 Well, thank you very much for coming out and I Want to point out that the first image on my title slide was put up there? Weeks ago and therefore demonstrates our incredible ability to predict the weather In truth I just like this photo a lot So it just happens to align reasonably well with the weather that we're getting so I'm not going to claim any skill and Predicting but it is kind of funny that that's the way it worked out So again, thank you for coming out on a Saturday, you know a lot of people like to take the weekend off and Vege out and I very very much appreciate you guys coming out to listen to what I have to say So let me give you a little outline I'm going to have some demos and things Demos are one of the scariest words in computer science because they almost always go south So I'm working without a net. So, you know, you may you may see some results of that But basically I'm going to try to explain to you what we mean when we say computational science, you know Science used to be sort of pencils and papers theories and test tubes. We added computers some time ago to that mix What does that look like? I'm going to talk about The quest I've been on with a lot of colleagues in the discipline and here at NCAR to get parallel algorithms and Systems and I'll explain what parallel means, but that's our way of tackling these big problems The recognition of the fact that we need a new facility to fit the big computers that we need to do this and how that took a basically ten years out of my life a Few a discussion of the few dark clouds on the horizon We're running we're running out of You know the trends that have powered us for the last half century and what are we going to do next? The next thing I'm going to sort of allude to is a possible path forward is artificial intelligence and And you know those of you remember the how 9,000 computer and all of the science fiction Associated this will probably see the pluses and minuses of that avenue Super computer in your pocket section. I will talk about these amazing little things that are called raspberry pies and how we're using them for education and Outreach and training to get people interested in this business And we'll talk about new ways to interact with data. So that will cover augmented reality and Using sound to represent data. So we will have Some Musical composition composed by the your system itself for you to listen to and then finally I felt obligated as part of this and since it's supposed to be an explorer series and I suppose that means I'm supposed to be the Explorer I thought I would just sort of convey to you why I find this stuff Worthy of devoting your life to you know, why am I doing what I'm doing? What do I care about it? And it's it's really because I think at some point you realize that this this whole set of phenomena is just very beautiful So I will go there. So how do supercomputers help science, you know, they're relatively new to the scientific Experience well you start with a problem in this case. We're interested in and how the earth works and What we do basically is Describe this through a set of mathematical equations Physical laws that we tend to believe are true We write them down and this is the only equation in this talk And you can't this set of equations you can't actually read them, but and that's a good thing actually All right, so then what we do is we take those equations and we translate them into algorithms and programs that a computer can understand And that's what that looks like and I won't be explaining that diagram either And you get input data. So it's a bunch of numbers perhaps that represents the initial state of the atmosphere what the temperature fields and pressure fields look like and then we feed that all into Supercomputer and here's our latest supercomputer, which is Cheyenne. I'll be talking more about that But basically the data in them and the algorithms go in and now come some results Generally speaking we get too many results Too many results to look at or think about and that has gotten worse and worse as we've gone along So you may have heard of the concept of big data or too big data those those issues come out there and then we move on to some kind of Analysis process that extracts out some curves and plots that we we show to each other and Then somebody comes along with a pencil and makes a check mark and says yes Maybe it's not a pencil it might be a fountain pen But the point is we come along and go yes that agrees with what we're seeing in the earth Or it doesn't and if it doesn't that's actually more interesting because then it sets off a series of very polite arguments amongst the scientists as to why that is and then that fuels perhaps modifications and So you can think of this as a kind of tail-chasing exercise that we've been doing since we got our first set of computers Working on this problem roughly around 1950 So that's how computers kind of help us do this stuff. So what does it mean when we say we're going to model their system? Well, this little picture Which includes I would point out a deer I'm not sure we have a deer in the equations, but the point is that there's a great deal of interaction between Different processes in the earth system you have ice floating around you have oceans you have water running off Excuse me while I try to navigate this arrow. You have water running off the land It snows in the mountains, which is what's happening