to the right as you go upward, and then you can see the S wave following behind. And the fact that – these are ordered by distance away from the event, and so you can see that the earthquake P wave and S wave is getting farther and farther apart as you get farther and farther away. This is called a record section. Well, so how do we make sure this stuff works and we don’t get fooled by all this noise and the teleseisms? Well, we have a testing and certification platform. It’s a library of historic earthquakes and historic noise. So we’ve got more than and we replay them every time we want to test a change to the algorithms. And we will not make a change to any part of the system unless that change runs through this gauntlet of 40 real earthquakes and 70 noise events, including teleseisms. And then we always test to say, okay, is this change better or worse? And if it’s worse, of course, we don’t make the change. Our ability to test, though, is limited because we do not have records of the biggest earthquakes that we’re trying to protect against. We don’t have an example of a real magnitude 7-1/2 in the bay. We don’t have a real 9 off the coast of the Pacific Northwest to use in our test suite. So that is one of the other limitations of our ability to test how the system will perform in the biggest earthquakes. Well, once we figure out that there’s an earthquake, where it is, and how big it is, we have to then predict the impacts. How hard is it going to shake, and where? And to do this, we use ground motion prediction equations. But ground motion prediction equations are approximations. There are factors that change how hard you’re going to shake. Local site geology. The path of the waves moving through the crust to get to you. The characteristics of the earthquake itself and whether it’s breaking towards you or away from you, or are you off to the side. So there are a number of things that go into variations in how you are going to shake for a particular earthquake. This diagram shows the reports of what people say they felt in the south Napa earthquake. The distance is along the bottom, and the intensity is the scale going up the vertical. The intensity here is modified Mercalli intensity – MMI. It’s a number that we use for characterizing how hard the shaking feels. It’s not magnitude. It’s a different kind of number. And to try make sure people understand that, we use Roman numerals to describe MMI intensity. Intensity II is about the point where you start to feel it. And you start to get damage at about V or so in intensity. But that brace there, in red, shows you the range of intensities that were reported by people 30 kilometers away from the south Napa earthquake. And some people reported it as a II, and some people reported it as a VI. And so there’s a lot of variability in the shaking by distance. And so that’s another limitation of what we can say about what you’re about to experience. The diagram on the left is the kind of information that would come out of the early warning system. Those eight-sided polygons are the various intensity levels. The outermost polygon is MMI II. The next one is MMI III. And so forth. And so we can do a prediction of the expected ground shaking, but, as this slide illustrates, that is going to be an approximation. The system is going to produce two basic products – one for people and one for things. The alert to people, right now, is going to be released if the earthquake is in our reporting area – the three states of the West Coast – or maybe it’ll be restricted to the metropolitan areas because that’s where we can do the best job. [clears throat] Excuse me. So if it’s in the region, if the magnitude is greater than 4-1/2, then we will report the polygon of MMI II – that is the polygon in which people should feel the earthquake. And that would be the basis for the alert area. Anybody within that zone would receive an alert. And the little moving thing there shows the evolution of a hypothetical event and how those MMI estimates would change over time as the fault ruptures further and the magnitude grows. The other product is for institutional users – those who are more sophisticated, are going to do their own calculation, their own estimate, and their own decision about what actions to take – whether to stop the train, whether to stop the factory process. So that will have a lower release threshold – probably magnitude 3-1/2. And we’ll send them more information. We’ll send them the location of the earthquake, either as a line or a point. We’ll send the magnitude. And we’ll send the whole suite of MMI contours. And we can also send them a map of – a grid describing the expected impacts as well. Well, a little bit more about the earthquake physics. This one is little bit difficult to grasp, but I’ll do my best to explain it to you. The bottom line here is that the warning time that you’re going to get depends on the threshold that you set for being notified. So let’s imagine that you had a factory. And you want to protect your factory from shaking. And you know that your factory will be damaged at an intensity level of V. Okay? So