Transcription
so I've done something cool with that, you know? Because in many places, literally, I mean, I'm introduced as Leica because the minute there's a silence, right, and I'm supposed to get up and go give a talk, and somebody's sweating, I, oh, I know it's my turn to get up and go. So I started shortening my name, and then people said no, we must have a last name also. So my last name is longer than my, so let me tell my whole name. It's Madula Gatur. So now you can understand why people sweat. So what I did? I, I shortened my first name Madhulika to Lia, and there's a story there. So many amongst here are Indians, and they would know. So the word Madu in Sanskrit means honey, so I decided, and so if you let people to choose, they will just use the first half. And I decided, uh, to take matters in my own hand, and I didn't want any Tom, Dick and Harry calling me honey. So, so that is the source of Lea. Then when I was told, well, we need last name, so I said okay, I'll give you a last name, and it's G. So it's Lea G. And the funny thing is, in India, if you want to show respect to someone, then you add the word g, je, you know, everybody like our current, um, P, Prime Minister Modi, G. So I let you call me Lia G, showing me respect without you knowing it. All right, that's, that's that. So let me get started. I'll try to go as quickly as I can because I think after that, uh, Marco, uh, Nick, not Marco, Marco is gone, but Farzad, Nick, and I will sit and kind of, um, share some questions with you, try to get your, um, impression. And sorry, I don't know, this is moving, this is the other problem. And, um, before I even do that, I just want to tell you this Capstone idea that was shared by this very longstanding colleague of mine, Carl Shriber, without whom probably the summer school wouldn't have existed. Um, came up with this idea, and I learned so much, and I hope each one of you have done the same thing. This was a brilliant idea, and it made me think that, you know, why is habitability part of some other division in the Science Mission Directorate at NASA? Because what you guys presented, comparative heliophysics habitability, it seems like we should own it. So I'm going to make a case, and I'll need your help with that. It was a really amazing, uh, presentation, and what you went through, uh, lot of fun, but lot of of knowledge gained at the same time.
So let me go through this quick update, and I'm going to talk a little bit about what's going on in the division right now, some of the fun stuff, missions, etc. And then I'm going to launch into what I'm doing these days and what I'm really amazed by, and share some of those on artificial intelligence. So this is, this is the, uh, you know, to give you a sense of the heterogeneous data that we utilize, that you have utilized in lot of your work. Um, we, this is fondly called the Heliophysics, uh, System Observatory, our mission fleet. There are 19 operating missions with 26 spacecraft, because some of the missions of multiple spacecraft, you know, literally crisscrossing the distance from Sun to Earth to all the way tittering in the interstellar medium with the Voyager, uh, spacecraft. And this is, this is what provides us the data that we need for all our science.
This quickly gives you what's coming up. This is a mission, uh, Heliophysics Mission timeline. This is really, um, cool because you can see between now, 2023 to 2025, we have a significant number of missions, and I'm involved. I'm the program scientist on two missions. PUNCH is one that's actually, actually going to be launching not April, these are, you know, sometimes these charts I pull out for from something else or a little, um, dated, but currently PUNCH, um, launches on February 2025. But there's just many more happening right now. We are all kind of very excited about ESCAPADE that is, uh, going to go up very soon. So some updates, um, you know, some you might be interested in. I don't know how sort of closely you follow NASA, uh, mission launches. Our Explorer Program, uh, Solar Terrestrial Probes Program or Living With the Star Program. They are, these are the programs that create missions, and Explorers are the most fun because here a single principal investigator, a researcher, can kind of generate his or her own idea of science and then propose that and then get selected. So this is, I'm just sharing some of that. Uh, there's new stuff going on in the world of Research and Analysis where we are really trying to create a digital library, a infrastructure for our data, and we are also putting more emphasis on the tools of artificial, uh, intelligence. Uh, so these are samples. This is not, not um, really, uh, from this year. This is 2023, but 2024, you know, most of these don't vary too much, and this is Research and Analysis. So when you graduate and you find a job in this country, of course, then you can start proposing to some of these, um, research elements as you are seeing. And so, uh, you can see that there is a multidisciplinary component there where habitable worlds are part of it. So it's, it's not kind of completely out of the picture; we recognize it. But we sort of manage that portfolio with many other divisions because they, of course, play a pretty significant, uh, role.
