Photo credit: www.smithsonianmag.com

The Evolving Role of A.I. in Weather Forecasting

Recent advancements in artificial intelligence (A.I.) have significantly improved its ability to predict extreme weather, marking a pivotal moment in meteorology. The potential of A.I. in weather forecasting came into sharp focus during events like Hurricane Lee last September, where traditional models accurately projected the storm’s trajectory just six days before landfall in Canada. In contrast, an experimental A.I. system named GraphCast made an accurate prediction nine days in advance. Furthermore, another instance involved Hurricane Beryl, where on the same day a European agency suggested Mexico as a target, A.I. indicated Texas as the likely landfall—an outcome that proved correct days later.

These instances highlight A.I.’s rapidly evolving capabilities in storm path prediction, but understanding its role requires a look back at traditional forecasting methods. For many years, meteorologists relied on numerical weather prediction, which harnesses data from satellites and ground-based stations. This information is processed by supercomputers that create simulations of atmospheric conditions. As computer scientist Amy McGovern from the University of Oklahoma explains, the complexity of the atmosphere necessitates an indirect approach, requiring significant computational resources and time to generate reliable forecasts.

Despite its effectiveness, numerical modeling has limitations, including lengthy computation times that can lead to delays in issuing timely forecasts. Consequently, the integration of A.I. into forecasting methods has gained momentum over the past two decades. Initially, A.I. was used in post-processing to refine inaccurate numerical models. More recently, hybrid models that combine traditional numerical methods with machine learning techniques have emerged.

A notable evolution in A.I. forecasting is the rise of data-driven models that operate independently of physical laws. These models leverage extensive historical weather data to identify patterns and generate forecasts quickly, often on standard computing setups rather than high-end supercomputers. This agility allows A.I. forecasts to be produced in seconds, enabling meteorologists to explore a wider array of potential weather scenarios.

However, the reliability of A.I. models in predicting extreme weather events remains a crucial concern. McGovern, who heads the NSF A.I. Institute for Research on Trustworthy A.I. in Weather, Climate and Coastal Oceanography, highlights that A.I. is currently more adept at enhancing warning times for severe weather such as tornadoes and hailstorms. For instance, utilizing A.I. could push average tornado warning times from 15 to upwards of 30 or 45 minutes, significantly improving public safety.

The Promise and Challenges of A.I. in Weather Forecasting

As promising as A.I. technologies are, there are still inherent challenges, especially regarding outlier events influenced by climate change. Traditional models, infused with physics, can often better predict unprecedented weather phenomena compared to A.I. models that lack this foundation. McGovern emphasizes the importance of having a degree of physical context in forecasting to bolster confidence in predictions, particularly for extraordinary climatic events.

The discussion surrounding A.I. does not imply a replacement of human meteorologists. Rather, as McGovern clarifies, A.I. is seen as a tool to augment their capabilities, helping to turn forecasts into actionable intelligence while maintaining the crucial human element in decision-making processes related to public safety.

The Global Landscape of A.I. Weather Forecasting

Various entities are employing A.I. in weather forecasting, including both governmental bodies such as NOAA (National Oceanic and Atmospheric Administration) and a range of private companies. While some organizations are more open about their methodologies, others are still in thenascent stages of integrating innovative techniques into their forecasting practices.

The European Centre for Medium-Range Weather Forecasts is also advancing its efforts with a model named AIFS, leveraging A.I. to improve predictions. This system resembles others like Google’s GraphCast and strives to enhance hurricane forecasting, known as tropical cyclone tracking in Europe.

Looking Ahead: The Future of A.I. in Weather Forecasting

As meteorological institutions continue to develop A.I.-integrated forecasting systems, education plays a key role in preparing the next generation for this technological shift. Initiatives aimed at introducing A.I. concepts to K-12 students and collaboration with institutions like Del Mar College are critical in fostering a diverse environment in the A.I. field.

Despite the rapid advancements, human expertise will remain vital for the foreseeable future. McGovern predicts that while A.I. may never entirely replace meteorologists, it will be instrumental in providing more options and ultimately enhancing public safety through better weather preparedness.

This interview has been edited for length and clarity.

Full image credit: Illustration by Emily Lankiewicz / ABI imagery from NOAA’s GOES-16 Satellite / NASA / Matthew Dominick

Source
www.smithsonianmag.com