Home Consumer What to Know About Google’s Breakthrough Weather Prediction Model

What to Know About Google’s Breakthrough Weather Prediction Model

By
Gizmodo.com
-
A tornado near Pender, Nebraska, in 2015. Photo: NOAA Weather in Focus Photo Contest 2015 | Brent Koops

By Isaac Schultz

The Sun’ll come out tomorrow, and you no longer have to bet your bottom dollar to be sure of it. Google’s DeepMind team released its latest weather prediction model this week, which outperforms a leading traditional weather prediction model across the vast majority of tests put before it.

The generative AI model is dubbed GenCast, and it is a diffusion model like those undergirding popular AI tools including Midjourney, DALL·E 3, and Stable Diffusion. Based on the team’s tests, GenCast is better at predicting extreme weather, the movement of tropical storms, and the force of wind gusts across Earth’s mighty sweeps of land. The team’s discussion of GenCast’s performance was published this week in Nature.

Where GenCast departs from other diffusion models is that it (obviously) is weather-focused, and “adapted to the spherical geometry of the Earth,” as described by a couple of the paper’s co-authors in a DeepMind blog post.

Instead of a written prompt such as “paint a picture of a dachshund in the style of Salvador Dalí,” GenCast’s input is the most recent state of the weather, which the model then uses to generate a probability distribution of future weather scenarios.

Traditional weather prediction models like ENS, the leading model from the European Center for Medium-Range Weather Forecasts, make their forecasts by solving physics equations.

Faith Based Events
“One limitation of these traditional models is that the equations they solve are only approximations of the atmospheric dynamics,” said Ilan Price, a senior research scientist at Google DeepMind and lead author of the team’s latest findings, in an email to Gizmodo.

The first seeds of GenCast were planted in 2022, but the model published this week includes architectural changes and an improved diffusion setup that made the model better trained to predict weather on Earth, including extreme weather events, up to 15 days out.

“GenCast is not limited to learning dynamics/patterns that are known exactly and can be written down in an equation,” Price added. “Instead it has the opportunity to learn more complex relationships and dynamics directly from the data, and this allows GenCast to outperform traditional models.”

Google has been tooling around with weather prediction for a while, and in recent years have made a couple substantive steps towards more precise forecasting using AI methods.

Last year, DeepMind scientists—some of whom co-authored the new paper—released GraphCast, a machine learning-based method that outperformed the current medium-range weather prediction models on 90% of the targets used in testing. Just five months ago, a team mostly consisting of DeepMind researchers published NeuralGCM, a hybrid weather prediction model that combined a traditional physics-based weather predictor with machine-learning components. That team found that “end-to-end deep learning is compatible with tasks performed by conventional [models] and can enhance the large-scale physical simulations that are essential for understanding and predicting the Earth system.”

The resolution achieved by GenCast is roughly six times that of NeuralGCM, but that was expected. “NeuralGCM is designed as a general purpose atmospheric model primarily to support climate modelling, whereas the higher resolution of GenCast is often expected for operational medium range forecast models, which is GenCast’s specific target use-case,” Price added. “This is also why we emphasized a wide range of evaluations which are crucial use cases for operational medium range forecasts, like predicting extreme weather.”

Thunderstorm cells wreak havoc on eastern Florida as Hurricane Milton makes landfall.
Thunderstorm cells wreak havoc on eastern Florida as Hurricane Milton makes landfall. Image: NOAA / CIRA

In the recent work, the team trained GenCast on historical weather data through 2018, and then tested the model’s ability to predict weather patterns in 2019. GenCast outperformed ENS on 97.2% of targets using different weather variables, with varying lead times before the weather event; with lead times greater than 36 hours, GenCast was more accurate than ENS on 99.8% of targets.

The team also tested GenCast’s ability to forecast the track of a tropical cyclone—specifically Typhoon Hagibis, the costliest tropical cyclone of 2019, which hit Japan that October. GenCast’s predictions were highly uncertain with seven days of lead time, but became more accurate at shorter lead times. As extreme weather generates wetter, heavier rainfall, and hurricanes break records for how quickly they intensify and how early in the season they form, accurate prediction of storm paths will be crucial in mitigating their fiscal and human costs.

But that’s not all. In a proof-of-principle experiment described in the research, the DeepMind team found that GenCast was more accurate than ENS in predicting the total wind power generated by groups of over 5,000 wind farms in the Global Power Plant Database. GenCast’s predictions were about 20% better than ENS’ with lead times of two days or less, and retained statistically significant improvements up to a week. In other words, the model does not just have value in mitigating disaster—it could inform where and how we deploy energy infrastructure.

“The development of GenCast, a machine learning weather prediction (MLWP) model, marks a significant milestone in the evolution of weather forecasting, as highlighted in the recent Google DeepMind paper,” said an ECMWF spokesperson, in an emailed statement to Gizmodo. “GenCast is one of the latest machine learning models reviewed in a series of high-profile scientific papers about MLWP coming from around the globe, which highlight the ongoing (r)evolution in weather forecasting.”

The ECMWF statement pointed out that the GenCast paper also compared the model’s performance to ENS 11-mile (18-kilometer) resolution. Now five years later, ENS runs at a 5.6-mile (9 km) resolution. “The GenCast paper presents innovative science from a machine learning point of view, but these improvements have got to be tested on how well they perform in extreme weather events to fully appreciate their value,” the statement concluded.

What does all of this mean for you, O casual appreciator of climate? Well, the DeepMind team has made the GenCast code open source and the models available for non-commercial use, so you can tool around if you’re curious. The team is also working on releasing an archive of historical and current weather forecasts.

“This will enable the wider research and meteorological community to engage with, test, run, and build on our work, accelerating further advances in the field,” Price said. “We have finetuned versions of GenCast to be able to take operational inputs, and so the model could start to be incorporated in operational setting.”

There is not yet a timeline on when GenCast and other models will be operational, though the DeepMind blog noted that the models are “starting to power user experiences on Google Search and Maps.”

Whether you’re here for the weather or the AI applications, there’s plenty to like about GenCast and the broader suite of DeepMind forecasting models. The accuracy of such tools will be paramount for predicting extreme weather events with enough lead time to protect those in harm’s way, be it from floods in Appalachia or tornadoes in Florida.

12/6 3pm: This story has been updated to include comments from ECMWF.


Disclaimer

The information contained in South Florida Reporter is for general information purposes only.
The South Florida Reporter assumes no responsibility for errors or omissions in the contents of the Service.
In no event shall the South Florida Reporter be liable for any special, direct, indirect, consequential, or incidental damages or any damages whatsoever, whether in an action of contract, negligence or other tort, arising out of or in connection with the use of the Service or the contents of the Service. The Company reserves the right to make additions, deletions, or modifications to the contents of the Service at any time without prior notice.
The Company does not warrant that the Service is free of viruses or other harmful components

RELATED ARTICLESMORE FROM AUTHOR