Cambridge researchers unveil faster and more accurate AI weather system that rivals supercomputers

[Skye Jacobs]

In a nutshell: Aardvark Weather, an AI-based system, promises to significantly enhance weather forecasting by delivering predictions dozens of times faster while using thousands of times less computing power than current methods. This system has been developed by researchers at the University of Cambridge, with support from the Alan Turing Institute, Microsoft Research, and the European Centre for Medium Range Weather Forecasts.

The speed and efficiency of modern forecasting systems are vital, as traditional methods rely on powerful supercomputers and extensive teams of experts, often requiring several hours to produce forecasts.

Recent innovations from tech giants such as Huawei, Google, and Microsoft have demonstrated that AI can significantly improve specific aspects of the forecasting process, including numerical solvers, which are crucial in weather forecasting as they simulate how atmospheric conditions evolve over time. These companies have achieved faster and more accurate predictions by integrating AI into these solvers.

As one example, Google has been developing AI models for weather forecasting and is currently marketing two models to its enterprise cloud customers. Developed by Google DeepMind, the models use historical weather data to predict future conditions 10 to 15 days in advance.

Aardvark represents a significant advancement by replacing traditional forecasting processes with a single, streamlined machine-learning model. Using a standard desktop computer, it can process data from various sources, including satellites and weather stations, to generate global and local forecasts in minutes.

"Aardvark reimagines current weather prediction methods, offering the potential to make weather forecasts faster, cheaper, more flexible, and more accurate than ever before," explained Professor Richard Turner from Cambridge's Department of Engineering, who led the research. "Aardvark is thousands of times faster than all previous weather forecasting methods."

Despite operating with only a fraction of the data used by existing systems, Aardvark surpasses the U.S. national GFS forecasting system in several key metrics and remains competitive with forecasts from the National Weather Service, which typically involve multiple models and expert analysis.

"These results are just the beginning of what Aardvark can achieve," noted first author Anna Allen from Cambridge's Department of Computer Science and Technology. She said the end-to-end learning approach can be easily applied to other weather forecasting problems, such as hurricanes, wildfires, and tornadoes. It can also be used for broader Earth system forecasting, including air quality, ocean dynamics, and sea ice prediction.

One of the most interesting aspects of Aardvark is its flexibility and simple design. Because it learns directly from data, it can be quickly adapted to produce bespoke forecasts for specific industries or locations, whether predicting temperatures to support African agriculture or wind conditions for European renewable energy firms. This contrasts sharply with traditional systems, which require years of work by large teams to customize.

This capability has the potential to transform weather prediction in developing countries, where access to expertise and computational resources is limited. "By shifting weather prediction from supercomputers to desktop computers, we can democratize forecasting, making these powerful technologies available to developing nations and data-sparse regions around the world," said Dr. Scott Hosking from The Alan Turing Institute.

Aardvark is expected to play a significant role in expanding the scope of weather forecasting. Turner mentioned that the model could eventually accurately predict eight-day forecasts, surpassing the capabilities of current models by three days. This advancement, along with Aardvark's adaptability and efficiency, positions it as a transformative force in meteorology.

The next steps for Aardvark include developing a new team within the Alan Turing Institute that will explore deploying the technology in the global south and integrating it into broader environmental forecasting initiatives.

Permalink to story:

Cambridge researchers unveil faster and more accurate AI weather system that rivals supercomputers

 

[NumberNine]

Prepare to agree to conpletely irresponsible terms when installing this. Insurance companies will be very interested.

 

[Xelions]

Me: "Let's check the weather today.... Oh sh*# theres a cat6 cyclone coming, wtf when did this happen. Hey Allegra what's a cat6"

AI Allegra Weather: "Category 6 is a level of wind speed and strength that'll f*#k you up! If it were real biatch. Got ya... you'll believe anything "

Need an AI that pulls pranks

 

[mbk34]

Why would you want to do better weather forecasting on your desktop PC? I just want better forecasting from the national weather providers. Here in London today it was forecasting rain all day and yet I'm doing the gardening in shorts and T-shirt.

 

[AnilD]

The real game-changer here isn’t just the speed—it’s accessibility. Moving weather prediction from supercomputers to desktops could completely reshape how vulnerable communities prepare for floods, droughts, or storms.