So this is another kind of new initiative that is going on, which is, um, Transform to Open Science. TOPS. If you attend AGU meeting or many other big, uh, meetings, you will hear lots of presentations here. And it's not like we were not doing, and my question was like, what were we doing before if not open science? But the reason we have created this as an initiative is to put greater emphasis on sharing data, sharing model, you know, um, sort of accelerate major scientific discoveries through, through supporting this adoption of open science. You have to work hard with community; it's a large community, and it's not just Heliophysics Division, right? At Science Mission Directorate, there are five divisions in the Science Mission Directorate, Heliophysics being one. So we want to broaden participation by, again, historically excluded community. So there are multiple goals here, and this is really being driven as one, one entity, one unit. It's not like every division does what it wants to. So this is, this is kind of important. And I'd say go to some of these websites and familiarize yourself with what is available from, in this case from NASA, but you know, NSF will have its own. So it just doesn't have to be NASA, but I'm, I'm giving the NASA overview because this is a, happens to be a NASA summer school.
So, uh, this is kind of another area where we have invested some thinking and resources, uh, which is Heliophysics Big Year. And the reason for this is really to globalize our science. And even in this room, you know, we have had sort of interesting heated discussions with deans, with faculty members, with the idea, aidea of the choice of the word heliophysics as a discipline. Heliophysics discipline will be 20 years old next year. When our division adopted the name as Heliophysics Division in 2005, before that we were called Sun Earth Connection Division. You know, I was running the Living With the Star program, which is an integrated program, all, all the subdisciplines within heliophysics. It seemed like we needed to give heliophysics sort of an academic qualification, and that required the first thing in any kind of discipline is you must have textbooks. We didn't have textbooks in heliophysics, and that was created at the summer school. And now the textbooks that you are seeing is the origin of the first many summer schools, which got compiled and, uh, rewritten. So, but still, the word heliophysics, when you write it, what doesn't recognize it? I mean, it's not a word, right? You get a squiggly line. Like, if I write my whole name, I'll get that. I think it knows me as Lea better right now. It's easier, probably. So we have to kind of solve all those issues. It's still not very well known. Our science may be one of the more difficult ones to communicate to general public. So the whole idea was to kind of communicate and globalize our science, and it was capped around some really amazing events, uh, that happened: two solar eclipses. I mean, that, that is, we have to be really, really lucky to kind of have that happen in our country. So the annular eclipse of October 14th. How many of you saw that? October 14th? Okay, few. And how about the real one, the total solar eclipse? Well, I was there; I didn't see it because, you know, cloud came in the way. So we started this effort in September of 2023, and the goal is to end it with Parker Solar Probe's closest approach to the Sun. Like, how amazing is this, right? The division can boast the farthest object in space, the Voyager spacecraft, and the closest – this is what we were missing – the closest object in our Solar, uh, closest to the Sun object in our solar system. And that'll be Parker Solar Probe perihelion in December of 2024. And then, of course, to cap all of these, there is all the excitement of solar cycle 25, solar maximum. We are seeing auroras in a way we haven't before. It seems like Earth's magnetic field is doing something. I mean, don't believe me, it's just a thought in my head, because all of a sudden, you know, kind of seeing aurora in Colorado, this part South. It seems like almost routine. It's not, but it seems that way. A lot of people have seen auroras. So how many of you have seen auroras in May, June, July, you know, the two big ones? So let's see. I mean, again, I missed that because I was in DC. I, I choose my places very strategically.
So these were the paths of the total, U, total and annular eclipse. So you can see lot and lot of people actually observe these, and we did have lot of presence in all these places doing citizen science and outreach to communicate our science in as simple terms as we can. And that is the job of each one of you. It's not up to us bureaucrats or faculty members. Each one of you have to talk to people to tell what you do, and get the word heliophysics in there, you know. So it'll, it'll get there eventually. So this is the Heliophysics Big Year, you know, human centered cross-coupled system. We've done a lot during eclipse, and I'm not going to show you statistics or anything. We are investing a lot of resources on citizen science. So again, I would urge you to go to the NASA website and look at stuff and see whether you have ideas that you can actually propose to, and you can talk to Vincent; he is known as the Aurora guy. And then, of course, I showed you kind of all the missions that are happening in 2024 and 2025. There's a lot to talk, talk about. It's never a dull moment in heliophysics. Not going to talk about this, but I have this right, and you'll have access to these. These are some of the citizen science projects that we are supporting during the Heliophysics Big Year. Take a look and see how you may contribute. And this is again various different websites that you can go and figure out if you find something interesting and how you may contribute.