to us right now And all the sun shines on all this and kind of fires it all up And now all these processes have to be accounted for in a complete description of the earth system and that's a tall order And you will see here that also that which is a human Power plant let's say is with emissions that also has to be factored into the research If we're going to have any relevance to what's actually going on now how we actually do this is We chop this all in the earth into a bunch of blocks. You can think of those as You know, let's say little chunks of latitude and longitude You see this block we will consider as a single point and we will model the system Relative to that set of points and as you can see what we also have to do Since the atmosphere has a vertical extent. We have a kind of stack of these points Similarly in the ocean so if you think about all this we digitize this all up And we apply these local laws in these boxes And then we calculate essentially the rates of change of the variables that we care about and those rates of change Then allow us to kind of march forward in time So for example, this is air column and water column 1969 scientists here at NCAR who's who's still coming into work Warren Washington and And his colleagues put together the first Model at least for at NCAR that looked like a global atmospheric model This is a piece of visualization of that model. It had Basically only the atmosphere one single point covered 500 kilometer box on the side Which means that the state of Colorado got one box Now here's a question for you. You know that Colorado is half flat and half mountains. So a really good question is What did that box get to describe Colorado? Was it mountains or? You know prairies Whichever one it got it was half wrong, right because you didn't you didn't resolve the other thing So that's a problem, but this was the first model. This was state-of-the-art basically 50 years ago and The vertical we had six layers that was all that that model had Now when I got here roughly 20 years later, we'd made a little progress We'd whittled this down to where Colorado got four Points basically and we had increased the number of layers from six to 18 And one of the things that was holding us back was the fact that These models were using algorithms that were not particularly local and did not particularly Speed up fast enough to allow us to do more complicated simulations so You know when I when I arrived here I'd been inculcated with this idea that we need to think in parallel and And so parallel computing is really about Dividing work up amongst lots of different workers and having them work together that's why I have a beehive because Bees don't send one be out at a time to look for honey They all go out and look well they don't look for honey They look for pollen and they bring it back and make it in the honey But the point is they send them out in waves and they all work together for the common good So it turns out that this whole thing of how to keep thousands of workers or Let's say hundreds of thousands of workers working on a task together cooperatively We call that parallel computing so with that in mind. This is the part where I quiz the audience Okay, so These are two household items you probably have become familiar with So my question is which one of these is a parallel device and which one is not anybody All right the comb and why Because a comb you allows you to comb many hairs at one time If you combed your hair with a toothpick you would find out The disadvantage of a cereal comb, right? So essentially somebody a long time ago says I think I can parallelize The toothpick by putting a lot of them on a single thing and I will call it the comb and That that person you know probably made a lot of money This on the other hand Hammer although somebody I showed this one since somebody mentioned showed me a picture of a two-headed hammer And I was now I was like well there their limits to everything Okay, but you know typically you only drive one nail at a time with a hammer, and so it's it's not so here's another one My wife is a knitter so she can answer this question You know which of these is parallel a loom that's supposed to be a loom. It's safe for ages nine and above for some reason I Guess that's so you don't weave your hair into it or something But the bottom line is which of these now is the fat which of these is parallel The loom right very good you guys are You've now learned everything I know about parallel computer So knitting you do one stitch at a time it takes forever to make a sweater with a loom It's an industrial process and actually the interesting thing about the loom is that back in the industrial revolution? people Realized it was really a painting to create patterns with a loom By changing the threads and so forth to make the weave the pattern in and so they said we need to automate that and so they started to create programs for looms that were driven by paper tapes And so those paper tapes told the loom what to weave in to automate the process of making patterns and Coincidentally the first Mechanism back when I was an undergraduate for loading a program into a computer was a paper tape That's not a coincidence the computers inherited some of this technology about cards or paper tapes from the weaving industry So there you go So this is my first serious loom When I got here at Incar This is the I say the coolest looking