Google’s new AI weather model, GenCast, is proving to be pretty impressive. It’s so accurate that it can compete with traditional weather forecasting methods. In fact, it even outperformed a top weather model during tests using 2019 data.
GenCast is still not a replacement for traditional models, but it’s adding a valuable tool to the toolbox. It could help make weather predictions even more accurate and help warn people about dangerous storms.
The model uses decades of weather data, from 1979 to 2018, to spot patterns and predict future weather. Unlike older models, which use complex equations to simulate the atmosphere, GenCast uses machine learning. Both types of models give a range of possible outcomes, but GenCast has shown it can be better at predicting things like tropical cyclones and extreme weather. It even gave an extra 12 hours of warning on average for cyclone paths.
However, GenCast was tested on an older version of a leading weather model, ENS, which has since improved. Still, GenCast was able to beat the older version, and similar tests with newer data have shown similar results.
Another benefit of GenCast is speed. It can produce a 15-day forecast in just eight minutes, while traditional models might need several hours. This makes GenCast much more efficient and less resource-hungry.
But GenCast isn’t perfect. It could be improved in areas like increasing its resolution for more detailed predictions. Right now, it predicts weather in 12-hour intervals, while traditional models predict in shorter time frames, which can be important for things like wind power forecasts.
Despite all the excitement, some meteorologists are still unsure about relying on AI. While GenCast is impressive, it’s not a substitute for physics-based models just yet. Still, GenCast’s code is open-source, so anyone can try it out.
In the future, it’s likely that AI models like GenCast will be used alongside traditional forecasting methods, making weather predictions more accurate and accessible to everyone.