Weather forecasting drives billions of economic decisions — from grid operations to evacuation planning. Better forecasting could improve supply chain planning, disaster warnings, and renewable integration. The industry has decades of satellite observations and ground measurements, making it ripe for AI-driven advancements.
And it’s already happening. But how exactly does AI get used in weather forecasting, and how does it actually lead to improvements?
In this episode, Shayle talks to Peter Battaglia, senior director of research at Google DeepMind’s sustainability program, which launched a new AI-powered weather forecasting model in November 2025. They cover topics like:
- Why precipitation is so much harder to predict than temperature
- How the weather industry works, with governments creating global models and private companies refining them for specific use cases
- What AI models can see that traditional supercomputer simulations can’t
- Novel sources of data like cell phones, door bells, and social media
Resources
- Latitude Media: Where are we on using AI to predict the weather?
- Latitude Media: Could AI-fueled weather forecasts boost renewable energy production?
- Catalyst: Specialized AI brains for physical industry
Credits: Hosted by Shayle Kann. Produced and edited by Daniel Woldorff. Original music and engineering by Sean Marquand. Stephen Lacey is our executive editor.
Catalyst is brought to you by Uplight. Uplight activates energy customers and their connected devices to generate, shift, and save energy—improving grid resilience and energy affordability while accelerating decarbonization. Learn how Uplight is helping utilities unlock flexible load at scale at uplight.com.
Catalyst is brought to you by Antenna Group, the public relations and strategic marketing agency of choice for climate, energy, and infrastructure leaders. If you’re a startup, investor, or global corporation that’s looking to tell your climate story, demonstrate your impact, or accelerate your growth, Antenna Group’s team of industry insiders is ready to help. Learn more at antennagroup.com.
Transcript
Tag: Latitude Media: covering the new frontiers of the energy transition.
Shayle Kann: I’m Shayle Kann, and this is Catalyst.
Peter Battaglia: We don’t really understand how the AI models forecast it, but they are capable of treating the hurricane as almost like a large macroscopic scale object that is moving.
Shayle Kann: They have spatial awareness in a way that the old models didn’t.
Peter Battaglia: Yeah. It’s a really interesting area, I would say, of the science of how AI works to understand exactly how they see the world in that sense.
Shayle Kann: Coming up, where the winds are blowing using AI for weather forecasting.
I’m Shayle Kann. I lead the early stage investing practice at Energy Impact Partners. Welcome. All right, so here’s a statement that I suspect would be pretty non-controversial. AI will improve weather forecasting. It’s obvious, right? And it seems like it must be true. I certainly would’ve agreed with that statement had you asked me before this conversation you’re about to listen to. But to me, the interesting question is why exactly? Through what mechanism can AI improve weather forecasting? For that matter, how do we actually do weather forecasting today? And if it does get better, what are some of the likely outcomes that it will enable? It’s an interesting set of questions for me for two reasons. First, weather forecasting itself is important to a whole host of other categories I care about. Obviously resilience, but also energy and a variety of others, agriculture, et cetera. But also, it’s interesting because I think it’s exemplary of a whole host of next wave applications for AI.
LLMs are, of course, finding their way through everything that requires language. Now there are world models starting to show up to try to revolutionize robotics and things in the physical world. But what about things like weather where we have used some machine learning historically, but can we do better with transformers and the new architecture of AI that we’re seeing in other categories? Let’s find out. My guest today is Peter Battaglia. He’s a senior director at Google DeepMind, where he is leveraging the big brain inside the DeepMind to improve weather forecasting. Here’s Peter. Peter, welcome.
Peter Battaglia: Thanks. Happy to be here.
Shayle Kann: All right. Let’s talk weather forecasting. I want to start maybe by having you school me a bit on something that I realize I don’t know, which is how do we do weather forecasting currently and maybe a little bit of history. Have there been major shifts technologically in how we forecast weather historically? So maybe walk me through the history such as it is of how we forecast weather, and then what do we actually do today?