supercomputer ever I ever programmed you have to say This is the as far as I know the only supercomputers in the inner museum of art It and it looked really cool. If you ever saw a Jurassic Park, there's a Evil supercomputer that's sequencing the DNA of the dinosaurs who run amok That that was this machine Actually the descendant of it, but it's the same architecture It was a really pretty avant-garde it was had 64,000 processors which was unheard of it at the time and it was able to do 28 billion math operations per second and As you will come to see that is a really pathetic number But in the day in 1987 I was I was just amazed by it by this thing and I couldn't actually believe that it That it actually was not a prop for a science fiction movie. It was actually a computer it was What we call single instruction multiple data the best way to understand that kind of architecture is all the processors work in lockstep Like a bunch of soldiers marching together. So everything at every computer Processor in the machine executes the same instruction at the same time And that That means that there is one master processor, which is telling all of them what to do Right. It's a very sorry to go into a German exit. All right. So You know my job was to take in cars climate model and port to that and that was largely Difficult because we really didn't have 64,000 degrees of freedom in the models. We were working with So this was a limited success but As as a project, but I learned a lot about Getting after that what we needed in order to make progress so we started looking around for scalable algorithms and You know, eventually the idea of working with something called spectral elements. It's a numerical method. It's very accurate It's quite local. So the laws of physics are local laws they just tell you how to update your local grid point and the idea was to get something that had those properties in an algorithm and This one there were lots of intense calculations in this little patch and then little amounts of communication with their neighbors and We had to put that on a sphere, but they're all square So immediately dealt with a problem that everybody's heard of which is putting a round peg in a square hall So how do you put a cube on to a sphere? Well, you take a something like a Rubik's Cube and you mathematically inflate it To to make it fit on the sphere and that's what you you get this kind of thing. We call a cube sphere You know one of the nice things about that compared to the other models at the time Oops, I'm jumping ahead Was that you get quite quasi uniform distribution of points this color is just showing you that the way points are distributed? Bunches up only a little bit and then this is a fun game And and literally we we thought about this problem in the way you would you would think about a packing box You know, it's mashed this thing out flat and then let's try and figure out how we can Break it up into little pieces so each processor can work on a on a piece and in this case we're looking at eight processors with eight different colors and An algorithm that I won't explain but that we use to divide that up So the whole idea was to get the same amount of work on each processor So roughly when they finish doing something they finish at the same time so they can then work on the next thing And so my next really significant machine was an IBM machine called the blue G now and It was not damaged in shipment. It actually did lean like that It's not you know, it looked like perhaps maybe somebody dropped it or something, but this this machine Really had some nice properties to it. It was very scalable It's interesting now then you look at the number of processors. It actually has dropped down to just 2,000 processors But look how much faster it is. You remember I was working on 28 million. This is now a hundred times Faster than that in 2005 So this machine and this is a very small test system. There are many times bigger machines at Lawrence Livermore for example had A machine with many of these racks, but our whole objective with this Was to map this algorithm with the round peg in the square hole cube sphere thing map that onto this machine and and Use these algorithms trying to make it as parallel as possible and I'm trying here desperately to lean to make it look as if it were straight up and down So very interesting machine Now as we started to look at this parallelism thing We realized we're going to need a bigger facility if we're going to make the next jump in computing that we think we need to make In order to take these parallel algorithms and now apply them on really big machines with lots of processors The machine room which is downstairs and you can look into it and you'll see it's pretty much empty at the moment That's where all our soup. That's where this picture was taken. That's in the downstairs That room was built in the 70s and just couldn't handle the size of machine order megawatts machines that we needed So I brought this up with the idea that we'd order a new facility and it would be delivered to the couple weeks and And I discovered that that's a lot longer process So I brought up the need for a facility in October of 2003 and in October of 2012 the facility was commissioned and put into service and there is a cat You know a huge number of people that got involved because this is involved a great deal of engineering design And it's outside Cheyenne, Wyoming