Published on November 16, 2023, 4:13 pm
DeepMind, a leading artificial intelligence (AI) company, has unveiled GraphCast, an AI-powered weather forecasting program that can predict weather variables up to 10 days in advance. According to DeepMind, this system is the most accurate 10-day global weather forecasting system in the world.
The traditional approach to weather prediction relies on Numerical Weather Prediction (NWP), which involves using physical equations and computer algorithms. However, this process is time-consuming, requires expertise, and consumes expensive computing resources. DeepMind’s GraphCast, on the other hand, uses machine learning and historical weather data to learn the cause-and-effect relationships that shape weather patterns.
What sets GraphCast apart is its ability to achieve a remarkable 90 percent accuracy rate. It only needs data on the weather six hours ago and the current state of the weather to predict conditions six hours into the future for up to 10 days ahead. This efficiency makes it more advanced and faster than traditional forecasting technologies.
The system was trained on four decades of weather reanalysis data from the European Centre for Medium-Range Weather Forecasts’ ERA5 dataset. This dataset is based on historical imagery, radar readings, and information from various weather stations. By utilizing this wealth of information, GraphCast can fill in gaps in historical data by using NWP.
One notable aspect mentioned by DeepMind is the computational intensity involved in training the model. Despite this challenge, GraphCast runs efficiently on a single Google TPU v4 machine. As technology advances and more recent data becomes available, DeepMind expects even greater accuracy in predicting weather patterns related to climate change.
In addition to its ability to predict long-term weather variables accurately, GraphCast can also forecast severe weather events such as tropical cyclones and extreme temperature waves with great accuracy further into the future. This capability makes it an invaluable tool in a world facing increasingly extreme weather conditions.
DeepMind also highlights another recent achievement by Google, the MetNet-3 system. MetNet-3 is currently the most accurate 24-hour regional forecasting system. DeepMind believes that these AI-powered systems can empower the global community to tackle environmental challenges by providing fast and accurate forecasts.
Fast and accurate weather forecasts are more crucial than ever in a world experiencing ever-increasing extreme weather events. With GraphCast’s impressive capabilities, it has the potential to transform how we predict and prepare for weather conditions, ultimately helping us mitigate the impacts of climate change.