you got that scenario? I’ve got a factory. I’m worried about intensity V because I think that’ll damage my factory. But I also need 10 seconds to shut it down. So you’ve got a decision to make about the right balance of time and threshold for triggering. Now, if you say, I’m going to be really conservative, and I want to shut down – make sure I have time to shut down. I am going to set my threshold fairly low at MMI II. So, in this picture, you can see that this is the first alert from this earthquake, and it comes out 5.2 seconds after the earthquake began. And it reports that the earthquake has reached a magnitude of 5-1/2. Now, in this hypothetical, the red star is us. We’re about Do we shut down our factory? Well, I said we wanted to be conservative. Why would we do that? Because the damage sustained is going to be worse – much worse than the cost of shutting down. That’s a cost-benefit that every sophisticated user is going to have to make for themselves. What’s the outcome of choosing a low threshold? Middle bullet – low threshold means you get more alert time, but you’ll also get more false alerts. If this earthquake stops here and is only a 5-1/2, you will shut down, and you will not experience the damaging shaking. And that’s a tradeoff that you’re going to have to think about and make. If you set the threshold higher, you will have fewer false alerts, but you’ll also have less time. Let’s look at some examples. What if I set my threshold – instead of an MMI II, what if I set it at III? I’ll wait until there’s a little bit heavier shaking estimated to reach my location. Well, then I get 22 seconds – I should back up – in the first one, I get 25 seconds of warning. This one, I’ve used 3 seconds, and the earthquake has grown, and the expected shaking at my location has gone up. What if I picked MMI IV as my trigger threshold? Well, now I’ve only got 7 seconds – or, 17 seconds to react. But remember, I said I’d get damaged at V, so what if I just wait until the earthquake grows and the estimate of the impact at my location goes all the way up to V – the thing I’m really worried about? In that case … [beeping sounds] [laughter] He really wants to talk to me. [chuckles] In that case, we’ve lost our early warning. Because it took too long for the earthquake to grow to a point where we’re certain that the shaking is going to be heavy at our location. So every sophisticated or technical user is going to have to make that kind of cost-benefit decision. Well, what about the people? What are they going to get? Everybody – since we’re playing with cell phones here – everybody expects to get the alert on their cell phone. And we have done the work to define what a cell phone alert will look like. It will have a distinctive sound. But that sounds is to be determined. We had hoped to use the same sound they use in Japan, and we asked the Japanese if we could use that, and they said, mmmm, no. [laughter] They wanted to reserve that sound just for their system. We thought it would be great to internationalize it. They want to keep it a Japanese sound. Okay, so we’re back to the drawing board. But luckily, our partners in the California Office of Emergency Services are going to work on that problem. They’re going to go with sound engineers and social scientists to develop a sound that will be distinctive for earthquake early warning and will meet all sorts of different criteria for not being so annoying that people shut it off or not being so melodic that it just sounds like another ring tone. So the important part, though, in that distinctive sound, is it’s going to be the basis for training people. What we would really like is a Pavlovian response to that sound, that you just jump under your desk without thinking about it. And then, we don’t want people to have to pull out their phone, wake it up, read a message – by then, they’ve consumed a lot of the warning time. And so it will use voice to say, earthquake, earthquake, expect shaking soon. Drop, cover, hold on. Protect yourself now. So the social science tells us that – don’t want to give you any fancy information. Oh, there’s an earthquake 60 miles away that’s going to give you MMI VII. Yeah, that is not what you need. You need clear instructions. [chuckles] What to do. You can worry about the details later. So this is what an earthquake early warning would look like. And, not only would this be what your phone did, but it would be what would appear on a television or in a radio broadcast or on a sign driving down the freeway. Whatever the modality of the alert is, we want it to be consistent so that people recognize it immediately. Well, there’s a problem with this. I’ve just described what a phone would respond like, but the bad news is, that can’t be done today. The best way to send alerts through cell phones is using IPAWS – the Integrated Public Alert and Warning System. That is the government system used for emergency alerts for weather, for terrorist attacks, for Amber alerts. So you’re probably familiar with that. You’ve probably received them on your cell phone. The piece of it that actually sends it to your cell phone