Prediction of the world : Where We Stand in Earthquake prediction । Prediction 2020, part-2

Prediction of the world, Prediction 2020,

Prediction of the world
Prediction of the world

Welcome to the second episode. If you missed the first episode, read it now…..

So let’s go to the main discussion.

So the basis for theearthquake early warning, it’s not to say theearthquake will happen.
Prediction of the world, Prediction 2020,
 It’s to say anearthquake has happened,

 now the shaking willhappen in such time. So you can already seethat the further away you are from the fault, the greateradvance knowledge you may have.

Prediction of the world, Prediction 2020,

 If you’re veryclose to the faults, you may not have asecond of warning. So this is not even a pointalmost of having warning. Although I’d like to saythat that one second,
 and I always thinkabout this eye surgeon with a laser in your eye,and maybe that one second we’ll just shut off thelaser and that might actually save your eye.

 So even one second is important. In Japan, in thenuclear power plant, they have seismometersthree kilometers down. And this three kilometers downwith a 3 kilometers per second, wave speed, gives youone second warning, is sufficient to shutdown the operations.

Prediction of the world, Prediction 2020,

 So even those fewseconds matter. But for the person, you andI, or [INAUDIBLE] next week, sorry, you might wantsome level of warning. And so a lot ofthe questions have been if the earthquakejust happened, it may last tens of seconds,maybe 30, 70 seconds.
 As soon as theearthquake happens, I want to know howbig it is, I want to know how much shaking it is.

 Because the longer wewait for that warning, the less Warning we’ll have. And so a lot of the questionshave been both in the– the use of data forearthquake early warning, but in terms ofphysics of earthquakes,
 is this beginning pulse,this beginning onset, will carry the informationof the size of the event. Another way to saythis is we call that the determinism in earthquakes.

Prediction of the world, Prediction 2020,

 Is the beginningof the earthquake carries enoughinformation that we know what the end of theearthquake is going to be like, therefore we cantell what magnitude,
therefore we can tellwhat type of shaking. And so a lot ofthe questions are–

 There’s a lot ofskepticism, again related to this Parkfield failedexperiment about earthquake determinism. And so in publications,I was recommended to not use this term, it’spolitically not correct.
 But all it says isthere any information at the beginning that tells usthe evolution of earthquake? And so it’s just a word, andI think there are sides to it,

Prediction of the world, Prediction 2020,

 and I’m showingyou the both sides. And we’ll take one side, we’llsee at the end of this talk. But basically thereare two camps, that’s why we have debates.

If we look at theonset of earthquakes, I know what the finalsize of the earthquake will be before it even ends. And then you have theno camp, which says look, all earthquakes look– [INAUDIBLE] thisstarted the same way.

 And I’ll explain those plots. But the bottom lineis that we have different data, differentways to process it, different school of thinking.

Prediction of the world, Prediction 2020,

 Those are probably the sameearthquakes and we’re just– we’re not coordinated enoughto be very systematic yet in our approaches to analyze thedata in a very objective way.
 So on this camp,these are plots that show time of the earthquakesto some normalization. This is in log scale of thepeak ground displacement. And the blue– The colors are showing you themagnitude of the earthquake.

 And basically you see thatthe way the displacement grows with time is not thesame for magnitude 9 or a magnitude 4 Others say– This is a 2014 study that saidsmall earthquakes start faster than big earthquakes. This study actuallyshows the opposite,
 so I’m not sure thevalidity of this. But still, it’s aboutthis growth, basically. This group showsthat in [INAUDIBLE].

Prediction of the world, Prediction 2020,

By the way, in earth science,we have so many scales that we put everything inlog time, spatial scales. We like to collapse thingsinto this lab log scale.
 So this is the log of the peakground displacement in meters. This is the log sincethe P wave started.

 And so you can seethese are bandwidth with earthquake magnitudes. You have a lot of– I don’t think we’re reallygood statistics, to be honest. Because we think thateverything is log normal, so we think median and meansand we try to respect this but it’s not necessarily robust.

 Nonetheless, if we takethe median of these curves, of P waves, which are thefirst waves that arrive, you can see that the beginning,whatever the magnitude you are at, the kind oflook one in the same. But this is one dataset, this is one team.