if you go up there, you can go in and take a tour We have a nice visitor center That's a picture of the facility and if you go up there, this is really the people piece And the machine sits back here in the back in that in that kind of box Very proud of this The first machine we put in there was 30 times faster than the machine we had on the NCAR campus at the time Here in the Mesa lab this machine we named Yellowstone in tribute to Wyoming Now this number is always problematic because people are like is that the national data? What is what is this number? this This is one and a half quadrillion math operations per second. Okay, so You know a quadrillion is a million billion So if you think of a one and a half million billion as opposed to 28 billion That's how much performance we've gone up and in the way the performance moves in this business Roughly a factor of a thousand every decade There's nothing like that anywhere else in technology yet You know, I always like to say between the Wright brothers if you were to do a thousand factor of improvement in flight in This in a decade it would mean be that World War one would have had f-18 fighter jets deployed Which would have given us a decided advantage. I Suppose just flying by would rip the wings off the other airplane But the point is factor of a thousand, you know airplanes have made a lot of progress over the years But it's nothing compared to computers So, you know, I mentioned here that it has 72,000 processors So we finally have gone by what this connection machine was first like And I say it has a fat tree interconnect and people are like, what is that? So it's It's not quite this But it is related to this if as an analogy if you think of the leaves as processors in the branches in the trunk as communication fabric So here are processors down here at the bottom and then switch elements, you notice the links between the switches get faster So this thing now looks like an upside-down tree and that's why it's called a fat tree and Really this interconnects what makes a supercomputer computer these processors that we use not all that different from the processors You might find in this laptop On a per computing element basis. What makes a supercomputer these links that allows them to work together in parallel So the an individual I'll show you a picture This is the top of Yellowstone and this stuff that looks like orange vermicelli down here this All of that stuff are those those cables each cable is about 8,000 times faster than DSL line and I always say in this picture that it illustrates the importance of labeling your cables Because if you don't You're in deep trouble trying to You don't want to be doing this Okay, so I want to show you what What a pettus gaol so pettus gaol means a machine that can do a quadrillion math operation So I just want to show you what a pettus gaol model can do this is a model for prediction across scales and You will see some things forming in here like this That is a cyclone you will if you watch that it'll smack Somalia and It will actually form an eye See there's a little eye there. There's another hurricane over here smacking into Southeast Asia see the eye forming there. So this is a four kilometer resolution model Some people when I show this picture say well, okay, this is satellite imagery. What does the model look like? This is the model so this This level of realism is the result of applying that kind of computing power to the problem with these very simple equations applied Just to the local grid points these grid points now Colorado probably has on the order of at that resolution has something on the order of 10,000 grid points Instead of one or four So that does now if you're a real expert, you know what these dates mean Round Halloween in 2012 Anybody know what happened then? No, that's that's Steve Thomas. That's a ringer. I'll give you five bucks later Steve Thomas is one of my colleagues have worked on the spectrum and stuff. So he's feeding Yeah, Sandy if you actually watch this and I won't show it again But if you watch this Sandy doesn't happen here the scientists that was doing this experiment was trying to understand the sensitivity of hurricane formation or cyclone formation To initial conditions. So this is a reality That didn't happen in which Sandy didn't happen the other two hurricanes really did happen or One was called merge on and the other was called son tin son tin in Vietnam or John hit Somalia but What what he is Investigating here is something that you may have heard of the the butterfly effect or the sensitivity to initial conditions this is a reality in which the initial conditions are a little bit different and The two hurricanes happen in one spot of the world the way they really did roughly but Sandy doesn't happen in this So I'm sure the people in New Jersey would like to click on this reality and have replay it that way but It shows you that Which way the atmosphere goes and what it actually ends up doing is very sensitive To the state that it it's in and that's really one of the reasons why our ability to predict the weather is somewhat limited And I'll just oh yeah, here we go So I just wanted to this is not actually four kilometer resolution, but that's the kind of grid that we use in this impasse model and it's a kind of a Geodesic grid you see a bunch of Hexagons like a honeycomb And that's sort of a feeling for what kind of grid is underneath