Goodbye, unreliable weather forecasts? Google DeepMind's AI model sets new standards for 15-day forecasts

DeepMind has developed a new generative weather forecast model
It was trained using 39 years of global weather data
It is more accurate than the current forecasting standard 97.2% of the time

If you’re still surprised and frustrated by inaccurate weather forecasts, Google has good news: DeepMind, its AI research lab, has unveiled a generative model that can predict the weather with incredible accuracy up to 15 days in advance.

To put it simply, GenCast is an AI model similar to ChatGPT. The difference with GenCast is that it has been specially adapted to the geometry of the Earth. Using current weather data, it is able to create future weather scenarios and suggest the most likely outcome.

The DeepMind team trained GenCast using archived weather data collected between 1979 and 2018. This included temperature, wind speed and air pressure readings from around the world. Based on this, the model was able to learn global weather patterns.

To test its effectiveness, Google then compared the model’s predictions with the industry’s current best forecasting tool, Ensemble Forecast (ENS).

The model should produce a forecast for 2019 with 1,320 combinations of variables and lead times. GenCast was more accurate than ENS in 97.2% of cases. For predictions made more than 36 hours in advance, this rose to 99.8%. This accuracy also applied to forecasting extreme weather events such as tropical cyclones.

What’s even more remarkable is that, according to DeepMind, a single Google Cloud Tensor Processing Unit v5 – the circuitry Google uses to accelerate machine learning tasks – took just eight minutes to produce a 15-day forecast. Traditional ensemble forecasts take several hours and run on supercomputers with thousands of processors.

No replacement for meteorologists

A laptop screen shows DeepMind's AI weather model predicting the path of a typhoon

(Image credit: Google DeepMind)

Traditional ensemble forecasts project weather patterns based on complex physical calculations using collected data. GenCast uses the same data and generates the most likely scenario based on all the insights from historical weather data.

GenCast won’t put meteorologists out of work. The model is only as reliable as the data it is trained on, and in a changing climate, past weather patterns may not prove a reliable basis for predictions far into the future. There are also a number of atmospheric variables that GenCast cannot account for, meaning manual equations are still required to arrive at a reliable forecast.

It is likely that GenCast will become another tool in the meteorological arsenal, with inputs taken into account alongside other data sources. Human experts are still required to assess the most likely scenario.

GenCast is also not the first use of deep learning to obtain more precise weather forecasts. FourCastNet is a data-driven forecast model from Nvidia, while Huawei’s Pangu Weather model has proven in testing to be more accurate than numerical weather forecasts. Atmo, a San Francisco company, is also developing AI models to improve forecasts based on real-time weather data.

DeepMind sees applications for GenCast that go beyond standard forecasting. By indicating the likelihood of different weather events, GenCast allows officials to plan for different outcomes and allocate resources accordingly. Google also sees a future where wind power forecasts could influence renewable energy planning.

Currently, the DeepMind team continues to research and develop GenCast as one of several AI-based weather models, including its deterministic medium-term forecasts. The model is open and DeepMind will soon release real-time and historical forecasts for integration into other models.

Breaking News

Dellupodisabato

Goodbye, unreliable weather forecasts? Google DeepMind’s AI model sets new standards for 15-day forecasts

Leave a Reply Cancel reply