New Technology Listens for Early Earthquake Warnings

Pete Miller [00:35]: Welcome to Predict and Prevent. I’m your host, Pete Miller. And today our guest is Itamar Zabari, co-founder, CEO, and CTO of Astroteq. Itamar brings a rare combination of deep software engineering expertise — honed in the Israeli Defense Force and the space industry — and a willingness to pursue what most scientists have long considered impossible.

Astroteq is using data on cosmic radiation from deep space, combined with satellite imagery and AI-driven machine learning, to forecast earthquakes up to 25 days before they strike. To put that in perspective, some governments have spent hundreds of millions of dollars building a system that delivers just 12 seconds of warning.

In this episode, Itamar breaks down the science behind this extraordinary capability, explains what governments, utilities, and industries can actually do with days rather than seconds of advance notice. Itamar says that earthquakes can be predicted because the earth actually does warn us, but it’s just whispering and you need to learn how to listen.

This is a conversation you won’t want to miss.

Pete Miller [01:55]: Itamar, you’re the CEO and CTO of a company that uses cosmic rays from deep space to predict earthquakes, which honestly sounds like science fiction. So, tell us a little bit about your background and how you ended up leading such a very interesting and somewhat unconventional venture.

Itamar Zabari [02:13]: So, I come from a technological background. Ever since I was very young, I liked tough problems to solve. I served in the Israeli Defense Force in their software division, and we wrote very, very complex software. And then later on in life, I kind of moved from technology into business. So, for me, it was always more about how can technology impact the day to day. It was always about that; it was never technology standing by itself. That’s kind of the rule that guided me throughout life. Like even in the Israeli army, I led a couple of revolutions in the way that they’re designing software. So, for me, it was always about that.

In the past 10 years, I was working in the space industry, participated in a venture that launched three nanosatellites in space. I was in charge of writing the whole software for both the satellite, ground devices, apps, and pretty much everything in between, and we proved that nanosatellites can be used for voice over IP. Like I had an app that we called a “chatellite” and we ran voice calls and we sent messages through them.

And then while I was working on that, my wife did her Ph.D. and she came to me, she said, listen, we found something potentially good. I’m like, what is it? She’s like, potentially we can forecast earthquakes up to 14 days in advance using cosmic radiation.

I’m like, wait, wait, what? Like, it’s crazy, right? It’s like, and you know, she showed me that and the results like it was general publication, like general research, it was global, it was it wasn’t like localized or anything. But bottom line was, potentially that can, that can happen. And it seems to be very, with very high certainty, it was sigma five.

So it was like, okay, Higgs Boson for sigma five or four, if I’m not mistaken, that’s real, then maybe that’s real. So it’s might worth my time, right? Cause it’s, it sounds revolutionary. And I kind of started diving into, into the data, into the patterns. And I saw some patterns there. I was like, I think it’s, I think it’s true. And then I thought we should set up a company and start pursuing it because we tried to do that through the academia. And it was very hard to get budgets for it. And it was very hard to lead anything because it’s very kind of a publication focused. Let’s not save the world focus, but publication focus: start and publish, move on, right?

Which has its merit because without it, we wouldn’t have existed, but it’s very hard to take it beyond that. And so we started with also Astroteq, back in the day it was called AstroTectonic. And yeah, I mean, for me, it was always about the changing people’s lives. If we can forecast days like earthquakes, days in advance, and we can save people’s lives. That was the bottom line, right?

And then while we were working on that, there was the massive earthquake in Turkey, 7.8, two years ago, 58,000 people died there, if I’m not mistaken. And we were trying to fundraise and the seismologists are blocking us everywhere. Investors are excited, they’re going to seismologists, seismologists telling them it’s impossible to forecast earthquakes. Anyways, so long story short, I found it very interesting. I thought that the impact would be huge, so I decided to, with Noemi, to set up that company and see what happens next.

Pete Miller [06:04]: It’s funny, Itamar, because my father, who’s passed away, but he was a doctorate in geology and he ended up running a gold mining company. And I talked to him about earthquake. He goes, there’s no way to forecast them. I wish he was alive so I could go, actually, there really is a way to forecast them. You mentioned your wife, Noemi, that she’s an astrophysicist. And she you said, can you just delve a little bit more into kind of her research and what was the discovery? And you kind of mentioned you tried in academia, but at what point did you decide that this was a commercial solution?

Itamar Zabari [06:40]: So that research was actually born because of her professor. There is this CREDO group, a group that specializes in focuses on cosmic radiation and possible use cases of them. And he was in Chile in 2010, just before the massive earthquake that took place there. They were doing cosmic radiation research.