Prediction of the world, Prediction 2020,

 These guys are saying, I’m usinganother data, I don’t see this, and debate goeson and on and on. And so, it’s reallyinteresting to think about there’s a lot ofdiversity in earthquake and so really, howshould we study them? We have all these scalesand [INAUDIBLE] scales and temporal scalesand so,
 I think we’re not reallyfully integrated yet, but I think there’shope to do that. I’m not yet fullyintegrated,

 but I’m just raising the question and Ithink eventually we should think about all the scales. And so– pardon for thelack of resolution here, there’s one way to do it, whichis the camp that I just showed you earlier, is you say,
 I haveall these diverse earthquakes and I’m just going tostack them together. I’m going to look atthe average earthquakes. And I was telling youeverything’s about lognormal. So what we do is a lot aboutcombining those messages and then trying to make thisaverage view of the earthquake.

Prediction of the world, Prediction 2020,

 And so, if you actuallythink this medians and median [INAUDIBLE] on all theseearthquake magnitude, this is time,
 this isgetting kind of like what the P wave would look like. These people found atriangle function that makes it to science right away. Triangles are weirdin earthquakes.
 Everything should be lognormaland should not have any kinks. But what they sawis if we average everything over everythingthen it looks like the same.

 So that’s saying there’s just acommon behavior of earthquakes. I showed in some of mywork that if I do the same, there’s some differencesbut they’re only related to earthquakedepth not earthquake sizes. So in terms ofpredicting magnitude you’re kind of out of luck.

Prediction of the world, Prediction 2020,

 So that’s one side you can–strategy for science or saying, take all the earthquakes,average everything, find the commonalitybetween earthquakes, simplify the systemas much as possible.
 Or you can embracethe complexity. Earthquakes are sodifferent from one another, and I was tellingyou it’s lognormal. So there’s a lot ofdifferences of those scales.
 So this is just showing youa little snippet of what data sets we’re working on ishow different earthquakes look like.

 And if they are so different,what can we say about it? Is there any way wecan predict earthquakes or the size of the earthquakeif they all look different?
 Another example. We got thousands and thousandsof these functions that look like earthquakes,
 butreally they vary a lot. So part of theresearch that I started doing this summer with a really,really good team of students, I was so fortunate.

Prediction of the world, Prediction 2020,

 I really loveworking with students who have no a priori onseismology whatsoever and start askingthese questions that are so fundamental,without having the history and the package of whythey should not be saying determinism.
 So this term was reallyvery creative, not stubborn, but you know very determinedin solving the problem.
 In actually found areally interesting pattern in the complexityof earthquakes. So it’s is no longer aboutlet’s simplify everything into averaging everything.
 He said let’s lookat this complexity, embrace it and find patterns.

Prediction of the world, Prediction 2020,

 And so I willsummarize it in such. We have an earthquake here. It’s a big earthquake, probablya magnitude 7.5, maybe 8 given the duration here,60 second of duration. This function couldbe like a P wave, so you could see thatin a seismometer.

 I’m happy to talk moredetails about that later. And I’m showing you twotypes of earthquakes. We have the big one in orange,the small one in turquoise, teal, beautiful color scheme. And they look different,
 right? So one is reallycomplex, lots of bumps. They’re not the simpletriangle function. If you look atindividual earthquakes, there are a lotof richness in it.

 And then the smallone has one bump. And so he basically said,let’s count the bumps. And he made bumps withGaussian functions, lognormal, we love it, and hecomposed these functions by just adding and summingall of these log functions and counting them aswhat we called subevents.

Prediction of the world, Prediction 2020,

 They are not a fullearthquake, they’re just a small earthquake,a small event that belongs to the earthquake. And so it seems likelarge earthquakes have a lot of bumps, subevents,and small earthquakes have fewer bumps. So that’s increased complexitywith earthquake size, which is not that surprising. What was interesting is we couldmodel this very simply in-house with simple fracturemechanics models, where we created earthquakeswith a set of parameters I can explain later.
 But basically what we foundis the big earthquakes that a lot of bumps and thesmall earthquakes– smaller earthquake has one bump. OK.

 So the number ofbumps or some subevent grows with earthquake size. This x-axis here isshowing you the size of the main event,the main shock, and these are thenumber of subevents.

Prediction of the world, Prediction 2020,

 You can see that withgrowing size, this is Tohoku, and by all means, an outlier foreverybody to study earthquakes, but the biggestone is the outlier. But if you look at ensembleof the number of subevents, you have a growingnumber of subevents. So big earthquakesare more complex because they havemore subevents.

 However, it’s not we don’thave a log scale here. And so actually, thesize of the subevents grows also with thesize of the earthquake.