 

[ibnMuhammad]

The real game-changer here isn’t just the speed—it’s accessibility. Moving weather prediction from supercomputers to desktops could completely reshape how vulnerable communities prepare for floods, droughts, or storms.

I still don't see how that is any different or more advanced than using a supercomputing with access to far more data, processing faster, and then disseminating info to all receivers?

In the end, one still needs an Internet connection.

 

[pnntmp]

the model could eventually accurately predict eight-day forecasts, surpassing the capabilities of current models by three days

If the most advanced model still in development may eventually be able to predict up to 8 days ahead, how do some people know what "climate change" will do in 80 years???
Either "climate change" guys are a bunch of crooks lying to everyone, or they have some super-secret knowledge not available to us. Which is the case?

 

[yRaz]

I still don't see how that is any different or more advanced than using a supercomputing with access to far more data, processing faster, and then disseminating info to all receivers?

In the end, one still needs an Internet connection.

The real advantage is cost. Instead of a needing a multimillion dollar supercomputer, the power bill that comes with it and the labor costs of maintaining it, you can do all this on a Threadripper machine in one room. This brings cost down significantly. As DeepSeek has shown, a well optimized model doesn't need the absurd amount of data(and the hardware to process it) that all these massive AI companies keep claiming they need. And to put some perspective on this, A single 4090 is more powerful than the supercomputers that were doing very effective climate modeling just 20 years ago. Earth Simulator in 2004 was capable of just 36TeraFlops, a 4090 can do over twice that(~85TeraFlops). The next biggest one from 2004-2009 was IBM BlueGene was capable of 480TeraFlops, or about 5 4090s.

People can say "oh, that's just tech progress" and while that's somewhat true, there weren't people in early to mid 2000's running supercomputer simulations on their computers from the 80's and 90s. Frankly, up until about 5-6 years ago, having a supercomputer in your house that could run off of a single, standard 240V socket was an absurd idea.

Now we can argue until we're blue in the face about whether exaflop scale supercomputers are better than having multiple, smaller computers for individuals to use, but they certainly do have their advantages. For weather modeling, specifically, where things can change in minutes, renting time on a supercomputer could be made irrelevant before the calculation is actually done. Renting time on supercomputers ISN'T CHEAP, either. Those things cost billions. So for a small team of climate scientists or even a local news station, the idea of paying ~$30,000 for a local, power efficient supercomputer can easily be cheaper than renting time on a supercomputer in a relatively short period of time.

And this brings me to a final point, localized AI will at somepoint bring computers out of the cloud and back into the home. It's not really a matter of if AI will become good, it's when. I still think we're 5 years away from being able to interact with an AI naturally, but from what I've seen over the last year makes me think maybe we're getting closer. I do see a day where a local super computer that runs AI that you can ask questions in a startrek like manner becomes an essential appliance in your house. I see backlash against Google, Microsoft and Facebook no longer being just a tech-persons idea, but something the general public is starting to see. Current politics have accelerated this, but even my 80 year old mother has been aware of the absurdity of these massive corporations owning our data and running our lives for years now.

These companies have the resources to do AI *the right way*, but don't because it would prevent them from siphoning money away investor money from venture capital and private equity into their own pockets. We have seen things recently that tell use that AI doesn't require these massive machines and amounts of data. People are already releasing their own models trained on sub-million dollar hardware and it's only a matter of time before you can pick which one you like most, download it and run it at home on an "appliance" that costs less than a used car. AI gating keeping is coming to an end, the cats already out of the bag and I think there will be a shift relatively soon, less than 10 years, where nVidia will have to start selling GPUs to consumers again because running your own AI at home will be the next trend. A start to an era where your family has an AI or a robot that you all interact with.

 

[ibnMuhammad]

A single 4090 is more powerful than the supercomputers that were doing very effective climate modeling just 20 years ago. Earth Simulator in 2004 was capable of just 36TeraFlops, a 4090 can do over twice that(~85TeraFlops). The next biggest one from 2004-2009 was IBM BlueGene was capable of 480TeraFlops, or about 5 4090s.

This isn't accurate.