And they were collecting cosmic radiation data. And a huge earthquake took place. They saw a huge anomaly before the earthquake, but they never gave it a second thought because they were looking at something completely different. And then 10 years later, 15 years later, think, no, yeah, 10 years later, they picked it up. They said, let’s, let’s investigate this, this huge anomaly just before the earthquake. Can they be related?

So, this is basically the trigger that led to that specific research. And that research was mainly kind of a statistical approach.  But it was a very solid one. The results were solid. They were peer-reviewed. So for me, was OK. So given it’s true, the implications are huge. Like for sure, governments would love to have something like that for sure. The insurance and risk focused industries would enjoy the value that this system can provide.

For me was like a very, very clear understanding that governments for sure and insurance was double for sure, because they are suffering both from governments because they’re ensuring governments, but also ensuring heavy industries, infrastructure like oil, gas, etc. So, whether they like it or not, they’re exposed to it, even if they’re putting extra fees for it or whatever.

At the end of the day, it made a lot of sense to me that we can drive huge value to them by helping them reduce the potential damages that they’re going to pay for. And that means that both sides will be happy, both the client because it suffers less damages and insurance company because well, they don’t need to pay so much money, right? And the client is going to have less service interruption. And that’s the sole purpose of that specific entity. They would like to continue to do business without suffering damages.

So for me, it was obvious that we can get there, but the question was, can we give them the three most important points? Can we tell them that a major earthquake is coming? Can we give them a specific timeframe of when it’s going to arrive? And can we localize it? And the answers to all the threes are yes, yes, and yes.

Pete Miller [09:23]: So just to set the stage for our listeners, can you kind of contrast that with the current warning system, sort of on a time basis and on an accuracy basis and things like that?

Itamar Zabari [09:35]: Yeah, definitely. So, there are kind of earthquake forecasting systems that are experimental or in the academia. Most of them are not as accurate or statistically kind of probability and statistically focused. And they can tell you that, for example, there’s a group in China that is called ETA, they can tell you that an earthquake will most likely take place in the next week in a specific, very specific area. But in order to give you that coverage, they need to install three very expensive detectors underground. So for every 100 square kilometers, you need to spend around 100,000 euros, give or take. So if you want to really distribute it, you’re going to spend hundreds of millions of dollars in order to cover seismic-prone  areas. So that’s kind of out of the window. Potentially it is doable.

The other systems that exist today are actually detection-based systems. They’re not really forecasting detection. It’s usually based on seismographs. And the problem with that is that it’s waiting for the earthquake to take place. And it measures two different, like it measure the waves that arrived, the first wave, and then depending on how far you are from the epicenter, how lucky you are, you might have, if you’re in Japan, a minute, or just a few seconds depending really on. So of course it’s amazing to have at least somewhat of a few seconds of warning but the potential for damage reduction and then life savings much, much lower if you just have a few seconds of notice time versus an hour or two or a few days.

So that’s what we have today and just so you understand it is so crucial for countries that they pay hundreds of millions of dollars for such systems. So ShakeAlert cost about $300 million, ShakeAlert in California, $300 million all in all, if you combine all of the costs and when it works, it provides you up to 12 seconds of notice time. Japan just paid across almost a billion dollars to set up a new system underground on the shore to detect tsunamis coming. So, those kind of problems are, you know, countries are actively seeking for new solutions in order to solve those problems.

Pete Miller [12:10]: So, 12 seconds and you’re at 14 days. So that’s…

Itamar Zabari [12:14]: Actually, if you want to be, we can actually see it up to 25 days in advance.

Pete Miller [12:22]: Wow. So, tell us sort of, you know, in statistics there’s correlation and causation and all the most good necessary and sufficient and all that. Tell us how, like, just give us a little insight into how to, can you walk us through kind of the science? How does monitoring cosmic rays actually tell you that an earthquake is coming?

Itamar Zabari [12:39]: Okay, that’s a tricky part. I need to be very careful of how to answer my answer because we just finished writing 100 patents that we’re about to file and I want to make sure I’m not tripping myself here. In a nutshell, there are publications that already out there that shows that there are specific precursors prior to, in magnetic field or in the atmosphere, prior to an earthquake. The question is, can you really understand from those precursors if an earthquake is really about to happen or not? And if so, where exactly, what’s the magnitude and what is the timeframe for its arrival? That’s the tricky part.

Cosmic radiation can give us a lot of information, but that’s just not enough. You need to have multiple data channels, kind of stacked in stacks, different machine learning models dealing with different data sets. And then once you do your magic, which unfortunately I cannot really specify right now, you can basically combine it into, is there going to be an earthquake in a specific area? If so, where exactly in that area? And how certain is the system that it’s going to happen, right?