Peter Battaglia: Yeah. So I have to admit, I’m actually a relative newcomer to the area of weather forecasting myself. So we had gotten involved in this several years ago, and it was sort of built out of a research program that was trying to model complex simulations, including fluids. And the earth’s atmosphere is a fluid. And one of the big challenges that we were sort of interested in exploring was modeling the atmospheric fluid, which is weather forecasting. So I should say that I sort of have gone through this journey of learning about weather forecasting. So the stuff that I’ll say, hopefully it’s accurate, but forgive me if I make mistakes. So I think my understanding about the field is that really a lot of the … Historically, weather forecasting was very important for agriculture and other use cases that were very important for day-to-day life. But I think it was about maybe 100, 150 years ago that you had agencies or bureaus that were starting to do marine forecasting or more systematic collecting observations systematically and treating it as a science.
But then probably about 50 years ago or so, you started to see the emergence of large government public weather agencies. So I think NOAA in the US was formed in the ’70s. I think ECMWF, the European Center for Medium Range Forecasting was also formed in the ’70s. And these are two of the big prominent weather bureaus, but most governments have a weather bureau and weather has traditionally been viewed as a public good. So this is something that they collect tax money and then fund their weather service. And the idea is that a lot of the … It’s not only is forecasting the weather useful, just again, like you’re going to have an umbrella or wear a coat, but for things that are more like there’s a dangerous storm coming, flood, extreme heat, extreme cold, those types of things, and also a lot of decision making like agriculture, energy, transportation.
And I think in general, weather forecasting has traditionally been understood to be something that’s a very good investment on a dollar. So the public tax money that’s invested in the National Weather Bureaus has significant economic returns on those investments. So that’s kind of, I think my understanding of the history of the kind of standardized or official weather forecasting business in a sense. Maybe one other thing I can say about the way to think about the industry, I think my understanding is you can kind of think about the weather forecasting industry is divided into … It’s almost like a pipeline, really. There’s this sort of government official weather bureaus that are issuing these global forecasts that are predicting all sorts of weather variables, but typically more at like a coarser spatial resolution. And then you have this big post-processing chain where they take the sort of base forecast and then they specialize them for different use cases.
So for example, when you look at the app on your phone and you see the chance of precipitation, that’s not coming directly from NOAA, that’s coming from other intermediaries that are taking local weather station data and other historical information, trying to tweak it and improve it and make it especially useful for your use case. And you see that sort of in energy and all sorts of other applications of weather forecasting.
Shayle Kann: Is there a corollary to how that has worked historically to what’s happening with LLMs today? Not to jump into the AI stuff too early, but just in the sense of like, is the global forecast, let’s say NOAA’s forecast, is that a big mega model that spits out this one big forecast? Yeah. And then what people are doing in the processing world is saying, okay, I’m going to take that model, but then I’m going to fork it is the wrong word, but I’m going to fork it and add a bunch of additional data into it to try to make it better at a smaller spatial resolution. I’m just trying to picture what it actually is.
Peter Battaglia: Yeah. So I mean, I didn’t say much about where your actual weather forecast comes from in terms of technically. So maybe if I say that, then maybe it will sort of open the answer to that question. So again, fluids, like the atmosphere is of fluid. And in physics, we have fluid equations called the Navier-Stokes equations, and they govern fluids at all scales, like from the largest scale structure of the universe, which actually turns out to also be a fluid down to what’s happening in your blood, basically. There’s turbulence that determines sort of how your blood flows and it has kind of important implications. Now, all things in between that, right? You have weather and stream flow and other types of things like that, all fluids. And what happens, engineers have figured out that, well, so the fluids are very complicated to simulate. So in order to simulate them accurately, they need to approximate the solutions so that they can run them on very large computers because they’re so complicated they would never run on a computer natively.
You have to sort of break up the computation and approximate certain things in order to actually model everything that’s happening, for example, in the atmosphere. So that’s called numerical weather prediction. The numerical is just saying that they’re making a numerical approximation to these Navier-Stokes equations. And traditionally, it’s been run on supercomputers. So a lot of the big supercomputer centers have either do a lot of weather forecasting or even built to do weather forecasting. And in many ways, it’s been sort of a triumph of science and engineering that we’ve been able to decade on decade predict not one day, two days, but like 10, 12, 15 days into the future. It’s hard to even sort of imagine the scale of that type of predictive accuracy was sort of unimaginable a hundred years ago. People just didn’t think in two weeks we can kind of know what the weather’s going to be.
