The future of climate projection
29:05 Share

Hi, everyone. It's Russ Altman here from the Future of Everything. We're starting our new Q&A segment on the podcast. At the end of an episode, I'll be answering a few questions that come in from viewers and listeners like you.

If you have a question, send it our way either in writing or as a voice memo, and it may be featured in an upcoming episode. Please introduce yourself, tell us where you're from, and give us your question. You can send the questions to thefutureofeverythingatstanford.edu. The future of everything, all one word, no spaces, no caps, no nothing, at stanford.edu.

S-T-A-N-F-O-R-D dot E-D-U. Thanks very much. They make up a substantial part of the momentum budget of the jet stream, which is this planetary scale phenomenon. Everyone's familiar with it. If the jet stream, for instance, were to slow down or speed up, the time that you would take to get from New York to London would be very, very different, right? So the momentum budget of the jet stream is in part set by these gravity waves. ♪

This is Stanford Engineering's The Future of Everything, and I'm your host, Russ Altman. If you're enjoying the podcast, and if it's helped you in any way, please consider sharing it with your friends, family, neighbors, and colleagues. Personal recommendations are a great way to spread news about the show.

Today, Aditi Shashadri will tell us how climate projection allows her to look at large areas of the Earth over long periods of time to understand how the climate is likely to evolve. It's the future of climate projection. Before we get started, if you're enjoying the show, please remember to share it with family, friends and colleagues. It's a great way to spread the news about the show and to make sure that everybody is clued in on the future of everything.

So when we think about climate, we think about the weather. We think about what temperature it's going to be today, whether it's going to be windy, whether it's going to rain. But on a bigger picture, we have to think about the climate as it develops over the globe. What are going to be the average temperatures? Are we going to see icebergs melting or forming?

What are going to be the jet streams that move air in large volumes from one part of the world to the other part of the world? That falls into the area of climate projection. Climate projection tells us things about when we're going to have cyclones. Are we going to have 90 cyclones this year or 50 or 150? It tells us about the polar vortex and where the jet stream might be and what that might mean for winter storms.

It might tell us about coastal interactions of the climate with the land and many other climate-based phenomenon. Well, Aditi Shashadri is a professor of Earth Systems Science and a fellow at the Wood Institute for the Environment at Stanford University. She's an expert at climate projection and uses both physics and data to model how the climate changes over time and over large distances.

She's going to tell us about how this works in terms of modeling the physics in boxes of weather, where she measures the temperature, the pressure, lots of parameters. She has lots of boxes, and she watches as climate elements move between and among the different boxes. She's also going to tell us how she got some special free data from a Google experiment that had really nothing to do with climate projection, and yet was a windfall for her.

Aditi, you're an expert in climate modeling, climate projection, and that's different from weather forecasting. So I want to start out just asking, what's the difference between climate projection and weather forecasting? Right. Well, climate projection

projection or prediction and an NWP, which is a numerical weather prediction, is actually very, very similar in that they solve the same sets of equations. And the equations are basically just conservation equations. So if you have a little box of the atmosphere, mass, momentum and energy in that little box has to be conserved and that anything that goes in has to either go out or increase that stuff in the box. However, there's some very important differences in the case of

Climate projection, there's no data that goes into that model typically. The model is just free running, it's just integrating forward in time, often for hundreds of years. In the case of weather prediction, the model uses data assimilation, meaning that as the model is stepping forward in time,

Any possible sources of data are given to it. And also, weather prediction is expected to be somewhat accurate, at least in like a 10 to 14-day range. I mean, when it says it's going to rain, there's a 50% chance of rain in three days' time, we tend to take that number somewhat seriously. It's expected to be literally true to some extent. In the case of climate prediction, yes.

asking the question, how hot is it going to be on June 1st, 2027 is a meaningless question because it's beyond the window of chaos, basically. And so climate protection is more about trying to understand the statistics in a changing climate. So what is a meaningful question is what might June on the average look like 50 years from now or

what is the 95th percentile of rain expected to be sort of on the average in June, 50 years from now. So they're very similar, but they have important differences. Okay, that's great. So thank you very much. And we're talking about climate projection here today. And so you said, well, first of all, you said window of chaos. And that is a great phrase. We have to, I just want to say, let's press submit on window of chaos. But I don't

But I guess it has this technical meaning. Okay, let's go there. What did you mean by window of chaos for your modeling? Right. So that's not a technical term. I love it, though. You should patent it. But what it means is that if you have two models or if you have the same model and you initialize it with chaos,

ever so slight differences in initial conditions, like tiny infinitesimal round off errors in your initial conditions, and you run the models forward in time, they begin to diverge from each other in that one goes one way and one goes the other way. I see. And these errors in the initial conditions in the case of the fluid equations of the atmosphere, the errors double about every five or six days. And

And at the end of 10 or 14 days, the models have completely lost track of the fact that they started off with very similar initial conditions. Okay. So it's a little bit like the butterfly effect that people talk about where a tiny little difference down the line. Okay. So thank you. It's exactly the butterfly effect.