Okay, now I get to kind of the next phase of my talk, which is application of artificial intelligence, machine learning, deep learning to utilize this big, you know, system of fleet, you know, spacecraft that you saw, Heliophysics System Observatory, that data, and accelerate science discovery. And so, to talk about the scale of data, I mean, if you can even see the little sort of squares right around, these are most of our missions. And as you can see, SDO, Solar Dynamics Observatory, whose images you see all the time, kind of occupies the biggest sort of quadrant there. DKIST is the ground-based solar observatory which is not full disk; it's seeing the Sun through a soda straw, unlike SDO which is the whole Sun, but that's even greater, much greater. So that's what's happening when you look at remote sensing observations. So right now, you know, uh, the AIA images – not all of the instruments on SDO, but just Atmospheric Imaging Assembly – they have about 8.7 petabytes of data uncompressed; that is about 20 petabytes. Talked to Google recently and said okay. And so think, think of the resolution of AIA: it's 4K by 4K, or HMI 4K by 4K every 12 seconds, I believe. Uh, so the question was, you know, what would it take to host this data on GCP, on Google Cloud? There is some, you know, competition calculation for us and said, oh, this will be the second largest data set for us if we did it. So that, that's kind of what we are dealing with. How do we do this in a way data is accessible in form and it can be used more efficiently, effectively, and sometimes in its entirety if it's needed? This is the other kicker: we take approximately two gigabytes of data every 15 seconds, and this is 100 plus – I don't know what the total number is, 20, 130, something crazy like that – operating missions. Now, this is all of Science Mission Directorate right now data, and, and we do this every hour, every day, every year, and as you can see, the collection rate is just growing exponentially. So handling, sorting, managing this data, this is a massive, um, you know, challenge. And you can imagine that our data – why do we do missions? For observations, for data. And so our data is one of our most valuable assets, and its strategic importance in our research and science, uh, is huge. Sometimes you have to show something in very big, bold letter so it just kind of drives the point home what we are dealing with now.
Uh, this is where I get to sort of, you know, the idea of artificial intelligence. And I know many of you kind of having probably experimented with it or even a little bit worried. The word "artificial" throws people off a lot, but there's nothing artificial about it, you know? I mean, AI is basically probabilistic statistics we created. So I have started using the word "augmented intelligence." Much like a telescope – we couldn't see what we see, we didn't know what stellar structure was, right? And we have done that with the aid of telescope. Or think of microscope; you can go anywhere. These tools are augmented intelligence, and AI just happens to be one. And let's use it for good. And you know, space has so much data, as I just showed, that, um, application of AI and deep learning in particular in different ways can provide amazing information, especially when we start looking at contiguous data sets over, uh, decades. Now this is why I like this, um, uh, GIF that I am trying to show. So some other values of AI is that it can repeat onerous human tasks, things that used to take scientists, you know, months, years, we can do it in a jiffy. But it also helps us make sense of complex patterns which really is beyond our human senses. We don't think about it, but we observe the universe with our limited perception, and AI is actually providing us with tools to infer patterns that we can't see. And so you're looking at this, right? This, this multi-dimensional GIF is incredible. When you see those dots randomly moving, you can't see the pattern, but when you create actually a training module, it can actually pull that out. It's very simple, but it, I, I think it, it just really makes you think what good tools can do for us when it comes to data.
So diving right into it, so since about 2017, I took an excursion from Heliophysics Division at NASA and went to NASA as in the Silicon Valley for about four years, you know, trying to work in the world of public private partnership, when I was literally introduced to the area of the potential for AI. So these are still pretty early days. So we created something called Frontier Development Lab. You can do Google Search, find it. We, we are as we are maturing, we give it many different names. This year's activity was called Helio Lab under FDL. So we've been focusing more on that. But this is the – I'll give you some, I'll try to give you some. So, um, Nick, tell me when I should stop so that I, we have enough time for discussion. I have a lot, so tell me how long. I'll try, if I take extra two, allow me. So I'll randomly kind of show you. So this is, this is actually we asked the question: can AI help us essentially, uh, find the complex structures in the solar wind? The answer is hell yes, and it can do it in much, much shorter time scale than it would take, uh, an ordinary researcher with our traditional tools. And what you're seeing, you know, what, what it, it identified is this, um, different sort of clustering of magnetic field structures in the solar wind, and those four top ones are some of the classic ones, right? So we have on the top panel, from left to right, we have interplanetary coronal mass ejection, first detection 1971 done by traditional, uh, approach; interplanetary magnetic field enhancement, which is the second panel, first detection 1983; and remember, these are all from our prior missions which are, of course, retired, no longer exist. The third one comes from actually Parker Solar Probe, which is Switchback, and many of you are working on all of this, right? Course detection 2019. And I put a question mark there. You know, this was just sampling of data by students, early career PhD students or postdocs, over eight weeks of summer, and we come up with something like phenomenal. So there's so much work to be done if you want to identify new things. This is one approach, not the only approach, but we should not be scared of experimenting.