there So here's Cheyenne, that's our newest supercomputer computer way we deployed that Right at the beginning of 2017 It is five point four quadrillion operations a second. So it's three times faster than Yellowstone You will see we've doubled the number of processors here roughly to 145,000 Now you want something excruciatingly difficult to do get a hundred and 45,000 of anything to work together You know It's it's like exponentially difficult cat herding All right, so this this scaling applications to that level of supercomputers is quite quite a challenge So here's some of the dark clouds How much smaller can we make a transistor? The first transistor that I'm showing up there in the upper left-hand corner Was basically the size of your thumb. It was invented in Bell Labs in New Jersey Where my wife's mother worked Think she she actually invented the transistor if I remember and then they took the credit But anyway, the point is that may not be true, but But she made she she was part of it, you know In the lab there anyway, so this thing about the size of your thumb this is an electron micrograph of So my electron microscope picture of a single transistor in a modern what they call a field effect transistor in a modern You know chip This is about 200 atoms across And if you look at this this little thing at the bottom that's the installation that basically keeps the thing from shorting out So the gate and the substrate there's a voltage potential across there and that little thin thing is only 10 atoms across That's not a lot of atoms If you misplace an atom somehow, it's 10% thinner That's pretty scary This is a problem that was that it's been dealt with over the past few years This was getting so thin That electrons were quantum tunneling through this barrier and leaking through and that was causing the chips to heat up This was dealt with and I won't explain how Within the laws of physics, I should say because it's a it's a product on the market, but you know to deal with that Is one of the reasons why our computer? Cheyenne isn't drawing way more electricity than it is because they they dealt with that kind of issue So when we talk about quantum mechanics, there really are Quantum engineers who have to worry about designing these kind of structures Well, my question is though at some point you can't keep shrinking this indefinitely You know we're we shrink this down from an inch and a half down to 200 200 atoms across at some point We're going to be played out That's a worry So are we at the end of modeling? There's some other problems here the processors have not been getting faster We've just been getting more of them, and I told you 145,000 is horrible to Control what if you have to control 10 million of them at some point the the processor Synchronization issues the ability for them to talk to each other becomes problematic We've kind of used up a lot of the parallelism that we have available So that's not available to us as much to speed up and our models are getting more complicated You know scientists love putting all kinds of physics in about you know They want to put the deer deers and and flowers and whatever else they can think of into the model to make it More complex all that ends up with subroutines and things that could break And so the complexity is getting us and then these things spew out a lot of data So I'm just a beam of sunshine here but You know unless we think about new ways forward Just like I kind of thought about massive parallelism as a way forward We need to think about new ways forward to keep moving ahead So basic question can artificial intelligence do more than be this a jeopardy And that's an actual picture of it clobbering to human champions You know so the question is how can we apply that and what is AI anyway? So this is sort of a taxonomy here AI is kind of an umbrella term that refers to Teaching computers to perform human tasks There's a branch of this it's called expert systems, which is Essentially working with human specified rules and working autonomously on problems So for example an expert system might be a cost estimation system that automatically comes up with the best configuration for a computer installation for example Machine learning these are algorithms that you know are able to actually Learn from data how to perform tasks and I'll show you an example of that Deep learning is something in which people have taken neural networks. So these are computer models Computer algorithms that model the human brain And a set of neurons and you know that there we have multiple layers of neurons in our brains Deep learning just means we have a deep set of neural network connections that allow us to Create a learning machine basically Allah a human brain So here's an example of this stuff being applied to real to real data. There's some little red Squares here what people are doing is like instead of having people look at pictures and say you know as I did with the Globe there. I said, oh, look, there's a hurricane over here. I'll kind of machine look and say oh, this is a hurricane over here So these little boxes are the machine identifying some feature I won't go into that but features that a Human would have had to look at and say that's an interesting block So a group at Lawrence Berkeley laboratory and nurse have been doing this kind of stuff And they've been trying to use it to identify things like tropical cyclones fronts Atmospheric