You said a number of other really interesting things. So one of the things you said is that part of an essential component of this modeling is, I think you said, like a volume or a cube of air, like a cube of atmosphere. And so one question is, in your area, what is the size of that cube? Like, I care about the wind in my backyard, so I might want you to make me, like, cubes that are a foot, one foot by one foot, or, you know, 30 centimeters by 30 centimeters. Right.

What kind of scales do you need in order to get accurate climate modeling?

Right. So there's no way that I'm going to be able to tell you anything about the wind in your backyard. So a typical, like a state-of-the-art climate model now will have a grid box that is about 100 by 100 kilometers. Okay. Very large. Very large. Very, very, very large. So if you have cities that are very adjacent to each other, the model doesn't know the difference between them. Okay. And you've actually put your finger on a crucial issue in modeling the climate or indeed modeling any fluid. Right.

which is that there are obviously processes that happen on smaller length scales than 100 by 100 kilometers. And there's also processes that happen on much smaller time scales than the time step of a model. So this is where a lot of people are putting their thoughts and efforts and time at the moment, which is how do we put in some meaningful representation of stuff that's happening that's below 100 by 100 kilometers in length scale or below...

the time step of a climate model, which will be about 30 minutes or so. 30 minutes. So if you think about, for instance, rain, clouds, I'm looking at clouds outside my window, they form on very, very, very tiny length scales. I mean, you have connection going up very rapidly in a tiny little, in a little,

fraction of the 100,000 kilometers. So clouds are a really, really good example of a process that involves all the stuff that's happening at tiny length scales and tiny time scales, which we have no hope of modeling or indeed no hope of ever modeling because of computational constraints.

And by ever, I mean sort of in my lifetime. Right. So, okay, that actually gets to my next question, which was going to be, are we doing 100 kilometers cubed because that's the key length scale that you care about? Or is it limited because of the computational capabilities? So, like, it sounds to me like if you could do 10 kilometers or less, you would. Because obviously there's a lot of 10-kilometer cubes in a 100-kilometer cube algorithm.

I'm guessing that this is computational limitations? It is computational limitations. So the thing with climate is that we need to say something about the climate 50 years from now, 100 years from now, right? And we need meaningful statistics on those things. And so to get those sorts of meaningful statistics, we need to run at 100 by 100 kilometers. That's what's possible. And it's not 100 by 100 kilometer cube. Oh.

Sorry. It's a small point. That's okay. The atmosphere is actually very much bigger on horizontal scales than it is on vertical scales. And so the spacing in the vertical is not 100 kilometers. The entire atmosphere that is modeled is about 100 kilometers. So it's very much smaller than that in the

Also, they're little pancakes, little, little pancakes. They're little pancakes, yes. I love it. Okay, gotcha. Okay, now the other thing you said is that the time that you use in your simulations, in other words, you do a bunch of calculations and then you calculate again for the future, that time is 30 minutes. Where does that come from? How much of that is based on computational constraints? And how much of that is because that's the kinds of questions you're asking?

Right. So with any numerical Navier-Stokes solver, sorry, Navier-Stokes is the equations of continuity, the mass momentum and energy conservation. So with any application, it doesn't have to be fluids on the earth, it can be any application. The inherent length and time scales that the equation solver can deal with are coupled together.

by factors that set numerical stability of your scheme. And so if you specify your horizontal resolution, you're also kind of specifying the time step. I see. And so both of those things are set by how much computing we can throw at it, basically. Okay. So the 30 minutes is related to the fact that if you did it, if you waited an hour, things would have changed too much and your equations might not be accurate anymore. Well, they blow up.

Right, they would blow up. And you could do 10 minutes, but you don't need 10 minutes. Is that true as well? You could do 10 minutes, but you don't need... It's sort of an optimization between what is feasible computationally and what is stable as well. Okay, great. So my final question for like setup is you also made reference to that when you're looking at this volume, there's...

parameters within this volume. I think you said like air is coming in and air is going out and you have to make sure that you don't create air or destroy air. So give me a sense of the kinds of, I'm guessing temperature is one of those things, but I don't know that. So what are the parameters in one of these little pancakes that you're actually modeling or measuring? Density, temperature, the winds in three directions. So U, V, and then W.

vertical wind. Yes. Pressure, temperature, and density are all linked to each other by the equation of state. And then, so those are the sort of dynamical variables. And then we also care a lot about water vapor, right? So how much water vapor is contained in that box? Is it raining out? If it rains out, then the amount of water vapor goes down, right? And cloud fraction is another thing.