So, for example, we’ve managed to bring it to a place where in Chile, we will only notify the authorities if they only care about 7 and above in magnitude, we will only notify them half a percent of days a year. So that’s about five days, give or take five, six days per year. And when we notify them, about 50% of the alarms that we will produce will be 5.5 to 6.5 in magnitude. And then another 30 to 35 % will be 6.5 to 6.9 in magnitude. And then the reminder is 7 and above, because 7 and above, they’re so rare that you have so many days of forecasts, right? And then suddenly, an  earthquake that is 7 and above is happening, right?

So, it’s almost like being able to predict where a lightning will strike in three days. First of all, will lightning strike? If so, where and how powerful it will be? Almost. So, it’s definitely doable. We did it, right? It just takes a lot of data, a lot of work and a lot of creativity to get there.

Pete Miller [15:10]: Yeah, that’s amazing. So, I mean, just to spell it out for our folks who are primarily involved in risk management and insurance, with these warning windows of 14 or five days in one day, what can governments and utilities and industries actually do to protect lives and assets that they can’t do with just sort of seconds of warning?

Itamar Zabari [15:32]: Yeah, so we actually moved away from one day, five days, 14 days. We have much more pinpointed forecast today. So, we’ll be able to tell you in the next 48 hours if it happens or in three to five days or in six to eight days. So, it actually gives you proper time to prepare yourself for it. And you can do so many things.

Just a day before, you can shut down your factory, right? If your factory is working while a massive earthquake is taking place, assuming that your factory stays without crumbling, then you have machinery that is operating, it’s moving. So, if a machinery is moving while a massive earthquake takes place, it will for sure be much more damaged than if it’s shut down and secured.

And then of course you have people that are there that will get hurt. And so they’re very simple steps. Shut down, shut down pipes, evacuate people, leave skeleton crew. And then just those things can reduce your damages by a factor of one to seven up to one to 160, depending on the industry itself. So, if you have a nuclear power plant and you know that an earthquake is going to happen in three to five days, you will move the power plant to safe mode or even shut it down just to make sure because if it’s like 7 and above you might prefer to shut it down. And then you can avoid what happened in Tohoku in 2011 where they had only 60 seconds before the earthquake hit and they didn’t manage to do much with it and total costs amounted to about half a billion.

if you normalize it to today’s valuation. Almost half a trillion from one earthquake, not to mention the nuclear leak, the radiation leak that spread around the world in the oceans and harmed fishery and animals, et cetera. So, you can do so many things just by having one day, three days, depending on the type of earthquake.

Pete Miller [17:45]: So, let’s extend that a little then from an insurance and risk management perspective. For insurers, how does this kind of predictive capability change the equations? What kind of conversations are you having with insurers about loss prevention?

Itamar Zabari [18:00]: So, I think that when we started to talk to the insurance markets, they’re like, forget about it, the earthquake cannot be forecasted. But as technology matured, machine learning, AI, predictive technologies, they started to think,

in those in preventive ways. So, for example, Tokio Marine is very, very focused on preventive. They have essentially moved from compensation to prevention as much as possible. And so once the insurance entity has made the transition from reactive to proactive, then the conversation becomes much easier.

And so, assuming they’ve done that, you can integrate it into your policies. So, if you’re insuring your employees or your factory, whatever that may be, then part of the insurance policy would be you have to use also take AI’s forecasting system, just like you’re using, you have to have sprinkler systems or anti-theft technology in your car or whatever that may be. It’s already, we’re not inventing anything. The insurance market is utilizing existing technologies in order to reduce its risk.

And so this is kind of how we’re looking at things. If you properly embedded into your product, you can dramatically reduce the damages that you pay and you give your client a better service because your client is suffering less damages. And so, it’s a win-win situation. They’re happy. You as an insurer are happy because your net profit is higher, right? And this is kind of where we’re looking to work. We would like to create a win-win situation for everyone. And then beyond that, of course, you have the claims departments where they need to prepare for the disaster. So, they’re doing it today very nicely when it comes to different types of disaster they can actually forecast. Same thing. Do it with earthquakes. Prepare your loss adjusters, right? So you know that it’s going to happen in a specific area. Reserve them. You very few loss adjusters in the world. You don’t have many of them. Reserve capacity for satellites. Take images before the earthquake takes place. Take images after. And then quickly assess the damages. And then because you know how to assess the damages faster, or you can do that, then you can close the cycle of compensation much sooner.

One of the more interesting points were that when I was in Lloyd’s Lab, I was sitting with of people from the insurance industry and they were like, you know, there were a couple of times where earthquakes caused us to lose reputation in such a such a strong way that we have to pay tens of millions and years to recover our reputation after a massive earthquake took place. We were just not prepared for it. It took us so long to bring in the people, to assess the damages, to understand really what’s going on there. And this is exactly where we’re coming into the picture. If they know that something is about to happen, they can be better prepared.