A single RTX 4090 has a peak theoretical FP32 (single precision) tflop of approx 82.6.
While the original Earth Simulator supercomputer from 2002 had a peak FP64 (double-precision) tflop of 40 (real-world linkpack result: 36 tflops.

So theoretically, the RTX 4090 FP64 perf would be 1:64, so approx ~1.3 tflops.

It's more fair to compare nVidia's server grade A100 gpu with a peak tflop of 9.5.

But even more powerful is AMD's server grade gpu: Instinct MI300X, which I believe has a theoretical peak of ~82 tflops.

Regardless, I generally understand what you're saying.

 

[gamerk2]

If the most advanced model still in development may eventually be able to predict up to 8 days ahead, how do some people know what "climate change" will do in 80 years???
Either "climate change" guys are a bunch of crooks lying to everyone, or they have some super-secret knowledge not available to us. Which is the case?

Weather is a chaotic system; small changes to initial conditions lead to vastly different outcomes. That's why most models run multiple simulations with minor variations across them to get a range of outcomes, and why weather is notoriously hard to predict beyond the next day or two.

Climate is different in that, on a planetary scale, its a much simpler system. More greenhouse gases trap more heat leading to a hotter planet; we've known how greenhouses work for well over 10,000 years now. The only debate is how and when we start hitting self-reinforcing heat growth (where increased warming leads directly to more warming). The only real debate to be had is how fast and what the localized effects will be; the fact the planet is getting warmer at a rate never seen in history is pretty much indisputable at this point.

 

[Cabuzzi]

Weather is a chaotic system; small changes to initial conditions lead to vastly different outcomes. That's why most models run multiple simulations with minor variations across them to get a range of outcomes, and why weather is notoriously hard to predict beyond the next day or two.

Climate is different in that, on a planetary scale, its a much simpler system. More greenhouse gases trap more heat leading to a hotter planet; we've known how greenhouses work for well over 10,000 years now. The only debate is how and when we start hitting self-reinforcing heat growth (where increased warming leads directly to more warming). The only real debate to be had is how fast and what the localized effects will be; the fact the planet is getting warmer at a rate never seen in history is pretty much indisputable at this point.

Here we go....

If climate change is really man-made and the "science is settled", please tell me what percentage of the 0.8 degrees or whatever is man's responsibility and what percentage is due to the natural cycle of the sun.

You can't, and no one else can, so please stop assuming we know how much the planet is going to warm in 10 years down to the fidelity of less than a degree. It's literally preposterous, and every single "model" eventually fails.

 

[gamerk2]

Here we go....

If climate change is really man-made and the "science is settled", please tell me what percentage of the 0.8 degrees or whatever is man's responsibility and what percentage is due to the natural cycle of the sun.

You can't, and no one else can, so please stop assuming we know how much the planet is going to warm in 10 years down to the fidelity of less than a degree. It's literally preposterous, and every single "model" eventually fails.

Actually, we can, on the timescale of a few decades. That's why most of the models look at least two to three decades out, since it's hard to nail down exactly when certain events will occur. But make no mistake: They *will* occur, and based on the past few years its quite possible we're starting to hit the exponential part of the warming curve.

Secondly, the natural cycle of the sun is *very* well understood. That's why you get a warming graph with peaks and valleys in a cyclical pattern. And if you look at them, the average has been rising pretty steadily since the post-WW2 era; that's the effect we're having.

While the sun does and will continue to warm with time, the effect on Earth is only noticeable on the timespan of hundreds of thousands of years, and can be disregarded.

 

[Cabuzzi]

Actually, we can, on the timescale of a few decades. That's why most of the models look at least two to three decades out, since it's hard to nail down exactly when certain events will occur. But make no mistake: They *will* occur, and based on the past few years its quite possible we're starting to hit the exponential part of the warming curve.

Secondly, the natural cycle of the sun is *very* well understood. That's why you get a warming graph with peaks and valleys in a cyclical pattern. And if you look at them, the average has been rising pretty steadily since the post-WW2 era; that's the effect we're having.

While the sun does and will continue to warm with time, the effect on Earth is only noticeable on the timespan of hundreds of thousands of years, and can be disregarded.