And then there's all sorts of other stuff that you can throw in. I mean, you can trace exactly the amount of CO2 in that box, or you can trace the amount of ozone in that box and things like that. Okay, okay. But then we're passing into the realm of chemistry, which makes the model very much slower. Yes, yes. But from a dynamical perspective, which is what I think about, I'm a fluid dynamicist, we're interested in how quickly the air is moving, so the wind, right?

along three directions and then we're interested in things like density, pressure and temperature. Great. So now I know because I've looked at your papers before we chatted, I know that one of the and I know this also just from reading the news, that one of the things that affects temperatures a lot is the oceans. And so do your models have to contain water or some kind of representation of the oceans or not?

Right. So if you wanted to do everything right, you would have the deep ocean coupled to the atmosphere, which also makes things very, very, very slow. There's various ways of doing versions of that. I mean, there's a thing called a mixed layer ocean where you just have the top like 50 meters or 100 meters of the ocean that is coupled to the atmosphere. Another cheat is that you could just specify sea surface temperature so that the temperatures at the base of your model are

A nice thing to do is to actually take a deep ocean model and we call it spin it up. So like run it forward for some hundreds of years and then use that spun up state as the base of your atmospheric model. That's a sensible thing to do.

But I think any question that you ask me is going to open up a whole Pandora's box because with the deep ocean, we think that the deep ocean currents will actually come into equilibrium over more like a thousand year timescale. And so running the deep ocean coupled to an atmosphere over a thousand year timescale probably

isn't going to happen anytime soon. It's just, it's again somewhat complicated. But we think that on decadal timescales, 10, 20, 30 years, the ocean shouldn't play a huge role in terms of atmospheric temperature changes. Of course, it has a role in things like El Nino, El Nino and La Nina, things like that.

but in terms of trends, the ocean changes very, very, very slowly. Okay, so that makes sense. So for a 10 or 20-year period, you can pretend like the ocean is constant, even though you know it changes, but it's changing very slowly. Okay, thank you. So now that took some time, but I think we now have a nice vision of what you're doing. And here's the big question, and this is what gets you excited. What are the kinds of questions your lab can now ask

with the capability to do these kinds of models? Right. So the capability to do these kinds of models has existed for a while. I teach a class on climate modeling and we talk about the first climate model which sort of happened in the 70s. So that's been there for a while. I think there's been a huge resurgence in interest in climate modeling of late.

because of various reasons. So going back to one of the things we said earlier, we think a lot about how these so-called sub-grid scale processes, by which I mean stuff that happens on length scales and time scales smaller than a climate model grid box can meaningfully resolve.

There's various such processes. There's convection, rain, clouds. I think a lot about this process called atmospheric gravity waves. That's great. I wanted to ask you about gravity waves because it sounds cool. So tell me about gravity waves. Okay. So I tend to speak with physicists a little bit from time to time. Gravity waves are not gravitational waves, which are wrinkles in space-time, right? This is not that. Right.

Atmospheric gravity waves are really ubiquitous in the atmosphere. They're forced anytime you have air moving over any obstacle, such as a mountain. There's all this orography at the surface of the Earth, right? And then they're forced whenever there's a storm, like a huge convecting storm, a tropical cyclone, jets and fronts. They're forced pretty much everywhere in the atmosphere. And they propagate dangerously.

both in the vertical and in the horizontal. So you can really tell there's all these little tiny islands in the oceans, right, in the Pacific. And you can tell that those islands exist thousands of kilometers away because gravity waves are forced over them. They travel over thousands of kilometers. And then at some point in the atmosphere, they break, just like waves break on the beach. They break in the atmosphere and they deposit tiny bits of momentum where they break. So they slow down the flow in the box where they break in the model of

And they're a really nice example of how multi-scale geophysical fluid dynamics can be because they themselves can vary over like a meter to planetary length scales. They can go from 10 to the 1 to 10 to the 5 meters. So that means you'll capture some of them in your boxes, but some of them will be much smaller than your box. Exactly.

