Powered by

Advertisment
Home Extreme Weather

AI can now forecast weather 10,000 times faster than humans

Nature magazine has recently highlighted the presentation of two new weather forecasting systems based on Artificial Intelligence (AI).

By groundreportdesk
New Update
Western disturbance brings rain and snow to north India

Nature magazine has recently highlighted the presentation of two new weather forecasting systems based on Artificial Intelligence (AI). These systems have the potential to greatly improve the quality of forecasts.

In fact, for the moment they already demonstrate a precision in their analyzes comparable to that of the methods currently used by meteorologists. The spectacular reduction in the time required to carry out the forecast and the ability to anticipate specific phenomena that are difficult to predict are some of the advantages of these two methods.

Compared to the hours of work required by the conventional system now used by professionals, the so-called numerical method, the new AI-based models are much faster and only need seconds. So much so that one of the new models, developed by researchers from the Huawei company in Shenzhen (China), obtains results 10,000 times faster and with a degree of precision similar to that of the integrated operational forecasting system of the European Center for Meteorological Prediction, the most powerful in the world today.

Any computer can use the new model, which has been fed with 39 years of meteorological data from the planet to forecast the conditions of temperature, wind, pressure, and humidity at different heights.

Better forecast extreme events

The second of the new AI-generated systems comes from the Tsinghua University School of Software, also in China, and the University of California (USA). Scientists from both universities have developed a model that predicts the precipitation that will occur in a specific location in a very short time, just six hours in advance. In fact, the model has been dubbed Nowcast.

The objective in this case is to more accurately forecast extreme weather events, capable of causing damage or endangering the population, such as hurricanes, tornadoes or tsunamis. It is a system that combines physical rules and deep learning. It works from the existing data and respects the underlying physical processes.

Tests carried out in specific cases to forecast rainfall inland areas of only 2,000 square kilometres yielded really encouraging results since this new system improved the forecast compared to the classic system in 70% of the cases. In any case, the researchers admit that there is still room for improvement in this model.

The fact that the speed of the process is improved so much is key to reducing the economic budget that is currently needed for the computer equipment of the meteorological centres. With less money, it will be possible to obtain the same quality of information and more quickly, say the researchers.

However, the experts consider that, at least for now, AI ​​will have to coexist with the conventional predictions directed by real professionals, for now, difficult to replace. Especially in the case of extreme events, for which there is not a long history of data available precisely because of their exceptional nature, human intervention is still necessary.

The climate catch

AI systems, lacking historical or predictive data, may face limitations in forecasting continuously record-breaking climate change-induced events. Colorado State University researchers Imme Ebert-Uphoff and Kyle Hilburn said that in the face of entirely new weather conditions, AI models can generate "highly erratic predictions," similar to patterns seen in other forms of AI.

This has the potential to undermine the progress made by meteorologists in achieving accurate and reliable forecasts, which are crucial for governments to issue timely public safety announcements and organize evacuations to protect vulnerable communities.

AI-based weather forecasting

The field of AI-based weather forecasting is rapidly evolving. Just two years ago, scientists suggested in a paper published in a Royal Society journal that AI weather models could produce results equal to or even better than numerical models. However, the researchers stressed that several fundamental advances would be needed before numerical weather models become obsolete.

Promising developments in the field, such as Pangu-Weather and NowcastNet, indicate progress towards these advances. The computational speed exhibited by models like Pangu-Weather has the potential for significant benefits, according to researchers at Colorado State University.

However, there are challenges that AI systems may face, particularly as the planet experiences rising temperatures and more intense extreme weather events due to climate change. Experts caution that AI models could have a hard time accurately simulating these heightened events, as they learn primarily from historical weather data. As extreme weather events become more unprecedented and less represented in historical records, the behavior of AI systems can become unpredictable and produce erratic predictions.

Studies evaluating AI performance

The paucity of extreme events in historical records poses a challenge for AI weather models, and studies evaluating AI performance in capturing extreme events with limited data have returned mixed results. Much remains to be explored regarding how AI models will work in a warmer climate.

Russ Schumacher, a Colorado state climatologist and scientist at Colorado State University, suggests that hybrid models that combine AI and numerical components may encounter less difficulty with record-breaking events. However, for AI-powered models alone, their response to situations outside the historical record remains unclear.

These considerations are crucial as researchers continue to develop AI weather models. Evaluating performance on routine daily weather forecasts and high impact events is essential. While AI weather models hold promise, they may not completely replace conventional approaches. Numerical models, AI models and human experience all have their strengths and play a valuable role in the synthesis and communication of weather information.

"In my opinion, ideally we get to a point where the field of meteorology can draw on the strengths of all approaches," says Russ Schumacher.

Keep Reading

Follow Ground Report for Climate Change and Under-Reported issues in India. Connect with us on FacebookTwitterKoo AppInstagramWhatsapp and YouTube. Write us on [email protected].