Instead the earthquakedynamics has this interesting andintrinsic link scale that is characteristicto the earthquake’s size. All of these dotsare a subevent that has a size shown in herein log scale attached to the size of the mainearthquake shown here.

Prediction of the world, Prediction 2020,

 These lines, thisis a one to one line saying the subeventhere has the same size as its masterearthquake, in which case this earthquake probably onlyhas one subevent to form it.
 This line here is showingyou that earthquakes would have about 100bumps, 100 subevents, to form the earthquake. And so what you can see, itis not an infinite amount. The subevents don’thave the same size.
 These events actually growwith earthquake magnitude. And so what was interesting,the current scale is representingthe distribution. And so,
 these smallevents tend to have mostly one main subevent,versus the large event have a lot more. And if you were todraw kind of a scaling, it’s not one to one butsome very strong scaling.

 So that means in terms ofstructure of the earthquake, a small event has a smallbump and the big event has a big bump at the beginning.

Prediction of the world, Prediction 2020,

But if you go to thoselarger magnitudes, if you measure thisfirst bump, then you can predict or estimate thefinal size of the event. Without the earthquake, theearthquake is not done yet, you’re just making an inferencebased on this organization of the seismic signal.
 And so we took thisindependent data set, we looked at this recentIndonesia earthquake in 2018, 7.5. It was large. Was not really on amap fault,
 per se, it was not really onthe plate boundary. It’s one of thoseearthquakes that just tell us we don’t know much.
 [INAUDIBLE] just onplate boundaries. Earthquakes happenin unforeseen places, so we can’t really justpredict an earthquake tomorrow, but these are theearthquakes that remind us how young our field is.

Prediction of the world, Prediction 2020,

 So the top functionis showing you the same, what will be a Pwave as a function of time. And then we fit in allthese Gaussian functions, joining these dashed lines. And at the bottom, thisis one of our attempts to say why it would be,given this bump,
 our estimate of the magnitude ofthe final earthquake. This green line is showingyou the past magnitude.
 This says, if Iintegrate here this, I had a smallearthquake, shown here. And if I go on, a lot ofthe earthquake has gone.

And so their predictionwas shown in blue, with a random first algorithm. And then what theexperimental data was showing is shown in red. And these are thevalidation part. And so it’s really stunning thateven though it’s not regular,
 these predictions were ableto show a weaker earthquake compared to a big earthquake. So I thought it was one ofthe most promising study for machine learningto predict earthquakes. In-house,
 our teamat Harvard has been able to predict thelocation of aftershocks. And so as I was tellingyou, we have the main shock and then we have manyaftershocks, which is something we forget about.
 If your building is alreadyshaken by the main one, you will have manymore earthquakes to go.

Prediction of the world, Prediction 2020,

 The building is going to have todeal with many more earthquakes and aftershocks. And so what thisstudy was showing is that if you lookat those black dots,
 the fault is showing atthese yellow patches. The red is showing basicallythe change in stress, and then you can predictfrom the slip that happened on the previous earthquake. And they worked withmultiple earthquakes here.
 And what they found isthat the traditional way of looking at where anearthquake would be, given a change ofstress in the crust, failed to predict many,many of the aftershocks.

Prediction of the world, Prediction 2020,

But basically theseapproaches were successful atpredicting the location of the next earthquake. And so this isshort-term forecast. It’s not saying years aheadwe’ll have an earthquake.
 It’s saying wehad an earthquake, we know there willbe aftershock. And it’s saying where thenext aftershock could be. Other things thatmachine learning has been really successfulat in seismology is detecting those very weak,weak ground motions that are buried in seismic noise.

 And it has a reallygood [INAUDIBLE],, that was one of the first to usea deep neural network to detect earthquakes. And we all workedin Oklahoma, we worked on the data set thatalready had a few thousand events, but we wereable to detect something like 20 times more events.

Prediction of the world, Prediction 2020,

And we have hundreds andhundreds of those events that we can look at thewave forms and say ha, actually therewas an earthquake, we just had not detected it withnatural and more established methods.
 And so in terms ofprediction, I think there’s a lot of hopewith the new methods, and mostly the pastrecording of data,
 where we canpotentially tease out some of these short-termterm forecasts. And I will stop here.
 Thank you. [APPLAUSE

One thought on “Prediction of the world : Where We Stand in Earthquake prediction । Prediction 2020, part-2

Leave a Reply

Your email address will not be published.