The AI weather forecasting tool, Graphcast, has been trained using around 40 years of historical weather data from the European Centre for Medium-Range Weather Forecasts (ECMWF).
According to a report by the Financial Times, by analyzing how weather systems developed and moved globally in the past, the system is expected to make accurate predictions up to ten days in advance.
ECMWF meteorologist Matthew Chantry mentioned that the AI advancements in weather forecasting are “much faster and more impressive than expected.”
The AI takes the current state of the atmosphere worldwide as input and then generates a ten-day forecast within minutes, as stated in the report. This is significantly more efficient than the complex physical models currently relied upon by meteorologists, which can take hours to run on supercomputers.
In extensive tests, Graphcast reportedly outperformed ECMWF’s forecasting system in 90 percent of measurements, including temperature, wind speed, and air pressure, as reported by the Financial Times. It also surpassed weather models developed by Huawei and Nvidia. The system successfully predicted the location of Hurricane Lee’s landfall in Nova Scotia nine days in advance, three days earlier than conventional methods.
Despite facing challenges in predicting rapid intensifications like Hurricane Otis, experts see promising potential in combining the data-driven approach with traditional weather forecasting models, according to the report. The UK’s Met Office plans to develop a similar AI system and integrate it with its physics-based supercomputer models.
READ MORE: Nothing Phone 2 Breaks Ground: First to Integrate iMessage, Sparks Privacy Concerns