is a different system that’s inside the cell carriers’ systems. And it’s called WEA – Wireless Emergency Alert. So IPAWS gets it and then passes it off to WEA, and WEA is operated by Verizon, AT&amp;T, T-Mobile, and Sprint. So whichever carrier you have – if you’ve got one of the off-brand ones, it still works. That system was designed for the things that I just described, like Amber alerts and weather. It was not designed for speed. Earthquake early warning needs to be really fast. And WEA is not fast enough for earthquake alerts. So we’re working with the cell carriers to remedy that situation, but that solution is a ways away. Let’s take a look at some of the limitations of using a cell phone for receiving alerts. So obviously, we can’t reach everybody. There are areas where cell phone coverage is not available. So those people would be left out. Well, what about sending text messages? I get text messages all the time. Too slow. Doesn’t work. Well, what about an app? I’ve got the Twitter app. I’ve got the Facebook app. This thing’s tweeting at me all the time, telling me stuff that I didn’t even ask for sometimes. So why not that? Again, not fast enough. Those technologies – if we tried to notify a million people, would take a long time. In fact, we don’t even know exactly how long. We can’t get that information. I think it may not just be available, and I’ll talk about that a little bit more in a sec. And then, the WEA system that I just described is cell broadcast. But, again, it is not currently fast enough, but we are actively working on making it faster. We’re working with both FEMA directly and with the organization that makes standards for the cell carriers. So we’re hopeful that it can be fast enough someday, but it is not fast enough now. So here’s a – kind of a complex, busy grid showing the possible alert technologies for mass alerting. And I didn’t mention the internet. You get a lot of information over the internet, but the internet is very fragile. In strong shaking, it’s not likely to survive. And so it may not be the best way to try to send emergency alerts. I’ve already described IPAWS and the WEA system, the fact that it’s too slow, but we’re trying to speed it up. In addition to WEA, IPAWS will also distribute through EAS – the Emergency Alert System – to television and radio. That’s even slower than WEA. So that’ll be a heavy lift to make that work. That organization of cell carriers that I talked about, we are working on implementing a different technology in the cell systems called ETWS – the Earthquake and Tsunami Warning System. That’s what they use in Japan. To get it operational in the United States, they estimate will take three to seven years. So that’ll happen, but, as you can see, it’s a ways away. Sorry – I talked about push notifications already. But there is another one coming that’s very promising, and that is using broadcast. Radio, television – broadcasts are all over the place. And it turns out that you can put data into those broadcast streams. And so the Cal OES – the California Office of Emergency Services – is prototyping putting the ShakeAlert alerts in the broadcast of public television stations. Now, you will not get the alert if you’re watching the station. It’s digital data in the signal. And so, the part that we’re missing is a receiver. You need a radio to receive that digital data and turn it into an alert. That’s, by the way, what happens in your car radio when it says what the – what the radio station you’re listening to is or what the song name is, that’s digital data inside the radio stream. But that, too, is going to take a little bit of time to completely develop, and you’ll need a purpose-built receiver. So back to the 2018 limited public rollout. What can we expect? On the project side, the alert generation side, the part that USGS has direct control over, we’ll install additional stations. We’ll continue to improve the software of the system to make it more reliable. We’ll build those public-facing secure servers. We’ll keep doing research and development to speed it up. And we’ll also develop a plan for CEO – Communication, Education, and Outreach – so that people will understand the system and know what to do when they hear an alert. On the public alerting side – always remember that word “limited.” We’ll do some experimental apps in order to actually measure how much an app will help. At what point does an app saturate? We know that, in the case of Japan, they have an app that has, they claim, 5 million users, but they do not guarantee you will receive the alert in a timely fashion. If you want that, you have to pay extra. [laughter] Okay? And that limits the number of users so that you can probably deliver it to maybe a few tens of thousands of people fast through an app. But beyond that, it probably slows down to the point where it is not useful. We’ll continue with our work with FEMA and the cell carriers to speed up IPAWS and WEA. But, again, that’s probably a few years out. And I’ve just described the DataCasting model for getting that out with the limitation that you would need to buy a purpose-built