GenCast: The AI Revolutionizing Weather Prediction

By Nikhil Sudarshan

The Lawrenceville School, NJ

On December 4, 2024, Google published a paper outlining the functions of its new AI model. It is a revolutionary piece of technology able to not only accurately predict weather events, but also do much better than the present forecasting methods. How does it do this, and what makes it superior?

GenCast works by creating a range of possible outcomes and the probability of each outcome occurring, training from a large library of previous weather data called ERA5 and using it to find patterns. From there, it ranks what is most likely to happen as well as listing the relative confidence level for each end result.

On the other hand, weather forecasting systems set in place today combine many different measurements into a single prediction, but this does not provide any measurement of the confidence of this result. This allows GenCast to gain an edge over these traditional systems because it is more nuanced, clustering a wide range of probable outcomes into a tight band to show what will probably happen over time. In fact, in 1320 trials from past weather data, GenCast consistently outperformed the European Centre for Medium-Range Weather Forecasts’ ensemble 97.2% of the time (Price & Wilson, 2024).

From all the different calculations needed to generate these predictions, one may think that GenCast copious amounts of time and much more computing power than ENS. However, it is really the opposite. GenCast can generate a 15 day forecast in around 8 minutes on a regular TPU (tensor processing unit), which is much faster than the ENS, needing hours on a supercomputer to predict an end result (Price & Wilson, 2024).

This fast generation of highly accurate predictions will allow many more lives to be saved and damage to be mitigated, especially regarding natural disasters. It will let more and better defensive measures be taken earlier, as well as preparing people, in advance, for the worst. Not only this, but renewable energy sources’ usage will drastically increase since these energy options heavily rely on the weather.

However, as with any source, GenCast still has its limitations. Although forecasts can be generated up to two weeks in the future, the inconsistency of the Earth’s climate will make it hard for these forecasts to be reliable and trustworthy at the start. GenCast is also really good at tracking where natural disasters will go, but especially for hurricanes, the intensity is not predicted, which is a big factor in deciding what the best course of action is to take.

GenCast marks the start of better forecasting, paving the way to deal with natural disasters better and allowing advances in other areas. Even so, it still has a long way to go, and there are still many steps to walk before we are able to completely predict the disorder of daily life. 

----Works Cited

Broad, W. J. (2024, December 4). Google Introduces A.I. Agent That Aces 15-Day Weather Forecasts. The New York

Times. Retrieved December 21, 2024, from https://www.nytimes.com/2024/12/04/science/google-ai-weather-forecast.html

Price, I., & Willson, M. (2024, December 4). GenCast predicts weather and the risks of extreme conditions with state-of-the-

art accuracy. Google DeepMind. Retrieved December 21, 2024, from https://deepmind.google/discover/blog/gencast-predicts-weather-and-the-risks-of-extreme-conditions-with-sota-accuracy/

Price, I., Wilson, M., Sanchez-Gonzalez, A., Alet, F., Andersson, T. R., El-Kadi, A., Masters, D., Ewalds, T., Stott, J., Mohamed,

S., Battaglia, P., & Lam, R. (2024, December 4). Probabilistic weather forecasting with machine learning. Nature. Retrieved December 22, 2024, from https://www.nature.com/articles/s41586-024-08252-9