Tomorrow the sun will come out and you will no longer have to bet your bottom dollar to be safe. Google’s DeepMind team released its latest weather forecasting model this week, outperforming a leading traditional weather forecasting model in most tests submitted to it.

The generative AI model is called GenCast and is a diffusion model, which is also the basis for popular AI tools such as Midjourney, DALL·E 3 and Stable Diffusion. Based on the team’s testing, GenCast is better able to predict extreme weather conditions, the movement of tropical storms and the strength of wind gusts across the Earth’s vast lands. The team’s discussion of GenCast’s performance was published in this week Nature.

The difference between GenCast and other diffusion models is that it is (obviously) weather-oriented and “fitted to Earth’s spherical geometry,” as some of the paper’s co-authors described in a DeepMind blog post.

Instead of a written prompt like “Paint a picture of a Salvador Dalí-style dachshund,” GenCast’s input is the current weather condition, which the model then uses to generate a probability distribution of future weather scenarios.

Traditional weather forecast models such as ENS, the leading model of the European Center for Medium-Range Weather Forecasts, make their forecasts by solving physical equations.

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

GenCast’s first seeds were planted in 2022, but the model released this week includes architectural changes and an improved diffusion setup that have better trained the model to predict Earth’s weather, including extreme weather events, up to 15 days in advance .

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

Google has been involved in weather forecasting for some time and has taken some significant steps towards more precise forecasts using AI methods in recent years.

Last year, DeepMind scientists – some of whom were co-authors of the new paper – released GraphCast, a machine learning-based method that outperformed current medium-range weather forecast models on 90% of the targets used in tests. Just five months ago, a team consisting largely of DeepMind researchers released NeuralGCM, a hybrid weather forecasting model that combined a traditional physics-based weather forecaster with machine learning components. This team found that “end-to-end deep learning is compatible with tasks performed by traditional (models) and can improve the large-scale physics simulations that are essential to understanding and predicting the Earth system.”

The resolution achieved by GenCast is approximately six times that of NeuralGCM, but that is to be expected. “NeuralGCM is designed as a general-purpose atmospheric model primarily to support climate modeling, while GenCast’s higher resolution is often expected for operational medium-range forecast models, which is GenCast’s specific target use case,” Price added. “For this reason, we have placed emphasis on a wide range of evaluations that are critical use cases for operational medium-range forecasts, such as forecasting extreme weather conditions.”

In the most recent work, the team trained GenCast using 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 with different weather variables and different 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 predict the path of a tropical cyclone – specifically, Typhoon Hagibis, the costliest tropical cyclone of 2019, which struck Japan in October. GenCast’s predictions were extremely uncertain at seven days’ lead time, but became more accurate at shorter lead times. As extreme weather results in wetter and heavier rainfall, and hurricanes break records in terms of their speed of intensity and time of formation, accurate forecasting of storm paths is critical to reducing their financial and human costs.

But that’s not all. In a proof-of-principle experiment described in the study, the DeepMind team found that GenCast was more accurate than ENS in predicting the total wind energy produced by groups of over 5,000 wind farms in the Global Power Plant Database. GenCast’s predictions were approximately 20% better than ENS’s with lead times of two days or less and maintained statistically significant improvements up to a week. In other words, the model is not only useful for disaster preparedness, it could also shed light on where and how we use energy infrastructure.

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

“This will enable the broader research and meteorology community to participate in, test, conduct and build on our work, accelerating further advances in the field,” Price said. “We optimized versions of GenCast to take operational input into account, so the model could be integrated into the operational environment.”

There’s no timeline yet for when GenCast and other models will be ready for use, although the DeepMind blog noted that the models “are beginning to improve user experiences in Google Search and Google Maps.”

Whether you’re here for the weather or the AI ​​applications, GenCast and the broader suite of DeepMind forecast models have a lot to offer. The accuracy of such tools will be critical to predicting extreme weather events with enough lead time to protect people in harm’s way, whether from flooding in the Appalachians or tornadoes in Florida.