Google recently released a revolutionary new atmospheric circulation model, NeuralGCM, whose computational efficiency is an astonishing 100,000 times higher than traditional physical models. This breakthrough is equivalent to the total progress in high-performance computing in the past 25 years. This means that climate change forecasting will usher in a new era, where scientists can predict future weather more quickly and accurately, and delve into the impact of climate change on different regions around the world, such as the probability of extreme weather events such as droughts and floods. and intensity.
Recently, Google made an amazing breakthrough in the field of weather forecasting. They developed a new atmospheric general circulation model called NeuralGCM. The computational efficiency of this model is a full 100,000 times higher than the traditional physical model, which is equivalent to the progress of high-performance computing in the past 25 years.
Google CEO announced the results on social media and pointed out that NeuralGCM will provide scientists with a new climate change prediction tool. This helps researchers understand the impact of climate change on different regions at a time when global temperatures are rising sharply, such as which areas may suffer from long-term droughts or the risk of flooding in coastal areas.
Traditional weather forecasting models usually rely on the laws of physics, dividing the earth into cubes with a side length of 50 to 100 kilometers, and calculating weather changes in these areas. However, this method is too large and many important climate processes are ignored. Differently, NeuralGCM uses neural networks to learn the physical principles of small-scale weather events from existing data, greatly improving the accuracy of simulations.
NeuralGCM was trained on weather data from 1979 to 2019 and demonstrated weather forecast accuracy exceeding existing state-of-the-art physical models within 2 to 15 days. In terms of climate prediction, NeuralGCM's performance is also quite impressive, especially in temperature prediction, whose error is only one-third of that of traditional models.
In addition, NeuralGCM is extremely efficient in terms of running speed and calculation cost. Compared with traditional models, it is 3,500 times faster, and the calculation cost is 100,000 times lower than X-SHiELD. It only requires an ordinary computer to run.
The launch of NeuralGCM marks a major leap forward in the field of climate modeling. It not only provides new possibilities for future weather forecasts, but also provides stronger support for our research on climate change.
Paper address: https://t.co/zyXhW8deko
Highlights:
? The computational efficiency of the NeuralGCM model is 100,000 times higher than that of traditional physical models, and it can simulate 22 days of weather in 30 seconds!
NeuralGCM's accuracy surpasses existing state-of-the-art models in weather forecasts ranging from 2 to 15 days.
? Its computing cost is 100,000 times lower than the traditional model, and it can be run efficiently using ordinary computers.
The emergence of NeuralGCM has brought unprecedented efficiency and accuracy to climate prediction and weather forecasting, providing a powerful tool to deal with climate change, indicating more accurate climate prediction and more effective response strategies in the future.