For decades, morning weather forecasts have relied on the same kind of traditional models. Today, weather forecasting is about to join the ranks of industries revolutionized by artificial intelligence.
A pair of papers published Wednesday in a scientific journal Naturetouts the potential of two new AI prediction approaches. These systems may produce faster and more accurate results than traditional models, researchers say.
They are part of a new wave of AI models that are sweeping the global meteorological community. And they have the potential to transform the prediction industry.
But experts warn that climate change could pose unique challenges for burgeoning AI weather models.
AI systems learn how to generate accurate forecasts based on historical weather data. However, certain types of weather events, such as heat waves and hurricanes, are becoming more intense as the planet warms, and in some cases extremes rarely seen in the historical record. This can make it difficult for AI weather models to accurately simulate unprecedented, record-breaking events.
These are issues that AI experts are still investigating. Yet the new Nature According to the paper, the world of AI weather forecasting is developing rapidly.
The first paper describes a model called Pangu-Weather. The model predicts various global weather variables such as temperature and wind speed up to about a week in advance. Developed by researchers at Chinese technology company Huawei Technologies Co. Ltd., the model can deliver results up to 10,000 times faster than traditional models.
Researchers have found that they can accurately track the path of tropical cyclones. And it’s slightly more accurate than the European Center for Medium-Term Weather Forecasts, one of the world’s leading weather centers.
Still, Pangu weather has some limitations. The researchers did not examine the performance of precipitation, which is a key weather variable and one of the most difficult to capture accurately in models.
The second paper, on the other hand, deals mainly with rainfall. This article describes an AI system known as NowcastNet. This program specializes in short-term forecasts up to just a few hours ahead. Researchers have found that NowcastNet can outperform many of its major competitors.
Pangu-Weather and NowcastNet are among the latest in a recent wave of new AI weather models, many of which were developed by private companies rather than the government agencies that have traditionally controlled weather. These programs differ from traditional forecasting systems in some fundamental ways.
Conventional weather forecasting relies on a system known as numerical weather forecasting. It is a kind of mathematical model that uses complex equations to predict how the weather system will change over time and space. These equations describe the actual physics behind the movement of air and water in the atmosphere and oceans.
With so much mathematics and physics involved, numerical weather models require a very high level of computational power. Therefore, it is costly and time consuming to execute. It also limits the microscopic processes that these models can accurately capture. For example, things like the physics of individual clouds are difficult to simulate in models that make large-scale global predictions.
Scientists have come up with various ways to circumvent these problems in traditional models. One strategy is a method known as parameterization. This is how scientists replace the real physical equations in the model with simplified programs that generally capture the process without forcing the real physics into the model.
But enthusiasts argue that artificial intelligence could replace these workarounds, yielding faster and more accurate results.
AI models don’t need to represent real-world physics in mathematical form. Instead, it ingests large amounts of historical weather data and learns to recognize patterns. These patterns are then used to make predictions when presented with new data about current weather conditions.
For decades, scientists have been working to integrate AI components into traditional weather models to make them run faster and cheaper. Also, some companies are now developing full AI models such as Pangu-Weather and NowcastNet, which can completely replace numerical model systems.
It’s a rapidly evolving field. Just two years ago, in a paper published in the Journal of the Royal Society, scientists suggested that AI weather models “could” produce results similar to or better than numerical models.
“It’s not unthinkable that numerical weather models will one day become obsolete, but many fundamental breakthroughs are needed to reach this goal,” the researchers said.
Emerging approaches such as Pangu-Weather and NowcastNet suggest that such breakthroughs are underway. And there is potential in this area, Colorado State University researchers Yme Ebert Apukhov and Kyle Hilburn said in comments about the new study, also released Wednesday. Nature.
In principle, the much faster computational speed exhibited by models such as Pangu-Weather “could be of immense benefit,” they write.
On the other hand, AI systems still have some potential hurdles, especially as the planet warms.
Experts warn that AI models may run into problems simulating extreme weather as climate change intensifies extreme weather.
Heatwaves, droughts, hurricanes, wildfires, and a myriad of other climate-related phenomena are all becoming more extreme as temperatures rise, some of which plunge into unprecedented territory. Heat records were broken around the world last week alone, with scientists warning the planet is likely experiencing the hottest day in human history.
Accurately predicting extreme weather events is one of the most important functions of weather models, allowing decision makers to make public safety announcements and protect vulnerable populations in sufficient time. You will be able to promote evacuation. But AI models learn how to use historical weather data to generate forecasts. And as the weather becomes more extreme, there may be fewer examples of such violent events in the historical record.
This means AI systems may not have enough data to accurately simulate unprecedented extreme conditions in the future. In fact, it can be difficult to predict how they will react given weather conditions that are completely unfamiliar to them.
The behavior of AI systems “is often unpredictable when the program operates under conditions never experienced before,” warned Ebert Uphof and Hilburn in their comments. “Therefore, extreme weather events can cause highly erratic forecasts.”
Other experts have expressed similar concerns.
The authors of a 2021 Royal Society paper note that the “rarity of extreme events” in the historical record poses challenges for AI weather models. There have also been several studies that have attempted to evaluate the performance of AI systems in terms of capturing extreme situations with limited data, but the results have been mixed, with some showing good performance. , they point out that some have slumped.
Russ Schumacher, a Colorado climatologist and Colorado State University scientist, said, “The question of how AI models perform in a warming climate is a very interesting question, but to my knowledge, “It’s not been studied very thoroughly at the moment.” By email to E&E News. Schumacher’s own research group has applied artificial intelligence to models that predict storms and other dangerous weather conditions.
A hybrid model that includes both an AI component and a numerical model component could create fewer difficulties in record-breaking events, he suggested. But for models driven entirely by AI, “it’s not entirely clear how AI will respond to situations that are completely off the historical record,” he said.
These are important assessments that researchers should consider as they continue to develop AI weather models, he added. They need to explore not only how models perform in routine daily weather forecasts, but also for dangerous and high-impact events.
In general, he suggests that AI weather models have potential. But they may not completely replace traditional approaches, he also noted. Numerical and AI models may ultimately have different strengths, and human experience will continue to be valuable in synthesizing and conveying information about weather.
“I think the meteorological field should ideally reach a point where it can take advantage of the strengths of all approaches,” he said.