That’s crazy. That relies on knowing what’s happening on the other side of the earth as the sort of prevailing winds carry the moisture and the temperature and all that kind of stuff.
Shayle Kann: And I imagine there’s like an exponential increase in complexity the further out into the future you get just because there are a variety of possibilities of what actually happens today and each one of those needs to be taken into account when I’m trying to predict what’s going to happen tomorrow and so on and so forth as you move into the future.
Peter Battaglia: That’s the butterfly effect, right? It’s that a butterfly might or might not flap its wings and then that will determine like a week later whether there’s a hurricane or not. So the idea is that little tiny changes or little tiny effects or missing effects could cause huge changes in weather over time. And that’s exactly right. So the atmosphere of fluid is thought to be chaotic, which means that’s sort of the definition of chaos. It’s that little tiny changes can have huge, large impacts later. That’s what makes it so hard. And there’s coupling across scales. Again, the butterfly flaps its wings, but then you have like up in the top of the atmosphere stuff is happening. Now, that is exactly why it’s very, very difficult to sort of find solutions to the exact equations that govern the atmospheric fluid. So we have to make approximations and we use supercomputers and we have all kinds of tricks.
I should also say another important thing to recognize is that when a weather forecast has been generated, the actual prediction of the future is only half of the process. It’s the second half. The first half of the process is figuring out what the weather currently is. So if you, we have satellites and we have weather stations and balloons and ships and all sorts of information that are taking measurements of like what the weather is all over the earth. But again, using the butterfly as an example, you would have to know like where every butterfly is in principle to act to perfectly forecast the weather. So if you sort of think that through, you realize that it’s not really … Weather forecasting is always going to be fundamentally uncertain to some level. We’re never going to be able to make perfect observations of the weather everywhere on earth with the precision required to perfectly predict the weather a week out.
And so when weather forecasting, that’s why you have a chance of rain versus like it’s definitely going to rain and you have like a range of temperatures, especially as you go out in time. And that’s again, sort of what makes weather forecasting so hard. So the first step in weather forecasting isn’t actually predicting, it’s taking all the satellite data and all the stations and all the different observations and estimating the current state of the weather across the earth. And once we have that estimate, then we can make the prediction with the supercomputer. And that second part is what our team and a lot of the teams in the field who are working on AI-based weather forecasting have been especially focused on. But my guess is that over time, we’re going to see other parts of the weather forecasting process, having more and more AI methods that are coming in and trying to advance them.
Shayle Kann: Before we get into the AI methods, it seems like we have generally, even pre-AI, we’ve been getting … I mean, you tell me if the curve has been linear or exponential or flat, but it seems like there’s been, I don’t know, fairly linear improvement in our weather forecasting ability for decades. We’re getting more precise. We are also getting better at predicting further out into the future, as you said, like a week, two weeks, et cetera. To the extent that that’s true, I’m sure it’s all these things, but how much of the improvement that we have seen historically has come from, I don’t know, A, as you said, just having better ground truth data on the current state of the weather, B, more compute, as you said, has been running in supercomputers. So we get more and more powerful computers. We can just run more and more complicated Navier-Stokes equations, or C, additional tricks basically, that allow you to do better predictions without adding more compute.
Peter Battaglia: Yeah, that’s a great question. So I’ll just admit, I don’t know the answer to that. I think all three contribute. So I can say on the first one, data, yeah, there’s better satellites that are flown and there’s more better systems for collecting balloon observations or these different sort of things. So we definitely are getting better data and we know that that improves the quality of the forecast. We are also getting better models. That’s definitely true as well. We’re building bigger supercomputers. They can operate at finer resolution. Just I think in the last less than 10 years, the ECMWF, which has the best weather forecast, they increased the resolution, meaning that they had finer detail and space in their forecasts and that allowed the forecast to be more accurate. So you see both adding just raw compute power, but also improving the quality of the models and the approximations has also made, I think, a pretty dramatic impact.