Pete Miller [21:23]: So you mentioned from reactive to proactive, which I think is, I mean, I Tokio Marine and some others are really into that. As you scale globally, so what does that transformation look like? And when you think about regions of the world, which regions in your mind need this most urgently?

Itamar Zabari [21:44]: Places like Mexico, Chile, definitely needed more. Indonesia, all of the places really, really suffer. Morocco, essentially – Turkey – the problem is that either the quality of construction is not good enough, just the standards are not high.

And so, for example, Chile just recently moved from standards such as California into Japanese standards. They have many more earthquakes. They are much stronger than California. It just didn’t make sense that they have standards of California. They should have better standards, And so areas that the quality of construction is not high, those are the areas that are the most vulnerable and definitely the ones that need us the most.

But at the same time, very well-developed countries as Japan, the more developed the country is, the more painful the destruction is, right? From an economic point of view. So even if you don’t lose so many people, which is great because you’re well prepared, each and every damage that you suffer is more expensive than if the country would have been poor. Because your technology is more expensive. So if we look at Taiwan, for example, they have one of the most advanced chip manufacturing in the world and it’s also a seismic area. if they have a seismic event that hurts their technology, the whole world is being brought to a halt almost. are no chips. So, this is the more advanced and developed your country is, more technology focused it is, the more painful it is. And this is kind of where we come in to try and help them reduce the risk and build resilience.

Pete Miller [23:32]: I was in an earthquake in Taiwan several years ago, so I can attest to that. What markets are you operating in now? And where would you like to expand?

Itamar Zabari [23:41]: So right now we are working towards Turkey. So, we have a local partner that we are now working in order to, we’re finishing the technical diligence hopefully in the next two weeks to a month, depending on what else they need to see. And we’re working with Tokio Marine in Japan. Right now, the focus is to kick off a POC or a pilot depending on, what exactly they would like to achieve there. I just like two days ago came back from Chile. I sat with the National Emergency Office there. So where do I want to go? I want to go to Mexico, Italy, California, everywhere that there is seismic activity. And I know that there is the will to implement such technology.

Pete Miller [24:25]: Let me talk to you a little bit about your solution combines hardware and software consistent with not giving away too much information. But can you give us a high-level description of all that’s required to deploy in an area for effective coverage?

Itamar Zabari [24:43]: So, it really depends on the level of accuracy that our clients would like to receive. The more accuracy, the more stations we need. But in a nutshell, our infrastructure is, I’ll probably need between eight to 10 stations to cover Japan, something like that. A bit more for California, like about 14 stations. And beyond the stations we are using, like you said initially, it’s a multi-channel, multi-messenger approach. So, we’re using additional data channels, which are satellite focus from star imagery, thermal imaging, and then additional channels. then, what we, our approach in general is to have the most easiest to deploy infrastructure.

So it’s very easy for us to deploy stations and to use as many global data sets as we can put our hands on because it’s simply global and we have and it’s just easier to integrate those data channels. This way it allows us to scale quite quickly. So, if a client, for example, needs us in a specific area within six months from the moment that we said yes and sign something, to the moment we can actually forecast earthquake there. So, time frame is very, very short and as you’re moving forward, it will become shorter and shorter.

Pete Miller [26:02]: For millennia, earthquakes seem to be random events. So tell us why you think it’s different now, and that really it is possible to forecast earthquakes.

Itamar Zabari [26:13]: Earthquakes can be forecasted. You know, it’s, I completely understand seismologists. I really do. They’ve been working in a specific field for many years. They’ve been using specific sets of tools for many, many years. And through those tools and that specific kind of angle, you cannot really forecast earthquakes.

But we’re now doing that. We went in a completely different direction, we have multiple data channels. This combined allows us to very, very accurately forecast earthquakes with a very, very low burden of alarm. Alarm rate is very, VERY, low. And so just to…Keep an open mind. This is possible. We’ve done that. And we’ll be happy to speak with the people who are interested in embedding our system and its capabilities into their services.

The earth does tell us that earthquake is about to happen. It’s just whispering. It’s not kind of, it’s not shouting it, it’s whispering and it’s whispering it in through different data channels. And so you really need to listen to the whispers and connect them in order to create a coherent picture, and understand something big is coming. But it’s definitely doable.

Pete Miller [27:33]: This is so fascinating to me, I have to tell you. That’s an amazing thing. And what a great thing. You can save people’s lives. And you know, we’re all about predict and prevent. So, this is kind of right up on that and at scale and very impactful. Thank you very much for your time.

Itamar Zabari [27:50]: Thank you very much, Pete, much appreciated.