Nvidia this week unveiled a new batch of Earth-2 AI-powered weather models that could dramatically change traditional physics-based weather forecasting.
The announcement did not come in time for a historically massive winter storm that swept through the US last weekend, but the new models could make forecasting easier, faster, and more accurate in the future for climate scientists and meteorologists, many of whom were gathered at the American Meteorology Society event this week where Nvidia made its announcement.
This is particularly the case with the new Earth-2 Nowcasting model, which leverages AI analysis of images and other sensor data from satellites and radar “to make country-scale forecasts into kilometer‑resolution, zero-to-six-hour predictions of local storms and hazardous weather in just minutes,” according to an Nvidia blog post.
Mike Pritchard, director of climate simulation at Nvidia said during a pre-announcement briefing that Earth-2 Nowcasting is the first AI-based weather model to outperform traditional, physics-based weather-prediction models on short-term precipitation forecasting by simulating storm dynamics directly.
The family of new models also includes a Global Data Assimilation model, which Pritchard said “produces initial conditions that are used to launch forecasts, combining the messy observations from satellites and weather balloons and stations into smooth fields that can be used to launch forecasts, and it does this in minutes on GPUs instead of hours on supercomputers.”
In addition to those two models, the new Earth-2 Medium Range enables 15-day forecasts, and Pritchard said it “outperforms the current leading open model, [Google Deepmind’s] GenCast, across more than 70 weather variables. It does this thanks to a new AI model architecture called Atlas... Philosophically and scientifically, Atlas is a return to simplicity. We're moving away from hand tailored niche AI architectures and leaning into the future of simple, scalable transformer architectures that are having transformative results in drug discovery, self-driving [vehicles] and robotics.”
He added that now Nvidia is showing that “these methods employ a critical mass of performance and engineering tooling can produce state of the art results in weather forecasting.”
Pritchard said The Weather Company is evaluating Earth-2 Nowcasting for localized severe-weather applications, while the National Weather Service is evaluating the new models to enhance its operational workflows. Other organizations also have started testing the new models.