New AI systems could speed up our ability to create weather forecasts

The first, developed by Huawei, details how its new AI model, Pangu-Weather, can predict weekly weather patterns around the world much faster than traditional forecasting methods, but with comparable accuracy.

The second demonstrates how a deep learning algorithm was able to predict extreme rainfall more accurately and with more warning than other leading methods, ranking first about 70% of the time in tests against similar existing systems.

If adopted, these models could be used alongside conventional weather forecasting methods to improve authorities’ ability to prepare for severe weather, said Lingxi Xie, senior research scientist at Huawei.

To build Pangu-Weather, Huawei researchers built a deep neural network trained on 39 years of reanalysis data, which combines historical weather observations with modern models. Unlike conventional methods that analyze weather variables one at a time, which could take hours, Pangu-Weather is able to analyze them all at once in seconds.

Researchers tested Pangu-Weather against one of the world’s leading conventional weather forecasting systems, the European Center for Medium-Range Weather Forecasting (ECMWF) Operational Integrated Forecasting System, and found that it yielded similar accuracy.

Pangu-Weather was also able to accurately plot the path of a tropical cyclone, despite not having been trained with tropical cyclone data. This finding shows that machine learning models are able to grasp physical weather processes and generalize them to situations they have never seen before, says Oliver Fuhrer, head of the numerical forecasting department at MeteoSwiss, the Swiss Federal Office of Meteorology and climatology. He was not involved in the research.

Pangu-Weather is exciting because it can predict the weather much faster than scientists were able to before and predict things that weren’t in its original training data, Fuhrer says.

Over the past year, several tech companies have unveiled AI models that aim to improve weather forecasting. Pangu-Weather and similar models, such as Nvidia’s FourcastNet and Google-DeepMind’s GraphCast, are prompting meteorologists to « rethink how we use machine learning and weather forecasting, » says Peter Dueben, head of Earth system modeling at ECMWF. He was not involved in the research but did test Pangu-Weather.