And then they also have a wide range of variability in time. So they have a huge range of frequencies. So they can be very, very, very, very quick or they can be kind of slow. Okay. So yes. And the thing is that while they are subgrid, meaning that they are not

The entire spectrum of these waves is not well captured in a model. They seem to play a really significant role in forcing the atmospheric circulation. So they make up a substantial part of the momentum budget of the jet stream, which is this planetary scale phenomenon. Everyone's familiar with it. If the jet stream, for instance, were to slow down or speed up, the time that you would take to get from New York to London would be very, very different, right? Yeah.

So the momentum budget of the jet stream is in part set by these gravity waves. I see. As an example, they play a significant role in the polar vortex, which is another feature of the climate that is very close to my heart. I've been studying it for 10 years, but we'll talk about that some other day. Great. This is the Future of Everything with Russ Altman. More with Aditi Shashadri next.

Welcome back to the Future of Everything. I'm Russ Altman, and I'm speaking with Aditi Shashadri from Stanford University. In the last segment, we got a feeling for how climate projection works. There are these little pancake-sized boxes at the scales of kilometers where we measure physical phenomenon related to the climate. We have a lot of those boxes, and we can update the climate in those boxes every 30 minutes, and we can do this for months or

or years, or even longer. As a result, we have a nice picture of how the climate might evolve in the future. In this segment, we're going to talk about some of the application areas. What do you do with this? What kind of things can you study? Things like the polar vortex. In this section, I wanted to ask you about the kinds of climate phenomena that you and your group can study using these tools. So what's exciting these days?

Right. So obviously, we've been working on gravity waves quite a lot. I have been thinking about the polar vortex, as I said, for 10 years. That was what my PhD thesis was on. So the polar vortex is this, it forms every winter over both poles. It exists in the stratosphere, which is the second layer of the atmosphere above where we live. And about every other winter in the northern hemisphere, it does this beautiful, exciting thing where it can break up.

So there's actually two kinds of things that it does. One is that it can break into two little vortices called daughter vortices. And the other thing it can do is kind of just shift off of the pole. It lives on top of the pole, but it can shift off of the pole. And when it does that, the jet stream in the troposphere where we live tends to move a little bit north and south, thus changing where the storms hit over the East Coast, as an example. So you can have really extreme winter weather following one of these events.

So that's something that my group studies quite a lot. I have a student recently who's been very interested in tropical cyclones. So a big open question over as long as I can remember has been what sets the number of tropical cyclones that the Earth sees every year? Oh, yeah. It's kind of fixed, actually. It's fixed at about 90 tropical cyclones every year. And we don't really know why that could be.

So my student Adam has been studying that. He's probably going to finish up with the next year. So this is a good ad for what he's been doing. Yes, good. Way to go, Adam. Right. So with Adam, we've been thinking about what sets this number of tropical cyclones. And he had this very nice result recently where he linked the number of tropical cyclones that we see on the Earth to the latitude of this large band of rainfall in the tropics. It's called the ITCZ, the Intertropical Convergence Zone.

So we've been working off this idea that the intertropical convergence zone gives you these precursor disturbances or I don't know what to call them, seeds. Yeah, little cyclone seeds. Right. And we find that when the latitude of the ITCZ is further north, you get more tropical cyclones. And when the ITCZ is further south, you get fewer tropical cyclones. Right.

And it's an idea that worked out remarkably well in the observations. - Oh, that's exciting 'cause that might actually lead to like, okay, this year based on where our ITCZ is, we're saying that it's gonna be more than 90. So hunker down versus this might be a slightly lighter year. Very interesting. - And also, I mean, it might help in the context of a changing climate because if we have reliable projections of where the ITCZ is gonna be in a given ocean basin,

you can think about how many tropical cyclones you would expect years in advance. If you have some idea of the statistics, as I said before, of where the ITC is going to be. Very good. So this is fantastic. And so who is the reception? So when you do this kind of modeling, and especially you get a nice result, as you just described,

Are there agencies like governmental or others who are the receptacles to your findings? Like, I'm sure that these models then might have practical consequences for like policymakers. And so are people listening and how do how do scientists like you kind of disseminate your findings to the relevant folks?

Right. I do talk to the climate modeling folks quite a bit. I have friends in, you know, GFDL and NCAR and places like that, which actually build climate models. And recently I've become a part of a, not recently, it's been two or three years, but a program that's actually funded by Schmidt Futures.

now called Schmidt Sciences, whatever, to lead to sort of systematic improvements in global climate models. And we all talk to each other and we talk to people at modeling centers.