rivers. These are streams of moisters that come up from the tropics and smack into the Let's say California is things called the pineapple express fall into that category They hand label the data so they go Computer that's a hurricane and that's a hurricane and that's a hurricane and it goes okay Okay, I got it. I got it then you show it another thing you show it a picture of a dog And it says that's a hurricane. He's a bad computer now They give it a hundred thousand hurricanes and and eventually it it's a it gets very good at recognizing All right, so here's a here's an actual application of this stuff and What this is is some work that a colleague Here at NCAR DJ Garnier has been working on and that's showing a neural network as a value add to a model output So you run the model it shows there's a line of thunderstorms from stretching from Denver to Nebraska What's the chance of some of those having hail? Well, this neural network has been trained With other high-resolution models to identify damaging hail conditions and it looks at this Model output and gives you a better probability of damaging hail so What is this convolutional neural network? Okay, so it's cats and dogs here right but the idea is there are multiple layers of these neural networks There's a big pile of cats and dogs and we've taught taught it By comparing local features and then less local features in the next layer and more broadly applicable features you know so within each layer of the neural network we're in a sense looking at abstractions of more Larger patches of what makes a dog's face different than a cat's face So all these things have weights that they apply To the inputs to determine the outputs that will be passed on to the next layer The output of this thing is like Okay, is this a cat or a dog and you know The really funny part about this is that they're very good at identifying cats and dogs But we don't have quite as clean a picture. We have blood, you know weather looks like green and yellow and blobs They're they're not well-defined As dogs and cats both have two ears, but thunderstorms don't have anything like ears So what does that actually translate into? So it's a it's an interesting thing that We've actually now got Neural networks that can do a better job at predicting damaging hail and This has been operationalized in various weather prediction systems and Validated relative to not only The verification criteria that people use but also the experience of the professional forecasters Now one of the interesting I find this really interesting so There's blade any blade runner fans in the audience. Okay. Do neural nets dream of electric hailstorms One of the things you can do with this thing is you look at this little network that you have run it You can ask the neural network by running it backwards. What's your idea of a perfect hailstorm? It's like what's your dream a hailstorm and it coughs up something that looks like this now This is at different levels the atmosphere, but warm moist air a certain amount of rotation at the mid-level and You know the top the top of this thing getting colder all of that stuff Looks like a super supercell which is what we intuitively associate with a hailstorm so from a I Guess the best way to say it is the The neural network is learning something that we think makes physical sense. It's not just saying Well, it has to be Tuesday for it to be a hailstorm, you know, it's not bringing in extraneous factors It's actually capturing the real structure of a real Hailstorm and that's very reassuring in the sense that it's on the right track scientifically So parametrization An emulation this is where we get rid of human crafted physics and We stick one of these neural network emulators in its place and then we use that to advance the time level now There's no plan To replace Isaac Newton with a robot and I would say that You know we yeah, yeah, I had to put that in there just Physical law is still right. I think you know our the sort of hypothesis. I'm working against with this stuff is that You know, we'll use physical law that we know is right like conservation of mass and momentum Conservation of energy those sorts of things where it's appropriate But the places where we're doing guesswork are where it's drudgery to find hurricanes by hand We're gonna bring machine learning in to help deal with that stuff and so machines Historically have been labor-saving devices and things that amplify what we can do. That's what we call mechanical advantage You know the whole way a screwdriver works is by making us somehow stronger by by using the tool And that's what I think machine learning will do So super computer in your pocket. Yes, this is this is Everybody that has a smartphone kind of doesn't realize how Amazing it is although judging by the way people seem to walk around worshiping them perhaps we do This this you will see a demonstration of this outside This is a Thing that we started doing for education outreach These cost thirty five dollars each one of these cards it's about the size of a Phone it has a phone processor in it actually a Broadcom arm processor This whole thing costs about 200 bucks fully configured We've been using this To teach students and to teach faculty how to How to work with super computers believe it or not we put the same software we put on Cheyenne people can program them in the same way and They're really a tremendous way to lower