Just that last paragraph tells me you don't understand the effect of the solar cycles, especially saying it can be "disregarded".

What it all boils down to is the audacity that we think we have the calculations down to the precision required to get to 0.8 of change over many years, then say we need to turn this world on its head (impacting primarily the poor, low, and middle class) in order to "fix" it. That is the travesty waiting to happen. That, and the amount of already rich people who will get richer propagating the panic that comes with their predictions.

This needs to stop. It is the biggest boondoggle in the history of humanity.

 

[gamerk2]

Just that last paragraph tells me you don't understand the effect of the solar cycles, especially saying it can be "disregarded".

Not really? The 11 year solar cycle period is perfectly well understood, and is (and has been) accounted for when evaluating temperature data.

What it all boils down to is the audacity that we think we have the calculations down to the precision required to get to 0.8 of change over many years, then say we need to turn this world on its head (impacting primarily the poor, low, and middle class) in order to "fix" it. That is the travesty waiting to happen. That, and the amount of already rich people who will get richer propagating the panic that comes with their predictions.

This needs to stop. It is the biggest boondoggle in the history of humanity.

Counterpoint: The increased damage (just in terms of dollar cost) caused by a warming planet is *already* an order of magnitude more, every year, then the cost to mitigate the source of the problem. And that's *before* considering long-term societal costs (especially when mass migration towards northern latitudes are considered).

 

[pnntmp]

Weather is a chaotic system; small changes to initial conditions lead to vastly different outcomes. That's why most models run multiple simulations with minor variations across them to get a range of outcomes, and why weather is notoriously hard to predict beyond the next day or two.

Climate is different in that, on a planetary scale, its a much simpler system. More greenhouse gases trap more heat leading to a hotter planet; we've known how greenhouses work for well over 10,000 years now. The only debate is how and when we start hitting self-reinforcing heat growth (where increased warming leads directly to more warming). The only real debate to be had is how fast and what the localized effects will be; the fact the planet is getting warmer at a rate never seen in history is pretty much indisputable at this point.

There's nothing simple about climate, your understanding of it is simplistic.
Greenhouse gases have the effect you describe in greenhouses, hence the name. Earth is about a gazillion orders of magnitude more complicated than a tiny isolated volume. Earth's climate depends on hundreds of variables which also depend on each other in wildly nonlinear ways, so isolating the effect of one of them (e.g. CO2) is absolutely impossible. Attempts to dumb this unfathomable complexity down to some neat dependency between CO2 and temperatures are preposterous.

 

[gamerk2]

There's nothing simple about climate, your understanding of it is simplistic.
Greenhouse gases have the effect you describe in greenhouses, hence the name. Earth is about a gazillion orders of magnitude more complicated than a tiny isolated volume.

But that doesn't change a very simple fact: More CO2 = more trapped heat. And there's no world where trapping more heat leads to anything but increased global temperatures.

You are right that trying to determine what higher global temperatures will look like at a local level is, frankly, a waste of time and effort.

Earth's climate depends on hundreds of variables which also depend on each other in wildly nonlinear ways, so isolating the effect of one of them (e.g. CO2) is absolutely impossible. Attempts to dumb this unfathomable complexity down to some neat dependency between CO2 and temperatures are preposterous.

*If* multiple of those variables were changing by statistically significant amounts at the same time, then you would have a point. But with CO2 over 25% higher then pre-industrial levels and most everything else remaining more or less statistically insignificant, we can isolate to a reasonable degree the effect CO2 has on climate.

You are right that even in this best case you aren't going to get a linear correlation, because of noise and just general uncertainty. But again: That doesn't change the very simple fact that more CO2 leads to more trapped heat.

 

[Northy]

In the 1950's scientists understood how emissions would effect the atmosphere. Everything that was said then is true now, and everything is happening exactly as predicted by scientists since way back then. As slowly and as surely as they predicted.

People like to deny it because it messes with their money. The irony is they can deny it until they die of old age, but the real problem doesn't start this decade, or next... it's in 50 years. By then there's nothing we can do. We're fk'd.

But really there's already nothing we can do, the major polluters are out of all our control anyway and there's 8 billion people who want everything...

Enjoy the ride, try not to choke or burn on the way down.