A machine learning-based weather prediction program developed by DeepMind researchers called “GraphCast” can predict weather variables over the span of 10 days, in under one minute. In a report, scientists highlight that GraphCast has outperformed traditional weather pattern prediction technologies at a 90% verification rate.
The AI-powered weather prediction program works by taking in “the two most recent states of Earth’s weather,” which includes the variables from the time of the test and six hours prior. Using that data, GraphCast can predict what the state of the weather will be in six hours.
In practice, AI has already showcased its applicability in the real world. The tool predicted the landfall of Hurricane Lee in Long Island 10 days before it happened, while the traditional weather prediction technologies being used by meteorologists at the time lagged behind. Forecasts made by standard weather simulations can take longer because traditionally, models have to account for complicated physics and fluid dynamics to make accurate predictions.
Not only does the weather prediction algorithm outperform traditional technologies to forecast weather patterns in terms of pace and scale, GraphCast can also predict severe weather events, which includes tropical cyclones and waves of extreme temperatures over regions. And because the algorithm can be re-trained with recent data, scientists believe that the tool will only get better at predicting oscillations in weather patterns that coincide with grander changes that align with climate change.
Soon, GraphCast, or at least the basis of the AI algorithm that powers its predictions, might pop up into more mainstream services. According to Wired, Google might be exploring how to integrate GraphCast into its products. The call for better storm modeling has already paved a path for supercomputers in the space. The NOAA (National Oceanic and Atmospheric Administration) says it has been working to develop models that will provide more accurate readings on when severe weather events might occur and importantly, the intensity forecasts for hurricanes.
A machine learning-based weather prediction program developed by DeepMind researchers called “GraphCast” can predict weather variables over the span of 10 days, in under one minute. In a report, scientists highlight that GraphCast has outperformed traditional weather pattern prediction technologies at a 90% verification rate.
The AI-powered weather prediction program works by taking in “the two most recent states of Earth’s weather,” which includes the variables from the time of the test and six hours prior. Using that data, GraphCast can predict what the state of the weather will be in six hours.
In practice, AI has already showcased its applicability in the real world. The tool predicted the landfall of Hurricane Lee in Long Island 10 days before it happened, while the traditional weather prediction technologies being used by meteorologists at the time lagged behind. Forecasts made by standard weather simulations can take longer because traditionally, models have to account for complicated physics and fluid dynamics to make accurate predictions.
Not only does the weather prediction algorithm outperform traditional technologies to forecast weather patterns in terms of pace and scale, GraphCast can also predict severe weather events, which includes tropical cyclones and waves of extreme temperatures over regions. And because the algorithm can be re-trained with recent data, scientists believe that the tool will only get better at predicting oscillations in weather patterns that coincide with grander changes that align with climate change.
Soon, GraphCast, or at least the basis of the AI algorithm that powers its predictions, might pop up into more mainstream services. According to Wired, Google might be exploring how to integrate GraphCast into its products. The call for better storm modeling has already paved a path for supercomputers in the space. The NOAA (National Oceanic and Atmospheric Administration) says it has been working to develop models that will provide more accurate readings on when severe weather events might occur and importantly, the intensity forecasts for hurricanes.