Image: © knelson20/Stock.adobe.com
After claiming to hit a breakthrough by predicting the structure of nearly every known protein, DeepMind is now turning its AI models to weather forecasting.
Google-owned DeepMind claims its latest AI model can make accurate, fast predictions of the weather and give earlier warnings of extreme storms.
The company claims its AI model – GraphCast – can predict weather conditions up to 10 days in advance, in a more accurate way than standard industry methods. DeepMind also said this model can make prediction in less than one minute.
There are estimates that 10-day weather forecasts are only accurate about half of the time, compared to a 90pc accuracy rate for five-day forecasts. Improving weather prediction presents benefits for both citizens and various industries, such as renewable energy and event organisers.
DeepMind also said its AI model can track cyclones with great accuracy, identify flood risks and predict the onset of extreme temperatures.
“GraphCast takes a significant step forward in AI for weather prediction, offering more accurate and efficient forecasts and opening paths to support decision-making critical to the needs of our industries and societies,” DeepMind said in a blogpost.
“By open-sourcing the model code for GraphCast, we are enabling scientists and forecasters around the world to benefit billions of people in their everyday lives.”
The company said GraphCast is already being used by the European Centre for Medium-Range Weather Forecasts. This institution is currently running a live experiment of the AI model on its website.
Combining old and new methods
DeepMind said its AI model uses deep learning to create its weather forecast system, instead of the usual method of physical equations called Numerical Weather Prediction (NWP).
The company said GraphCast is trained on “decades of historical weather data” to help it predict how weather patterns evolve and that it combines elements of traditional weather prediction. Despite this, DeepMind claims in a study that the model is rather small compared to other AI models, containing 36.7m parameters.
“This trove is based on historical weather observations such as satellite images, radar, and weather stations using a traditional NWP to ‘fill in the blanks’ where the observations are incomplete, to reconstruct a rich record of global historical weather,” DeepMind said.
DeepMind said in its study that the data from traditional weather-prediction sources is “invaluable” for this type of AI model and said its new approach should not be viewed as a replacement for traditional weather forecasting methods, as they have been rigorously tested in many real-world contexts and offer “many features we have not yet explored”.
“Rather our work should be interpreted as evidence that [machine-learning weather prediction] is able to meet the challenges of real-world forecasting problems and has potential to complement and improve the current best methods,” the company said.
Last year, DeepMind claimed to achieve a scientific breakthrough when its AlphaFold model predicted the structure of nearly every protein known to science – more than 200m in total.
At the end of last month, DeepMind claimed the next version of AlphaFold can predict nearly all molecules in the Protein Data Bank – a database for the 3D structures of various biological molecules.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.
