Researchers at Google’s AI research lab, Deepmind, have developed a new AI weather forecasting model called GenCast. In a paper published in Nature, the company explains that this model provides faster and more accurate predictions up to 15 days in advance. Google Deepmind also explained how this new model works in a blog post. Here are all the details
Google DeepMind’s GenCast weather tracking AI model: How it works
This model, which outperforms the current leading system, uses probabilistic ensemble forecasting to predict a range of possible weather scenarios, offering a more comprehensive picture of the upcoming conditions.
The researchers are also making GenCast’s code, weights, and forecasts publicly available to benefit the wider weather forecasting community. This new model can be particularly crucial as climate change increases the frequency of extreme weather events, making reliable forecasts more important than ever.
Unlike older models that just gave one “best guess,” GenCast can provide many possible weather predictions, like having 50 different weather reports. This can help in understanding how the weather may change and what different things could happen.
GenCast uses a special kind of AI that’s good at creating things, like pictures, videos, and music. But instead of creating those things, GenCast creates different weather possibilities. It’s like having the AI imagining many ways the weather could go.
To teach GenCast how to do this, scientists showed it 40 years of weather data, like temperatures, wind speeds, and how much the air pushes on things. This helped GenCast learn all about how weather works all over the world.
To check GenCast’s efficiency, scientists tested it using weather data from 2019. They compared the model to the best weather prediction system that is currently being used and GenCast was better at predicting the weather most of the time.
Researchers looked at lots of different weather things, like temperature and wind, and GenCast was more accurate in almost all of them, especially when predicting further into the coming days.
GenCast is also good at knowing when it’s not sure about the weather. Just like when it is not sure if it will rain, the model will say “maybe” instead of being inaccurate.
The best part is that GenCast is super fast. It can predict the weather for 15 days in just 8 minutes. The old way of predicting weather usually takes hours.