And I think that sort of blurs into your third category of other tricks. I think in general, without getting into the details of how the numerical models work, you can kind of think about them as a backbone that’s making a sort of general prediction at a core scale. And then you have a lot of other parameterizations and trickery under the hood and making finer and finer grain predictions and also updating the backbone to be consistent with the fine grain. And those are all being advanced in parallel. And these engineering teams and scientific teams are sort of working together to make these better. The last thing I would also say too is, again, going back to that post-processing part of the weather forecasting pipeline, it’s not just, again, that these large NOAA and ECMWF and these other large agencies, it’s not just that they are improving the forecast, it’s that other parts downstream and post-processing are improving what they’re doing.
So actually the first advent of AI and machine learning in weather forecasting, or at least some of the earliest was not trying to overhaul the whole weather forecast process itself, but using more and more statistical methods and linear regression and non-linear regression, neural networks, and other types of earlier machine learning techniques to improve not the base forecast, but the specific application. So maybe we can calibrate your chance of rain better if we have a slightly better downstream model.
Shayle Kann: So, okay. So we’ve been improving, we’ve been in more recent years applying the earlier versions of ML to continually improve. I guess I’m curious from your perspective what the biggest gaps are. I mean, obviously we don’t have the ability today to generate a perfect forecast three months into the future. It could always get better, but apart from just that element of it, are there any areas where you feel like actually it’s really hard to do X? Is it like precipitation is a bugaboo or something else, right?
Peter Battaglia: Yeah. I mean, I think there’s sort of two ways to answer that. You’re always going to be limited by the quality of your data. So if you don’t have good data about something, you’re just bad, garbage in, garbage out sort of thing. So these models take an estimate of the state of the current weather and then predict what’s going to happen. If your estimate isn’t very good because your raw observations weren’t very good, you’re not going to get a very good forecast. So improving, just collecting more data and using the data you have collected to form a better estimate of the current weather, that’s definitely going to always improve things. So it’s sort of a known gap, right? Now we don’t know exactly what the ceiling is. We don’t know if we do this satellite or that station observation, how’s it going to improve things?
We might have an idea, but we don’t always know and sometimes we have to just test it out. But the other thing I would say is that you have different features of weather, which are harder or easier to predict. So an obvious one is temperature. So temperatures sort of vary smoothly. If you look at a map of the temperature across the earth, it’s not up a mountain, it’s going to be colder and in a valley it’ll be different, but it sort of varies smoothly. What doesn’t vary smoothly is precipitation. So like a rapid, violent thunderstorm that sort of emerges out of nowhere and there’s like high wind and low pressure and all this kind of stuff, like where exactly that front will be, like where exactly the precipitation will happen, what exactly the wind and these kinds of things are much, much harder because the detail of like everything’s happening at a finer scale.
Even if you look at a radar map, you can see this. It’s not like precipitation sort of varies smoothly over the earth. You see like there’s a little thunderstorm right there or like there’s a rain and then a few miles over, nothing. So that type of very high resolution, complex patterns of precipitation, for example, wind as well, those are much harder to predict because you’re effectively predicting a lot more information. You can’t just sort of summarize it by saying, “Oh, every 25 kilometers or whatever, the temperature is this, and then everything else is just kind of interpolated in between.” You have a lot of stuff happening at a much finer scale, finer scale than a lot of our models even capture. And then we have to do secondary steps to try to resolve those finer details. All
Shayle Kann: Right. So let’s talk about AI then. I mean, you mentioned this is one thing, right? Whenever we talk about AI quote fingers applied here, there is ML as a subset of AI is related entity. We’ve been doing ML already. So I guess the first question that I have is, as you think about leveraging AI now and into the future for weather forecasting, what version of AI are we talking about? Like what version or versions are you actually … What are the actual capabilities and/or model structures that are interesting here?