The next one, this is kind of also very interesting. We know how important surface magnetic field, photospheric magnetic field is for any kind of problem in space weather forecasting, right? But the challenge is to create a homogeneous set of magnetograms. So over the last 40 years, we have had lots of ground-based data, lots of incredible space-based data, but they are not homogeneous and therefore absolutely not suitable for long-term, you know, studies. So the question again we asked: well, can we take all these data, you know, first of all sort of render them ready, so you are actually going to, um, cross-calibrate them and, uh, denoise them, do all of that, so you can, that's easier to do right now. And then the next question was: what else can AI do? So lot of the ground-based observations are lower resolution. Are missions like SOHO MDI, it's much lower resolution than SDO HMI. So the question is: can we, for example, take a low resolution data and upscale it? And we can do that too. And so I'll show you, and I, I don't have much time to talk about it. The paper is here if you want to read. So there is the input data, right? SOHO MDI, lower resolution, that was the model input. The target is a SDO HMI, which is much higher resolution, and the ML output, as you can see, is beginning to mimic the SDO HMI, so have upscaled it. There's another ground-based example of super resolution: that is the GONG data that many of you use, GONG input, and then there is the ML GONG output. What else do I have? Uh, this is a very cool one.
So this is one of the first things we tried, and I'd say go look at the charts and get some of these papers and start looking at it. So Solar Dynamics Observatory launched in 2010, and it has three instruments. One of its instrument, EVE, lost one of its sensors called MEGS. And you have seen in the presentation that you have heard from many of the faculty members here, it, uh, there was an electrical malfunction and MEGS stopped producing data. And MEGS is one of the more important sensor data because it measures spectral irradiance, and spectral irradiance really directly contributes to the atmosphere. So the question we asked: okay, you know, Solar Dynamics Observatory has AIA with nine, ten filters, you know, collocated, taking observations at the same times; these are also extreme ultraviolet images. So can we utilize this data to create a, a pipeline? You know, you have to train, validate, and then predict, and so that it would output the spectral irradiance as if today, as if MEGS was still operating. And that, that's what you're seeing. We actually created that, and the result is better than physics-based models. It, it's, it's pretty incredible. But what, what's, what's important here to remember is that we just can't do this, um, you know, out of the blue, right? We have to experiment, test, train all that, but we get spectral radiance only if you also have the AIA images. So you need your input in some form or the other.
This is another one. This is a recent one. This is actually creating, um, lots of interesting branching out: how can we observe the Sun and solar radiance from any perspective? And, um, I don't know if Robert Jari, who is now, um, I think a Fellow at H, uh, he was one of the summer school attendees during the, uh, during the COVID time virtually, and he's contributed hugely to this work. Okay, and we only have data in the ecliptic, which is SDO and STEREO, but with that we can actually create a 4 pi UV radiance as you are seeing in this picture. It, it's again just math, neural radiance, uh, fields, where you can create, you know, from multiple input images that renders a 3D visualization. There's lot of stuff you have to do and understand, but you are actually seeing the full volume of the Sun, um. This is the radiance that we calculate, just like we did for SDO, and you can do this at any spot in the solar system. But remember, as long as you have data, you can do it, right? The irradiance extraction and all of that. So collecting some observation is still very important.
And so I know that Nick has stood up, so I'm not going to go very far. I'm just going to show these global geomagnetic perturbation forecasting using deep learning. I'm just going to show you the results, not that we are getting results, you know, that's better than any state of the art. This is giving us 30 minute prediction of geo-effectiveness globally for the first time, predicting dB/dt just utilizing data. There's more. There is actually, you know, inferring neutral density because we don't have actually neutral density measurements in the ionosphere, and we are doing that by inputting the solar input directly as opposed to proxy, taking all the OMNI data at L1, taking the magnetic field data at ground, and what we are getting is significantly better than any of the models we have today. So I'm, I'm just, I'm very quickly going to talk to this.
So this is kind of where we are right. It is important to think how we are going to mature AI/ML. So on the top, what you see those are like, um, solar terrestrial interaction, safeguarding life, to these are our core kind of objectives in the division: discovering the secrets of the universe, you know, fundamental physical processes, searching for life elsewhere, you know, that's the exploration. Heliophysics where we, r, is in that red box, and we really need to kind of create, um, uh, increase the readiness level of these kind of tools so that it can be globally used, globally applied either for discovery or for exploration. And this is just one example that I gave you, right? This is the super resolution, so you can see what space it occupies, and we, we are making significant progress. That's all I can tell you. But this is a space that I'm working in and trying to mature and support. And I think this is my last slide, very likely. This is the Heliophysics, uh, Division AIML strategy, um.