GenCast predicts weather and the risk of extreme conditions with cutting-edge accuracy

Recent artificial intelligence model improves prediction of weather uncertainties and threats, providing faster and more true forecasts up to 15 days in advance

The weather affects us all – shaping our decisions, our safety and our way of life. As climate change causes more and more extreme weather events, true and reliable forecasts are more significant than ever. However, the weather cannot be predicted perfectly, and forecasts are particularly uncertain over periods longer than a few days.

Because a perfect weather forecast is impossible, scientists and weather agencies exploit probabilistic ensemble forecasts, in which a model predicts a range of likely weather scenarios. Such aggregated forecasts are more useful than relying on a single forecast because they provide decision-makers with a more complete picture of possible weather conditions in the coming days and weeks and the likelihood of each scenario.

Today in the newspaper published in NatureIntroducing GenCast, our up-to-date high-resolution (0.25°) AI ensemble model. GenCast provides better forecasts of both daily weather and extreme events than the leading operating system, the European Center for Medium-Range Weather Forecasts” (ECMWF) ENS, with a maximum of 15 days’ notice. We will be publishing our model code, weights and predictions to support the broader weather forecasting community.

The evolution of AI weather models

GenCast is a critical advancement in AI-based weather forecasting that builds on our previous weather model, which was deterministic and provided a single best estimate of future weather. In contrast, a GenCast forecast consists of a collection of 50 or more forecasts, each representing a possible weather trajectory.

GenCast is a diffusion model, a type of generative artificial intelligence model, that underlies recent rapid advances in image, video and music generation. However, GenCast differs from them in that it is adapted to the spherical geometry of the Earth and learns to accurately generate a intricate probability distribution of future weather scenarios when the latest weather condition is given as input.

To train GenCast, we fed it four decades of historical weather data from ECMWF ERA5 archive. This data includes variables such as temperature, wind speed and pressure at various altitudes. The model learned global weather patterns with a resolution of 0.25° directly from processed weather data.

Setting a up-to-date standard for weather forecasting

To rigorously evaluate GenCast’s performance, we trained it on historical weather data through 2018 and tested it on data from 2019. GenCast demonstrated better forecasting skills than ECMWF’s ENS, the best operational ensemble forecast system on which many national and local governments depend every day decision .

We tested both systems comprehensively, analyzing forecasts of various variables at different lead times – a total of 1,320 combinations. GenCast was more true than ENS for 97.2% of these targets and 99.8% for turnaround times longer than 36 hours.

An ensemble forecast expresses uncertainty by making multiple forecasts that represent different possible scenarios. If most forecasts indicate that the cyclone will hit the same area, uncertainty is low. But if they predict different locations, the uncertainty is greater. GenCast strikes the right balance by avoiding both overstatement and understatement of confidence in its forecasts.

It takes a single Google Cloud TPU v5 device just 8 minutes to generate one 15-day forecast in the GenCast set, and each forecast in the set can be generated simultaneously and in parallel. Conventional physics-based ensemble forecasts, such as those produced by ENS, with a resolution of 0.2° or 0.1°, require many hours of work on a supercomputer with tens of thousands of processors.

Advanced forecasts of extreme weather events

More true extreme weather risk forecasts could lend a hand officials protect more lives, prevent damage and save money. When we tested GenCast’s ability to predict extreme heat and chilly and high wind speeds, GenCast consistently outperformed ENS.

Now consider tropical cyclones, also known as hurricanes and typhoons. Receiving better and more advanced warnings about where they will make landfall is invaluable. GenCast provides excellent predictions of the tracks of these deadly storms.

Better forecasts could also play a key role in other aspects of society, such as renewable energy planning. For example, improvements in wind energy forecasting directly raise the reliability of wind energy as a source of sustainable energy and potentially accelerate its adoption. In a proof-of-principle experiment that analyzed forecasts of the total wind power generated by groups of wind farms around the world, GenCast was more true than ENS.

Next-gen forecasting and climate understanding at Google

GenCast is part of Google’s growing suite of next-generation AI-powered weather models, including Google DeepMind AI-powered weather models deterministic medium-term forecastsand Google research NeuroGCM, SEEDSAND flood models. These models are starting to improve user experience on Google Search and Maps and improve forecasting precipitate, fires, flood AND extreme heat.

We greatly value our partnership with weather agencies and will continue to work with them to develop AI-based methods to improve their forecasting. Meanwhile, established models remain crucial to this work. First, they provide the training data and initial weather conditions required by models like GenCast. Collaboration between artificial intelligence and established meteorology highlights the power of a combined approach to improve forecasts and better serve society.

To foster greater collaboration and lend a hand accelerate research and development in the weather and climate community, we have made GenCast an open model and made it available code AND weightsas we did with our deterministic medium-range global weather forecasting model.

We will soon release real-time forecasts and historical forecasts from GenCast and previous models, enabling anyone to integrate weather inputs into their own models and research processes.

We are keen to engage with the broader weather community, including academic researchers, meteorologists, data analysts, renewable energy companies, and food security and disaster response organizations. Such partnerships offer deep insight and constructive feedback, as well as invaluable opportunities to make commercial and non-commercial impact, all of which are critical to our mission to apply our models for the benefit of humanity.