the barrier for doing for doing science The interesting thing is this little thing would have been on the top 500 fastest super computer list 20 years ago 1998 This this right here and that machine probably costs 10 million dollars or something So when you next time you misplace your cell phone You should realize it's two or three million dollars worth of 1998 super computer You just lost so you might want to think about keeping track of it, but it does tell you something about how far we've come All this stuff that we do on her phone really was was sort of stuff confined to supercomputing labs 20 just 20 years ago So it begs the question where we're going to be in 20 years, you know This by the way is a simulation of Hurricane Harvey run on on the Raspberry pi cluster that you'll see out in the reception area It didn't quite do as well as the NOAA supercomputers which are much bigger the weather the national weather service predictions But it showed Houston getting hit by roughly three feet of rain and You know reasonably good forecast this is using initialized data from NOAA and then running that forward model To do the forecast so this machine essentially is a weather Weather Simulation platform runs are essentially the model that that would be used in production at NWP Center We're using these to train faculty. I think I mentioned that We targeted a Minority-serving college in Miami called Miami Dade College It serves a greater Miami area. It has a hundred thousand students in it these Are a group of people who went down there for this training We've now done this two years in a row the faculty are standing in the back But you know for there is a kind of digital divide that a $200 computer can bridge And I've seen it. I've seen these things build resume line items for people who got jobs at Apple and Google and Stuff like that So it's a it's a game-changer for students in terms of getting access to an experience with high-performance stuff All right, so bring on the demo. We were So Pokemon Planet Our Medio AR so we started looking at augmented reality and one of one of our student interns who actually is in the audience You know he Nihont started working on this project to put Augmented reality and I'll show you what that is in a second Our simulations which normally you'd have to come to our visualization lamp to look at Start to put them as 3d virtual objects on to Computer like a handheld computer like your phone or or a tablet So what I'm gonna do is I'm gonna press this easy button here Which will switch this over You're now looking at this tablet Hopefully Well, you know how this goes I'll try to take some slack up because I have a feeling that that's gonna Give us some problems So what I'm gonna do you hopefully oh I had reverted back. It lost the signal sneaky thing. Okay. There you go And if I actually can get my hand away from the device, I'll move this around And See, I've just created a virtual globe. I can make it bigger. See I'm I'm making that bigger just by That and if you look you see it's moving That's a movie. It's not a still picture. I Can look at the North Pole, but it basically anywhere I want So There's that hurricane. I mentioned that it hit See, there's another hurricane hitting That's a cyclone hitting India So the nice thing about this experience. I can also do this. This is Odile This is not moving. This is a static thing. This hurricane hit Baja, California in 2014 you can look at this Moisture is a function of height so I can look at that from the side From any angle or I can look at it from the top down. You see that it it looks a bit like a Danish, you know There right a roll. So that's the hurricane aspect of it So all of this Basically allows us to communicate the science we're doing in in 3d and Interactively on a handheld device now When you leave here, there are some cards and you can freely download this for Apple or and or iOS or Android platforms depending on which of those two religions you subscribe to and you can play around with this You can download these you can get access to these kind of apps Oh, well, there's a VR form of it in which you can basically view it with no background One of the things is it's actually In my experience that well, I we have people pictures of people holding there So it's a month having it pop up like that as a much You know, it just adds something to the experience, but you can do it either way I think it gives you a better sense of the 3d is of what you're seeing You just see it on the flat background the 3d aspect of it So as you can see from this Switch back as you can see from this picture The globe industry is about to go out of business If it hasn't already, but I mean really you do not you know, you don't need to invest in These kind of physical objects anymore, but for us this really liberated these simulations So apart from that kind of demo, what can this stuff do? So one of the things we're looking at is using these kind of augmented reality movies as outputs from Let's say a fire model. This happens to be a model of of yarnail Hill fire that took the lives of 19 firefighters and The idea here would be being able to run a fire model put a movie which has got this 3d-ness in it and then allowing people You know as responders to look at that interactively in three dimensions This is another example of how this can be used. This is from the weather channel They did not build a tank on the set and