Google introduces GenCast, an AI model for weather forecasting
AI is Google’s new favorite hammer and the next nail on its path is weather forecasting. The company is introducing GenCast, a “high resolution AI ensemble model”, which was detailed in a paper published in Nature.
Accurate weather forecasting is important for anything from your day-to-day life to disaster preparedness and even renewable energy. And GenCast beats the current top system, ECMWF’s ENS, in forecasts up to 25 days in advance.
GenCast is a diffusion model, similar to those you may have seen in AI image generators. However, this one is tuned specifically for Earth’s geometry. It was trained on four decades of historical data from ECMWF’s archives.
To test it, Google trained GenCast on historical weather data up to 2018 and ran 1,320 different forecasts for 2019 and compared its output against ENS and the actual weather. GenCast was more accurate than ENS in 97.2% of cases, going up to 99.8% more accurate for forecasts for 36 hours ahead or longer.
Here’s a demo. Google tasked GenCast with forecasting the path of Typhoon Hagibis, which hit Japan in 2019. You can see the path that the typhoon took in red, in blue are the possible paths predicted by Google’s AI model. At 7 days out, they are pretty spread out, but they narrow in on the actual path as the typhoon gets closer to landfall.
GenCast predicting the path of Typhoon Hagibis
Giving local authorities more time to prepare for severe weather is one use case. GenCast can also predict wind speeds near wind farms, the weather over solar farms and so on.
GenCast is an “ensemble model”, which means it produces 50+ predictions with different probabilities. One such prediction spanning a 15-day forecast can be generated in 8 minutes on a Google Cloud TPU v5, says Google. The multiple predictions can be done in parallel. Meanwhile, a traditional weather forecast model takes hours on a supercomputer.
Google is releasing GenCast as an open model and is sharing its code and weights. The company plans to continue cooperating with weather forecasting agencies and scientists going forward to make future forecasts even better.