device. In addition to these, we are doing some live pilots. Now, this is a big list. I’m not going to run down through the entire list. But we are working with a number of organizations who are doing real, live implementations. Now, not all of these implementations are guaranteed to be completed in 2018. But they will ultimately be completed. And they’re doing very interesting things, like notifying children in schools through a prototype in the L.A. school district. And closing water valves in the Pacific Northwest. We’re working with companies up there doing that. And so there are a number of these pilot projects that will come to market in 2018, and others, probably, in later years that will lead the way in showing how the technology can be used and how to do those kinds of implementations. And in fact, we expect to be creating a brand-new earthquake early warning industry that doesn’t currently exist. In fact, one of these companies – Earthquake – or, yeah, Early Warning Labs is a start-up specifically for using ShakeAlert data and making products to protect its clients. And we hope that there will be many other companies doing that sort of thing. We do have to do some education and outreach. We’ve got a mechanism for doing that. We have a multi-state committee that is working on the public education part – what materials to use and how to message this for greatest effective use. And it actually also includes British Columbia. We’re talking to the Canadians to coordinate our efforts because they’re interested in developing earthquake early warning as well. So last slide. Summary. Full-time alerting is limited – or, full-scale alerting is limited by several factors, as I’ve been describing – funding to complete the system and operate it is not yet in hand. The sensor network is incomplete. Mass notification technologies that exist today are too slow. And people need to be educated. But, despite all these limitations, ShakeAlert will begin limited public operations in 2018, using those paths that work, that are fast enough, and through the pilots that I’ve described. I think that’s it. [laughter] [ Applause ] All right. We’re going to do a opportunity for questions, but we do ask that you either use the mic on the stand, or it’s over there on that [inaudible]. Or Jim Straus, UC-Berkeley, one of our partners, will come to you with a mic if you raise your hand. - And we would like to thank some of the pilots who are in the audience tonight. For example, PG&amp;E is in representation here, so they’re helping to keep your Bay Area safe in the case of earthquakes. So, first question. - In the case of Japan, can you give us, as an example, in the Fukushima earthquake, what was in place, and what worked? - The Japanese system has been operational for the public since 2007. In the Tohoku earthquake that damaged the Fukushima reactor, the system worked. It was an offshore event. It – I’m not sure I’ll remember all of these, but it believe it triggered hit land where they had sensors. An alert was issued, and thousands of people were warned. That’s the good news. The bad news is that the Japanese system only reports an epicenter and does not account for the length of the fault that I described. And so that fault ruptured southward in a magnitude 9. And so they actually under-alerted. They should have alerted the Tokyo area because the rupture was moving in that direction, but they did not do so. So it was mostly a success, but it had some limitations. - One … - They slowed down trains? - Yes. Yeah, in fact, Japanese work on early warning was first motivated by the opening of the bullet train – the Shinkansen. That happened in 1990 – no, 1964, when the Shinkansen opened. And so the Shinkansen has always had earthquake early warning built into it. It’s gone through several iterations, but the public warning system followed much later, in part motivated by the very devastating Kobe earthquake that killed 6,400 people. - Okay, we have this question up front. - Well mass notification – why don’t we think about putting loud sirens on cell phone poles in populated areas? Put enough of them around, you wouldn’t have to worry about opening your cell phone or listening to the radio or TV. You’d hear the warning signal. - That’s correct. And that’s what they do in Mexico City. They have a network of over But that siren system was not primarily built for earthquake early warning. It was built for crime and other security measures. So earthquake early warning is just part of that. It’s very expensive to do so. And we are exploring the use of sirens. The city of San Francisco has more than 100 sirens. They’re too slow. [laughter] Yeah. - Two things. One, I’ve seen a lot of very compact seismic stations. What’s the problem with the environmental impact statements? They do not seem like they’re facilities that should really get bogged down in a lot of problems with impact statements. - Well, I agree with you. - And secondly … - Can I answer the first one, then you can ask the second one? - Sure. - We are told, when we try to push back against those requirements as well, that if you put a hole in the