Thunder on Indonesia seen from International Space Station
NASA Earth Observatory / International Space Station (ISS)
The AI weather program running for a second on the desktop matches the accuracy of traditional forecasts that take hours or day on a powerful supercomputer, claiming its creators.
The weather forecasts, since the 1950s, depend on the physics-based models that exert extracts from the comments made using satellites, balloons, and weather stations. But these calculations, known as numerical seasonal prediction (NWP), are extremely intensive and rely on vast, expensive and energy-lung super computers.
In recent years, researchers have tried to streamline this process by implementing AI. Google scientists created an AI tool last year that could replace a small piece of complex code in each cell of a weather model, allowing computer power dramatically. Deepmind later took it even further and used AI to change the entire forecast. This approach has been adopted by the European Center for Medium-rang Weather Forecast (ECMWF), which Launched a device Last month, Artificial Intelligence is called Forecasting System.
But this gradual expansion of the role of AI in the weather prediction has reduced by changing all traditional number of crunching-a new model has been created Richard turner The University of Cambridge and their colleagues is to be replaced.
The Turner says that the previous work was limited to the forecast, and passed at a step called Inquiology, where data from satellites, balloons and weather stations around the world collided, cleaned, manipulated and merged, manipulated and merged, which could lead to the forecast. “This is actually half of computational resources,” says Turner.
Researchers created a model called Aardvark Weather, which replaces both the first, forecast and early stages. It only uses 10 percent of input data that perform existing systems, but can achieve the results that assess their method in the latest NWP forecast, report turner and their colleagues.
Creating a complete forecast, which will take hours or days on a powerful supercomputer for NWP forecast, can be done in about 1 second on a single desktop computer using Aardvark.
However, Aardvark is using a grid model of the Earth’s surface with cells that are 1.5 degrees square, while ECMWF uses a grid with ERA5 model cells Small as 0.3 degreesThis means that the model of Aardvark is very thicker to take complex and unexpected weather patterns, David shults At the University of Manchester, UK.
“Many unresolved things are going on which can blow your forecast,” Shults is called. “They are not representing the extreme. They cannot solve it on this scale.”
Turner argues that Aardvark may actually beat some existing models in taking unusual events such as Cyclone. But he believes that AI models fully rely on those physics-based models for training like him. “It doesn’t work at all if you take away their training data and simply use observation data to train,” they say. “We tried to do so, and went to the physics model-free, but did not work.”
He believes that the future of weather forecasts can sometimes be scientists working on an accurate physics-based model, who are then used to train the AI model that repeat their output rapidly and with low hardware. Some are even more optimistic about AI’s possibilities.
Nikita Gurianov Oxford is believed to be at the University of Oxford, over time, the AI will be able to make weather forecasting that actually crosses NWP. They will be trained on observation and historical weather figures alone, making a completely independent accurate forecast from NWP, they say. “This is a question of scale, but also a question of cleverness. You have to be clever from how you feed the data – and how you structure the nerve network.”
Subject:
