 |
| Artificial intelligence (AI) model-based methods show some potential in accelerating weather forecasting by orders of magnitude. Illustration photo. (Source: nchmf.gov.vn) |
According to a recent research report published in the journal Nature , Chinese researchers have developed an artificial intelligence (AI) model with 3-dimensional neural networks to forecast global weather with medium to high accuracy.
In their research, the large meteorological model research and development team of Huawei Cloud has proposed a 3D neural network that adapts to the Earth's coordinate system to process complex and heterogeneous 3D meteorological data.
Trained on nearly 40 years of global data, the relatively large Pangu-Weather meteorological model obtained 100 million-level parameters within two months.
Pangu-Weather takes just 1.4 seconds to complete a 24-hour global weather forecast, including humidity, wind speed, temperature, sea level pressure, and more. The system's prediction speed is 10,000 times faster than traditional numerical methods.
In a specific example, for super typhoon Mawar last May, Pangu-Weather performed brilliantly by predicting the path of the storm five days in advance.
The model also shows better forecast results, based on reanalysis data from all tested events, when compared with numerical weather prediction (NWP) methods.
Today, daily weather forecasts, severe natural disaster warnings, and climate change predictions are all made using this method, which relies on high-performance computing and complex physical models. However, this method requires a lot of machines and computing time.
Tian Qi, author of the report and senior AI expert at Chinese cloud service provider Huawei Cloud, said the conventional NWP method requires 4-5 hours of computing on a supercomputer cluster with 3,000 servers to forecast global weather for the next 10 days.
However, the accuracy of these new AI-based methods is still significantly lower than that of NWP methods. Bi Kaifeng, co-author of the report, also acknowledged the limitations of AI-based weather forecasting systems, saying that they still rely heavily on reanalysis data and need to improve their ability to predict extreme weather conditions.
“We believe that AI-based methods should coexist with conventional numerical methods to provide more accurate and reliable weather forecasting services,” said Qi Tian.
Source
Comment (0)