ground the size of a pencil, you need this kind of impact report. - Yeah, well, I mean, it – that – in impact statements, there are often very long processes of comment and things like that. And this does not seem to be the sort of projects that would trigger those delays and lawsuits and all the other stuff that tangles up impact statements. They would look like they’d be pretty pro forma. - Well, I agree with you. And we’re trying to get to that point. Now, the impact studies we need to do are not the same as you would do for a housing development or a … - Sure, yeah. - … or a dam, but they’re still onerous and can cost us thousands of dollars and months of work. - The second is, how have Japan and Chile managed to confront some of these problems like the speed of dissemination? I mean, they’re working with much the same cell phone technology and broadband carriers and all the rest. How is that they have skinned this cat? - Well, I don’t think I would put Chile on the list of active earthquake early warning systems. There are many countries that do – Mexico, China now has one, and, as you mentioned, Japan. Korea is building one. There are portions of India where they’re being built. Italy, Romania, Turkey – all building or have built systems. In some cases, they’re not public. The only country that I’m aware of that has effective alerting through cell phones is Japan. And they implemented that ETWS – Earthquake and Tsunami Warning System – the system that we’re trying to bring our system up to that spec, but again, it would take three to seven years to do so. Yeah. - They’re using different cell phone technology or … - Yes. - … what hurdle have they surmounted that you have not surmounted? - Well, it’s just a different way of designing a cell phone system and the behavior of all the physical components and the protocols that are used for sending alert messages. I could go into excruciating detail about how that works, but it even bores me, so … [laughter] - Okay, another question from the back, then we’ll take one at the front. - Okay, thanks for a great presentation. The question I have is, it seems that your data is well-suited for machine learning or deep learning. - Mm-hmm. - Have you guys – I’m sure you started, you had your requirements before it was very popular and we had the computing power for it. But have you been looking into that? - There have been various papers in seismological journals about that. And there is some current activity. In fact, there is a – I don’t know if he’s a postdoc or a grad student at Caltech who is going to take a look at whether that will help us or not. - Yeah, I guess the – having the speed is probably one of the critical things. - Yeah. - And how long do the Amber alerts take to get out, just so we have a ballpark? - Well, that’s part of the frustration is we’re not really sure. The “we” being me. I mean, I’ve been on phone calls with the cell carriers for a year and a half, and I’ve been asking repeatedly, how long does this stuff take? And either they don’t want to say or they don’t know. And so that’s a bit of a challenge. And in trying to get past that problem, we are doing a test by developing an app in the – I’m sorry, that’s the app thing. We’re trying to figure out how fast an app can go. But in addition to trying to figure out how fast WEA can go, we’re talking with IPAWS about doing a through-and-through test – an experiment that would include citizen scientists with their phones that would measure when the alert arrived so that we can directly measure the speed of the system. Now, we expect that it will vary by carrier, vary by region. But if I had to guess what the outcome of that was going to be, I would say it would say it would be from tens of seconds to minutes. - Which would be too long. Okay, thank you. - And you may have been in a situation where you were in a room when an Amber alert was broadcast. A lot of phones go off, but sometimes phones go off a minute later. Or two minutes later. Yeah. - Okay, we have a question up front. - Yes. Hi. Thank you for your lecture. I just wanted some clarification about the S wave and the P waves. - Mm-hmm. - Now, it looks, up there, that they’re created some distance from the epicenter. The S is created first, and then the P wave after? - Okay, both waves are generated simultaneously at the rupture of the fault. This picture is a few seconds afterwards when those waves have had some time to start to move outward from the earthquake rupture. And so this is meant to illustrate that the P wave is now getting out ahead of the S wave. You know, I don’t know what the scale of this diagram is, so I can’t put numbers on it, really. But those two waves are generated simultaneously. - The same place at the same time. - Yeah. Sometimes the analogy I use is, imagine a race between a Porsche and a Volkswagen Bug. At the starting line, they’re dead even. [laughter] If they go, a block later, the Porsche is out ahead of the Bug. After two blocks, it’s farther out ahead. After three blocks, it’s farther out ahead. And so the gap between the two – the faster and the slower – grows as time