Peter Battaglia: Yeah, that’s a good question. So these days, I mean, AI is a pretty casual term. I find myself just using the word AI just to mean a lot of different things because I think it’s kind of easier and usually people kind of know. But the way I would say it is the difference between AI and so machine learning is sort of the statistical inner core of AI. It’s trying to taking data and trying to capture the patterns through a training process and then kind of use some inductive assumption that like, “Oh, what we’ve seen in the past is going to be similar to what we see in the future.” Modern AI I think is a broader family of things. It sort of involves agents and your interactions with them and a lot of language models are often sort of associated with AI. What we use in our weather forecasting models, and a lot of folks out in the community are using as this field is advancing and this AI-based weather forecasting is developing, we’re still mostly using fairly traditional machine learning, supervised learning.
So supervised learning just means you take a data set that has a pair of examples, an input example and a target example, and you train a model to try to take input examples and accurately predict the target examples. And so if you think about weather as, again, like I said before, you’re estimating the current state of the weather and then the next step is to predict what’s going to happen next.
A supervised learning method can be trained to do that. And that’s what we’re doing in our models and like most folks that I see are doing as well. And then the only extra step is that just makes one prediction, but then we feed the output of the model back into itself and then we have it make another prediction. So the output becomes the input and then it generates another output. And if you just sort of chain those steps together, you get your first input and then you get a sequence of outputs that represent future steps in time. And so supervised learning, we use a lot of, these days in terms of AI architectures, transformers and graph neural networks are what we use. People use convolutional neural networks, but I don’t feel that neural network architectures these days tend to be the sort of exciting part.
It’s usually more of like the training and the sort of data, how you handle the data and that kind of thing.
Shayle Kann: So, but you mentioned transformers, because I guess if we had been having this conversation five years ago, I imagine that you still would’ve told me about supervised learning, for example, right? That’s not new. Transformers had been invented by that point, but had not been broadly applied the way that they are today. So what is it that this new wave of AI unlocked by things like by transformers and convolutional neural networks and so on? What does that enable above and beyond what you would’ve been able to do five years ago?
Peter Battaglia: Yeah, that’s a good question. So the way I look at it is, so transformers are very similar to graph neural networks. Both of them are … So actually, let’s take this back. So we used to use often convolutional neural networks. And the idea here is it learns a little function that’s sort of local in an image and then it sort of applies that same function everywhere and then you stack up sequences of these layers and that eventually lets you like the information on one side of an image communicate with the information on the other side of the image because it’s a hierarchy. A transformer architecture allows you to make a direct connection between the information on one side of the image and the other side of the image, the same way that graph neural network does. And I like to think about it like graph neural networks because what it’s like saying is, well, in a graph you have nodes and you have edges or connections between the nodes and a longer connection between nodes is for nodes that are farther away and shorter connections are for nose that are closer.
So if you use the graph neural network analogy to describe the older convolutional networks, it’s like the graphs are all small. Everything’s kind of close. It’s like nearby in an image. Graph neural networks allow you to choose how far away you want information to interact. And in transformers, it can be understood as a graph that has connectivity across any spatial scale. So in language models, the way a transformer works is it says, when I want to make a predictions about the next word, I want to be informed by the most recent word, but also every word that has happened in the text before. And that’s important because in language, the next word is not predicted by just the previous word. It actually is predicted by stuff that happened earlier in the sentence or in the paragraph or in the book. And so the ability to have information across large spatial scales that interact with one another, that ability allows you to … It opens new patterns of computation and allows you to represent functions that have traditionally been harder to represent, but it allows you to make better predictions of the next word or the next … In our case, you can kind of think about it as the next spatial point or a far away spatial point.
And that allows the model to be more flexible and capture richer functions.
Shayle Kann: There’s a good comparison there. What do you think of as being … I guess you just described something that is similar about what you can do in weather forecasting thanks to a transformer architecture to what you can do with large language models, which is what most people are going to be most familiar with in the new wave of AI. What’s different?
