Nvidia's weather AI models open up a new approach to meteorology, insurance and disaster risk management in the context of increasing climate change.
On July 26 (local time), at the annual meeting of the US Meteorological Society held in Houston, Nvidia introduced three open source artificial intelligence models to support building better and faster weather forecasts.
This is part of the long-term strategy of the AI chip company to provide open source software, operating optimally on Nvidia's hardware platform, for many fields from chatbots, self-driving cars to climate science.
In the field of weather forecasting, Nvidia is aiming to replace traditional physical simulations, which take a lot of time and money, with AI-based models.
According to the company, after being trained, these AI models can achieve equivalent, even superior, accuracy compared to old methods, while processing speeds are significantly faster and operating costs are lower.
Mr. Mike Pritchard, Climate Simulation Research Director at Nvidia and Professor of Earth Systems Science at the University of California, Irvine, said that one of the most important practical applications of AI weather models lies in the insurance industry.
Insurance companies often need to assess risks from extreme phenomena such as large storms or widespread floods to build products and set premiums.
However, detailed forecasting of unusual events has always been very costly.
Traditional weather forecasting often uses the gathering method, in which the system must run in parallel with many different scenarios to see how a phenomenon can unfold.
To detect rare risks, the number of scenarios needs to increase sharply, causing calculation costs and processing time to be very large.
According to Mr. Pritchard, AI helps solve this bottleneck. “When trained, AI is 1,000 times faster. That allows operating large-scale simulation systems that were previously very difficult to implement. Insurance companies can now run systems with about 10,000 members,” he said.
The three Earth-2 models that Nvidia announced this time include: a 15-day weather forecast model; a model specializing in monitoring and forecasting strong storms across the United States for six hours; and a model used to integrate data from many different weather sensors, helping to create a more accurate starting point for other forecasting technologies.
According to observers, if these models are widely deployed, AI could create a turning point in how humans monitor, forecast and respond to extreme weather.