passes. - But the Bug is going to cause more damage. - But the Bug is going – yeah, maybe I should make it a truck or something like that. - So maybe you can answer – you mentioned PG&amp;E is here, so I was talking to – somebody was telling me that PG&amp;E actually has one of the largest Wi-Fi systems in existence, and that is through the network of smart meters and that there was negotiations – the person I was talking to was about trying to get emergency services access to that network. Is that something that is part of this? Do you know anything about it? - I have heard something about that, but I don’t know if I want to put somebody on the spot or not. Stu, you want to say anything? - You want to be on the spot, Stu? - I want to be on the spot. [laughter] [inaudible] so, yes, PG&amp;E – is this on? - It’s on. You just have to talk closer. - We do have an network of smart meters and … - [audience comments] - More closer. Is that – there we go. - Yes. - So we do have a network of smart meters. Most of them have – see them in your homes – outside your homes. And we’re constantly working on developing those smart meters, making them even smarter. There’s going to be a new generation where we’re putting accelerometers in the meters so they will act like miniature seismic stations in everybody’s house. So we’ll get even denser recordings of strong ground motion, and we’ll have a better idea of the variations in ground motion in communities around the Bay Area. So that’s to achieve this vision that we had many years ago – what we called micro-zonation, where we can really see how ground motion varies with location in the area. And we are an active partner with the USGS in ShakeAlert and earthquake early warning, and we’re, you know, interested right now in different ways to implement. So we are one of the guinea pigs. And right now, our primary focus is going to be on life safety, in terms of getting that message out to our employees. First job is to survive the earthquake, so then we can get out and start to do the restoration work following the earthquake. We’re investigating different ways to push that message out and train people to know what to do, as Doug was saying. So it’s almost a Pavlovian response. You hear the alarm, you duck and cover. You don’t think about what to do next. - Is that system currently a two-way communication system? Or is it one-way only? - It’s one-way. - Okay. - It’s one-way. But give us time. [laughter] Give us time. Thank you. - Thanks, Stu. - Thanks, Stu. - So recently, people in Hawaii were falsely alerted that a nuclear missile was about to hit them. - You had to bring that up, huh? [laughter] - So how would you prevent that, and how would you prevent hackers from messing with your system? - Okay, so two parts for that question. The first is, the error was human error. Somebody poked the wrong button the human console to generate that alert. Our system is fully automated. That doesn’t mean it’s going to be perfect. No system is. So there can be false alerts for many of the reasons that I’ve already described, so don’t expect the system to be perfect. So we’ll do the best we can to reduce false alerts, to keep them to a minimum, but we can’t guarantee that the system will not generate false alerts that then pass through and through to end users through the WEA system. So every technology has a dark side [chuckles] and a – and a positive side. Your second – the second piece of your question was about … - Hackers. - Hackers. - Hackers, yes. So we have instituted a lot of cybersecurity controls on our system. And in fact, we’re going through a very excruciating process right now to satisfy the federal government requirements for cybersecurity. There was a red box around that alert layer in the slide that I showed you. That red box represents that secure environment. And, you know, messages are encrypted. They’re secured, and we’re doing everything we can to ensure that the system can’t be hacked. The – if you all want to go out to an individual sensor and start jumping up and down simultaneously [laughter], we might get fooled. I don’t know. You might want to give that a shot. [laughter] - Okay. I think we’ll have one more question. - Oh, I thought you had one over there. This occurred to me as I was sitting over there. In terms of training for using whatever the system turns out to be, I have been, just coincidentally, in the last couple weeks, I’ve been in half a dozen different meetings where the first thing you do when you go in the door is, cell phone, off. And I’m just wondering if that – somewhere in the training process is where that potential weakness in the whole system might be covered. - Mm-hmm. That’s an interesting observation. I can think of a few possibilities. One is, that if buildings are wired into the system, and through PA systems, that would be a second way to deliver the alert. And by the way, in psychology, people really do want to verify that an alert is true before they take an action. You’ve probably been in a room where a fire alarm has gone off. And what do people do? Generally nothing. They look around. Is there