Peter Battaglia: So that’s a great question. I think what’s interesting is that the way that … So in language, the text is understood to be, is treated as a sequence. It’s like token, token, token. We are also modeling sequences in weather, but we’re not allowing our models to look too far back in time. So because weather is actually different from text in a fundamental way. In fact, most physical processes are. They are what’s called Markov in that the most recent state of the system determines the subsequent state. So like I said, in text, that’s not the case. Right now, I’ll just pause … You didn’t know what word I was going to say next, right? It kind of depends on the context, a bunch of words behind it earlier. With weather forecasting in principle, if you know exactly what’s happening right now, you can fully predict what’s going to happen next.
You don’t need to look further back in the past. So we actually use transformers not to model the interactions in weather over time, like the sequence of text, but in space. So in text, you actually don’t have a sense of spatial structure, right? You just have one sequence of text. It’s this word, word, word. And when you read, you just see word, word, word. In weather, you have spatial structure, you have weather all over the earth at the same time. And it’s all, especially the close weather, it sort of determines and can be used to predict what’s happening next at our current location. And so we use transformers and graph neural networks to capture the short and long range spatial dependencies. And those interactions between what’s nearby and what’s about to happen next are what determine weather, and that’s how we sort of make these predictions.
But one thing I should also add is that similar to how I was saying earlier that it’s kind of impossible to measure everything that’s happening on the earth and the fine detail in weather, you have to make approximations. These models do too. And this actually brings us to a very fundamental difference between how AI models are making their predictions and how traditional models are. So AI models can take the statistical structure of weather patterns. So for example, if I’m looking at a hurricane that’s traveling over the earth, right? In a traditional model, the way it simulates that is in very fine detail, it kind of figures out like, what’s the pressure and the temperature and the wind and the moisture and what are those things? What’s going to happen next is determined strictly locally. AI models, because they can look at a much larger spatial range, they can use what traditional methods use or they can use other approaches.
Because when you look at a hurricane, it almost looks like an object sliding over a globe, right? That’s not how a traditional model models it. And we don’t really understand how the AI models forecast it, but they are capable of treating the hurricane as almost like a large macroscopic scale object that is moving because they can see all the structure of the hurricane and they can see sort of what’s happening in the recent past.
Shayle Kann: They have spatial awareness in a way that the old models didn’t.
Peter Battaglia: Yeah. And we don’t know. It’s a really interesting area, I would say, of the science of how AI works to understand exactly how they see the world in that sense.
Shayle Kann: It seems like to me, on one hand, a much harder problem than an LLM, because you’ve got the entire physical world and the data is sparse, as you said, and there’s lots of complex interactions. On the other hand, it’s determinative in a way that LMs are not, right? There is no correct next word necessarily. That’s right. It’s a best guess as to what the best next word should be, but in the case of weather forecasting, there is a correct prediction to make. And there is a universe of historical data that you can draw upon to do it. So I’m back and forth on weather this is a harder problem or an easier problem than making a really, really good LLM.
Peter Battaglia: I think depending on where your allegiances lie, that’ll probably be … I think it’s probably more of an opinion question, but yeah, I think you’re absolutely right. So one thing I might say it a different way is that it’s still, like I said before, it still is fundamentally uncertain from the standpoint of the information That’s available to the model. Now, yes, the physics is truly deterministic underneath, but because the model doesn’t, again, see the butterflies or see the little fine scale stuff, from its perspective, it actually is a random process because if it doesn’t know whether the butterfly flapped its wings or not, how could it know whether the hurricane’s going to form or not? Now, so from the perspective of the information available to the model, it is also an uncertain random process to some extent. But yeah, I think what you’re saying is exactly right.
So I think the underlying structure of text is random in different ways. Again, I’m going to pause and then I’m going to say a word, furniture.
Weather doesn’t work like that. It doesn’t just have something pop into its head that’s completely different than what’s present in the historical record, like you pointed out. So I think that the structure of the uncertainty is different in whether it’s more constrained in a way. So in text, you can imagine if you’re watching … Same thing with video, you’re watching a movie and someone’s going to come through the door. You have no idea what they’re going to be wearing as a shirt. It could be wearing blue, red, anything. There’s no way to predict it. And weather you can always have some idea, you just don’t know the fine details. On the flip side, weather is an extremely complicated process. This fluid, chaotic fluid system has interactions from small scales to large scales and is happening all over the earth at once. So in some sense, instead of just predicting the next word, you’re predicting millions of variables at once.