But in terms of the policy applications, I think people like me need to do very much better in terms of how we communicate. I mean, this is a theme across many of my guests and including myself is that, as you know very well, scientists are not trained to do policy. We're trained to do science. And yet so much of what we do is of interest to policymakers. And that's why I asked. And it sounds like they are watching. And of course, at this stage in your career, you're just trying to get the models to work and do useful things. And then

But as you get these results, such as Adam's results, I can imagine that you will be drawn into these conversations because people will say, oh, this might affect how we actually allocate resources or something like that. Absolutely. So tell me about what are the interesting things on the frontier that are getting you excited every morning?

Right. So an exciting frontier is using a combination of observations. So I, for instance, am using new observations from this fleet of balloons in the upper atmosphere. It came actually from a project that was set up by Google and had nothing to do with atmospheric science or climate. They were just putting balloons into the upper atmosphere to provide internet to everyone below. Yes.

But I got my hands on this data and we're actually using it to provide constraints on atmospheric gravity waves. And it's kind of magical because we got the data completely free of cost. As an example, if you were to put a dozen balloons into the upper atmosphere on a scientific ballooning campaign, it would be off the order of tens of millions of dollars.

And this is a thousand balloons and it's providing pretty much global coverage, not quite, but pretty much global coverage and it's entirely free. Wow.

And new methods such as data-informed methods, machine learning, to infer some of the stuff that I talked about. How do these smaller scales impact larger scales? Yes, this sounds exciting because your initial conversation was all about physics. And it sounds like you're now merging in some new sources of data. And that's actually what you said the weather people were doing. But it sounds like now you climate people are doing this as well. Exactly. Exactly.

Very, very exciting. I did want to ask you also, and I forgot to ask, so I'm glad we have a little bit of time. There's a lot in the news these days about geoengineering of the weather, putting things. It sounds to me like your tools are relevant to those questions. If I put a bunch of stuff in the air on purpose to try to, for example, remediate global warming or something like that. Tell me, tell

Does your work interact with that? And if you have opinions or concerns, I would love to hear your take on this climate engineering discipline that seems to be emerging. Right. So you would definitely need to do a great deal of very careful modeling to try and understand the side effects of doing something like this. So I've said that the atmosphere is multiscale. It is also highly non-local. So if you were to do something, it would affect...

goodness knows what somewhere else. So the most common form of geoengineering is putting sulfate aerosols into the stratosphere. The stratosphere is a bit of a focus for me. And the consequences of something like that would be far-reaching and some of the side effects may be completely unforeseen. So a great deal of modeling and a great deal of understanding, both physical and modeling-based understanding should be built up before we attempt to

So here's a, I don't know if this is too much of a pointed question, but are your models ready to run? Can we add some sulfate to your models? We did talk a little bit about chemistry and I heard you say loud and clear that that makes a lot of the models slow down because you then have to add chemistry. So is this something that's likely to be, um,

something you're looking at and is it within range? And, and, and because it seems to me that you're absolutely right, that we really do need to do extensive simulation before we make such major decisions about messing with these large scale phenomenon. Right. So it's something that can be done now with the existing generation of, of models. And even if you don't have interactive chemistry, you can put in some version of the effects of doing something like this in temperature and

and see what happens downstream, right? And there's also people who are thinking a lot about emulators of models rather than like running the model itself. You train an emulator and then you run the emulator instead of the model. So you move from physics to kind of AI type approaches? Correct.

which the emulator is only as accurate as the model that you train it on. It does well in the regime in which you trained it, and generalizing is a problem. I mean, generalizing is when the model sees something that it hasn't seen before, right? Yes. Yeah, so we're at the point where we can do geoengineering experiments with models, for sure.

It's not something that I do myself. It's something that I think studying it is a very good thing to do. And it sounds like you are contributing to the understanding of how to do these models best. And so if I were a betting person, I would bet that in the next several years, you might be drawn to it just because of the compelling opportunity. But we'll see how that plays out. Thanks to Aditi Shashadri. That was the future of climate projection.

Thank you also for listening to this episode. You know, we are approaching 300 episodes in our archives. And so you have a huge selection of discussions on a wide variety of topics that will get you clued in on the future of everything. In addition, if you're enjoying the show or if it's helped you in any way, please consider rating and reviewing. In fact, we like to get a five if we deserve it. So please consider doing that. It helps a ton.

You can connect with me on X or Twitter at RB Altman, and you can connect with Stanford Engineering at Stanford ENG.

If you'd like to ask a question about this episode or a previous episode, please email us a written question or a voice memo question. We might feature it in a future episode. You can send it to thefutureofeverything at stanford.edu. All one word, the future of everything. No spaces, no underscores, no dashes. The future of everything at stanford.edu. Thanks again for tuning in. We hope you're enjoying the podcast.