any smoke? Are other people leaving? I’m not going to be the first one to go. That’s a real barrier to effective response to alerts. And so, you know, if there’s a building alert that goes off, and your cell phone goes off, that would be a verification that might cause people to react positively. As far as the issue of shutting off your cell phone, the battery dies – there are many things that really can’t be remedied in that kind of situation. People will probably be enabled to opt out. There’s actually some discussion about that. And the behavior of the phone is part of what’s being negotiated. It could be that, even if you turn your ringer off, for example, a severe alert might still activate your phone. There’s always that balance between privacy, annoyance, and getting the alert to folks. And so, you know, that’s actually going to a discussion that will happen in the FCC – the Federal Communications Commission – about the behavior of cell phones related to alerts. - As the daughter of an emergency manager, I always leave for the fire alarm. I don’t know about the rest of you all, but I always leave. I think – did we have one more question? One more question? - Many years ago, I managed an exploratory research program at the Electric Power Research Institute. And on behalf of our nuclear power people who worry about stuff like this, we gave a little bit of money to a team that was – Berkeley, and I think Stanford – Berkeley and somebody. And they thought they had a glimmer of an idea for earthquake prediction. - Mm-hmm. - Did that ever go anyplace? - There have been lots of people with glimmers of ideas about earthquake prediction. [laughter] And to my knowledge, none of them have panned out. And so I think I’ll leave it at that. I’m in the skeptics camp. I don’t think earthquake prediction will ever be actionable. - You mentioned the Mexico system and that it’s siren-based. Are they using any other message delivery? And didn’t – and are you coordinating with the Mexican program at all? - Yes. We talk to the Mexicans frequently. We sent a team down to evaluate how the system performed and how people responded after the last couple of quakes that they had. Their system is different in many ways from ours. In some ways, their problem is easier. Their system was originally built – they went live in 1980 – no, 1992 as a response to the And their system originally was, the earthquakes we know are going to happen off the coast. And our population is in Mexico City. So if we put sensors along the coast, we can protect Mexico City. That was a very simple and effective use of earthquake early warning. Since then, they have spread southward into Oaxaca and have a more general system. But our problem is quite different from that. Our problem is, our earthquakes can happen anywhere, and our people are everywhere. And so we have to build a system that’s capable of solving that problem. And so, for that reason, it’s different in many ways from the Mexican system. - Okay, one last question, and then we’ll … - This is – this has to do with funding. I thought the ShakeAlert wasn’t getting any funding, but then there was some California congressmen that were supporting it. What’s going on with – Kevin McCarthy is in California, and he’s a great Trump buddy. Is there – are we – what’s the status of support? - Well, I’m not sure I want to get into politics. - No, you don’t have to. I’ll do that. [laughter] - I was looking for my slide that shows the funding picture. And I’m not sure I can find it in this particular representation. I think that’s it. Go. Okay. So there it is. And it’s not that we don’t have any funding, but it’s less than we need. You can see that it started out quite modest in 2013. It’s grown, but we’ve now plateaued at 10.5 per year. And, again, that’s over and against 38 million originally estimated, and we know that number is now bigger. Plus, just to operate it every year, our original estimate was 16, and we know that number is really going to be bigger once the system is fully built out. So as far as the funding picture goes, we’re sort of plateaued. The city – or, the state of California has put in that$10 million one-time.

They may make an additional request.
And they’re actually looking at

ways to fund the system in an
ongoing way. In fact, watch the news.

There will probably be
some news in February

at the state level.

So, as I said, this project
is a partnership – is a collaborative

effort among many players.
So federal government,

state government,
may all help to fund this thing.

But we’re not there yet.

- [inaudible] the numbers,
it comes out to 37.6 million.

Is that – that’s not
per year, apparently.

- Which one are you looking at?
- [inaudible] million, the California 10,

there’s 1 million in Oregon,
and the 10 million from Moore,

and U.S., 5.6.

you get 37 million.
- Yeah, but there’s much more

going on – much more being paid for
by those dollars, including the research

and the operational costs in those
multiple years across the – yeah.

- Okay, let’s thank our speaker.
Thank you, Doug.

[ Applause ]

[background conversations]

[ Silence ]