So I think it’s probably better to have this as a debate over beers with your friends in their LLM lab rather than something that can be adjudicated just on the basis of these things.
Shayle Kann: I guess there’s one other question on this sort of comparison to LLM world. Notoriously, the big LLMs are trained on the internet. Your training data set is all words on the internet. And so that’s one of the reasons that they’ve been among the first major AI models in this new wave to commercialize is because there is this gigantic body of training data that you can draw upon. Now we’re hearing lots of folks who are in robotics world, for example, facing the challenge of there just isn’t an equivalent data set. You can try to train on YouTube videos or whatever, but it’s not quite the same thing. In the case of weather forecasting, it seems to me like in theory, you have an incredible historical … You can look at every historical weather measurement. If you had access to that data, if NOAA does, every input data point they ever took historically, and then the subsequent next measurement, which dictated what happened after that, I would think that would be an incredibly rich training data set.
Well, two questions. Am I right about that? And is that actually available?
Peter Battaglia: Yeah, those are good questions. So the first thing I would say is actually, it’s not just language. The first big visual neural networks came out after ImageNet, a big corpus of image data. The first language models, even a decade ago, they were starting to build large text databases, protein folding, big databases of protein. So actually you see AI and machine learning are still very, very sensitive to the availability of high quality data and large amounts of it. And one of the best ways to advance the field is to go collect high quality data and make it sort of standardized and available. So for weather, I think that there’s sort of good things. There’s good news and bad news. So for one thing is, we were very fortunate when we started. I think all the folks who were working on AI weather forecasting have benefited tremendously from work that was done by the ECMWF, the European Center for Medium-Range Forecasting.
They built this data set called ERA5, and they’ve been building these … ERA5 I think was the fifth generation of the era dataset. It was a record of Earth’s weather going back for decades. And I think it was originally released going back into the 1979, and then they actually opened it back into the ’60s. And they didn’t design this dataset to be supporting machine learning. I think it was more to just have an authoritative record of the climate on earth over year on year. And it’s at a six hour resolution and 25 kilometer spatial resolution. So it’s very, very rich. It just happened to be perfect for machine learning for weather. And it was just a really well curated data set. The folks at ECMWF are just brilliant and sort of organized and systematic and they had made this available and it allowed a lot of people to stand on the shoulders and build great new AI methods.
Now, one thing is though, because we had different satellites and different stations over time, it’s not actually all … The data set is standardized, but it’s not derived from the same underlying observations. So the quality going back in time actually gets worse in terms of it’s not as accurate of a record of weather just because, again, as I said before, the input data wasn’t as good. Now, the other thing that’s not great about weather is that weather takes a while to happen. So we have to sort of just wait for more weather data for more weather to happen to get more data. The weather data we have so far, we’re sort of stuck with it. Now, tomorrow we’re going to have one more day of weather data, but when our models are taking six hour steps, we have to just kind of wait. So you’re sort of like, we got a lot of data, but at the same time, there’s not much room to get more of the same kind of data.
And I think we and a lot of others are now looking to more unusual or underexplored sources of data to support building richer, better models.
Shayle Kann: Right. Is there a distributed network data … In theory, if you had access to, I don’t know, everybody’s cell phone, everybody’s iPhone, there’s probably a bunch of sensors in the iPhone. Presumably you could pull something from that that would have some signal for you.
Peter Battaglia: Yeah. Yeah. You can geek out on all these things. My favorite one is like, I have a video doorbell. It sits there and watches weather all day. Cars, your car, right? It’s got a thermometer in it, it’s got your windshield wipers. Some of them now, they’re rain sensing. So it’s sensing rain, or it’s like the lights go on automatically when it’s dark. So they’re sensing weather it’s cloudy or these kinds of things. So I get very excited about the possibility of using all these kinds of things. The other one that’s even weirder is people go on and they tweet about the weather. They talk about the weather on social media and those types of observations, those are still observations. Well, we don’t know if they’re very good, but there’s actually a very wide range of pretty unusual under explored data sources. But even before we get there, I think you can start to think about there’s a lot of companies that are trying to build very cheap weather stations.