immerse this car underwater. This is all augmented reality showing Forest storm surge how deep underwater your car is going to be Some people just do not get the idea of what six feet of water means to their car This gets it across in a visceral way that perhaps you should you know move to higher ground now this is the This is the next of my Demos okay, so what what you're going to have here is the sound of climate? so We we have created a kind of climate orchestra that in which the following Sections of the orchestra represent different things so temperature will be played today by the clarinet Precipitation will be the marimbas The sea ice will be represented by a violin and the co2 level of the future climate Will be represented by piano chords and the pianos will shift from major to minor chords To deliver a sense of ominous Yeah, I wish I were making this so So that that whole thing, okay is that let me see I lost my oh there it goes Okay, so red here is temperature And we're moving through time as you can see the year is going by This is just temperature and co2 Remember as the piano and it's shifting into more and more tense sounding chords This is sea ice. So we're looking at sea ice. I Swear this sounds like the beginning of a song Anyway, the green is sea ice and you can see that the sea ice will be disappearing sometime and it drops away and The piano tune is just left So the co2 again is now we're looking at precipitation This is the one that's the most like Philip Glass because The precipitation doesn't do anything Really my apology my apologies to Philip Glass who's listing it Okay, so there's not much of an anomaly associated with precipitation in this ensemble of climate simulations This is them all playing together. This is all three together So this is a yet another way of communicating and Data so why am I why am I doing all this because It isn't your you know, it's like they said though osmobile. It's not your your father's osmobile This is not your father's super computing, you know, we're moving into an era where we're trying to communicate with people who themselves Have a kind of supercomputer from 20 years ago We're also trying to be mindful of fact that there are visually impaired people who can't see what we're talking about when we show augmented reality so Trying to trying to reach people in more immediate ways and communicate what's in this massive amount of data much more directly and then I just want to end with this meditation on some of the things that Have motivated me so Basics, there's a book. I recommend you read it if you're interested in in science it's called a beautiful questions by Frank Wilczek. He's a Particle physicist he's developed the theory of the strong nuclear force who's Nobel Prize winner And he's asked these questions that have come up before in our history of science Why is the world mathematical? Why does mathematics work so well? Why when we do this math? Is it so beautiful and and The origin of beauty is something he focuses on a great deal in that book and This is the one that's fascinated me. I mean you write down these equations and in a piece of code you run them and outcomes Hurricanes spontaneously I've been asked people like how do you insert the hurricanes into the model and it's like I Just write down what Newton says and I push a button and the hurricane just appears on its own But thunderstorms the Gulf Stream everything like that. This is a little simulation that reflects that This is showing us high resolution couple climate simulation So the atmosphere is not visible here, but the colors show the sea surface temperatures of the ocean If you look down here, you can see this is the Gulf Stream this thing that looks like an artery Those meanders that's pumping warm water up the east coast of the United States You can see the ice is declining and then begins to come back in November all of this stuff comes out of the equations and You know you look at this on other planets. This is Jupiter This looks like something Van Gogh would have painted The interesting thing there is if you look at that that little tiny hurricane there Is about half the size of planet Earth? So they have really big and that's not even to mention this thing. So they're really big hurricanes on Jupiter Here's a here's a moon of Saturn called Titan These are it's a grainy picture, but it's a miracle. We have any pictures of the surface of Titan This is showing these dark areas. Those are oceans made out of liquid natural gas interesting thing is we've taken climate models change the working fluid to essentially liquefied natural gas and Gotten reasonable climate models for Titan Ice caps this giant thing that looks like a Cinnabon is actually the North Pole of Mars So we're not the only ones with polar ice caps. These are partly out of dry ice and partly out of water But just extraordinary beauty This of course is the planet that has it's all it all the earth Which is of incomparable beauty And then you know this is showing we're changing the earth And we're making it into something slightly less friendly to human beings. This is 1990 moisture in the atmosphere This is what the models show the future to look like within a lot more moisture and if you look at the size and the strength of the Kinds of tropical cyclones that are coming out of that model. It's a little bit fright and with that I want to thank you and take any questions you might have