People can put them on their roof. These kinds of things could really help both with the kind of core weather forecast and probably a lot of the applications that people want to use weather for. So
Shayle Kann: My takeaway from that is that despite what I said, there being this amazing historical record of every weather measurement that’s ever been taken, you still feel kind of data poor or training data poor, I guess.
Peter Battaglia: You’re always data poor, right? The story of modern AI is you’re basically always kind of data poor because one of the most incredible facts about modern AI is just how well it scales with data. More data just means better models. And I was a person who was very skeptical of this. I didn’t think that it was going to scale like this and I would make my sort of logical arguments, but it turns out I was wrong. And I think a lot of people were wrong and that the folks who really understood that data could really add value even at extremely large scales were right and pursued that course and brought us to where we are today.
Shayle Kann: Interesting. Okay. So I guess I want to finish by talking about what might come. If you draw a line forward a few years into the future, I know you pick your time, three years, five years, 10 years, whatever it is, and you and everybody else who’s working on AI weather forecasting succeed, where might we be? What might be possible in a few years that’s not possible today?
Peter Battaglia: Yeah. I mean, I think weather affects everything and it has different things. Energy is obviously a very sensitive to weather. Some things are only kind of loosely affected by weather. So one thing I would like to see, and I think is very exciting, is a wider range of use cases of weather. So for example, we know that even people make different choices about what to put in their refrigerator or what clothes or whatever, these different choices are going to go on a trip, what they expect the weather to be. I think that you can start to make more subtle and informed guidance and suggestions for people on the basis of what a more accurate weather forecast. And that’s kind of like at the consumer level, but I also think that at the kind of industry level, there could be a huge opportunity. So for example, in energy, we see there’s, if you have a wind farm or a solar farm, you’re making forecasts about the weather and then you’re kind of using that to figure out if you’re going to have energy to sell and how you’re going to price it.
But I have a feeling that there’s a lot of headroom, a lot more to be gained in how we plan how to operate our electrical grids, how we predict what the electrical demand is going to be. Is it going to be hot? Is it going to be cold? Is it going to be humid or the air is carrying more mass, which requires more energy to heat or cool? I think that we just haven’t really scratched the surface of the opportunities. I think that supply chains and logistics and even just like lots of choices that driving and these types of things, I think really could be better informed by better weather forecasts. And I don’t think we’ve even begun to kind of get into this. And we need to see the quality of the forecast get better and more customized to these use cases to start unlocking that.
The other one obviously is just better crisis handling crises. Some of our recent work on tropical cyclone forecasting, I’m very proud of. And we think hopefully with better forecast, we can warn people earlier, more accurately. We can, in some cases where there’s a disaster like a wildfire, maybe we can even go and intervene and try to stop it before it gets completely out of control. So again, this is not something that we can do today, but I’m very optimistic that … And in general, maybe this is coming through, but my team and I really believe in the power of technology can have a lot of positive benefits. So we often are trying to look for ways that we can put technology to best use and sort of leave our kids with a world that’s better than the one that we grew up in. All
Shayle Kann: Right, Peter, this was super interesting and useful for me both in the context of thinking about weather forecasting and just understanding how AI is getting applied in various industries, what the challenges are, what the opportunities are. So really appreciate your time.
Peter Battaglia: Sure. Yeah, it was great to be here and your questions were great too, by the way. It was really fun.
Shayle Kann: Peter Battaglia is the senior director of research at Google DeepMind’s Sustainability Program. This show is a production of Latitude Media. You can head over to latitudemedia.com for links to today’s topics. Latitude is supported by Prelude Ventures. This episode was produced by Daniel Woldorff, mixing and theme song by Sean Marquand. Stephen Lacey is our executive editor. I’m Shayle Kann and this is Catalyst.
